1 BEHAVIORS IN PERSONS WITH DEMENTIA : A DYADIC EFFECT OF CAREGIVERS STRESS PROCESS? By JUDY LYNN MCGEE CAMPBELL A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2008
2 2008 Judy Lynn McGee Campbell
3 To my husband David
4 ACKNOWLEDGMENTS I would like to thank m y disse rtation committee for their cont inued support, patience, and assistance during this process. In particular, Dr. Meredeth Rowe, my dissertation chairperson, has been very instrumental in my growth as a researcher and writer. Without her sustained efforts in guiding and assisting me along the way, this would not have be en possible. I also greatly appreciate the as sistance of Dr. Michael Marsiske with the analyses and methodology. Lastly, Drs. Ann Horgas and Jenni fer Elder were very influential early in my doctoral education and have continually read, made suggestions, and discussed ways to improve my dissertation. I am very appreciative of the phenomenal de dication of my diss ertation committee. I am extremely grateful for the support I ha ve received from the University of Florida Alumni Fellowship and from the Hartford Building Academic Geriatric Nursing Capacity Scholarship. Without this suppor t, I would not have been able to attend the University of Florida, and would not have b een as successful in my studies As a Hartford Scholar, my mentors, Dr. Meredeth Rowe and Dr. Ann Kolanow ski (Pennsylvania State University), helped make it possible for me to meet the goals of the program, and make connections with persons who will be instrumental in my career as a gerontological nursing researcher. I am also appreciative of the opportunity to participat e in the UF Gerontology Traineeship, and the mentoring in research associated with that program. Many other persons have functioned as mentor s, friends, and support system through this process; too numerous to men tion. In particular, Drs. Cla ydell Horne and Annette Kelly provided excellent mentoring, suppo rt, and assistance, and served as co-investig ators in Dr. Rowes research project that provided data for secondary analysis in my dissertation. Fellow students Dr. Brandy Lehman, Dr. Andrea Boyd, and many others have become friends, and are a large part of my success. Dr. Jason Beckstead from the University of South Florida provided
5 tutoring in advanced statistical techniques that contributed to my ability to do the dissertation analyses. Several others provided deeper unders tanding in statistical techniques, and experience in various aspects of the researcher role. The UF College of Nursing graduate faculty, administration, and staff have also contributed to my success. In particular, Cecile Kiley has been eager to assist in so many ways, and the staff members in Office of Research Support were all helpful. Staff and students associated with Dr Rowes research project have been helpful in many ways. It goes without saying that w ithout the support and love of my family and my God this journey would not have been possible. I am very thankful to have a fa mily that is willing to make sacrifices to further my potential for su ccess. My husband, David, has been by my side during this entire process, and I will always be grateful for his listening, helping, and understanding as I dedicated myself to my studies. Lastly, and of course not least, I would like th ank the caregivers and persons with dementia that participated in the study that served as a parent study for my dissertation. Their willingness to participate despite their hec tic, stressed lives is amazing, and I will dedicate my research to finding solutions for this vulnerable group.
6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ............................................................................................................... 4 LIST OF TABLES .........................................................................................................................10 LIST OF FIGURES .......................................................................................................................11 ABSTRACT ...................................................................................................................... .............13 CHAP TER 1 INTRODUCTION .................................................................................................................. 15 Statement of the Problem ...................................................................................................... ..15 Study Concepts and Relationships ..........................................................................................16 Behavioral Symptoms in Persons with Dem entia ........................................................... 16 Caregivers Perceived Stress ........................................................................................... 17 Caregivers Emotional Responses ................................................................................... 18 Correlates of Caregivers Perceive d Stress and E motional Responses ........................... 19 Dyadic Consequences of Caregivers Stress Process ......................................................21 Behavioral Symptoms in Pe rsons with Dem entia: A Consequence of Caregivers Stress Process? .............................................................................................................23 Theoretical Foundations ....................................................................................................... ..25 Caregivers Perceived Stress Em otional Responses ....................................................26 Caregivers Perceived Stress Behavioral Sym ptoms ...................................................26 Caregivers Emotional Responses Behavioral Sym ptoms ......................................... 26 Secondary Analyses of BehSx as a Cons equence of Caregivers Stress P rocess ................... 28 Specific Aim # 1: .............................................................................................................29 Specific Aim # 2: .............................................................................................................29 Summary ....................................................................................................................... ..........30 2 LITERATURE REVIEW .......................................................................................................33 Background and Significance .................................................................................................33 Dementia and Dementia Caregiving ............................................................................... 33 Dementia-Related Care Dyads ........................................................................................ 37 Testing a Model including Dyadic Effects ...................................................................... 40 Theoretical Foundations ....................................................................................................... ..41 Stress-Health Model ........................................................................................................42 Consequences of Dementia Caregivers Stress Process Model ......................................42 Expanded Consequences of Dementia Caregivers Stress Process ........................................ 44 Caregiver Consequences .................................................................................................44 Dyadic Consequences ......................................................................................................45 Consequences for persons with dementia ................................................................ 45 Consequences for the dyad as a unit ........................................................................47
7 Are BehSx in PWD a dyadic consequen ce of caregivers stress process? ...............50 Relevance of the Proposed Theoretical Model .......................................................................52 Dyadic Effect of Caregivers Stress Process: PWD BehSx ................................................... 53 Caregivers Perceived Stress .................................................................................................. 53 Predictors of Percei ved Stress/Burden ............................................................................54 Progression of Perceived Stress/Burden in Dem entia Caregiving .................................. 57 Treatment of Percei ved Stress/Burden ............................................................................58 Association between Perceived St ress and Em otional Responses .........................................58 Caregiver Emotional Responses to Stress .............................................................................. 59 Predictors of Emotional Responses ................................................................................. 60 Progression of Emotional Responses .............................................................................. 61 Treatment of Emotional Responses ................................................................................. 62 PWD Behavioral Symptoms ................................................................................................... 63 Influence of Caregiver St ress Process on PW D BehSx .......................................................... 64 Influence Related to Caregiver Coping and Stress Management .................................... 65 Influence Related to Caregiver Presenting Affect ........................................................... 66 Interventional Evidence for Caregiver Influence on BehSx ............................................ 69 Summary ....................................................................................................................... ..........71 3 MATERIALS AND METHODS ...........................................................................................74 Introduction .................................................................................................................. ...........74 Original Study .........................................................................................................................74 Design ........................................................................................................................ ......74 Subjects ...................................................................................................................... ......75 Entry and exclusion criteria ...................................................................................... 75 Sample and setting ...................................................................................................75 Recruitment and retention ........................................................................................ 76 Procedures .................................................................................................................... ...77 Methods ....................................................................................................................77 Vulnerable human subjec ts considerations .............................................................78 Data considerations ..................................................................................................79 Methodological Strengths of Original Study that Augm ent Dissertation Study ....................80 Dissertation Study ............................................................................................................ .......82 Dissertation Design ......................................................................................................... 82 Dissertation Subjects ....................................................................................................... 82 Dissertation Procedures ................................................................................................... 83 Dissertation Measures .....................................................................................................84 Neuropsychiatric Inventory ......................................................................................84 Neuropsychiatric Invent ory-Questionnaire. ............................................................. 86 Measurements from Original Study ................................................................................ 86 Mini-Mental Status Examination. ............................................................................87 Short version of the Zarit Burden Interview ............................................................ 87 Center for Epidemiologic St udies-Depression scale. ...............................................89 Positive and Negative Affect Schedule. ................................................................... 90 Composite negative mood score. ............................................................................. 91 Statistical Analysis .......................................................................................................... 92
8 Missing data .............................................................................................................92 Preliminary analyses ................................................................................................ 93 Model testing ............................................................................................................94 4 FINDINGS ...................................................................................................................... ........99 Univariate and Bivariate Descriptives ....................................................................................99 Change over Time in Model Variables .................................................................................102 Perceived Stress in Caregivers ......................................................................................103 Emotional Responses in Caregivers .............................................................................. 105 Behavioral Symptoms in Persons with Dem entia ......................................................... 106 Model Assumptions ..............................................................................................................107 Hypothesized Relationships ..................................................................................................109 Separate Effects of Perceived Stress and Em otional Responses ................................... 110 Mediation ................................................................................................................... .111 Model Parameters .......................................................................................................... 112 Perceived stress ......................................................................................................112 Emotional responses ............................................................................................... 112 Mediation model .................................................................................................. 113 Summary of Findings in re lation to Proposed Model ...........................................................113 5 DISCUSSION AND IMPLICATIONS ................................................................................ 133 Support of Proposed Theoretical Model ............................................................................... 133 Findings in Relation to Past Research .................................................................................. 135 Model Variables ............................................................................................................135 Relationships between Model Variables .......................................................................136 Directionality of Relationships between Caregiv er Stress Process Variables and BehSx ......................................................................................................................... 137 Strengths, Limitations, and Direct ions for Future Research ................................................. 139 Summary ....................................................................................................................... ........142 APPENDIX A MISSING DATA ..................................................................................................................143 Systematically Missing Items in the NPI-Q ......................................................................... 143 Solution for Systematically Mi ssing Data in this Study ....................................................... 144 B STATISTICAL ANALYSES ...............................................................................................146 Multilevel Analyses ........................................................................................................... ...146 Preparation of the Data .................................................................................................. 146 Estimation Techniques .................................................................................................. 147 Modeling Nested Data ...................................................................................................148 Modeling Change over Time ................................................................................................ 149 Comparisons over Time by Treatment Status ............................................................... 150 Modeling the Covariance .............................................................................................. 151
9 Building Models using Inform ation Criteria and PsuedoR2 ........................................152 Additional Considerations ............................................................................................. 153 Modeling Mediation .............................................................................................................154 C PRELIMINARY TREND ANALYSES ............................................................................... 156 LIST OF REFERENCES .............................................................................................................170 BIOGRAPHICAL SKETCH .......................................................................................................195
10 LIST OF TABLES Table page 3-1 Description of sample caregivers. ...................................................................................... 97 3-2 Description of sample persons with dementia. .................................................................. 98 4-1 Means, standard deviations, and t -tests by treatment status for observed scores across 9 data points for m odel variables. .................................................................................... 115 4-2 Pearson correlations between model variab les at Baseline (below the diagonal) and Month 12 (above the diagonal & underlined with dashes). ............................................. 116 4-3 Pearson correlations (and list-wise mean s) between m odel variables across all dyads and data points (list-wise N=308). ...................................................................................117 4-4 Model construction criteri a for Perceived Stress .............................................................118 4-5 Model construction criteria for Emotional Responses. .................................................... 119 4-6 Model construction criteria for BehSx. ............................................................................ 120 4-7 Final Model Parameters for each Variable as m odeled over Time and by Treatment. ...121 4-8 Model construction criteria for effects of Caregivers Stress P rocess Variables on PWD BehSx. .................................................................................................................... 122 4-9 Final Model Parameters estimated for BehSx as predicted by Perceived Stress and Em otional Responses. ...................................................................................................... 123
11 LIST OF FIGURES Figure page 1-1 Full Consequences model adapted from Schulz & Martires Stress-Health m odel. ...... 31 1-2 Reduced model for evaluation in current study. ................................................................ 32 3-1 Original Study subjects entry/exclusion from diss ertation study analyses. ....................... 96 4-1 Caregivers Perceived Stress as observed over Tim e by Treatment Status (means and standard error bars). .........................................................................................................124 4-2 Caregivers Emotional Responses as observed over Tim e by Treatment Status (means and standard error bars). ...................................................................................... 125 4-3 PWD Behavioral Symptoms as observed over T ime by Treatment Status (means and standard error bars). .........................................................................................................126 4-4 Caregivers Depressive Sym ptoms as observed over Time by Treatment Status (means and standard error bars). ...................................................................................... 127 4-5 Caregivers Negative Affect/Mood as obs erved over Tim e by Treatment Status (means and standard error bars). ...................................................................................... 128 4-6 Perceived Stress modeled over Time by Treat m ent Status; Demonstrates no effect of Treatment over Time. ....................................................................................................... 129 4-7 Emotional Responses modeled over Time; Demonstrates no effect of Treatment over Tim e. ......................................................................................................................... .......130 4-8 Behavioral Symptoms m odeled over Tim e by Treatment Status; Treatment improved model and interaction of Treatment w ith Time approached significance. ....................... 131 4-9 Behavioral Symptoms m odeled as predicted by Per ceiv ed Stress (C -top) and Emotional Responses (B-bottom); Combined model (center) inconclusive regarding mediation of Perceived Stress e ffects by Emotional Responses. .................................132 C-1 Outcome PWD BehSx plotted against M onths paneled by Treatm ent Status with Regressions. .................................................................................................................. ...158 C-2 Observed variation in PWD Behavioral Sym ptoms over Time within subjects, paneled by Treatment Status. ........................................................................................... 159 C-3 Behavioral Symptoms ov er Months paneled by Dyad with RegressionsControl dyads only. .......................................................................................................................160 C-4 Behavioral Symptoms over Months paneled by Dyad with regressions Experim ental dyads only.................................................................................................. 161
12 C-5 Predictor Caregiver Perc eived Stress plotted against Months paneled by Treatm ent Status with Regressions. .................................................................................................. 162 C-6 Predictor Caregiver Emotional Respon ses plotted against Months paneled by Treatm ent Status with Regressions. .................................................................................163 C-7 Observed variation in caregivers Perceived Stress over Tim e within subjects, paneled by Treatment Status. ........................................................................................... 164 C-8 Observed variation in caregivers Emo tional Responses over T ime within subjects, paneled by Treatment Status. ........................................................................................... 165 C-9 Perceived Stress over Months paneled by Dyad with RegressionsControl dyads only. ......................................................................................................................... ........166 C-10 Perceived Stress plotted over Months paneled by Dyad with Regressions Experim ental dyads only.................................................................................................. 167 C-11 Emotional Responses plotted over Mont hs paneled by Dyad with regressions Control dyads only. ..........................................................................................................168 C-12 Emotional Responses plotted over Mont hs paneled by Dyad with regressions Experim ental dyads only.................................................................................................. 169
13 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy BEHAVIORAL SYMPTOMS IN PERSONS WITH DEMENTIA: A DYADIC EFFECT OF CAREGIVERS STRESS PROCESS? By Judy Lynn McGee Campbell August, 2008 Chair: Meredeth Rowe Major: Nursing Sciences I evaluated a newly developed model in whic h dementia caregivers stress process is viewed as a predictor of behavi oral symptoms in persons with dementia. Multilevel modeling was used to assess both withinand between-per sons change in caregivers perceived stress, caregivers emotional responses, and behavior al symptoms in persons with dementia. Additionally, caregivers perceived stress and emotional responses we re modeled as predictors of behavioral symptoms over time, with emotional responses expected to reduce the influence of caregivers perceived stress on beha vioral symptoms in a manner sim ilar to formal mediation. My study was a secondary analysis of a car egiver intervention designed to reduce caregivers stress, worry, and subsequent emotional responses; the intervention was a technological innovation meant to assist caregivers in managing nighttime wandering of the care recipients. There was no exp ected influence on behavioral symptoms, such as agitation, aggression, and other symptoms. In my study, behavioral symptoms diverged over time according to treatment status, and the interact ion between time and treatment approached statistical significance. However, caregivers pe rceived stress and emoti onal responses were not significantly influenced by treatment status as expected. The caregiver s perceived stress and
14 emotional responses both predicte d behavioral symptoms; however the emotional responses did not fully reduce the effect of perceived stress on behavioral symptoms as expected. Results support the underlying premise of my model; that caregivers stress process influences the behavioral sympto ms in persons with dementia. However, further research is needed to develop relationships and the directionali ty of those relationships within the model. If my model is supported through further investigati on, researchers and clinic ians may be able to develop or utilize interventions th at are more holistic and dyadic, directed at the caregiver, but with results for both caregivers and persons with dementia.
15 CHAPTER 1 INTRODUCTION Statement of the Problem As the population continues to live longer, th e incidence of dem entia increases, with over 13 million individuals expected to have dementia by the year 2050 (L. E. Hebert et al., 2003a). Since the large majority of those with dementia remain in the community (Czaja et al., 2000), informal caregivers provide mu ch of the care of persons with dementia (Riggs, 2001). Community-based informal caregiving for person s with dementia is considered uniquely challenging when compared to othe r types of caregiving situations with greater stress and more problems associated with care duties (Ory et al., 1999). There are three salient aspects of dementia careg iving that were the focus of this study: care recipients behavioral symptoms, caregivers perceived stress, and caregivers emotional response to stress. These phenomen a have most often been related in previous work through the informal caregivers stress process model, wherei n behavioral symptoms may be perceived as a problem, and their behaviors are considered a prim ary stressor for the caregiver. Behavioral symptoms have generally been conceptualized in re search in this manner, as a predictor of stress for caregivers (Goode et al., 1998; Pearlin et al., 1990; Pinquart & So rensen, 2003b; Schulz & Martire, 2004). However, while the severity and/or frequency of behavioral symptoms may contribute to the caregivers perception of the behavior as a problem, caregivers own characteristics may also influence the perception of the behavior as a stressor. Additionally, researchers are beginning to recognize that th e relationship between be havioral symptoms in persons with dementia (PWD) and caregivers st ress process may be bi-directional; that is, caregivers stress may be as likely to cause behavioral symptoms as behavioral symptoms are to cause caregivers stress (Burge ner & Twigg, 2002b; Schulz & Martir e, 2004; Sink et al., 2006).
16 In this study, caregivers stress process will be investigated as one potential predictor of behavioral symptoms in persons w ith dementia in their care. Study Concepts and Relationships Behavioral Symptoms in Persons with Dementia Behavioral sym ptoms (BehSx) almost universal ly occur in dementia syndromes (Lyketsos et al., 2002; Steinberg et al., 2006). BehSx are also referred to as behavioral and psychological symptoms of dementia (BPSD) and neuropsychi atric symptoms to reflect the neurological changes and psychopathology in dementia that underlie the symptoms (Cummings, 1997; Lyketsos, 2007). BehSx encompass doma ins such as delusions, hallucinations, agitation/aggression, depression /dysphoria, anxiety, elation/e uphoria, apathy/indifference, disinhibition, irritability/lability, motor distur bance, nighttime behaviors, and appetite/eating disorders (Cummings, 1997; Cummings et al., 1994). BehSx in dementia are a complex phenomenon with multiple underlying causes. While BehSx are associated with the neuro-degenera tion that occurs in dementia (Cummings, 1997; Harwood et al., 2000; Mok et al., 2004), there are also social, environmental, and psychological contributing factors. For example, it has been proposed that BehSx refl ect either the persons stress related to the inability to effectively pr ocess and cope with environmental stimuli (Hall & Buckwalter, 1987; Smith et al., 2004), or th e persons needs related to physiological, psychosocial, or environmental conditions, that ma y be sought out and re solved (Algase et al., 1996). Lastly, BehSx in persons with dementia ha ve been proposed to represent expression or communication of something, such as the stress they feel or a need that they cant readily communicate (Kitwood, 1997; Touhy, 2004; Woods, 2001). This more contextual or relational view of BehSx is in contrast to that currently represented in the caregivers stress process, specifically, that of a problem to be managed. While caregivers certainly may perceive BehSx as
17 a problem or stressor, a more contextual view may allow an enhanced holistic understanding of and response to BehSx (Kolanowski & Whall, 2000). Such a response may allow contextual intervention to ameliorate the BehSx themselves, and in turn alleviate the caregivers perception of BehSx as a stressor, providi ng valuable relief for community -based dementia care dyads. Caregivers Perceived Stress Common causes of caregivers stress include both primary stressors directly related to caregiving, such as care recipient difficulties with activities of daily living, and secondary stressors such as caregiving-related work and fa mily issues; for example, loss of productivity at work, or conflicting responsibilities for care reci pient and for other family members. When caregivers perceive that resources are too low, or demands too high or too threatening, overall levels of perceived stress increase. Researchers have consistently demonstrated that informal dementia caregivers (typically family members) on average report higher levels of perceived stress than non-caregivers (Clyburn et al., 2000 ; Haley et al., 1987; Le inonen et al., 2001; Pinquart & Sorensen, 2003b; Schulz & Martire, 2004), and higher le vels than caregivers of persons with other chronic illn esses (Bertrand et al., 2006; Gara nd et al., 2005; Martire & Hall, 2002; Pinquart & Sorensen, 2003b). When levels of perceived stress are high, dementia caregivers have shown an associated increase in emotional responses, with subsequent detrimental mental and physical health changes (Pinquart & Sorensen, 2003b, 2004, 2007; Schulz & Martire, 2004). In caregivers, perceived stress is often conceptualized as the more domain-specific caregiver burden or caregiver strain, and these terms may be conceptually viewed as interchangeable with perceived stre ss in this context (Donaldson et al., 1998; B. G. Knight et al., 2000). The terms are defined similarly, with the capacity to adapt to stressors inadequate in relation to the demands of th e caregiving situation (O'Rourke & Tuokko, 2003; Pearlin et al.,
18 1990; Schulz & Martire, 2004; Zarit et al., 1980). Additionally, measurement instruments for caregiver burden and perceived st ress reflect overlapping latent c oncepts (Bedard et al., 2001; S. Cohen et al., 1983; Levenstein et al., 1993). In this study, care giver burden was considered conceptually comparable to perceived stress. Caregivers Emotional Responses Caregivers em otional responses to stress vary, and dementia ca regivers have higher levels of emotional responses such as depressive symp toms (Schulz et al., 2003), anxiety (Cooper et al., 2007), negative affect (Brummett et al., 2006; Hale y et al., 1987), and othe rs (Martire & Hall, 2002; Pinquart & Sorensen, 2003b). Depressive symptoms are frequently reported when caregivers emotional responses are included in research (Clyburn et al., 2000; Croog et al., 2001; Haley et al., 1987; Lee et al., 2006; Ma usbach et al., 2007; Schulz et al., 1995; Thommessen et al., 2002; Zanetti et al., 1998), and may include changes such as a flat affect, sluggishness, lack of attention and flexibilit y, and other psycho-emotional or depressive symptoms. Additionally, negative affective resp onses reflecting distress, fear, irritability, nervousness, and jitteriness may occur (R adloff, 1977; Watson et al., 1988). Dementia caregivers generally have incr eased depressive symp tomatology (Martire & Hall, 2002; Pinquart & Sorensen, 2003b). In an early review, all studies reported increased levels of depressive symptoms (Schulz et al ., 1995), and in the US Resources for Enhancing Alzheimer Caregivers Health (REACH) I trials 41% of caregivers had risk for clinically significant levels of depressive symptoms (CES-D 16 or > out of 60). Participants had a mean score of 15.4 (SD=11.5; median 13; range 0-56; N=1222) across all REACH sites (Schulz et al., 2003). When a clinical diagnosis is required, rates are not as high, but remain noteworthy. In a recent systematic review of the literature, dementia caregivers were identified as having high rates of clinically diagnosed de pressive disorders (22.3%), with increases in depression after
19 initiation of the caregiving role, and with women more at risk th an men (Cuijpers, 2005). An earlier review also revealed hi gh rates of both clinical depres sion and anxiety (Schulz et al., 1995). Dementia caregivers have also reported significantly higher negative affect, in addition to higher levels of depressive symptoms and lower le vels of life satisfaction, compared to controls (Haley et al., 1987). Negative affect has been as sociated with other measures of importance in caregiving. For example, in one study, negativ e affect, along with social support, explained 43% of the variance in the relationship betw een caregiving and sleep quality. Along with caregiver health and care recipient dependence in IADLs, negative affect was useful in predicting caregiver strain at 6 months after the care recipients str oke (Blake et al., 2003). Correlates of Caregivers Perceive d Stress an d Emotional Responses Perceived stress and emotional responses are often reported simultaneously in dementia caregiving research, and are recognized as im portant variables in dementia caregivers experiences. For example, in a study of 172 careg ivers, 68% of subjects reported high levels of perceived stress, and 65% of subjects had depres sive symptoms. Perceived stress was positively correlated with care recipient BehSx (r = 0.54, p < 0.01), as was depression (r = 0.35, p < 0.01). Additionally, perceived stress was re lated to coping style. In pa rticular, caregivers who used emotion-focused coping had higher levels of perc eived stress, as opposed to lower levels in caregivers who used problem-solving approach es (Papastavrou et al., 2007). Another study found that increased stress related to symptoms was, in tur n, correlated to caregiver worsening mental health, including depre ssed mood (Hooker et al., 2002). Several factors have been associated with perceived stress and emotional responses in dementia caregivers. BehSx were consistently th e strongest correlate with these stress process variables in a review of 228 studies, with amount of care and physical disability less salient, and
20 with spouses having more difficulty with thes e symptoms than adult children (Pinquart & Sorensen, 2003a). In an international, multi-continent review, the type and amount of care and supervision required were predicto rs of perceived stress, while Be hSx were more predictive than physical needs. Additionally, living with the care recipient increased depression and anxiety. Caregivers who were female and younger (the latter often confounde d with adult child relationship) were also more at risk for perceived stress, as were those who reported low selfassessed competence (Torti et al., 2004). Problem behaviors, income, and caregivers health and perceived stress were predictive of psychiatric morbidity in an earlier review (Schulz et al., 1995). A feeling of closeness with the care recipient was among up lifts caregivers could receive from their duties, which were associated with lower levels of depression (Pinquart & Sorensen, 2003a, 2003b), while negative spouse and soci al interactions, such as arguments, have been found to predict depressive symptoms in dementia caregivers (Haley et al., 2003) Caregivers background and care environment cont ributed to perceived stress in addition to symptoms of the PWD (Onishi et al., 2005). In one study, behavioral sy mptoms contributed less to caregivers depressive sympto ms than caregivers perceptions of their own health and poor performance (Zanetti et al., 1998). In a study that assessed care reci pient and caregiver characteristics for contribution to caregiver depression, BehSx, particularly angry or aggressive behaviors, was only one of several contributing factors (Covinsky et al., 2003). In a Korean study, the relationships of perceived stress with de pressive symptoms and anxiety were mediated by caregiver factors (Lee et al., 2006). Caregivers high levels of pe rceived stress and emotional responses have been associated with poor health outcomes, includi ng morbidity and mortality, in several studies; these findings were the basis for a model of stress-related h ealth issues in dementia caregivers (Schulz &
21 Martire, 2004). The association between caregivers high levels of stress and morbidity continue to be supported through research (Son et al., 2007). Many interventions designed to improve perc eived stress and emotional responses in dementia caregivers have been studied. Severa l reviews have indicate d that historically, interventions designed to improve perceived stre ss, depression, and psychological health have largely been unsuccessful. However, more recently developed and more comprehensive, multicomponent interventions, particularly when carried out for longer periods of time, have proven helpful (Pinquart & Sorensen, 2006; Sore nsen et al., 2002; Torti et al., 2004). Dyadic Consequences of Caregivers Stress Process The care relationship is by nature a dyadic process, which ha s the potential to overshadow previous interaction patterns (Lyons et al., 2002; Sebern, 2005). W hen the care recipient has dementia, changes occur that shake the core of the relationship and shape future interactions (Whitlatch, 2001; Woods, 2001). Primary informal caregivers and PWD develop a new dimension of their relationship, a connection or an interrelatedness, whic h provides the core of the care dyad and establishes rationale for the infl uence of one partners ex periences on the other partner (Kenny et al., 2006; Kitwood, 1997; Quayhagen & Quayhagen, 1996). Persons with dementia and their caregivers have been shown to have significantly corr elated quality of life (Thomas et al., 2006). This inter-connectedness can lead to negative dy adic influences. For example, caregivers ineffective coping with stress was related to shorter survival times in the PWD (McClendon et al., 2004; Schulz & Beach, 1999; Schulz & Martire, 2004). However, positive changes can also occur. In a qualitative examination of an intervention that involved dyadic cognitive remediation exercises, the authors concluded that the care dyad was a functional entity that had improvements in interactions and relationship quality that went beyond any individual benefits received by the interven tion (Quayhagen & Quayhagen, 1996). Considering
22 caregivers and PWD dyadically may allow a more relevant understanding of the dynamic process and interaction that o ccurs (Lyons et al., 2002). In research, dyads generally consist of paired individuals who are not independent, but linked into a relationship that is characterized by partner effects, mutual influence, or a common fate. Partner effects occur when one persons char acteristics affect the ot hers outcomes, mutual influence occurs when each persons outcomes influence the others, and common fate assumes that both members are exposed to the same causal factors. The two individuals are distinguishable as persons within a dyad by their roles, but have similarities or relatedness based on their mutual experience (Kenny et al., 2006) The caregivers and pe rsons with dementia within the care dyad, like other dy ads in research and practice, ar e not necessarily independent in their functioning, but each has an impact on th e other through daily ac tivities and routines (Kenny et al., 2006; Kitwood, 1997; Quayhagen & Quayhagen, 1996). Thus, the two exist within a functional entity that is dynamic, multi-dimensional, and influential for both its members (Quayhagen & Quayhagen, 1996). Researchers interested in dementia ar e beginning to emphasize methodologies and interventions that address both members of the care dyad (Bourgeois, 1993; Cavanaugh et al., 1989; Gitlin et al., 2003b; Haight et al., 2003; Lyons & Sayer, 2005; Lyons et al., 2002; Quayhagen et al., 1995; Whitlatch et al., 2006a; Whitlatch et al., 2006b; Zarit et al., 2004). Dyadic research in dementia care may seek to in clude the opinions or outlook of the PWD within dyadic assessments. For example, in the early stages of dementia, researchers evaluated an intervention promoting the planning of future car e. Findings indicated that caregivers, care recipients, and counselors found the intervention helpful and appropriate, assisting the dyad to gain knowledge and improve communication following diagnosis. In other instances, research
23 may be designed to develop and test interven tions targeting one or both members, with hypothesized outcomes for one or both members of the dyad. Outcomes may measure changes in the member not targeted by the intervention, or may include both PWD and caregivers. For example, four interventions directed at both dyad members (cognitive stimulation, dyadic counseling, dual supportive seminar, and day care) were evaluated regardi ng their effects on both caregivers and persons with dementia. The cognitive stimulation group lead to improved measures of cognition in PWD and decreased depressive symptoms in caregivers. Two other interventions improved outcomes for caregivers only; use of daycare decreased caregiver hostility, and the seminar improved caregiv er coping (Quayhagen et al., 2000). As research begins to establish the inter-conne cted nature of the dementia care dyad, it is relevant to focus on the important relationship between caregivers stre ss and PWD BehSx. The next step in understanding this relationship is to establish whether caregivers stress can influence the presence or seve rity of BehSx in the PWD. Behavioral Symptoms in Persons with Deme ntia: A Consequence of Caregivers S tress Process? Since caregivers in community-based dyads play a predominant role in managing environmental stimuli and providing for care reci pient needs, their stress may impede their ability to provide the optimal caregiving environmen t they strive for. For example, experts often recommend that caregivers adjust their own approaches, affect and demeanor to enhance the caregiving environment and reduce BehSx that are common in dementia (Cohen-Mansfield, 2000a; Colling, 2004; Haley, 1997; Hall, 1994; Kola nowski & Whall, 2000; Mittelman et al., 2004b; Quayhagen et al., 2000; Smith et al., 2006; Teri, 1999; Vitaliano et al., 1993). However, caregivers who are more stressed, with negativ ely affected emotional responses, may display altered interactions, attitudes, a ffect, or perceptions. The resu lt may be that the dyadic care
24 relationship suffers, and that ca regivers characteristics and st ress process may influence BehSx (Dunkin & Anderson-Hanley, 1998; Sink et al., 2006). In a systematic review of the literature from 1990-2001, Black and Almeida (2004) found associations between BehSx and caregivers psychological sequelae of care. While few articles met inclusion criteria, evidence did support correlations between BehSx and perceived stress (strong), psychological distress (moderate), a nd depression (poor-moderate). Furthermore, multivariate analyses added evidence that BehSx were predictive of these outcomes, often the most salient predictor. Noneth eless, longitudinal evidence was s carce, and the temporal nature of these relationships was not established w ith the limited evidence. Thus, it remains unknown whether BehSx are causative or simply correlated, with the potential for the opposite directional relationship (Black & Almeida, 2004). Other reviewers have also noted a lack of rigor and ability to lead to conclusions concerning relationships between caregivers stress and symptoms in the PWD, specifically with regard to directionality of the relationships (B allard et al., 2000; Dona ldson et al., 1997; Teri, 1997). For example, one reviewer highlight ed this phenomenon by noting that one study reported high rates of care giver depression (75%) among those whose care recipients were depressed, while another study reported that 100% of care recipients with depression had care givers who were depressed (Teri, 1997). In summary, BehSx arise within the milieu of the dyadic care relationship. BehSx may be conceptualized, at least in part as a function of the caregivers stress process, and may be an important gauge of the quality of the relationship or caregiving environment. The person with dementia may have a cognitively limited interpretation of caregivers non-verbal cues and verbal interactions, but may react based on that interpretation. BehSx in PWD reflect the inter-
25 connected nature of the relations hip, with significance for both members of the dyad, not only in the moment, but also related to more distal outcomes, such as placement decisions and the eventual characterization of behaviors as a pr oblem or stressor. U nderstanding the temporal nature of the relationship between caregiver stre ss process variables a nd PWD BehSx is central to care provision and recommendations for dementia care dyads. Theoretical Foundations In preparation for this study, the Stress-Health model was adapted to explain behaviors in relation to caregivers stress, and the newly revised m odel was adopted as the theoretical foundation for this study. The resulting Conseque nces of Dementia Caregivers Stress Process model is seen in Figure 1, with the concepts from the Stress-Health model sh aded in gray. In the revised model, consequences of caregivers stress are viewed broadly, with influence within the dyad, not just for the caregivers. Please see Chapter 2 for a full discussion of this model. In this study, a conceptual model of the Dyadic Effect of Dementia Caregivers Stress Process: PWD BehSx was tested (see Figure 2) Key concepts from the revised theoretical model, including caregivers pe rceived stress and emotional responses, and their relationships with the BehSx in PWD, were considered. Th e premise of the model is that caregivers emotional responses, coupled with increased levels of perceived stress, may also have extensive consequences within the dyad, beyond those effect s related to caregivers health. In other research, caregiver characteristic s, as well as caregiver-directed interventions, have been linked to several care recipient outcomes in dementia (de Vugt et al., 2004; McClendon et al., 2004; Teri, 1999; Teri et al., 2005), indi cating that caregivers have pervas ive influence w ithin the dyad. A brief review follows regarding the model relationships, with more detail in chapter 2. In the review that follows, caregiver perceived stress and burden are c onsidered interchangeable.
26 Caregivers Perceived Stress E motional Responses The relationship between caregiver s perceived stress and emoti onal response is part of the caregivers stress process. When caregivers apprai se their adaptive capacities as lacking in the face of demands, perceived stre ss leads caregivers to respond emotionally. The depression, anxiety, and other emotional respon ses in caregivers are posited as predictors in more distal outcomes for the caregiver, such as personal h ealth (Pearlin et al., 1990; Schulz & Martire, 2004) and may also influence outcomes within the dyad. Caregivers Perceived Stress Behavioral Symptoms Caregivers perceived stress wa s hypothesized to have an in fluence on PWD BehSx in this study. This relationship has not been well developed previously. In a few studies, caregivers stress has been identified as having influence on intera ctions within the dyad. For example, caregivers stress lead to an increase in the o ccurrence of care recipient behaviors in dementia (Burgener & Twigg, 2002b; Si nk et al., 2006). Wife caregivers of husbands with Alzheimers dementia were more depressed and more stressed, with less hope than non-ca regivers, and also had less r eciprocity and shared pleasures with their spouses (Galla gher-Thompson et al., 2001). Caregivers who were highly critical were also more strained/distressed, more likely to attribute care recipient ac tions as deliberate, and their care recipients had higher levels of BehSx (Tarrier et al., 2002). Caregivers Emotional Responses Behavioral Symptoms Caregivers em otional responses also may contribute to an increase in BehSx. Body language is an important communicator of emo tional responses. Virtual imaging techniques were used to establish body posture and body angles representing reported affective states, including aversion, openness, irritation, happine ss, and self-confidence, which were then accurately rated by volunteers (Gra mmer et al., 2004). In care dyads researchers have noted that
27 caregivers who are depressed or ha ve mood alterations may be irrita ble or visibly tense, appear sad or anxious, or perhaps even be overly cri tical of the person with dementia in verbal exchanges (Sink et al., 2006; Whitlatch, 2001; Woods, 2001). Caregivers who are depressed and resentful about their caregiving du ties are more likely to treat th e care recipient in a harmful manner, such as yelling or th reatening the person with dementia (Shaffer et al., 2007), or handling them roughly (Williamson & Shaffer, 2001). Lastly, a correlation between depressed dementia caregivers and depressive BehSx in care recipients has been proposed (Teri, 1997). In this study, emotional responses to stress perhaps as demonstrated by body language and interactions, is considered to function in a manne r similar to that of a mediator, contributing to BehSx severity, with the ability to at least partially explain the effect of caregivers perceived stress on BehSx. While this study will not incl ude a formal test of mediation, modeling will reflect the anticipation of such a relationship. Emotional responses have been indicated as a mediator in other relationships involving perceived stress. In dementia care dyads, emoti onal responses have been posited as a mediator between perceived stress and health (Schulz & Martire, 2004). Additionally, in a meta-analysis of dementia care interventions, reduction of psychological symptoms was identified as a mechanism by which interventions achieved de layed nursing home placement. For example, depressive symptoms, along with caregivers sati sfaction with social support and response to BehSx, accounted for 61% of the effect a coun seling intervention had on delaying institutional placement by 557 days (Mittelman et al., 2006). Cha nges in depression were also responsible for the counseling interventions indirect effect on car egivers self-reported health (Mittelman et al., 2007). In other settings, affective states, includi ng positive affect and depressive symptoms have acted as a mediator between a variety of cond itions/treatments and health, quality of life, and
28 mortality (Blane et al., 2008; Deichert et al., 2008; Grov et al., 2006; Hill et al., 2005; Pastor et al., 2003; Williams, 2005). While emotional responses to stress have not been specifically studied as a mediator between caregivers pe rceived stress and PWD Be hSx, the relationships reported here confirm the important role of emoti onal responses in the more distal effects of the stress process. Secondary Analyses of BehSx as a Cons equence of Caregivers Str ess Process A monitoring system recently developed to as sist caregivers in management of nighttime awakenings in persons with dementia was tested in a randomized clinical trial (pilot), providing an experimental design and longitudinal data intended to reflect care givers stress process (Rowe, 2003a). Measures of stress process variables were expected to change over time in the intervention group in this st udy; however, no direct influences on PWD BehSx were hypothesized. Data were collected regarding BehSx, allowing the conceptual model to be tested using the parent study data in a secondary analysis. Utilizing the monitoring system intervention groups from the parent study for comparison of variable trajectories may le nd support to the direction of cau sality in relationships from caregivers stress process variables to BehSx. This is an important feature of this study because of the potential bi-directional na ture of the relationships, wherein BehSx can eventually lead to caregivers interpretation of them as a problem or stressor. Ultimately, it will be imperative to distinguish whether the BehSx are an outcome or a predictor of th e stress process variables, or both. In the parent study, there was no training geared toward managing BehSx; the intervention was a physical adaptation of the environment meant to decrease wo rry and stress for caregivers. Since BehSx were not directly targeted, any im provement or change in BehSx would be more likely due to changes in the caregivers stress pr ocess variables. If in tervention group members improve and/or control group members worsen over time with regard to BehSx, this will provide
29 initial support for the direction of causality posited in the conceptual model proposed for this study. The overall purpose of this study was to test a model (Figure 2) in which both CG Perceived Stress and CG Emotional Responses have a direct positive relationship with PWD BehSx, and the Perceived StressPWD BehSx relationship is mediated by CG Emotional Responses. A secondary purpose was to describe the temporal patterns of each of the model variables, and evaluate the influence of Treatment Status from the parent study on the concepts and relationships in the model. Treatment stat us was considered a factor in modeling of the change over time for Perceived Stress, Emo tional Responses, and PWD BehSx. The specific aims and hypotheses are: Specific Aim # 1: To describ e temporal patterns of change in each model variable (CG Perceived Stress, CG Emotional Responses, and PWD BehSx) with consid eration of change a) within persons (level 1), and b) across persons (level 2) according to Treatment Stat us. It was hypothesized that experimental dyads would have improvement (or no deterioration) over time, on CG Short ZBIBedard, CG Emotion (comprised of Center fo r Epidemiological Studies-Depression & Negative PANAS), and PWD NPI-Q Behaviors, as compared to control dyads. Specific Aim # 2: To investigate the relationships between CG Perceived Stress, CG Emotional Responses, and PWD BehSx, with consideratio n of measures over 9 potential data points in a year, both within dyads and estimated across dyads within the population. It was h ypothesized that, both within and across dyads, both CG Perceived Stress and CG Emotional Response would have a direct positive relationship with BehSx in PWD. Within a given dyad, this meant that when the caregivers Perceived Stress and Emotional Responses were above their own average, the PWD
30 in their care would have higher Behavioral Symptom severity. Across dyads, this meant that when caregivers Perceived Stress and CG Emo tional Responses were, on average, higher, Behavioral Symptom severity in the PWD were also higher. It was also hypothesized that when both Perceived Stress and Emotional Responses in caregivers were modeled, the influence of Perceived Stress would diminish, in a manner similar to mediation. Summary If the m odel is supported through establishment of relationships, and directionality is supported through the modeling of change over time future studies may further support this view of BehSx. Intervention to assist the caregiver by decreasing the impact of the stress process may have far-reaching implications, including ame lioration of BehSx in dementia. This may potentially further relieve perc eived stress and emotional res ponses for the caregiver, and may have a positive impact on more distal outcomes, such as institutionalization of the person with dementia. Therefore, it is important for research to establish whether or not the caregivers stress process does have consequences within the dyad, in particular as related to PWD BehSx. If this precept is supported through research, caregivers and clinicians will be better positioned to direct interventions appropriately.
31 Figure 1-1. Full Consequences model adapted from Schulz & Martires Stress-Health model.
32 Figure 1-2. Reduced model for evaluation in current study.
33 CHAPTER 2 LITERATURE REVIEW Background and Significance The focus of this study was the community-bas ed dem entia care dyad. In this section, dementia, and the care of persons with dementia is discussed. Furthermore, the dementia care dyad is developed as integral to the care of pe rsons with dementia, and the problems associated with dementia caregiving are introduced. Dementia and Dementia Caregiving De mentia is a serious and cos tly health problem facing older adults and society at large, and will continue to increase its impact as the po pulation ages (L. E. Hebert et al., 2003a; Hoyert et al., 2005; Snow et al., 2004). De mentia is a general term refe rring to global loss of cognitive function that also interferes w ith social and occupational performance; the most common form of dementia in older adults is Alzheimers Disease (AD) (Smith & Buckwalter, 2005). Medical treatment may improve the pers ons functional life expectancy and reduce burden on informal caregivers and the health-care sy stem, an important contribution for this impending public health crisis. However, the most common dementia syndromes have no cure. Decline is usually steady, with worsening disability as time passe s, and medical treatments are limited in their effects (Doody et al., 2001; Geldmacher & Lin, 2006; Hart, 2004; Sink et al., 2005; Small et al., 1997). Once diagnosed, a person with dementia will have a shortened life expectancy; however, the decline may last up to 20 years. It has been estimated that this l ong phase of care costs the U.S. approximately $100 billion each year, and ex penditures will reach crisis proportions as the population ages ("Alzheimer's association: Statistics about Alzheimer' s disease", 2006; Riggs, 2001). Persons with dementia get sicker and their care is co mplicated by dementia, with
34 hospitalization often related to mismanagement of medications, malnutrition, dehydration, or injuries. American taxpayers reportedly spend about $7000 per year for an average Medicare patient (in 2006), and a person with even a mild dementia may double that cost. Furthermore, costs significantly increase each year past diag nosis (Geldmacher & Lin, 2006; Taylor & Sloan, 2000). Medicare expenditures related to AD are expected to increase 75% between 2005 and 2010, and Medicaid spending related to nursing ho me stays will increase 14 % ("Alzheimer's association: Statistics about Alzheimer's diseas e", 2006; Riggs, 2001). Approximately 80% of persons who have de mentia remain in the community; thus millions of Americans provide informal assistance to persons with dementia (Czaja et al., 2000). The unpaid informal caregivers of those with dementia are sometimes referred to as the invisible part of the dementia care system (R iggs, 2001). The Alzheimers Association reports that families provide almost 75% of care for those with the disease, with the average lifetime cost of family care for AD at $174,000 ("Alzheimer' s association: Statistics about Alzheimer's disease", 2006). When the cost of informal car egiving duties are considered, estimates are even higher, with a 1997 study reporting that informal caregivers personally provide services that would cost approximately $34,000 per year (in 1997) in the years prior to institutional care, spending almost $6000 on supplies and servi ces each year (Haley, 1997). Additionally, businesses that are a part of the American ec onomy lose approximately $61 billion per year related to lost productivity of informal caregive rs ("Alzheimer's associat ion: Statistics about Alzheimer's disease", 2006). As cognition declines, the costs of care increase for families. A 1995 study found that caregivers of older Americans had out-of-pocket expenses from over the previous two years that included $1350 for those without dementia, $2150 for those with mild/moderate dementia, and
35 $3010 for those with severe dementia ( p < 0.01) (Langa et al., 2004) In a study relating such costs to scores on the Mini-Mental State Ex am (MMSE), researchers found that when the baseline score was 20, preventing cognitive decline of five points saved $2424, while the same level of prevention when the baseline was 12 coul d save as much as $7407 (Ernst et al., 1997). Policy makers frequently seek to relieve so me of the burden for caregivers, with the anticipation of delaying placement of the person with dementia into extended care. This is sensible from a policy perspective since extend ed care often increases the burden for taxpayers. However, others argue that decisions about care should consider what is realistic and appropriate for persons with dementia, without placing informal caregivers into a situation that they can not realistically shoulder alone. Placement should be considered when it is meeting the needs of both caregivers and persons with dementia (Riggs, 2001). An impor tant consideration in this debate is that regardless of how difficult it becomes to care for an individual with dementia at home, many caregivers simply pr efer to continue this arrang ement. Additionally, while one recent study showed reduced burden after placemen t of the care recipient (Gaugler et al., 2007), other studies have indicated th at caregivers may not be relieve d of burden upon placement of the care recipient (Dellasega, 1991; Sc hulz et al., 2004). Even though placement may be the best solution, it is not a universal remedy to the di fficulties of caregiving, and caregivers remain integral to the caring process in most inst ances. Caregivers who do initiate placement face further decline of the person with dementia, cha nging roles, travel to the facility, and even feelings of guilt and ambiguity (Schulz et al., 2004). Understanding the consequences of dementia caregivers stress and finding ways to support them may be helpful to pe rsons with dementia and caregiver s alike, with the side effect of delayed placement and overall decreased costs of care. One potential avenue of supporting
36 community-based dementia care dyads is through the alleviation of BehS x that often occur in dementia and contribute to the caregivers stress process (Schul z & Martire, 2004). Additionally, at least some of the increase in costs of dementia care may be attributable to these BehSx. For example, a recent study estimated that for each increase of 1 point on the NeuroPsychiatric Inventory, a researcher-adminis tered tool that measures BehSx in dementia, total costs of care increase between $247 and $409 per year, de pending on the value of unpaid caregiving (Murman et al., 2002). Addressing Be hSx through evidence-b ased intervention may reduce financial costs and burden for informal caregivers. The impact of dementia is far-reaching, invol ving much more than financial costs. Caregivers frequently sacrifice both psycho-emotional and physical health in order to care for persons with dementia. Caregivers of those w ith dementia have substantially more negative personal consequences than those who care for physically impaired older adults, lo sing more of their personal life, including hobbies, vacations, a nd family time that could help to ease their burden. Caregivers with these high demands are at risk for serious illness and even death (Haley, 1997; Ory et al., 1999; Schulz & Beach, 1999; Sc hulz & Martire, 2004). The importance of caregivers in any discussion of dementia care cannot be over-es timated. Howard Fillit, MD, Director of the Institute for the Study of Aging in New York may have said it best: Caregiver burden is an exceedingly important co mponent of what happens in this disease. If we lose the caregiver, we lose the patient (Geldmacher & Lin, 2006). His assertion highlights not only the importance of the informal caregiver, but also the interconnectedness of the person with dementia and th eir primary caregiver. Perhaps in no other disease process in adulthood is the informal care dyad a more important consideration.
37 Dementia-Related Care Dyads The relationship that develops between inform al caregivers and care recipients with dementia is somewhat unique to caregiving s ituations involving adults. Even though all caregivers and recipients devel op a dyadic relationship, the demen tia care dyad has the declining cognition of the care recipient as a confounding factor. In any caregiving situation, protecting the basic relationship is important to enhance the caregiving experience and prevent the negative consequences of caregiving. Low levels of reci procity and higher levels of negative exchanges from the care recipient have been linked to nega tive appraisals of careg iving (Carruth et al., 1997). After a diagnosis of a dementia, though, underlying relationships undergo an almost immediate change, with decreases in intimacy and companionship, emotional distancing, loss of control of the situa tion, and decreased pred ictability (Quayhagen & Quayhagen, 1996). In spouse caregivers, the degree of marital happiness, equality in relationships, and expressions of tenderness may decrease (Eloniemi-Sulkava et al ., 2002). Relational spouses may have greater positive well-being and lower caregiver distress than those who approach caregiving with reactive or other interactive patt erns (Lewis et al., 2005). Dementia care dyads face changes in pers onal identities, personality, reciprocity, and communication that affect the relationship. Caregiving may have rewards for some, particularly if the previous relationship was satisfying (Snyder, 2000). However, most caregivers find the experience very stressful (Haley & Pardo, 1989). Being a care recipient in such a relationship has challenges as well, as persons with demen tia lose control of their own affairs (Davey & Eggebeen, 1998; Liang et al., 2001 ). Assisting either individual within a dementia care dyad requires recognition of the dyadic care relationship. Researchers have begun to acknowledging the importance of maintaining and strengthening the relationship in dementia car e dyads through interven tions to build and/or
38 protect the relationship. These ar e particularly important early in the disease, while activities that require reciprocit y or extensive communication skills ar e realistic (Haight et al., 2003; Sebern, 2005; Woods, 2001). Examples include: reminiscence and life review (Haight et al., 2003), communicating care preferences (Whitlatc h et al., 2006a; Whitlatch et al., 2006b), memory clubs (Zarit et al., 2004), and using memory aids to facilitate co nversations (Bourgeois, 1993). The findings from these studies show promis e for disrupting the traj ectory of decline in relationships that occur as a result of dementia. Another intervention for helping dyads with psycho-emotional adjustment early in the disease process was developed by Quayhagen & Quayhagen (1996), with substantial positive impact. The intervention, designed to promot e cognitive stimulation, co mmunication skills, and practical problem-solving techniques to combat th e effects of dementia, helped dyads return to relationship quality. Although the interventi on was launched into a context of depleted interaction skills, the dyads moved from a discordant phase, characterized by internal conflict, to equilibration, where the members of the dyad sou ght balance in their working relationship, then to regeneration, where skills were regained and/or affection was restored. Finally, there was an emergence of life quality in almost all of the care dyads. One caregiver remarked, It doesnt matter whether she really improves or not.. It has brought us closer together as a family (Quayhagen & Quayhagen, 1996). Such closeness early in the caregiving journey may be one key to success as the person with dementia declines further. As the condition of dementia progresses, the relationship may also influence caregivers perceptions of the experience of caregiving, with the quality of th e relationship and interactions serving as the lens through whic h appraisal of the caregiving situ ation is made (Lyons et al., 2002). For example, resentment and percep tions of being unappreciated may surface as
39 relationships suffer. These may in turn affect interactions in a nega tive way. Even in nondementia caregivers, quality of life indicators are thought to be dynamic and reciprocal within the care dyad, with the care recipients disease affecting burden, and burden in turn affecting the disease (Carruth et al., 1997; Glozman, 2004). Established dementia care dyads face a wi de range of phenomena related to their circumstances. A qualitative anal ysis based on direct observation of spouse dementia caregivers revealed 4 themes of caregiving, including activ ities of daily living, co mmunication, supervision & surveillance (often a 24-hour endeavor), and activity/stimulation. All the caregivers had given up most of their own lives to take on the role of caregiving, which included responsibility for running the household which had prev iously been shared with th e spouse, as well as providing support or help in care recipien ts activities. These caregivers were constantly on duty, rarely able to leave the care recipient al one, but demonstrated that they care about the person with dementia, as well as for them. The complexity and multi-dimensionality of the caregiving situation related to dementia was evident, and went far beyond overt care, with much of the stressful work of the situation invisible (Jansson et al., 2001). Caregivers face varied challenges as they struggle to provide care to a loved one with dementia, including lack of knowledge, feelings of guilt, isolation, uncertainty, loneliness (Gruffydd & Randle, 2006), and lack of personal time (Thomas et al., 2002). Perceived stress related to symptoms of depressi on and other behavioral changes in the care recipi ent contributed more to a perception of burden than physic al needs (Donaldson et al., 1998; Pinquart & Sorensen, 2003a, 2003b; Thomas et al., 2002). Care givers identified as skilled were more flexible and accepting of challenges, while thos e less skilled were more rigid and had more intense emotional responses to ca regiving (Farran et al., 2004).
40 Interventions that allow the ca regiver a sense of control (B uckwalter et al., 1999; Parks & Pilisuk, 1991), or improve their belief in their abili ty to give adequate car e (Fisher & Laschinger, 2001; Gilliam & Steffen, 2006; Gitlin et al., 2001) may improve the experience of caregiving in dementia, and thus have a major impact on the re lationship and interactions within the dyad, with potential for lasting effects. Testing a Model including Dyadic Effects In research, dyads consist of paired individuals who are not independent, but linked into a rela tionship that is characterized by partner effects, mutual influe nce, or a common fate. Partner effects occur when one persons characteristics affect the others outcomes (actor-partner); mutual influence occurs when each persons outcom es influence the others (cross-partner); and common fate assumes that both members are expo sed to the same causal factors. The two individuals are distinguishable as persons within the dyad by their roles, but have variables that may reflect their mutual experience (Kenny et al., 2006). Dyadic analysis, in the statistical sense, often refers to measuring an identical outcome in both members of the dyad (for example, marital quality in both husbands and wives), a nd then assessing influences of one member on another. Dementia care dyads do meet requirements of dyads usually considered in research; the interdependence, partner effects, mutual influe nce, and potential for common fate (Kenny et al., 2006). However, typically, the caregiver and care recipient in dementia care dyads have been considered independently in research studies, separate from the dyadic relationship, or dyadic effects have been assessed without specific con ceptual recognition of the dyadic nature of the relationships. Recent methodological advances ha ve prompted argument from some researchers that separating out members of a dyad for st udy does not adequately represent the actual situation (Lyons et al., 2002). Considering both dyad members si multaneously in research may
41 present methodological challenges when the care reci pient has dementia, rela ted to the difficulty in assessing relationship-based variables from the perspective of the person with cognitive decline. Nonetheless, the influence of th e dyad, with the caregiver in a prominent role regarding the tone of the relationship, must be recognized. Considering bot h caregiver and care recipient within the relationship allows a more relevant understanding of the dynamic process and interaction that occurs (Lyons et al., 2002). Several studies have targeted both caregivers and persons with dementia within the dyad. For example, dyadic interventions using psychoeducational counseling and occupational therapy for practical approaches to cognitive losses have been us ed to impact outcomes for both members of the dyad (Gitlin et al., 2003a; Gra ff et al., 2006; Nobili et al., 2004). A dyadic exercise training program along with caregiver education about beha vioral management strategies improved physical health and depression in persons with Alzheimers disease (Teri et al., 2003). In early stages of dementia, dyad-base d interventions are partic ularly beneficial for planning future care. These resu lts show that intervention geared toward one or both members of the dyad may have benefit for the other. In this study, BehSx in persons with dementia are conceptually regard ed as a dyadic effect of caregivers stress, with actor-partner eff ects from caregiver to PWD in relation to BehSx. While dyadic analyses in the traditional sense ar e not possible for testing these relationships, the dyad remains conceptually important. Theoretical Foundations The stres s of caregiving in dementia-related dyads has been associated with serious and extensive effects in caregivers, including psychological, social, and physical health changes for caregivers (Bergman-Evans, 1994; Haley et al., 1987; Roth et al., 2001). The Stress-Health
42 model (Schulz & Martire, 2004) extended previo usly developed dementia caregivers Stress Process model, incorporating previous research and conceptual work, to emphasize dementia caregivers health outcomes related to the stre ss process (Goode et al., 1998; Haley et al., 1987; Haley et al., 1996; Pearlin et al., 1990; Schul z et al., 1995). The Stress-Health model was adopted and adapted to explain th e consequences of caregivers stress process more broadly, and serves as the theoretical foundation for this study. Stress-Health Model Briefly, the Stress-Health m odel builds on prev ious work regarding the stress process in dementia caregivers, extending the effects of the stress process to include caregiver morbidity and mortality. Primary stressors related to care duties and secondary stressors related to family, work, and other aspects of the car egivers life as affected by the care duties are considered. The caregiver appraises the demands and their capacity to adapt, and when demands are overwhelming or resources inadequate, perceived st ress occurs or increases. The caregivers subsequent emotional and behavioral responses to stress can lead to health problems in the caregiver, or may increase the stressor itself or wo rsen the caregivers per ception of the stressor. Consequences of Dementia Caregivers Stress Process Model The Stress -Health model has be en extended in preparation for this study to include consequences of caregivers stress beyond those related to health of the caregiver, in the Consequences of Dementia Caregivers Stress Pro cess model (Figure 1). Identifying BehSx as a potential consequence of caregiv ers stress, this model also c onsiders the prominent role of caregivers in preventing or re ducing BehSx in PWD: identifying and meeting needs, identifying and avoiding triggers for stress, and providi ng a positive and engaging social environment (Algase et al., 1996; Buckwalter et al., 1999; Cohen-Mansfield, 2000b; Hall, 1994; Kolanowski et al., 2005). Lastly, this model acknowledges that although the person with dementia may have
43 deficits related to cognitive losses, he or she remains a whole, relational being who reacts to the caregiver and the care environment based on underlying personality and psycho-social characteristics, as well as the context of the situation (Kitwood, 1997; Touhy, 2004; Woods, 2001). The primary change from the core model is that a more comprehensive view of the influence of caregivers stress is incorporate d, with caregiver cons equences and dyadic consequences added as more global concepts, not necessarily measured directly, that serve to classify the potential outcomes of caregivers stress. For example, caregiver morbidity and mortality from the original model would be cl assified as caregiver consequences. Additionally, the feedback identified in the Stress-Health model could be considered circular; however, with regard to BehSx, the change in frequenc y and/or severity of PWD BehSx as a result of the caregivers stress process can al so be viewed as separate from the perception of BehSx as a problem when considered in light of all the stre ssful aspects of caregiving. Thus the original feedback connection between caregiv er emotional and behavioral responses and caregiver perception of BehSx as a problem and a stressor is removed. Rather, PWD BehSx and CG reactions to BehSx (Figure 1 double border ovals) are outcome or dependent variables, conceptualized as dyadic consequences of the caregivers stress process. While the distinction between BehSx as a dya dic outcome and perception of BehSx as a problem may be slim, dissimilarities exist. The latter reflects the care givers reflection on the BehSx as one of several or many stressors that contribute to the overa ll appraisal of demands versus their ability to adapt, wh ile the former is specific to the severity of the behavior and how they feel about or react to the BehSx. Even though the BehSx and reacti ons may contribute to the eventual perception of BehSx as a problem and a stressor, it is the outcomes that stem from
44 the stress process that are the focus of this mode l. An introduction to the model concepts and relationships follows. Expanded Consequences of Dement ia Caregivers Stress Process The basic precept of this newly revised model is that dementia caregivers stress process may have broad consequences. Caregivers stress process, with perceived stress and emotional and behavioral responses, may subsequently result in consequences for the caregivers themselves, or may have more global influence within the dyad. Caregiver Consequences Several ou tcomes specific to the caregiver may be classified as consequences of stress. The case for the caregiver outcomes of morbidity and mortality is clearly made by Schulz and colleagues (Schulz & Beach, 1999; Schulz & Mar tire, 2004; Schulz et al., 1997; Schulz et al., 1995), and there is extensive evidence to suppor t this model, including several meta-analytic reviews (Pinquart & Sorensen, 2003 a, 2003b, 2007; Vitaliano et al., 2003). Examples of specific health effects of dementia caregiving include changes in immunity, blood pressure, hormones, and lipid levels (Haley, 1997; V italiano et al., 2003; Wilkins et al ., 1999). Ineffective self-care routines and preventive practices, such as regular exercise, complicate this picture of compromised health in caregivers of those with dementia (Haley, 1997; Schulz et al., 1995). Additional caregiver outcomes may include soci al isolation (Nagatomo et al., 1999; Upton & Reed, 2006) and fatigue (Nagatom o et al., 1999; Teel & Press, 1999), among others. In addition to physiological and psycho-social sequelae, co gnitive performance of older adult caregivers may be influenced (Caswell et al., 2003).
45 Dyadic Consequences In the newly revised m odel, actor-partner effects may mean that consequences of caregivers stress extend to the person with deme ntia, or may include ou tcomes rooted in the dyadic relationship. Such outcomes are cl assified as Dyadic Consequences. Consequences for persons with dementia McClendon and colleagues (McClendon et al., 2 004) have previously suggested that the stress process m odel (Goode et al., 1998; Pearlin et al., 1990) be extended to examine the effect of caregiver characteristics on survival in the person with dementia. In addition to survival, serious consequences for the PWD may include abuse, mistreatment, potential for injury or becoming lost related to unsupervised activity, or premature institutionalization related to the caregivers stress rather than the needs of the pe rson with dementia. Each of these consequences is discussed briefly in this section. Care recipient abuse or mistreatment. Abuse is generally considered a risk factor for persons with dementia (Wolf, 1998) although there is some controve rsy in this regard (Gainey & Payne, 2006; Wijeratne, 1997). There was a st atistically significant higher prevalence of dementia in a sample referred for abuse than in a sample referred for other reasons (Dyer et al., 2000). In another study, 11.9% of those completi ng questionnaires reported abuse toward the person with dementia, while 33% re ported receiving abuse. The la tter group was more likely to direct abusive BehSx back at the care recipien t (Coyne et al., 1993). Lower functioning in the person with dementia, more years of caregivi ng, and higher levels of burden and depression, contributed to the possibility of abuse (Coyne et al., 1993; Williamson & Shaffer, 2001), and persons with dementia who have communication probl ems or disinhibition that may contribute to a cycle of violence are more at risk (Hansberry et al., 2005). Even when abuse is not overt, the care recipient may not be treated well. For exam ple, in a qualitative study about relinquishing
46 the caregiving role, one participant captured he r responses to the stre ss by saying that she found that she was not nice anymoreimpatient with her mother (Caron & Bowers, 2003). Injury or Becoming Lost. It is well documented that circadian rhythm and sleep disturbances are common in dementia (Motohash i et al., 2000). Ineffective sleep patterns because of nighttime caregiving duties or c oncern for the PWD can add to the caregivers susceptibility to the deleterious effects of caregiving. The resu lting sleep disruption, fatigue, and continual worry can lead to excessive caregiver burden, which can in turn lead to increased depressive symptomatology (Donald son et al., 1998; Rowe et al ., 2004; Teel & Press, 1999). The combination of caregivers who are fatigue d, sleepy, over-burdened and depressed with care recipients potentially unsafe activity (Klein et al., 1999; Rowe & Bennett, 2003; Rowe & Glover, 2001) may lead to the person with deme ntia becoming lost or injured. This serious consequence is certainly not limited to community -based care dyads, and is also fundamentally related to characteristics of the person with dementia, such as functional problems, memory, judgment, anger or agitation. None theless, injury or death of th e care recipient may result if a person with dementia does become lost (Logsdo n et al., 1998; Rowe, 2003b; Rowe et al., 2004), and falls or serious injury can occur even if the person doesnt get lost during unsupervised or unsafe activity (Kallin et al., 2005; Kolanowski et al., 2006; Oliver et al., 2007; Rowe & Fehrenbach, 2004; van Doorn et al ., 2003). Awareness of care recipients dangerous activity and the resulting concern about the person becoming in jured or lost, in the face of other stressors, may trigger institutional placement. Premature Institutionalization. Perhaps the most obvious and most often discussed distal outcome of caregivers st ress process is that of instituti onal placement of the person with dementia. When mounting caregiving stressors become severe, or reach a breaking point,
47 institutionalization is seen as a method of surrend ering, at least in part, the caregiving role, often after several crises have occurr ed (Annerstedt et al., 2000; Butche r et al., 2001). A majority of those placing a loved one cited the need for more skilled care than they could provide (Annerstedt et al., 2000; Buhr et al., 2006). However, in a prospective study, dementia was found to be predictive of nursing home admission even when controlling for functional decline (Banaszak-Holl et al., 2004). Caregivers who ex hibited a desire to institutionalize their care recipient had higher burden, more family dysfunc tion, and decreased social support (Spitznagel et al., 2006). Caregiver perc eived stress and associated depressive symptoms, along with problem BehSx, may result in placement as the only option for care (R. He bert et al., 2001). While placement may indeed reflect the best option for both caregiv ers and persons with dementia as the condition progresses and caregiv er health is compromised, placement earlier than is necessary may decrease survival rates (McClendon et al., 2006). Consequences for the dyad as a unit The caregiving relationship is by nature a dyadic process, and has the potential to overshadow previous interaction patterns (Lyons et al., 2002; Sebe rn, 2005). W hen the recipient has dementia, changes occur that shake the co re of the relationship (Whitlatch, 2001; Woods, 2001). Consequences of caregivers stress that affect both members of the dyad are primarily embedded within the care relationship, and may include changes in the relationship or in interpersonal interactions. Conversational interactions. While interpersonal interactions may be affected by caregivers stress process, there is very lit tle in the literature specifically regarding conversational interact ions within dementia care dyads. One case study reported on a woman with severe dementia in a nursing home, her car egiving daughter, and a researcher. Supportive conversations and initiation of the topic by the woman herself increased the amount of lucidity
48 displayed during the conversation; however, the daughter was not as sup portive and placed more demands on the woman than the researcher. S upportive conversation incl uded using therapeutic communication techniques, as well as ignoring errors and helpi ng to find words and complete sentences when needed. Demanding conversa tion included questioning and correcting. The author suggests that c ommunion, defined as a shared sense of reality and affective state, may be the most appropriate personcentered approach during inter action, with emphasis on staying within the person with dementias frame of re ference, listening and asking only open-ended questions, and leaving the initiativ e to the person with dementia (Normann et al., 2005). Formal caregivers of nursing home residents with dementia reported episodes of lucidity related to such non-demanding interactions, where trustworthiness and respect were maintained (Normann et al., 1998). Negative changes in the relationship and/or interactions. Caregivers stress process may influence their interpersonal interactions in specific ways. For example, caregivers level of anxiety was among predictors of abusive interactions in rural dementia care dyads (Compton et al., 1997). Caregiver role stress, comprised of personal distress, domestic upset, and negative feelings about the care recipient with de mentia, predicted psyc hological well-being and productive BehSx in persons with dementia, in cluding problem-solving, task performance, and social BehSx (Burgener & Twigg, 2002b). Dyadic outcomes such as mutuality, the positive aspects of the care relationship (Archbold et al., 1990; Lyons et al., 2007), and communality, the level of mutual responsiveness to the partner in the relationship (Williamson et al., 1998), may reflect aspects of ca regivers stress process. Caregiver coherence and care recipient attachment. Caregivers psycho-emotional state related to past losses may influence their caregiving relationships. In dementia care dyads
49 comprised of adult daughters who cared for mothers with mild to severe dementia, non-verbal interactions in the mother upon reunion after a br ief hiatus were correlated to the caregiving daughters coherence. Coherence, established during a previous interview, was defined as resolution of any losses they had experienced in a way that they could establish close attachments. In daughters who displayed more coherence, mothers displayed more signs of joy and responsiveness upon reunion, including joyful facial expressions, proximity seeking, contact-maintaining behavior, and responsiveness of the mothers, as well as reciprocal emotional attachment between mothers and daughters. The au thors suggest that this relationship is based on daughters emotional availability to their mo thers, without hindrance of emotional conflicts related to their attachment histories. Furtherm ore, the mothers almost universal ability to display attachment to a preferred other reflects the core social and emotional self, with earliest socio-emotional connections surviving the longest, namely, a working model of their capability of attachment. The authors conclude that mother s with dementia who live with a daughter who struggles with past history and related attachment issues, the result may be a fearful, intimidated existence (Steele et al., 2004). Positive vs. negative affect exposure. Caregivers affective manifestations may reflect their stress process. It has b een previously proposed that expo sure to genuine positive affect states improve how well any individual or social unit flourishes. Furthe rmore, experiences of negative affect are more detrimental than experi ences of positive affect are protective. The benefits of a high positive to negative affect ra tio cover a wide spectrum of valuable outcomes, including improved longevity. Positive thinki ng and actions can be triggered by pleasant feelings, and vice versa (Fredrickson & Losada, 2 005). This supports the importance of affective
50 responses in any relationship, but the care recipient with dementia may be particularly vulnerable to an imbalance toward negative affect e xposure related to thei r cognitive limitations. Are BehSx in PWD a dyadic consequence of caregivers stress process? BehSx have most commonly been studied as a s tressor in relation to the caregivers stress process, and have frequently been identified as the most important predictor for caregivers stress (Torti et al., 2004). However, whil e persons with dementia contribu te to caregivers stress, that same stress process may influence caregivers gene ral affect and ability to provide an optimal care environment, specifically, to be responsiv e, accommodating and pleasant in interactions. Because of their cognitive losses, persons with dementia may be vulne rable to the tone of interactions within the relati onship. Persons with dementia may perceive the caregivers stress/burden, depression, anxiet y, or negativity, and react acco rdingly. Such a convergence of interactions provides the foundati on for the influence of the care givers stress within the dyad, particularly as relate d to PWD BehSx. While health professionals and researcher s should not ignore contributions from the historical relationship or from other contextual factors, the caregivers stress process merits attention as a potentially influential predictor of BehSx and caregiver beha vior-related reactions. For example, in one study, high expressed emot ion, characterized as critical remarks or hostility toward the care recipient, increased the presence of disturbing BehSx in persons with dementia up to a year later. Authors concluded that environmental stimuli rather than organic causes resulted in BehSx (Vitaliano et al., 1993). It may be unrealistic to expect informal caregivers to provide an optimal care environment without first addressing their increased risk for perceived stress and psycho-emotional se quelae (Schulz & Martir e, 2004). In this person-centered conceptualization, dementia-related BehSx are recognized as an expression or communication of something rather than simply a problem to be managed
51 (Kitwood, 1997; Touhy, 2004; Woods, 2001). This con ceptualization is also consistent with other works that seek to expl ain dementia-related BehSx. Two models emerge as the most prevalent theoretical explanations of BehSx. A lowered stress threshold model (Hall & Buckwalter, 1987) posits an ever-increasing stre ss response related to environmental stimuli, which includes interactions with the caregiver A need-based model (Algase et al., 1996) views BehSx as representing phys iological and psycho-social n eeds that cannot otherwise be communicated by the PWD, and the caregiver as responsible for deciphering and responding to these needs. Both these models acknowledge the importance of the caregiver in the management of BehSx. This conceptualization also acknowledges that informal caregivers pr ovide services that put them at substantial risk. This model was not proposed to claim that informal caregivers should be held accountable for managing the ve ry BehSx that increase their stress, or for moderating their own reactions to BehSx. Rather, this model adva nces the notion that caregivers and persons with dementia are a dyadic unit that should not be separated in care practices or research. The model supports the development of caregiveror dyad-direct ed interventions that will have positive outcomes for both dyad member s. Discovering or promulgating effective methods to reduce caregivers perceived stress and emotional responses may also have more pervasive effects within the dyad. Additionall y, if BehSx and caregivers behavior-related reactions are reduced, the eventual perception of BehSx as a problem and stressor may also be reduced. Consistent attenti on to both members of the dya d in research questions and intervention designs and delivery may allow professi onals to more rapidly move past the hidden victim and lack of personhood conceptualizations to acknowl edge that both caregivers and persons with dementia may best be as sisted when they are considered a dy adic entity.
52 Relevance of the Proposed Theoretical Model Walker & A vant (2005) propose that a theory is most useful when it presents new insight, a better understanding, or a different view of the phenomenon. When the theory reflects an already-developed body of knowledge, there should be clarification, with new predictions or control possible where none previously existe d (Walker & Avant, 2005). Currently missing from the literature is a mode l recognizing BehSx as a produc t of the dyadic relationship including the caregiver, whose stress process ha s pervasive influence w ithin the dyad. This conceptualization will allow the science to move forward, with interventi ons geared at helping both caregivers and persons with dementia, rather than expecti ng caregivers to provide an optimum environment to alleviate BehSx without the personal support to fi rst decrease their own stress related to this intense care environment. Although some researchers have recognized this tenet within their studies, the work is most of ten presented from one dyad members perspective or the other. This model is a real istic natural advance in theory and science at this juncture. The newly developed theoretical model is an attempt to extend current views of the influence of caregivers stress wi thin the care dyad, while also providing a new wa y of looking at dementia-related BehSx within a holistic, dyadic car e context. This model is most relevant in community-based care dyads, reflecting the caregiving milieu as a function of the caregivers stress. The model recognizes that outcomes fo r both members are inextricably imbedded within the context of the informal care relationship. This model is not an attempt to fully explain BehSx in persons with dementia. However, if s upported through research, th is theoretical model supports adopting a view of BehSx as at least part ially a product of the dyad rather than solely related to the disease. This viewpoint may provide a more relevant framework for intervention, emphasizing the salience of caregiver-directed or dyad-directed intervention with emphasis on stress reduction, and recognizing th e care dyad by addressing the issues they face as a unit.
53 Dyadic Effect of Caregivers Stress Process: PWD BehSx A conceptual m odel with selected key relationships for beginning the process of testing the newly revised theoretical model wa s developed for this dissertation study. This reduced model, the Dyadic Effect of Caregive rs Stress Process: PWD BehSx model, posits that PWD BehSx are a dyadic effect of caregivers perceived stress, that may also be influenced by caregivers emotional responses to stress with the latter acting in a manner similar to that of a mediator (Figure 2). The review that follows will focus on the concepts and relationships in the reduced model to be tested in this study. Caregivers Perceived Stress Perceived s tress is defined as an appraisal that ones adaptive cap acities are lacking in relation to the demands of the situation; i.e., th at resources are inadequa te or demands are too high (Pearlin et al., 1990; Schulz & Martire, 2004). In comparison, Zarit operationally defined caregiver burden as the extent to which car ing for a relative has caused a perception that caregiver emotional or physical he alth, social life, or finances are suffering in relation to the caregiving role (Zarit et al., 1980; Zarit et al., 1986). Burden rela ted to caregiving is generally seen as subjective belief that resources are not or will not be sufficient to meet the demands of the role, and burden is viewed as a precursor to negative outcomes such as caregiver depression (O'Rourke & Tuokko, 2003). The similarity of the p erceived stress and the caregiver burden definitions is striking. These similarities extend to the most widely used instruments designed to measure these concepts. For example, in one version of the Za rit Burden Interview, participants are asked to rank items reflecting their levels of stress, anger, negative affe ct, privacy, uncertainty, loss of control, inadequacy, strain, time pressure, and de gree to which their health and social life have suffered (Bedard et al., 2001). In contrast, the Perceived Stress Scale (S. Cohen et al., 1983) asks
54 for rankings regarding stress, adequacy, coping with change and unexpected events, lack of control, time pressure, anger, confidence, and feelings of being overwhelmed. The Perceived Stress Questionnaire (Levenstein et al., 1993) is somewhat more subjective and comprehensive but includes some of the same responses, for ex ample, worries, pressure s, frustrations, time pressures, too many demands, inadequacy (or judgme nt from others), fear s, and overloaded with responsibility. Given these evident similarities, including studies that use dementia-caregiving-specific terminology and approach will add to the discussi on of the stress process in this population, and will be included in this review. Burden is a much more common measurement in dementia caregiving literature, and is more specific to ca regiving demands and stressors. While caregiver burden may not be a pure reflection of percei ved stress, burden and perceived stress do encompass many of the same latent concepts. In this review, the two terms are used interchangeably. The perceived stress associated with demen tia caregiving is pervasive among informal caregivers. Authors of a multi-national, multi-con tinent review state that the remarkable findings of substantial perceived stress in caregivers of those with de mentia, coupled with an aging population, are staggering in thei r impact. While the perceived stress of dementia caregiving could be described as universal across studies and geographic re gions, cultural considerations such as obligations of women in caregiving also had considerable influen ce (Torti et al., 2004). Predictors of Perceived Stress/Burden In a study of 1,874 community-dwelling care recipients, perceived stress has been correlated to basic ADLs, IADLs, depression, severi ty of dementia, and co-morbidities in care recipients, as well as depressi on in caregivers. Caregiver pe rceived stress ratings were significantly higher when care recipients were male, had fallen in the last 6 months, had
55 behavioral disturbance, or a diagnosis of demen tia or chronic obstructive pulmonary disease. Furthermore, caregiver perceived stress was si gnificantly worse when caregivers themselves were spouses rather than children, had poorer health, or fell in the age range just after retirement, 65-74. Multiple regressions identi fied independent predictors of caregiver perceived stress: care recipient fall history, depression, basic ADLs, a nd severity of dementia, and also care giver depression. These, together, expl ained 33% of the variance in careg iver perceived stress in this study (Kuzuya et al., 2006). Perceived stress was also significantly correlated to BehSx in persons with dementia, with aggressive BehSx most disturbing and memory-related BehSx least burdensome (Papastavrou et al., 2007). Caregiving duties. Types of care tasks required, numb er of hours spent in caregiving duties, amount of supervision required by the person with dementia, and living with the care recipient all influence dementia caregivers pe rceived stress (Torti et al., 2004). Based on a critical review of caregiving literature, Chou a nd colleagues (1999) developed and tested a model of caregiver perceived stress in Taiwanese deme ntia caregivers, where caregiving involvement and emotion-focused coping had a positive effect on caregiver perceived stress (more perceived stress as these factors increased). Filial obligation, caregiving self -efficacy, and problemfocused coping had direct negative effects (less perceived stress as factors increased). While the demands caregivers face increase thei r perceived stress, it was the person involvement in that care which lead to caregiver bur den/stress (Chou et al., 1999). Caregiver characteristics. Once the informal caregiver is involved, a confident approach to the associated problems is more beneficial to the caregiver than an emotional response to the situation (Chou et al., 1999). Personality traits of caregivers may influence perceived stress and the associated psychological symptoms more than the stressors. In one study, neuroticism, or the
56 trait tendency to negative affect and sensitivity to stress, accounted for 28% of the variance of caregiver perceived stress, wh ile care recipient behavior problems accounted for 4%. Neuroticism was also negatively associated with positive affect, and social support was positively associated with positive affect. Positive affect was in turn related to reduced perceived stress (Shurgot & Knight, 2005). This supports previous res earch that indicated neuroticism (high) and mastery (l ow) increased reports of percei ved stress related to caregiving (Bookwala & Schulz, 1998). Care receiver characteristics. BehSx of dementia are consistently reported as one of the main contributors to caregiver perceived stre ss. In a systematic review regarding this relationship, cross-sect ional studies supported th is connection, with pooled data suggesting a moderately strong association of BehSx to percei ved stress, distress, an d depression. However, the temporal relationship between the variables was not successfully es tablished, reportedly due to lack of longitudinal studies meeting inclusion criteria (Black & Almeida, 2004). BehSx are also consistently found more burdensome than phys ical tasks of caregiving, and emerge as the primary source of caregiver stress in North American studies (Tor ti et al., 2004). Care recipient anger and aggressiveness were associated with caregiver psychological burden, but not social burden or guilt (Ankri et al., 2005). In another study, perceived stress was significantly related to BehSx in persons with dementia and reactions in caregivers, with aggressive BehSx most disturbing and memory-related BehSx least burd ensome (Papastavrou et al., 2007). Communication. Using structural equation modeli ng, communication difficulties related to dementia were found to affect caregiver pe rceived stress. Communi cation problems were linked to demand burden, stress burden, and obj ective burden in the caregiver, but the relationship was mediated by BehSx. Communication problems mediated the relationship
57 between cognitive and functional status and BehSx in the persons with dementia, indicating that perhaps the inability to communicate could lead to frustration and then BehSx, which in turn contributes to perceive d stress in caregivers. Communica tion problems and BehSx also fully mediated the relationship between cognitive/functional status and time burden; however, there remained a direct link from cognitive/functional status to demand burden (Savundranayagam et al., 2005). These findings highlig ht the importance of communicat ion within the care dyad, and its relationship to the nega tive effects of caregiving. Progression of Perceived Stress/Burden in Dementia Caregiving There is sub stantial perceived stress early in the dementia caregiving process, while there seems to be some adaptation to the role over tim e, with lessening percei ved stress (Torti et al., 2004). In an Australian study, longe r duration of caregivi ng and practical coping were related to less perceived stress (McConaghy & Caltabiano, 2005). A Canadian study revealed that perceived stress increased over 6 months time when the care recipient had Alzheimers dementia, while perceived stress decreased over the same period when the presenting dementia was not Alzheimers (Bedard et al., 2001). Caregivers of persons with dementia who have poor health, limited social life, and lack of positive outlook on the situation were vulnerable to burnout related to caregiving burdens in one study. Women, particularly wives, are more at risk for burnout, and may be more emotionally involved, closer to the situation, than men. Furt hermore, those who cared for the person with dementia at home had virtually no differences in risk of burnout than those whose loved one was institutionalized; guilt and grief related to caregiv ing do not go away when the care recipient is institutionalized. Even siblings and distant rela tives are at risk for bur nout if they are closely involved (Almberg et al., 1997). When the relati onship between perceived stress and collapse of caregiving at home, or inst itutionalization, was e xplored, those caregivers who were not
58 living in the same home had grea ter collapse, as did those ca ring for someone with behavior problems (Gort et al., 2007). Treatment of Perceived Stress/Burden Caregiver perceived stress has pr ove n resistant to treatment. In a meta-analysis meant to clarify the state of science, Acton & Kang (2001) found that, overa ll, the interventions evaluated had no positive effect on perceived stress; in some cases perceived stress worsened in the intervention group or improved more in the co ntrol group. The interventions were grouped by type: support group, education, psycho-educa tion, counseling, respit e care, and multicomponent, with only the latter t ype showing significant effects on perceived stress. Citing the tenet of more is better with regard to treating perceived stre ss, the authors state that although blanketing caregivers with help in many areas seems logical, se veral interventions considered multi-component did not successfully demonstrate a difference. The authors propose that perceived stress as a construct is perhaps too multi-dimensional for use in evaluating interventions (Acton & Kang, 2001). Association between Perceived Stress and Emotional Responses Caregiver perceived stress has been shown to play a large role in m ental and psychological outcomes for dementia caregivers. For example, perceived stress has been associated with psychiatric symptoms across studie s in a review of the literature (Schulz et al., 1995). In a study that associated mental health and sleep, older wife caregiver s of persons with Alzheimers Dementia had significantly poorer mental health ou tcomes than controls. However, the caregiver perceived stress score was a better predictor of ment al health than the set of caregiving stressors, which included BehSx, years of caregiving, and hours spent in caregiving each day (WilletteMurphy et al., 2006). In a small sample of deme ntia caregivers in Australia, caregiving burden was found to influence satisfaction with life (47% of variance in the measure) and psychological
59 stress, with perceived stress diminishing over time when practical coping was utilized (McConaghy & Caltabiano, 2005). In a recent st udy of 172 dementia caregivers, 68% had high levels of perceived stress, and 65% had depressive symptoms, with almost half at risk for clinical depression, and these symptoms were further co rrelated with BehSx (Papastavrou et al., 2007). Berger and colleagues (2005) found differing rela tionships over time in their longitudinal study (2-yr.) of dementia symp toms, caregiver perceived stre ss, and caregiver psychological variables. In the face of declining cognitive and functional status, worsening severity of dementia, and increased behavi oral disturbances, perceived stress was essentially stable, increasing only slightly. Depression, on the othe r hand, did become more prevalent over time. However, there was a decrease in the amount of se vere depression even as slight and moderate depression increased. The authors conclude that some caregivers have an intense emotional adjustment in the beginning of the caregivi ng process that subsides over time, while the perceived stress and depression that is related to symptoms gene rally increase (Berger et al., 2005). Caregiver Emotional Responses to Stress Em otional responses stem from the perceive d stress or burden of caregiving, and may include such responses as depressi on, anxiety, negative affect, anger, irritability, and others. In a comprehensive review of the l iterature, it was found that virtua lly all studies of dementia caregivers reported high rates of depressive symptoms and anxiety (Schulz et al., 1995). For example, in a study of 44 dementia caregiv ers with matched controls, caregivers had significantly worse depression, affect and satisfaction (Haley et al ., 1987). More recent research has suggested that those caregiver s with less stress and more meaning and benefit from duties, and more support, had lower depression and impr oved life satisfaction (Hal ey et al., 2003). Mahoney and colleagues (2005) reported dissimilar results, with 10.5% of older adult caregivers
60 at risk for clinical depression, similar to that of the general older adul t population. Anxiety was more common in this sample, with 23.5% displa ying case levels. Th eir sample reflected caregivers from various settings, including living alone, with caregiver, or in an institution (Mahoney et al., 2005). In the national, multi-si te REACH trials, prevalence of risk for depression (CES-D > 16) wa s 30% (Burns et al., 2003). Predictors of Emotional Responses Caregiving duties. W hile the duties related to caregi ving would intuitively qualify as a predictor of caregivers emotiona l responses, the research findings mixed in this regard. Care recipient disability has been associated with emotional symptoms in caregivers (Berger et al., 2005). Caregivers of those with more severe dementia, with increased ADL and functional dependence, and increased hours sp ent caregiving per week were mo re likely to have depressive symptoms (Covinsky et al., 2003). In c ontrast, another study found that those non-caregivers with dependent relatives who had more severe di sability, including physical or cognitive decline, had higher levels of depressive symptoms, while caregivers and those wi thout severe care needs did not have increased levels of depression (Amirkhanyan & Wolf, 2003), suggesting that it is not the provision of care that causes depression but the needs of the person for which care is provided, and that providing such care may even be protective. In female caregivers in the natural environment, situati onal stressors increased negati ve mood, while perceived support reduced the impact of the stressor on mood (Atienza et al., 2001). Caregiver characteristics. Caregivers with positive feeli ngs about caregiving have been shown to have significantly lower depression scor es (C. A. Cohen et al., 2002). Subjects that appraised caregiving tasks as le ss stressful and associated meani ng and benefits with caregiving also had less depression (Haley et al., 2003). Optimism consistent ly impacted mental health and affect in a positive way, more than effective coping mechanisms. However, effective coping did
61 improve caregivers affect, and also reduced stress when dealing with difficult caregiving situations (Gottlieb & Rooney, 2004). Care recipient characteristics. BehSx in persons with dementia were predictors of caregiver depression, with angry and aggressive BehSx particularly influential (Covinsky et al., 2003). Irritability in the care r ecipient and more impairment in ADLs in the care recipient have also been found associated with depression and anxiety (Mahoney et al., 2005). In Alzheimers dementia, Neundorfer and colleagues (2001) used multi-level modeling to examine change in depression in caregivers over time. Higher dependency in IADLs and more depressive symptoms in the patient were related to caregiver depression at the beginning of the study. Over 5 years time, increase in patient depression and IADLs was associated with increase in caregivers depression, while the rate of acceleration in caregiver s depression was associated only with patients acceleration of functional decline. The aut hors compare caregivers in this study to spouses whose depressed partners ma y be critical, negative, and unappreciative (Neundorfer et al., 2001). Progression of Emotional Responses Several researchers have found an increase in depression over tim e in dementia caregivers, often related to BehSx in care recipients (Ber ger et al., 2005; Gaugler et al., 2005; Neundorfer et al., 2001). However, this finding is not consistently reported. Powers and colleagues (2002) examined stability of depressive symptoms us ing the Beck Depression inventory in caregivers over a two-year period in the face of cognitive de cline in persons with dementia. Depressive symptoms were substantially similar over time using intra-class correlations, and avoidance coping strategies were positively correlated to de pressive symptoms. The authors conclude that depressive symptoms are likely tr ait-based rather than dependen t on a state; and thus may not adequately assess response to an intervention (Powers et al., 2002) In another study, caregivers
62 were examined in relation to depressive symp toms prior to becoming caregivers, and existing depressive symptoms upon care recipient diagnosis actually decreased as dependency in IADLs increased, while those without pr evious depressive symptoms ha d increased symptoms as care recipient dependency increased (Neundorfer et al., 2006). Progression of emotional respons es may be independent of d ecisions regarding placement. Schulz and colleagues demonstrated that placement of a relative in long-term care did not ease the symptoms of depression and anxiety, and may even make these symptoms worse, particularly in spouse caregivers (Schulz & Martire, 2004) Furthermore, among those caregivers whose relatives resided in nursing homes, those who cared for someone with dementia were exposed to more psychological distress. This was despite no difference in coping strategies or number of visits to the facility (Levesque et al., 1999). Treatment of Emotional Responses Despite findings that suggest depressive sy m ptoms may be trait-like characteristics (Powers et al., 2002), depressive symptoms seem more amenable to treatment than caregiver perceived stress. For example, Buckwalter and colleagues (1999) were successful in alleviating depression in caregivers by teaching them techni ques to manage behavioral issues in care recipients with dementia using the progressively lowered stress threshold theory. The authors argue that the intervention helps caregivers regain a sense of cont rol over their lives that is lost when faced with dementia caregiving; this loss of control may exacerbate depressive symptoms. They also measured subjective affect, and f ound significant improvements in caregiver tensionanxiety, anger-hostility, fatigue-ine rtia, and confusion-bewilderment (Buckwalter et al., 1999). Another intervention, teaching caregi vers to match assistance to leve l of disability in dementia, was successful in reducing depressive symptoms ove r time, as compared to those who received a
63 periodic phone call only (Chang, 1999) In the REACH trials, risk of clinical depression was reduced in the enhanced car e group (Burns et al., 2003). PWD Behavioral Symptoms It has become eviden t in r ecent years that challenging BehSx are quite common in all stages of dementia (Tractenberg et al., 2002), with a total prevalence rate as high as 90% (Steinberg et al., 2003), and with the potential for up to 98% developi ng symptoms over a fiveyear period (Lyketsos, 2007; Steinb erg et al., 2006). Regardless of this high rate of occurrence, it is important to consider that troubling BehSx may not be related to cognitive decline alone. Although cognitive symptoms such as short-term memory loss and declines in judgment and other cognitive functions are hallmark symptoms of dementia, non-cognitive symptoms are pervasive in this population as well (Shah & Allen, 1999). Thus, BehSx are multi-dimensional and may reflect cognitive disability as well as psycho-social features. This study focuses on psychosocial aspects of BehSx. For example, poor previous interpersonal relationships and comorbid clinical depression may contribute to BehSx (Compton et al., 1997; Hamel et al., 1990; Schulz & Martire, 2004; Wolf, 1998). BehSx have been conceptualized as dyadic in nature, reflecting th e influence of actorpartner effects within the dyad. BehSx can thus, wh en considered distressful, be interpreted as a problem or stressor and contribute to the caregivers stress. For example, care recipient BehSx, particularly apathy, led to deteri oration of the marital relationship in those dyads where spouses cared for a husband or wife with dementia, rega rdless of cognitive or func tional abilities of the patient (de Vugt et al., 2003). Th ese findings highlight the strong clinical significance of BehSx within the care dyad, especially since apat hy has been shown one of the most commonly appearing initial symptoms (Lyketsos et al., 2002 ; Steinberg et al., 2003). Stress within the dyadic relationship related to apathy in the beginning of the disease process may set the stage for
64 negative interactions that will ha ve long-term effects within the dyad. More importantly to this study, caregivers stress and emoti onal responses may contribute to the severity of BehSx. The review that follows will highlight the evid ence that supports this relationship. Influence of Caregiver Stress Process on PWD BehSx Experts em phasize that it is the caregivers re sponses and approaches that must be altered to effectively deal with BehSx in persons with dementia, and that the person with dementia can not willingly change. Good communication, cr eativity, flexibility, accommodation, affection, reassurance, patience, consideration, and compassion are emphasized (Cohen-Mansfield, 2001a; Logan, 2004) to contend with BehSx in persons with dementia. Coping with difficult BehSx in dementia care giving was the primary theme that emerged in a descriptive study that select ed subjects from a major caregiver clinical trial. The authors reported that skilled caregivers were more awar e of behavioral sequenc es and responded with creativity and persistence, awar e that their responses could es calate or alleviate the BehSx. Skilled caregivers were more able to attribute BehSx to the condition rather than a personal affront. Creativity, flexibility, attentiveness, pe rsistence, emotionally responsive interactions, and monitoring their own responses to care recipient BehSx were seen as key to effective care of the person with dementia with difficu lt BehSx (Farran et al., 2003). In a large population-based (US) study, researchers sought to delineate caregiver characteristics which may contribute to BehSx in dementia. Citing a clear association between caregiver perceived stress, depression, and car e recipient BehSx, the au thors propose that the relationship is likely bidirecti onal. Their findings indicate d that younger and less educated caregivers reported more BehSx, as did those who were more burdened, more depressed, or spent more hours in caregiving. This relations hip persisted after controlling for dementia
65 severity, and the authors propose that this relatio nship is likely valid despite the cross-sectional nature of the study (Sink et al., 2006). In addition to links between caregiver perceive d stress, depression, and ways of interacting with care recipients, the potential influence of the caregiver stress process is evident in literature regarding a) caregiver coping and management strategies, b) caregivers presenting affect, and c) interventions meant to assist caregivers, but with direct outcomes related to BehSx. Influence Related to Caregiver Coping and Stress Management Caregiver coping strategies have been associat ed with survival in persons with dementia, which m ay in turn be linked to behavioral and functional decline. In particular, wishfulness intra-psychic coping was related to shorter survival time in the person with dementia, while instrumentalacceptance coping was not. The au thors propose that caregivers who use intrapsychic coping may be less responsive and ps ychologically available to the person with dementia, which may in turn lead to inadverten t contributions to decl ine in the person with dementia (McClendon et al., 2004). In another study, those caregivers labeled as non-adaptive were more neurotic, more burdened and depressed, felt less competent, an d reported more BehSx in general, and more hyperactivity symptoms specifically in the care recipient with dementia. Furthermore, changes in behavior patterns over time refl ected caregivers non-adaptive a pproaches, suggesting that the BehSx were in direct response to the caregiver interacti ons, such as impatience or irritation. The authors suggest that effective strategies may decrease caregiver perceived stress and also improve BehSx in the persons with dementia (Aalten et al., 2003; de Vugt et al., 2004). A separate study indicated th at trait-based optimism ha d more influence on coping effectiveness than situation-based factors such as ways of coping. Optimism improved coping, increased positive affect, and decreased negati ve affect, over and above the contribution of
66 coping effectiveness, which also decreased negati ve affect and improved mental health. BehSx such as stubbornness, repetitiveness, and frequent complaints in persons with dementia were correlated to significantly poorer mental health scores and less positive affect in caregivers (Gottlieb & Rooney, 2004). Influence Related to Caregiver Presenting Affect Even in cog nitively intact older adults, nega tive social exchanges have been associated with distress and sense of well-being, while positive social exchanges were linked only to wellbeing. Emotional exchanges, such as anger or criticism, signifying deterioration of the relationship, were the strongest negative predictors. When these occurred in relationships with spouses or family members they were especially hurtful. Alternativ ely, companionship was strongly associated positively with well-being and negatively with distress. These findings support the need for positive social interaction to promote psychological health in older adults, who may be especially sensitive to negative exch anges such as criticism (Newsom et al., 2005). In dementia care dyads, enmeshment or cohesion seemed to play a protective role for caregivers. Depressed and anxi ous caregivers had less expressed positive affect and less conflict resolution. Burdened caregivers expressed more anger. Depressed and burdened caregivers were more likely to have care recipients who showed signs of disengagement (Mitrani et al., 2005). More depressed spouse caregivers were al so found more likely to treat their partner in potentially harmful ways, such as threaten ing or handling roughly (Williamson & Shaffer, 2001). In burdened caregivers of persons with Alzh eimers disease, emotional lability and destructive BehSx in the patient led to anger-rese ntment and restrictions in caregivers personal time and social life. When social and persona l time were limited, depression and anxiety were outcomes. The authors conclude that perceived stress in caregivers may create a situation where
67 both caregivers and persons with dementia may experience nega tive changes in day-to-day interactions and morale in the home. (Croog et al., 2006). In Australia, in rather optimal cultural c onditions with regard to support and living conditions, caregivers approached BehSx by i gnoring the problem or providing tolerance and understanding. Rebukes and threats had been ab andoned as not effective or making matters worse. Researchers concluded that these care givers naturally practic ed methods recommended for optimal, dementia-friendl y care (Ward et al., 2003). In a study of families where affective responsiveness (closeness vs. anger, sadness, & criticism), problem solving, communication were impaired, caregivers reported both poor family functioning and higher ratings of strain and burden. However, it was not clear wh ether perceived stress/strain resulted in poor family func tioning, or vice versa (H eru et al., 2004). Caregiver affective approaches can also influence BehSx in a positive way. Caregivers singing during morning care routines which can be stressful to persons with dementia, were compared to background music and usual care wit hout music. Music, and even more so, singing, resulted in dramatic improvements in balance an d sensory awareness, physical strength, and use of space. For example, persons with dementia had improved posture, were able to participate more fully in activities, and seemed more happy (Gotell et al., 2003). Expressed emotion, considered the level of criticism and conflict displayed in the relationship, was higher in adult children caregivers than in s pouses, and in dyads where BehSx in the care recipient were a probl em. Caregivers attrib ution of behavior problems to the control of the care recipient was predic tive of patient-to-caregiver, car egiver-to-patient, and general criticism/overall conflict in the relationship. The authors caution that correlational analyses limit these interpretations (Spruytte et al., 2002). Howe ver, increased expressed emotion (criticism) in
68 caregivers did predicted an increase in BehSx in the care recipient over five years (Vitaliano et al., 1993) in another study. Caregivers taught to maximize communication with the person with dementia, based on abilities, contributed to significantly decreased BehSx, while symptoms increased in the control group. Suggestions ranged from simplifying la nguage and concepts to using primarily nonverbal communication in later st ages (Silvestri et al., 2004). In comparison with the degree of cognitive losses, persons with Alzheimers dementia do retain some ability to recognize non-verbal cues and use them to identify emotions in facial expressions, although this skill is lessened in co mparison to normal older adults. Furthermore, persons with decline were able to recognize several diverse emotions such as anger, happiness, and sadness. The authors emphasize that it is import ant that caregivers pay attention to their own non-verbal communication, keeping it congruent with what they ar e saying. They also propose that non-verbal responses from the person with dementia may be a response to accurate perceptions of emotions in othe rs (Bucks & Radford, 2004). In videotaped observations of couples that di d or did not have Alzheimers disease, noncaregiving couples had more interaction in ge neral and were more supportive of each other; caregivers were more depressed and more stress ed, and also had less re ciprocity and shared pleasures with their spouses. Non-caregivers had higher scores on the measure of hope, but there were no differences between caregivers and noncaregivers in shared values and emotional closeness. Caregiving wives were more facilitative during a comp lex planning task than during a meal, while husbands with AD tended to use interactions meant to build rapport. Husbands without AD became more facilitative during the planning task, but those with AD were not able to manage such facilitation (G allagher-Thompson et al., 2001).
69 Interventional Evidence for Caregiver Influence on BehSx A m eta-analysis review of car egiving interventions supports the influence of caregivers on persons with dementia. In addition to careg iver outcomes, a meta-analysis assessed the effectiveness of caregiver interventions on care receiver symptoms. Multi-component and interactive, participatory psycho-educational pr ograms for caregivers were found most helpful. Longer interventions and those delivered in group settings were most effective in improving care receiver symptoms (Pi nquart & Sorensen, 2006). In a review to establish which psychological interventions provided ad equate evidence for practice, interventions based on the PLST model and a standard antecedent-behaviorconsequence model emerged as most helpful. In all studies that contributed to the evidence, a problem-solving approach, with identification of trigger/antecedents, and modifications of the environment, the schedule, or interactions were recommended for treatment (Logsdon et al., 2007). Within the National Institutes of Health Resources for Enhancing Alzheimers Caregivers Health (REACH) initiative, one study demonstrated that an in-home skill-building program designed to teach caregivers management techniques and to enhance the environment for safety and ease in caregiving, could help both caregivers and persons with dementia. At 6 months caregivers had improved skills and persons with de mentia had fewer behavioral issues. At 12 months caregiver affect was im proved and BehSx remained in a decreasing trend. The authors suggest that caregivers sense of personal control through enhan cement of skills may result in less feelings of being overwhelmed, thus changing their affect, and their mo re skilled approaches could help to reduce Be hSx (Gitlin et al., 2005). A dyadic exercise intervention along with education of car egivers about managing BehSx was found to improve behavioral and functional out comes in the persons with dementia (Teri et
70 al., 2003). This study extends a previous body of evidence that education of the caregiver(s) can make a difference in reducing behavioral pro blems, and may even reduce institutionalization (Teri, 1999). In an intervention based on a stress and c oping framework, treatment consisted of a psycho-educational program designed to decrease the impact of Be hSx within the dyad. After 16 weeks, treatment subjects had significantly re duced their reaction to BehSx compared to controls, 14 % and 5% decrease, respectively. The fre quency of BehSx was also reduced, as was the composite frequency/reaction scores. The authors suggest that the reason frequency of BehSx was reduced was that they had likely im proved their coping abilities, with a resulting effect on BehSx in the person with dementia. However, there was no impact on secondary measures such as burden, stress, or anxiety (R. Hebe rt et al., 2003b). Significant findings for both members of th e dyad were described after a randomized control trial testing th e effectiveness of inte r-disciplinary collaborat ive care emphasizing nonpharmacologic management of BehSx in dementia BehSx and caregiver distress related to BehSx, and caregiver depression were significantly improved (C allahan et al., 2006). BehSx have emerged as important outcomes when both members of the dyad are targeted in intervention studies. However, research ha s not consistently found that caregiver-directed interventions had an influence on BehSx. For ex ample, BehSx, functional decline, and caregiver responses to these were assessed for longitudina l change after a psycho-educational intervention geared toward caregivers, yielding positive outco mes, but outcomes which varied according to relationship of caregiver to care recipient (Gerdner et al., 2002). Negative caregive r appraisal of BehSx, but not BehSx, were found reduced in a 4year longitudinal study using a counseling and support intervention (Mittelman et al., 2004a) and in a 3-year longitudinal psycho-educational
71 support intervention (Ostwald et al., 1999). An intervention designed to improve caregivers sense of personal competence and control led to fewe r declines in instrumental activities of daily living in care recipients, but not in BehSx. Nonetheless, women in the study reported enhanced self-efficacy with regard to handling BehSx, and spouses reported less upset about BehSx (Gitlin et al., 2001). In the REACH initiative, one study targeted behavior management through a primary care intervention. While BehSx were expected to improve, the measurement consisted of how bothered the caregiver was by the BehSx rather than behavior severity. Over 2 years, caregivers bother significantly decreased in a basic behavioral management intervention, but did not improve significantly more in a comparison grou p who received an enhanced intervention with additional treatment geared toward relieving caregiver depression and improving well-being. Depressive symptoms did improve more in the enhanced interven tion group (Burns et al., 2003). While the improvement of bother related to BehSx seems independent of improvements in depressive symptoms in this sample, BehSx in the enhanced care group were initially significantly lower than those in the basic group, and may not have had room for extra improvement. For instance, in the basic group bother was reduced from 19.6 to 14.8, while in the enhanced group, bother was reduced from 11. 8 to 9.2 (Burns et al., 2003). Even though baselines were controlled for in analyses, th e range for bother was 096, and a score of 9-11 could reflect the minimal amount of bother on less than half the possible 24 behavior items. Summary Dem entia and its sequelae can be considered an impending public health crisis. As persons with dementia lose cognition, a dyadic relationship develops between informal caregivers and care recipients that can have extensive influe nce. The stress that dementia caregivers experience related their role can have substantial, sometimes severe consequences. The resulting
72 perceived stress and associated emotional re sponses may influence outcomes not only for the caregivers, but within care recipients, or the dyad as a unit. While severity of BehSx in persons with dementia may result in eventual interpre tation of them as stressors for caregivers, and contribute to the caregivers stre ss process, caregivers may have, as a result of the intense process, altered interpersonal interactions with the person they care for. Persons with dementia are fully relational beings who can respond to interactions with the caregiver, and even when caregivers are vigilant about verb al responses in interactions, ne gative messages may be received from non-verbal mannerisms. It may not be re alistic to expect caregivers to alter their interactions and reduce BehSx wit hout first assisting them with their own stress, so assisting caregivers with stress/burden related to caregivin g may also improve the situation for the care recipient. Therefore, developing our knowledge and understanding regarding caregiver influence within the dyad, particularly with regard to PWD BehSx, may provide a basis for more appropriate caregiveror dyad-dire cted interventions to alleviate BehSx. If BehSx are alleviated, the entire caregiving milieu may be enhance d, and both dyad members may have substantial benefits. BehSx may have influence on serious, more distal outcomes such as institutionalization, and may also eventually translate into increased stress for the caregiver, resulting in an endless cycle that continually worsens the quality of th e caregiving milieu. For example, while caregiver factors such as perceived stress and depression are also factors, th e severity of BehSx have been cited in decisions regarding institutionalization placement (Buhr et al., 2006; Coehlo et al., 2007; de Vugt et al., 2005; Gaugler et al., 2005; Yaffe et al., 2002). Research to better understand antecedents to these BehSx may have influen ce for persons with dementia, caregivers, and society at large. Regarding BehSx as a dyadic consequence of the caregiv ers stress may provide
73 a unique insight into how BehSx might best be treated, with caregi veror dyad-directed interventions to alleviate caregiver stress, addi ng to the non-pharmacologic measures available to address BehSx. Such interventions might help to break the cycle, and ultimately improve quality of life for both caregiver s and care recipients.
74 CHAPTER 3 MATERIALS AND METHODS Introduction This study was a secondary analysis of data previously collected from the parent study, described herein as the origina l study. The disserta tion study was conceptualized early in the data collection phase of the original study, and changes were made to accommodate the dissertation study. Key components of the original studys materials and methods will first be summarized, and then the dissertation study plans will be outlined. Original Study Design The original study was designed as a random ized clinical trial to cr eate and test a new monitoring system for nighttime activity in person s with dementia (Rowe, PI, STTR, NINR, NIH 2R42NR004952-02A2). Data were collected over 12 months in a longitudi nal design, at baseline (month 0), and at months 2, 3, 4, 5, 6, 8, 10, & 12. The system is a newly developed technology that uses components similar to those of home security system s, with alarms on outer doors, motion sensors as needed within the home, a bed sensor, and bedsid e alerts so that the caregiver is aware when the person with dementia is out of bed and moving around the home at night. The overall aim of the original study was to test the efficacy of the monitoring system on proximal clinical outcomes such as improved caregiver sleep and reduced worry; and in the PWD, reduction of injuries and unattended exits from the home. Secondary outcomes in the caregivers were also assessed, such as sl eepiness/fatigue, perceived stre ss, depressive symptomatology, mood/affect, and desire to institut ionalize (Rowe et al., 2007).
75 Subjects Entry and exclusion criteria To be eligible, care recipients were required to have a m edical diagnosis of dementia, and needed to be cared for in the home without prov isions for professional care at night. The primary caregiver was required to speak English, and needed to have some concern about the person with dementia regarding their nighttime activity. Additionally, the caregiver could not have sleep conditions, medications, or cognitive/functional limita tions that would limit their ability to respond to system alerts (Rowe, 2003a). Sample and setting A convenien ce purposive sample of community-dwelling care dyads was recruited from three areas in central/north Florid a, and the primary setting of th e research was in the homes of the participants. Geographic areas were chosen to reflect a more dive rse population, and every attempt was made to seek minority participan ts, with over-sampling accomplished for AfricanAmericans, while Hispanics were more difficult to over-sample related to the inability of potential participants to sp eak English (Rowe, 2003a). The first 4 dyads recruited were used in the preliminary study phase to establish reliability of the newly designed system. These subjects were automatically recruited into the experimental group. Subsequent group assignment was random af ter verbal consent to participate, managed by a staff member not on the research team, for 45 of the dyads recruited. Two dyads were allowed to self-select into the control group beca use of difficulties recru iting in this population and their desire not to receive the system, a nd two voluntarily accepted assignment to the control group because their sleeping arrangements were inco mpatible with configuration of the system (Rowe et al., 2008).
76 Recruitment and retention Subjects for the original study were recruite d through advertisem ents in newspapers and flyers distributed at support groups and conferences. Resear chers contacted a variety of dementia-appropriate local clinics and organiza tions to explain the study, and those willing to recruit individuals for the study were provided with material p ackets and a means for potential subjects to call the researcher. It was up to the interested caregivers to contact the researcher, and no record was kept of those given informati onal packets. Advertisements also utilized email/web-based newsletters, the Alzheimers Association newsletter, and the IRB-01 web advertising, as allowed (Rowe, 2003a). On seve ral occasions, researchers presented the study to interested groups, and then inform ation packets were left with a contact person with in the group. In this difficult-to-recruit popul ation, retention of subjects was very important. Several measures were instituted in the original study to improve retention. For the experimental group, the caregiver was allowed the opportunity to keep the system at the end of the study, with limited access to technical assist ance after the research team completed their visits. For the control group, $15 gift certificates for local stores were gi ven out at each data collection point (total potential $135), and information pamphlets regard ing dementia caregiving were supplied at each visit. Alzheimers Association Safe Return prog ram fees were paid if registration was desired by the subject. Throughout the study, retention was encouraged through le tters, communications, prompt availability for problem-solving or to an swer questions, and birthday and holiday cards for subjects. Letters of appreciation were sent upon subjec ts leaving the study, and after completion of the entire study, cer tificates of appreciation were presented (Rowe, 2003a).
77 Procedures Methods In the origin al study, there were three investigators involved in the collection of data, each assigned to a geographic area intend ed to maximize diversity of the sample. At the outset of the study, intensive training was provide d regarding installation and tr ouble-shooting of the system. Each investigator had a research assistant assi gned to assist with scheduling and other tasks associated with the study. Regular monthly meetings were held to discuss any issues and progress of the study. The primary investigator was available to assist co-i nvestigators as needed with trouble-shooting. Computers were assigned to each investigator for the purpose of collecting data in an efficient manner. Assessment tools were ente red into a user-friendl y interface by a technical consultant, which allowed caregiver s to quickly complete tools usi ng an attached mouse. In the event that there were computer failures, invest igators carried paper copies that were then transported by hand and entered into the data storage program by the research assistant or investigator. In the field, data sets were sa ved to portable drives as well as to the laptop computer, and were e-mailed through secure server s to assistants for uploading into data storage sets at the earliest possible time. The data sets never included personal or identifying information. Data were stored in a Microsoft Acce ss data base until ready for analyses. Outliers in the data were checked for coding errors and validity prior to entry into SPSS for statistical analyses (Rowe, 2003a). Month 2 for those in the experimental group followed a minimum 2-week reliability phase that varied in length from par ticipant to participant; thus baseline and month two time points were not equidistant among participants. This ri gorous reliability testin g phase was necessary to manipulate sensors to suit the dyad, verify that the person with dementia did not exit the
78 critical exterior doors without an alarm sounding, verify proper system functioning, and confirm appropriate caregiver knowledge regarding operation of the system (Rowe, 2003a). Likewise, due to scheduling issues, subse quent months were not always exactly one month apart. Vulnerable human subjects considerations Persons with dem entia are considered a vulnerable population in ethical reviews. However, the primary impact of this study was on the caregivers. Intere sted caregivers, who contacted the research team for information, were given an explanation of the study and prescreened by phone for inclusion criteria (except cognitive status assessment). Those qualifying were visited in the home, where the caregivers MiniMental State Exam (MMSE) (Folstein et al., 1975) was conducted. One caregiver scored too low on the MMSE and was not recruited. Consents differed according to group due to the complexity of the experimental consent; thus recruited subjects were asked to sign informed consents after group assignment was determined. An adult caregiver, including spouse, adult chil d, or other relative, ga ve consent for him or herself, and gave proxy consent for the person with dementia. Persons with dementia were asked to give assent for participati on in the study (Rowe, 2003a). Approval was obtained from the University of Florida Health Science Center Institutional Review Board. Data collected regarding persons with dementia were primarily collected from the caregiver; thus the main impact for the more vulnerable persons with dementia was the completion of an MMSE. There were times wh en collecting the MMSE for the person with dementia was not possible, for example, when pe rsons with dementia were either non-verbal or seemed stressed by the researchers presence in the home. When it was determined that obtaining this measure of cognition was inapprop riate given the specific situation, it was not collected (Rowe, 2003a).
79 Participants were allowed to withdraw fr om the study at any time; however, no subjects withdrew for reasons other than medical illness, death, or change in caregiving circumstances such as institutional placement. One subject consented but did not have any data collected after family concerns arose regarding her participa tion. One subject moved without notice, and another was lost to follow-up for unknown reasons (Rowe et al., 2008). Data considerations Data for this study were stored and m aintained in a secure fashion. There were no personal identifiers such as name, address, etc. placed on questionnaires, forms, or data files from the field. Files for the study data are locked in a file cabinet in the office of the PI, and the key for ID numbers with personal information and informed consents remain in a separate locked file. Electronic data is stored in netw orked storage drives at the Univer sity of Florida Health Science Center, with only subject numbers as identifiers, accessible only w ith a password. Data from the original study will be maintained in this manner for a minimum of three years after completion of the study (Rowe, 2003a). The study was monitored by the Data Safety Monitor Board (DSMB), to observe for adverse events related to the system, and for pr eliminary review of da ta. Reviews scheduled after the reliability phase of the first four subj ects, after 10 homes reached the 3-month point, and after all treatment homes were in the trial for at least 3 months did not result in changes in the study. The PI of the original st udy generated a report at each point and alerted the DSMB that a review was needed. Although there were provisions for early clos ure (Rowe, 2003a), these were not necessary as interim analysis did not demons trate a worsening patter n in experimental group variables (caregiver sleep and injuries/exits in PWD).
80 Methodological Strengths of Original Study that Augment Dissertation Study According to experts, resear chers designing studies for community-based dem entia care dyads should consider the inextri cably intertwined nature of the caregiver/PWD dyad, and in intervention studies, should include positive effect s with clinical significance for both members of the dyad (Opie et al., 1999; Schulz, 2001; Zar it & Leitsch, 2001). The intervention in the original study had hypothesized clinical significance for both members of the dyad. More importantly to the dissertation study, the original study provides the dyadi c caregiving setting necessary for the dissertation purpose and aims. The model proposed for the dissertation will be most suitable for use in studying non-pharmacol ogic, dyadic interventions for those in the community, provided it is supported in the disser tation and in future research, and delineates relationships between caregiver and care receiver variables. The original study provides the milieu necessary for initial testing of this model. Another concern of experts is that dementia -related intervention studies should have more stringent designs to move the science forward (Ayalon et al., 2006; C. Beck, 1998; Burgener & Twigg, 2002a; Cohen-Mansfield, 2001b; Opie et al ., 1999). Several design issues strengthen the original study, and thus the data that will be used in the dissertation study. First, the experimental design and mostly random assignm ent to treatment status group strengthens the study considerably, reducing possibl e threats to validity. In deme ntia care dyads, however, the control condition, meant to have little or no impact on outcomes, is often problematic. Many caregivers respond positively to a program simply because they receive affirmation for their efforts, attention, and a sympathetic ear. C ontrol and experimental groups should be developed with these considerations in mind (Schulz, 2001; Zarit & Leitsch, 2001). In the original study, caregivers were not in contact with other through the study inte rvention, and caregivers in both groups were visited at each coll ection point, and thus provided equi valent doses of this natural
81 part of the intervention. Additionally, the longitudinal design a llows time for such a placebo effect to wane. The longitudinal design also lends strength to the datas abi lity to describe change over time, and facilitates the use of mo re powerful statistical analysis techniques such as multi-level modeling to describe both with inand between-person change (Lyons & Sayer, 2005; Zarit & Leitsch, 2001). This was particularly important for the dissertation study since data points in which there were no data available were easily accommodated using these techniques. Also, the capability of modeling both within-person and between-person changes over multiple waves in the study period will facilitate the ability to test the model (Fauth et al., 2006), since the relationships in the model were reflective of pe rson-level trends, and dyad-level relationships. Acknowledging relationships within dyads provided more relevant results to test the model than combining the data from all caregivers and PWD would have yielded. Lastly, in studies related to de mentia subjects, the degenerative nature of diseases that cause dementia may be problematic, since change s in outcomes related to cognition may obscure any changes in related measures over time. Dementia-related BehSx may be more stable, may be more resistant to intervention, and may require a sufficient period of assessment to adequately understand their change trajectory (Gitlin et al., 2003a; Mittelman et al., 2004a). Even modest improvements or lack of decline over time can ha ve a lasting impact within the dyad (Burns et al., 2003; Herrmann & Black, 2000; Zarit & Leitsc h, 2001). The multiple measurement points in the original study will allow for within-person trends to be pred icted, establishing change slopes over a year, and will also allow comparison by treatment group. While the intervention effectiveness for these outcomes is not the pr imary focus, demonstrating such change is
82 important for indirect support of the directionality of model relationships, as discussed in Chapter 1. In summary, the original study provides an ex cellent dataset for th e dissertation study. Change over time in the dissertation outcomes can be tested in relation to time-varying measures from the original study that will adequately reflect stress process concepts included in the model. The longitudinal design will allow more precise delineation of model pathways, with consideration of within-pers on (or dyad) change, and betw een-person (or dyad) changes according to group. The original study provides an appropriate community-based dyadic setting for the testing of the proposed model. Dissertation Study Dissertation Design The dissertation study is designed as a sec ondary analysis of da ta collected in the previously described original st udy. The dissertation study aim s pr imarily to test a conceptual model of caregivers stress proces s and its influence within the dyad, particularly as related to BehSx. Longitudinal data regardi ng variables that reflect the careg ivers stress process, collected in the original study, will be us ed in dissertation analyses. In addition, explicit design changes were made early in the original study to include a tool for measuring outcomes for the dissertation study, specifically, th e Neuropsychiatric Inventory-Ques tionnaire (NPI-Q) (Kaufer et al., 2000) for the measurement of PWD behavioral symptoms. NPI-Q data were collected at each data point in the original study specifically fo r use in the dissertation, and these data were used along with original study data to assess model relationships over time. Dissertation Subjects Entry and exclusion criteria in the d issertation study were the sa me as those utilized in the original study. However; of the 53 dyads, four subjects from the original study had only baseline
83 data and were excluded from the dissertation anal yses, which focused on change trajectories over time. Please see Figure 3-1 for the subject flow chart, and Tables 31 and 3-2 for complete demographic data on the sample caregivers and persons with dementia, respectively. These tables represent the entire sample except wher e indicated. Caregivers had a mean age of 62.81 and persons with dementia averaged 80.85 years. Over eighty percent of caregivers were female, and there were more male persons with dementia (54% male). Caregivers were more often wives (40%) or adult daughters ( 38%), with fewer husbands and sons in this role. There were proportionately more daughter caregi vers in the control status, but this difference did not reach significance. Caregivers were predominantly white (78%), with 18% African-American, and 4% Hispanic. With the exception of 2 dyads, caregivers and PWD within dyads had the same race. The caregivers most often had at least some college education. There were no significant differences in the demographic findings according to treatment status. MMSE (Folstein et al., 1975) scores for persons with deme ntia were obtained in about half the original sample, with no significant difference in mean scores according to treatment status. The average score on the MMSE was 13.83 (range 2-26), reflecting moderate le vels of dementia in the sample. Dissertation Procedures The dissertation study relies on the m ethods and rigor employed in the original study, and all procedures remained in compliance with IRB approval. Data for this study were stored and maintained under the auspices of the original study. IRB appr oval for changes to allow the addition of the NPI-Q for outcomes was obtained early in the data collection phase of the original study, and this dissertation study invest igator was added to the original study team. Subjects already enrolled were consented again after the IRB approval of the NPI-Q for the dissertation study, and were aware that the data would be used to assess additional outcomes separately from the original aims of the study. S ubsequently enrolled subjects were consented to
84 include outcome measures for the dissertation study (Rowe, 2003a). Although the NPI-Q was collected for re-consented subjects during subsequent data collect ion points, only 37 subjects had baseline data on the NPI-Q. Institutional Re view Board approval was obtained to utilize the original study data in a secondary analysis. The NPI-Q takes approximately 5 minutes to comp lete; thus it is anticipated that this did not deter persons from entering and completing th e original study, despite the somewhat intense battery of tools that were completed at each data point. Every effort was made to make this additional time commitment conveni ent for participants, incorporat ing it into the user-friendly computer interface and providing any assistance needed in completing the tool. Dissertation Measures The NeuroPsychiatric Inventory-Questionnaire (NPI-Q) (Kaufer et al., 2000) was chosen to m easure the dual outcomes of PWD behavioral symptoms and CG behavi or-related reactions, reflecting the theoretical concept of dyadic conse quences of the CGs stress process. The NPI-Q was recently developed as a more efficient, self-administered version of the original Neuropsychiatric Inventory (NPI). The tools use of the caregiver as informant, as well as its brevity (and thus less time burden for stressed caregivers), were the primary features that made the NPI-Q appealing for this study. However, th e NPI-Q is a more recent adaptation of the wellestablished, interview-based NPI; thus both will be discussed below. Neuropsychiatric Inventory The original NeuroPsychiatric Inventory (NPI) (Cumm ings, 1997; Cummings et al., 1994) was originally designed to help differentiate between types of dementia, and has been used extensively in medical research. The NPI is an informant-based, directed interview intended to assess a comprehensive set of ne uropsychiatric symptom domains that may have occurred in persons with dementia over the previous month, including: delusions, hallucinations, dysphoria,
85 anxiety, agitation/aggression, euphor ia, disinhibition, irritability/l ability, apathy, aberrant motor behavior, and more recently, ni ghttime behaviors and appetite/e ating disturbances (Cummings, 1997; Cummings et al., 1994; Lange et al., 2004). Participant caregivers rate any present behaviors with regards to severity on a 3-point scale (1-3), and frequency on 4-point scale (1-4). Care givers may then rate the distress they feel related to each positive symptom, from 0 (not distressing at all) to 5 (extremely distressing). Behavioral severity and frequency scores are mu ltiplied to obtain a maximum score for each type of behavior, and domains are totaled for a sum scor e. Distress scores are totaled for a separate score reflecting caregiver reaction to behaviors (Cummings et al., 1994). Reliability and validity have been established for the NPI (Cummings, 1997; Cummings et al., 1994; Forester & Oxman, 2003; Lange et al., 2004). Test-retest reliability for frequency/severity ranges from 0.51 to 1.0, and inte r-rater reliability has been demonstrated as high (93.6-99%). Internal consistency has been established for severity (alpha 0.87), frequency (alpha 0.88), and frequency/seve rity product (0.88). The NPI has shown content validity, and convergent validity with the Reisberg BEHAVE-AD (Behavioral Pathology in Alzheimers Disease Rating Scale) (Reisber g et al., 1987) and the Hamilton Rating Scale for Depression (Hamilton, 1967). Various researchers have esta blished differing factor structures among the behavior domains. The NPI is very well-accep ted, and has been cited over 250 times in the research literature (Cummings, 1997; Cummings et al., 1994; Forester & Oxman, 2003; Lange et al., 2004). The tool has demonstrated change in re lationship to treatment, and is one of the most commonly used measures in clinical trials. Clin ical significance has been established as a score equal to or greater than 4 on the behavior scale (Lyketsos, 2007).
86 Neuropsychiatric Inventory-Questionnaire. The NeuroPsychiatric Inventory-Questionna ire (NPI-Q) (Kaufer et al., 2000) was subsequently designed to allow for self-adm in istration by informal caregivers, with proxy assessment of behaviors in PWD, to improve the applicability of the NPI, using written instructions to direct the caregiver in rating it ems. Caregivers first answer yes/no regarding presence absence of behaviors in the domain, using guiding ques tions to improve understanding. If behavioral domains are endorsed, severity is ranked (from 1-3, total 36 ) and related caregiver distress may be ranked (from 1-5, total 60). Only se verity of BehSx is ranked in this tool to improve its brevity, one of its primary design cr iterions. The frequency is removed, with the rationale that severity more closely correlate s with caregiver distre ss, and having both is somewhat redundant since they are highly correlated (Forester & Oxman, 2003). The NPI-Q demonstrated convergent validity with the NPI using 60 dementia caregivers, as well as internal consistency and test-retest reliability. Correlations between the NPI total score and the NPI-Q total score was 0.91 for behaviors. Correlati ons were higher in the group of subjects with lower cognition. Prevalence of beha viors was slightly higher (5%) on the NPI-Q. For the NPI-Q domains, correlations ranged from 0.71 and 0.93 for behaviors and between 0.71 and 0.97 for caregiver distress. Test-retest correlations for beha viors and distress were 0.80 and 0.94, respectively (Kaufer et al., 2000 ). In this study, only the beha vior scale was utilized in the testing the model, and the Cronbachs alpha fo r the behavior scale in this study was 0.817. Measurements from Original Study The dyads treatm ent status group assignment in the original study will be used as a predictor in analyses in the dissertation study. Additional measures from the original study are described below.
87 Mini-Mental Status Examination. The Mini-Mental S tatus Examination (MMSE) (Folstein et al., 1975) was used as an inclusion criteria for caregivers, and was measured in care recipients to establish cognitive disability among PWD. The MMSE assesses orient ation, recall, attentio n, calculation, reading, language, and ability to follow commands, name obj ects, write a sentence, and copy an object. The maximum score is 30, and higher scores refl ect higher cognition. Reliability and validity were originally established in 206 patients w ith mental disorders and 63 normal patients. Normal subjects averaged scores of 27.6, while those with dementia av eraged 9.7; age did not contribute significantly to scores. Test-retest reliabil ity at 24 hours and 28 days were adequate, at .89 and .98, respectively. Two examiners ach ieved 24-hour retest re liability of .83. Concurrent validity was established with the W echsler Adult Intelligence Scale (WAIS), with Pearson r of .78 on verbal portions of the WAIS, and .66 on the performance portions. It was concluded that the exam distinguished those with cognitive impairment from normal subjects (Folstein et al., 1975). Some concerns have arisen regarding th e MMSEs potential edu cation bias, flooring effects and sensitivity for diagnosing and staging dementia. However, these concerns are not consistently supported, and the MMSE has been wide ly used and considered a reliable and valid instrument, quick to administer, for screening of cognitive status (Ashford et al., 1989; Harrell et al., 2000; Jones & Gallo, 2001; Mitrushina & Satz 1991; O'Connor et al., 1 989; Perneczky et al., 2006; Rapp et al., 2002; Schmand et al., 1995; To mbaugh & McIntyre, 1992). In this study, among the persons with dementia who completed the MMSE, the alpha was 0.821. Short version of the Zarit Burden Interview An abridged version of the Zarit Burden Inte rvie w (ZBI) (Zarit et al., 1985; Zarit et al., 1980) was selected to measure caregiving burden in the original study, and will be used to reflect
88 caregivers perceived stress in th e dissertation study. The original form of the Zarit (Zarit et al., 1980) tool included 29 items, but the 22-item revised version (Zarit et al., 1985) gained recognition and is the instrument most often used to measure perceived stress in dementia caregiving research. Considered valid and reliable, it has been used to demonstrate differences between groups of participants and to measure change over time (Bedard et al., 2001). Several versions of shortened interview scales have been proposed (Bedard et al., 2001; R. Hebert et al., 2000; Whitlatch et al., 1991), but the Bedard vers ion was designed to be more appropriate for longitudinal research (Bedard et al., 2001). The Short Zarit Bu rden Inventory-Bedard (ZBIBedard) was utilized in the original study. To validate the Short ZBI-Bedard, the data were analyzed from 413 caregivers at baseline, with follow-up at about 6 months. All care reci pients had cognitive impairment, and caregivers were primarily spouses (62.5%) and women (58.8%). The authors used factor analysis, change scores, and dementia diagnosis, along with item -total correlations, to reduce the 22-items down to twelve; the alpha coefficient was 0.89 in the final version. Co rrelations between the short and full versions ranged between 0.92 and 0.97. The aut hors suggested that a sc ore of 17 or higher may reflect high perceived stress, but cautioned that this can not be considered normative. The short version was confirmed as an adequate substitute for the longer version, appropriate for cross-sectional, longitudinal, and intervention studies. Questi ons regarding burden perceptions are answered on a scale from 0 (never) to 4 (nearly always), with a potential range of 0-48 (Bedard et al., 2001). The tool was further valid ated and found reliable in a population-based (Canadian) randomly selected sample, where a cu t-off score of 10 improve d sensitivity (to 75%) but lowered specificity (to 68%), leading authors to conclude that it was premature to assign a
89 cut-off score for too much burden (O'Rourke & Tuokko, 2003). In this study, the Cronbachs alpha was 0.889. Center for Epidemiologic Studies-Depression scale. The CES-D (Radloff, 1977) is a self-adm inis tered instrument that assesses subjects regarding the frequency of 20 depressive sympto ms during the past week, using a 4-point scale from rarely or none of the time to most of the time. Composite scores range from 0 to 60, with higher scores indicating more depressive symptoms. In the original tool, four items were worded to assess positive affect with the inte nt of breaking tendencies toward responding negatively, and were reverse-scored for contribu ting to the composite score (Radloff, 1977). This practice was later questioned, considered a violat ion of assessing the construct of negative affect using positive affect symptoms, with only the 16 negative item factor confirmed valid (Schroevers et al., 2000). In the original study, where data we re collected for use in this dissertation study, these four positive items were changed to reflect the opposite negative affect, to present a less confusing asse ssment to the stressed caregivers in this sample (Rowe et al., 2008). For example, feeling not as good replaced just as good as others, hopeless replaced hopeful unhappy replaced happy and not enjoying replaced enjoyed life. The validity and reliability of the CES-D is widely supported. The original tool was validated with large samples of normal adults from two communities in the US (average 9.25, n=2,514), and a patient sample (average 24.42, n=70) (Radloff, 1977). Valid ity was supported in a community sample of older adults (avera ge 9.1, n= 3,932) compared to those with schizophrenia (average 13.0, n=50), persons reco vered from depression (average 14.9, n=87 and those with acute depression (average 38.1, n=148) (Weissman et al., 1977). The CES-D has demonstrated criterion and constr uct validity; it has excellent internal consistency with alpha coefficients usually ranging from .85 to .91, moderate test -retest correlations, and has
90 demonstrated response to intervention (Beekman et al., 1997; R. G. Knight et al., 1997; Musil et al., 1998; Radloff, 1977; Rose-Rego et al., 1998; Weissman et al., 1977). A score of 16 or greater on the CES-D has been established to reflect significant symptomatology (Myers & Weissman, 1980; Radlo ff, 1977). This cut-off score resulted in 100% sensitivity and 88% specificity in one st udy (Beekman et al., 1997), and in 1711 older adults, scores greater than 16 dur ing a 9-months period was correlate d with greater use of health services and poorer perceived he alth (Callahan et al., 1994). Th e CES-D has been utilized in caregivers extensively to assess depressive symptoms (C. A. Cohen et al., 2002; Gallicchio et al., 2002; Hooker et al., 2000; Loewenstein et al., 2001; Rose-Rego et al., 1998; Schulz et al., 2004). For example, in dementia caregivers from a national caregiver rese arch initiative, 1229 caregivers were used to demonstrate associations between greater freque ncy and importance of religious practices and lower CE S-D scores (R. S. Hebert et al., 2007). In this study, the Cronbachs alpha was 0.923. Positive and Negative Affect Schedule. The Positive and Negative Affect Schedule (PANAS) ( Watson et al., 1988) was developed as a simple, brief self-report assessment of positive and negative affect, using items that were relatively pure markers of the m ood factor, with near zero load ings on the opposite factor. The 10 positive and 10 negative items on the tool are scored with a 5-point Likert scale ranging from very slightly or not at all to extreme. Various temporal de signations can be used, from now, this moment, to year, or in general (Wats on et al., 1988). In the original study reported herein, the number of items was reduced to 10, five each positive and ne gative, to decrease the burden of completing the tool in the sample of time-stressed dementia caregivers (Rowe, 2003a). The items chosen were those that had the greate st variability in a large sample of communitydwelling older adults (Diehl, 2005). In this di ssertation study, only the items reflecting response
91 to the stress process in a negativ e way were scored, including th e items distressed, scared, irritable, nervous, and jitte ry. The daily mean of thes e items was calculated and ranged from 0-5. These were measured each morni ng upon awakening over a one-week time period, and daily mean scores were averaged to give a mean negative affect score for the week at the measurement point if at leas t 3 days were recorded. The PANAS was originally tested in a sample of primarily college students, and found to have high internal consistency (alpha coeffi cients ranging from .86-.90 for various temporal designations), excellent convergent and discriminant validity with lengthier measures and with items compared to their own and opposite factor, c oncurrent validity, and re-test stability over a 2-month time period. When rated for the moment the items verified fluctuations in mood, while measures for year or in general dem onstrated trait-like stability; even moment-based measurements were found to be somewhat reflecti ve of ones general dispositional affect. The negative affect scale, but not the positive, was significantly related to perceived stress, and was stable across the 24-hour time period. The tool wa s comparatively valid and reliable in a smaller clinical sample (Watson et al., 1988) and in a large non-clinical samp le (n= 1,003) in the UK (Crawford & Henry, 2004). The negative and positive factor structure has been supported, (Crawford & Henry, 2004; Denollet & De Vries, 2006), and the negative affect portion of the scale has also been shown to split further into tw o factors, such as anxiety & anger, vs. guilt & shame (J. G. Beck et al., 2003), and upset vs. afraid (Killgore, 2000). In this study, the Cronbachs alpha for the reduced inst rument using only the negative 0.960. Composite negative mood score. In the dissertation study, a com posite Negativ e Mood score was established using scores from the CES-D and the Negative Affect scale fr om the PANAS. Z-scores were created from both measures based on the sample means, and thes e two Z-scores were averaged. The resulting
92 negative mood score was used in analyses regarding model relati onships to reflect caregivers emotional responses to stress. Statistical Analysis In this dissertation study, stress process vari ables and treatm ent status from the original study were examined in relation to BehSx. Systematically missing data for these outcomes is addressed, and preliminary analyses are described. Lastly, multile vel modeling was used to test the model presented herein. All data analyses used SPSS version 14, and SPSS syntax was used for multilevel analyses. This study used a significan ce level of < 0.05 in all statistical tests, with corrections as noted. Missing data Due to a co mputer glitch in the laptop-base d data collection inte rface, there was data missing in approximately 1-2 % of the potential NPIQ behavior items (9 data points; 53 dyads, 12 NPI-Q items (5724 total possible items); 81 values missing). This error occurred as selections were made by subjects on the electronic data co llection interface. When the caregiver answered yes to questions on the presence of 3 of the 12 items, irritability, m otor disturbance, and nighttime behaviors, they were mistakenly direct ed to the screen that assessed their distress, skipping the option to rate of the severity of be havior. This issue was addressed early in the study, but there remained some missingness even in later months from some study computers. These data were not missing at random, since an equipment failure caused errors in which data that were missing were related to observed data in the same assessment tool; this data could be considered to be systematically missing. Furthermore, although multi-level modeling using maximum likelihood can accommodate missingness acro ss data points such as is related to dropouts or skipped appointments (Fitzmaurice et al., 2004; Schafer, 2007), the missingness
93 described in this instance occurred within items that were used for sum scores collected at each data point, and thus could not be accommodate d by such sophisticated modeling methods. Conventional wisdom says that one could use one or several methods to calculate a score for these missing values: other subj ects answers for those items, that subjects answers at other months, and other answers on the scale for that su bject at that month; in this particular case caregiver ratings of distress related to the behavior could also be us ed to assist with imputation of the missing items. However, since the rate of missingness was low and ra nge of possibilities was so narrow, such complicated techniques were co nsidered inappropriate given the returns on the investment of effort. Simple averaging methods have been shown to perform adequately compared to more complex, state-of-the-art te chniques in similar situ ations(Schafer, 2007; Schafer & Graham, 2002), and were al so considered (see Appendix A). This missingness was ultimately remedied by calculating behavioral symptoms as a mean of the items available for that person at that data point In other words, if 9 items were answered, the total score was divided by 9, and then the mean was multiplied by twelve (# of items that should have been scored) to return the value to the more familiar scale. This adjusted value was identical to sum scores for those subj ects who answered all 12 items, but reflected what was known about the behavior s in persons where less than 12 items were answered. This method was used if at least 9 items on the scale were answered. This met hod was also utilized in instruments from the original study when an item was missing. Preliminary analyses Univariate and bivariate statistics were used to describe th e sample across all data points for each m odel variable. Means and standard de viations were calculated for the sample, and according to treatment status group. Histograms were examined for evidence of normality. Ttests were used to establish the outcome of group random assignment for model variables at
94 baseline. Pearson correlations were used to establish relationships between va riables at baseline. T-tests were also used to compare treatmen t status group means across data points, with Bonferroni corrections for multiple comparisons. Individual subjects outcomes were graphed over time and examined for evidence of non-linear patterns. Model testing The repeated m easures, the presence of mi ssing data, and the unequal time distances for data collection made multilevel modeling a natural fit for these analyses. A brief description of the analyses follows, and a more detailed acc ount of the estimation methods and analysis techniques may be found in Appendix B. Multilevel modeling techniques are very versatile, and models can be estimated in a number of ways. Of interest in this study was th eir ability to estimate vari able trajectories both within persons over time, and averaged across pers ons over time. In the data, the measurements over time (level one) were nested within persons and the persons were level two, with treatment status a person level factor that influenced measures within persons. Su ch longitudinal designs are considered a special case of hierarchical or multilevel modeling that addresses such clustering of data within persons rather than ignoring it by aggreg ating the persons data together (Cho, 2003). Days, indicating days from baseline, was a considered the c ovariate that reflected time. While measurement occasion or month of the study might also have been used to indicate time, these did not adequately reflect the study design, wherein measurements were not equidistant, particularly from baseline to m onth 2 in the experimental group compared to controls, and in the last six mont hs compared to the first six. Month of study was, however, used as the repeated term to provide structure for the covariances, and in gr aphical comparisons of modeled values.
95 Maximum likelihood was used to estimate models for each variable, in several stages: a) an unconditional means model without predictors, to establish variance for further modeling; b) an unconditional growth model with only time as a predictor, to establish whether there was enough variance within persons to model level 2 f actors; c) an unstructu red conditional model, with treatment status as a leve l 2 or person-level factor, to as sess for reduction in within-person variance; and lastly d) several conditional models with treatment status at level 2 and with the covariance structured. The covariance was structured by adding a random statement with month as the repeated variable and testing several covariance structures, including unstructured, antedependence, first order, compound symmetry, compound symmetry, heterogeneous autoregressive, first order auto-regressive, heterogeneous, toeplitz and toeplitz, heterogeneous. These structures were chosen based on their e xpected longitudinal vari ances and correlations across elements. The Akaike information criteri on (AIC) was used to compare the models with the default covariance model ( variance components or diagonal ); this criterion balances complexity with parsimony in indicating the best model fit. The covariance model with the lowest AIC was considered the fina l model for each of the variables. In order to evaluate the hypot heses related to relationships in the model, variables were created to reflect the caregiver s average and centered predictors. These variables were designated as fixed effects, and in the case of th e centered predictors, as random effects, to allow assessment of predictor influence on PWD BehSX within dyads and averaged across dyads (fixed effects) to reflect the populations effects. These analyses were also conducted using maximum likelihood estimation, and the covariance wa s structured using th e repeated statement and AIC as the criterion for ev aluating various structures.
96 Figure 3-1. Original Study s ubjects entry/exclusion from dissertation study analyses.
97 Table 3-1. Description of sample caregivers. t p p
98 Table 3-2. Description of sample persons with dementia. t p N 14 13 p Note
99 CHAPTER 4 FINDINGS The purpose of this study was to test a m ode l in which both CG Perceived Stress and CG Emotional Responses were hypothesized to have a direct positiv e relationship with PWD BehSx, and it was expected that when both variables were modeled as predictors, the effects of Perceived Stress would be reduced, in a manner si milar to mediation. A secondary purpose was to describe the temporal patterns of the model va riables, and evaluate th e influence of Treatment Status from the parent study on th e concepts trajectories over time. In this chapter, preliminary descriptive findings are presented, and study variables and relati onships are modeled over time, within and across dyads. The sample was pr eviously described in chapter 3, along with management of systematically missing data on the NPI-Q instrument (also discussed in detail in Appendix A). Appendix B contai ns detailed information about the multilevel analyses conducted in this study. Univariate and Bivariate Descriptives Sa mple means and standard deviations for m odel variables are displayed in Table 4-1. These are presented across each data point (months ) in the study, and then considering all data points available for all caregivers. The Short Zarit Burden Inventory (ZBI)-Bedar d was utilized as a measure of Perceived Stress. Caregivers scores, on average, ra nged from 19.09 to 21.86 on a scale of 0-48 across the data points. When all availabl e data points for all caregivers were considered, the mean was 20.74 (SD= 8.62) with a range of 0-47. The caregivers CES-D and Negative PANAS scal es had a significant moderate correlation (0.57; p= .000; N=49) at baseline and are conceptually similar. In order to reduce the total number of model variables, Z -scores were created using the gr and mean across all subject and
100 data points, and subjects Z -scores were averaged to form a composite Emotional Response measure. Caregivers emotiona l responses averaged between -0.086 to +0.069 across the study. When all data points for all caregive rs were considered, the mean was -0.002 (SD= 0.90), and caregivers Emotional Responses ranged from -1.25 to +2.86. The NPI-Q behavior scale was used to meas ure BehSx, and since it was added after the original study began, only thirty-f our dyads out of the total sample of 49 had baseline values. PWD BehSx scores, on average, ranged from 9. 04 to 11.58 across the study. When all available data points from all PWD were considered, the average was 10.55 (SD= 7.28), with a range from 0 to31. Bivariate relationships were described using t -tests. Baseline means for the treatment status groups were compared for significant diffe rences, for each of the model variables. The significance level was corrected due to the large number of comparisons (Bonferroni: there are 3 t -tests, 0.05 /3 = 0.017). After adjustments, there we re no significant differences at baseline for any of the model variables. T -tests comparing treatment status groups for each of the vari ables are also reported across all data points in Table 4-1. In general, differe nces between treatment status groups decreased from baseline to month 5, and then increased fr om month 6 to month 12. Finally, all available data points for all dyads were considered, and there was a significant difference in each of the model variables over the study period. These differences are simply aggregate comparisons, however, and do not reflect change over time. Observed scores for each of the model vari ables were compared over the study period according to Treatment Status, and the means ar e graphed in Figures 4-1 through 4-3. For Emotional Responses, the component measures, depressive symptoms (CES-D) and mood
101 (Negative PANAS) are also graphed over the st udy period for comparison, in Figures 4-4 and 45. Standard errors are also presented in the graphs. Control caregivers and persons with dementia had visually evident higher variability among scores th an experimental subjects over measurement points in the study, part icularly during the last half of the study. At month 5, in an unexplained inconsistent occurrence, the two gr oups had essentially th e same mean for PWD Behavioral Symptoms, with the control group symbol obscuring th at of the experimental group. It is noteworthy that regardless of treatment stat us, average BehSx remained above the score that reflects clinical significance (>4). Pearson correlations were calculated between th e three model variables at baseline and at 12 months (see Table 4-2). Perceived Stre ss and Emotional Responses had the highest correlation at baseline, and at the final data point, BehSx and Perceived Stress were most correlated. BehSx correlated more strongly to both predictors at Month 12 than at baseline. Correlations among model variab les across all potent ial data points were moderate and significant (see table 4-3), with the outcome BehS x more strongly correlat ed to Perceived Stress than to Emotional Responses. Lastly, distribution of the measurements was assessed over time. Distribution of Perceived Stress is adequately normal; how ever, both Emotional Responses and BehSx are slightly skewed to the right, with values bunched more densel y at the lower end of the range (median= -.2004 and 9.0, respectively). Histograms were examined a nd also reflected this skewness. However, the skewness statistics remained between 1 and -1 (Perceived Stress 0.23, SE= 0.13; Emotional Responses 0.75, SE=0.13; Behavioral Symptoms 0.59, SE=0.14), reflecting adequately normal distributions; kurtosis values were below 0.5 for each variable.
102 Change over Time in Model Variables Multilevel modeling was used to es timate separa te models of change over time for each of the model variables, as outlined in Aim #1. A brief description of the analyses follows. Please see Appendix B for more detailed information. Multilevel modeling was chosen to allow individual trajectories to be predicted at Level 1 (within pers ons), and to allow estimation by Treatment Status at Level 2 (across persons). Each variable was first assessed with an empty model, with no random effects. For each variab le, several models with both fixed and random effects were estimated. First, an unconditional means model determined whether there was enough variation at Level 1 to conduct further analyses (Model A). Next, an unconditional growth (i.e., change) model assessed the eff ect of time (Model B). Random effects were assessed for significant variance to model a Leve l 2 predictor. Within -person variance (random effects) and between-person popula tion effects (fixed effects) we re estimated with Treatment Status as the Level 2 predictor (Model C), a nd the covariance was structured by adding a repeated statement and testing alternative structures, which were chosen for their suitability to longitudinal data (Model D). All models estimated using the maximum likelihood procedures. Maximum likelihood estimation uses an iterative procedure to determine the estimates that represent the highest probability that the data came from the population, through examin ation of withinand betweenperson variability, and assigning sample means more weight when persons had high withinpersons variability compared to the sample, or were missing data points (Fitzmaurice et al., 2004; Singer & Willett, 2003). Variance components structure was used for random structure in initial modeling of each variable. Models were compared to the baseli ne unconditional models using change in the -2LL criterion. Additionally, the m odels ability to explain variance identified at both Level 2 (fixed effects; between -persons variance) and Level 1 (random effects;
103 within-persons variance) were considered in a ssessing models improvement. This reduction in variance through modeling is somewhat comparable to the use of R2 in traditional models. However, this so-called pseudoR2 is a bit different. This figu re represents the reduction of explainable variance in the models, so a fairly small ch ange can result in a large percentage of change (Singer, 1998). Lastly, cova riance structures were evaluated using fit statistic s that adjust for degrees of freedom automatically, the AIC and BIC. The AIC was determined to be the most appropriate criterion in this study (Fitzmaurice et al., 2004); howe ver, the two criteria often agreed, and most often did not improve upon the default ( variance components ). Multilevel models assume a linear trend, and deviations from this need to be addressed in modeling. Trends over time with in individuals were investig ated, and these preliminary assessments are addressed in Appendix C. The findings from these assessments supported the use of linear modeling. However, there was substa ntial intra-individual va riability in all three model variables, and the variation in direction and ma gnitude of change was such that withinperson modeling was warranted, and that it was prude nt to establish the fo rm of the trend over time within the modeling process. Therefore, change models were estimated with linear, quadratic, and cubic change (cubic not presented herein). Orthogonal quadratic change functions were computed to avoid multi-collinearity between the linear and quadratic terms, using the residuals from a regression of the quadratic term on the linear term. The residuals from this regression reflected the unique c ontribution of quadratic change, independent of linear change. Perceived Stress in Caregivers The m odel construction procedure for caregivers Perceived Stress is outlined in Table 4-4. The unconditional means model (A) revealed th at approximately 71% of the variance in Perceived Stress was unexplained. Intra-class correlations were calculated, and approximately 18% of the explainable variance was with in subjects (12.99 + 58.33=71.32; 12.99/71.32= 0.18),
104 with the large majority between subjects. In the unconditional change models (B), while the linear change model had improved criteria, the lin ear trend itself was nonsignificant. The form of the overall change trend most representative of the data was quadratic. The fixed quadratic trend was significant, and the random, within -person trend approached significance (shown separately in Table 4-7), indica ting that the general overall trend (across dyads) was quadratic. The quadratic change model resulted in an improvement over the unconditional means model, with a significant reduction in the -2LL, and with approximately 17% of the explainable withinpersons variance explained (betw een-person variance essentially unchanged). Using this model as a baseline for further comparisons, adding treatment status appr oached significance (p=0.08) in its improvement in the model (C), and the addi tion of month as a repeated variable and various covariance structures (D) did not converge or improve the m odel. Since the model with treatment added (C) approached significance in criteria reduction, and explained an additional 11% of the variance between persons, this mode l was accepted as the final model for Perceived Stress, indicating a better representation of the data than change alone. Parameters for the final model are described in Table 4-7. In the final model, the overall intercept at mid-study for Perceived Stress was 18.56 ( p=0.00), and the effect of treatment status was significant (p= 0.018), with control subj ects averaging 5.03 points hi gher than experimental subjects (at mid-study). The effect of linear time was not significant, but quadratic time was significant in the population ( p= 0.01); the quadratic estimate, however, was negligible, although it did approach significance within persons. Thes e findings indicate that, on average, subjects displayed a quadratic trend. The Treatment*Time interactions with bot h linear and quadratic time were not significant, indicati ng that both treatment status gr oups followed the same pattern of change over time. The random residuals, re flecting variation within persons, remained
105 significant (estimate 10.83, p=0.00). The intercept variance also reflected significant variability remaining between subjects (estimate 52.70, p= 0.00). Emotional Responses in Caregivers The m odel construction procedure for Emotiona l Responses is outlined in Table 4-5. The unconditional means model (A) reve aled that the unexplained varian ce was low (.82). Intra-class correlations revealed that 27% of the unexplai ned variance was within subjects (.22 +.60=.82; .22/ .82= .27) and 73% was between persons (.60/ .82= .73). In the unconditional change models (B), the form of the ove rall change trend was quadratic across subjects (seen separately in Table 4-7). The primarily quadratic change model resulted in an improvement over the unconditional means model and the linear cha nge model, with a si gnificantly improved 2LL, and with approximately 22% of the within-persons variance explained compared to 14% in the linear change model (between-person variance in creased 3%). Using the quadratic change model as a baseline, adding treatment to the mode l (C) did not result in significant improvement in the -2LL (decreased 1.932 with 2 degrees of freedom). Furthermore, there was only a 4% gain in between-persons variance explained, and the within-persons variance changed negligibly. Structuring the random effects (D ) by adding both month as a repeat ed variable and AR1 as the covariate structure improved the amount of betw een-persons variance explained to 8%, but the previously described within-persons varian ce explained (in the accepted change model) dissipated almost completely (decreased 21%). Furthermore, this covariance-structured model also lacked significance in the -2 LL change. Based on these analyses, it was concluded that the unconditional change model re presented the data best. The parameters for the final change model (unconditional change) are presented in Table 4-7. The overall intercept for Emotional Re sponses was 0.0002 and non-si gnificant at mid-study (p=0.999). Linear and quadratic change over time were negligible; however, the fixed quadratic
106 trend was significant ( p = 0.045), indicating that the genera l trend within the population was quadratic. Treatment and Treatment*Time interac tions were not included in the final Emotional Responses model because they failed to improve the model. The random residuals reflected significant within-p erson variance remaining (estimate 0.17, p= 0.00), and intercepts across persons were significa ntly variable (0.62, p= 0.00). Additionally, random linear and quadratic changes within persons were estimated. While th e change over time within persons was very small, linear change was significant (p= 0.018), and quadratic change was non-significant. Stated otherwise, the quadratic trend did not hold for all persons; some displayed linear trends more prominently that distinguished th em from the sample as a whole. Behavioral Symptoms in Persons with Dementia The m odel construction procedure for BehSx is outlined in Table 4-6. The unconditional means model (A) revealed that approximately half the variance in BehSx was unexplained. Intra-class correlations were calculated, and approximately 35 % of the unexplained variance was within subjects (17.29 + 32.27= 49.56; 17.29/49.56 = .35), with the re maining variance due to between-subjects effects. The form of the overall change trend wa s indiscernible in the sample when unconditional change models were run (B); thus the simplest form, linear, was utilized. The linear change model resulted in an improvement over the unconditional means model: a significant reduction in the -2LL, with approximately 2% of the within-persons variance explained, and approximately 3% reduction in between-person va riance. However, the fixed (averaged) linear trend itself was non-significant, and the random (within-person) effect over time needed to be removed to allow convergence of the model to occur. Random intercepts remained, allowing subjects to vary in intercepts, but they had si milar slopes over time. Adding treatment status to the model (C) provided furt her improvement in the model: the reduction in the -2LL was significant compared to the accepted change model, and an additional 7% of
107 between-persons as well as 1% of within-pers ons variance was explaine d. Providing structure for the random effects through adding a repeated measures statement and various covariance structures did not improve the model. These models either did not converge or increased the criteria, indicating poor fit compared to the default variance components structure. Parameters from the final model are descri bed in Table 4-7. The overall intercept (population mean) for BehSx was 9.084 at mid-study ( p= 0.00), and the effect of time was not significant. The effect of treatment status approached significance ( p = 0.056), with control subjects averaging 3.19 points hi gher than experimental subjects at mid-study. The linear time*treatment interaction also approached significance ( p= 0.075), with control subjects increasing approximately 0.06 per day from baseline ( p= 0.082), and experimental subjects decreasing -0.003 (NS). The random residuals reflected significant variance remaining within subjects measurements (estimate 16.61, p= 0.00), and the intercepts across persons were significantly variable (estimated variance 29.29, p= 0.00) at mid-study. The high variability remaining within and across persons supports the a ddition of either fixed or random predictors to explain the variance. Model Assumptions In this study, careful consideration of missingness due to dropout was important, since the outcome measure, BehSx, is known to contribute to institutional placemen t decisions, and thus dropout may have been related to current and futu re unobserved data. This generally means that data were missing not at random However, according to Singer & Willett (2003), missing data may be considered missing at random even though it is related to the measures that are missing, if those missing measures are expected to be st rongly correlated with va lues that have been previously obtained (p. 159). In this study, unobserved BehSx scores that may have been associated with dropout through institutional plac ement were expected to be correlated with
108 previously observed measures for the subj ect, supporting the pl ausibility of the missing at random assumption. Nonetheless, careful examinations of violations of the assumption of missingness were conducted in two ways. First, final models were run with dyads that completed the study and results were compared to those of the total sample. Second, completion of the study (vs. not completing) was used in comparative analyses to determine whether dropout was related to covariates or factors that should be considered in future modeling. Completion of the study was defined for these analyses as those dyads that remained in the study at month 12. Missed data points earlier in the study did not disqualify the dyad from designation as a completer. The final models for each variable were run using only those who completed the trial (67% of dyads ), and there were only minor differences. For example, in the BehSx model, the Time*Treatment interact ion moved to significant (despite fewer subjects/observations and supposed less power), a nd the treatment difference (intercept at midstudy) was not yet significant. Slightly more of the within-subject varian ce was explained, while less of the between-subject va riance was accounted for. Si milar findings resulted when Perceived Stress and Emotional Re sponse final models were run. Completion of the study vs. not completing was used as a factor in exploring timeinvariant sample descriptives from baseline for potential contributions to missingness, including: Treatment status Caregiver gender, race, relationship, working/not, education, and age PWD gender, race, diagnosis, age, and MMSE For example, this allowed determination regardin g whether variables (such as gender) influenced dropout later in the study. All chi-square and ANOVA tests were non-significant, indicating that these descriptive variables di d not influence whether dyads completed the study or not.
109 Additionally, within-person-averag ed perceived stress, emotional responses, and BehSx were not significantly different between those who completed the study and those who did not. Although it remains possible that drop-out relate d missingness is missing not at random these analyses lend support to the tenability of the missing at random assumption. Furthermore, selection models or missingness pattern models were considered unrealistic (no determinable patterns of covariate-related missingness). In this study, therefore, the drop-out related missingness was considered missing at random, or ignorable. In addition to addressing the missing data assumption, residuals from each of the model were assessed for meeting the assumptions of multilevel models as recommended by Singer & Willett (2003, pages 131-132). The probability plots of the residuals from the estimated final models against the observed estimates were insp ected, and displayed adequate normality. The model residuals were also assessed for equal variance at each data point (month), to support the homoscedasticity assumption. Visual inspections of the plots reveal similar range and variability at each data point. Hypothesized Relationships The relation ships hypothesized in aim #2 were also assessed using multilevel modeling. First, variables were created from each subject s data to reflect the ca regivers within-person variation and within-person m ean for both Perceived Stress and Emotional Responses. Specifically, within-person change was repres ented with centered variables, using each caregivers own mean to establish subjects devi ation or variation from their mean at a given data point, and these were modeled as both a random (within dyad) and a fixed effect (averaged across dyads for the population or fixed effect). The effect of centered variables within dyads was interpreted as the effect of the caregive rs current perceived stress level (compared to average) on BehSx in their care recipient with dementia. Across dyads, the effect of centered
110 caregiver variables was that, on average, when the populations caregiver stress variables were higher than usual, the BehSx in PWD were higher as well, with the opposite also occurring. This effect reflected how caregivers changing stress process might influence BehSx over time within the population. Additionally, within-person mean s were modeled as a fixed effect (means were averaged across dyads for the average population or fixed effect); each subject s contribution to the average was the same regardless of data point. The effect of mean caregiver variables was interpreted such that caregiv ers whose stress was, for exam ple, higher than the population average, had care recipients with dementia who also had more severe BehSx, across the study regardless of data point. Caregivers Perceived Stress and Emotional Resp onses were modeled as predictors in this manner separately first, and then in a comb ined model wherein Emotional Responses was expected to reduce the effects of Perceived Stress, similar to way a mediator reduces the effect of a predictor. This reduction was exp ected both within and across dyads. Separate Effects of Perceived Stress and Emotional Responses The initial step of assessing these relationships was to establish patte rns in BehSx for PWD within the dyads, which was also the first st ep in aim #1. Therefore, the model building represented in Table 4-8 begins with the accepted change mode l for BehSx from aim #1. Adding the caregivers Perceived Stress (model C) and Emotional Responses (model D) as both fixed effects (centered and mean) and random eff ects (centered) improve d the model criteria significantly, indicating that both variables are associated with BehSx (more on the interpretation in the next section). In the significantly improved Perceived Stress BehSx model, 33% of the betweenpersons variance and 13% of the within-persons variance was explained, that had previously
111 been unexplained in the BehSx change mode l. The significantly improved Emotional Responses BehSx model comparatively explained 27% of the between-pers ons and 6% of the within-persons variance previously unexplained. While not shown in the table both mean Perceived Stress and mean Emotional Responses were also modeled as moderators of the change over time in these relationships ; in both cases, the interactions were non-significant and the model criteria did not improve. Mediation Next, both variables were entered in a co m bined model with the Perceived Stress BehSx model as a comparison, in effect assessing whet her Emotional Response variables mediated the relationship between Perceived Stress and BehSx. The change in the -2LL from the Perceived Stress BehSx model was not signifi cant, and the model did not converge successfully. Thus, the combined model did not improve upon the Perceived Stress BehSx model. However, once the random effects of the centered variables were removed, model convergence was successful and the combined model was significantly improved in comparison to the Perceived Stress BehSx model. Thus, in this combined m odel, mean and centered Perceived Stress and Emotional Responses were fixed effects, and on ly the random intercept was modeled. Attempts to further improve the model (by providing st ructure to the random parameters) were not successful; model criteria did not improve. This model did explain an additional 4% of betweenpersons variance, but also explained 4% less of the within-persons variance compared to the Perceived Stress BehSx model, probably because the random effects included the intercept only. To assess for reduction of the effects of Perceived Stress in the combined model, the model parameters will be assessed.
112 Model Parameters The estim ated parameters for each of the models are shown in Table 4-9. Recall that time is centered at mid-study. Therefore, intercepts reflect the average number of days from baseline. Perceived stress When controlling for th e effects of caregiver s Perceived Stress (m ean and centered), the mid-study intercept of BehSx severity and the effects of time on BehSx were not significant in the population. When a caregivers mean Per ceived Stress is one poi nt higher than the population average, BehSx in the PWD increases by 0.42 ( p= 0.00). In addition, on months when caregivers current level of perceived stre ss, on average, are one point higher than usual, PWD levels of BehSx severity increase by 0.33 (p= 0.00). Within dyad s, the residual random effects of BehSx are highly variable (estimate 14.69, p= 0.00) when the caregivers Perceived Stress average and deviation from average are cons idered. Across dyads, ther e is also significant variability remaining among intercepts (estimate 21.14, p= 0.00), and the random within-dyad effect of the caregivers Perceived Stress compar ed to their average on BehSx was not significant at mid-study. In other words, at this particular point in the st udy, the individual caregivers level of Perceived Stress had not influenced the BehSx of the PWD in their care. Emotional responses The BehSx intercept at m id-study was 10.67 (SE = 0.73; p= 0.00) while controlling for caregivers Emotional Responses (mean and centered) The effect of time was non-significant. When mean Emotional Responses differ by 1 point, BehSx also differ by 3.57, and when caregivers emotional responses vary, on average, from their usual by one point, BehSx vary by 2.14. Though these effects may seem high compared to those of Perceived St ress, recall that the Emotional Responses variable was based on z-sc ores with a range cons trained between 1 and -1; thus these values reflect an approximate ma ximum change in BehSx based on caregivers
113 Emotional Responses. The residual random effe cts, or variability of BehSx within dyads, remains high (estimate 15.78, p= 0.00). Variability of BehSx intercepts across dyads is also significant (estimate 23.10, p= 0.00), and there is no significant within-dyad effect of current deviation of caregivers emotion on BehSx of the PWD they care for at mid-study. Mediation model In the com bined mediation model, mid-study intercepts and the effects of time are again not significant, and Mean Emotional Responses no longer have significant effects on BehSx. When caregivers Mean Percei ved Stress differs by one point, the BehSx vary by 0.32. Although this is slightly reduced from the Perceived Stress BehSx model, the non-significant effect in Mean Emotional Response negates the possibility of reduction of the e ffect by the presence of Mean Emotional Response. However, caregiver s current level of Emotional Responses, on average, did have a significant effect, and ma y have contributed to these changes. When caregivers measures were, on average, increas ed by one point, BehSx in PWD, on average, increased by 0.29 and 1.29 for Perceived Stress ( p= 0.00) and Emotional Responses ( p=0.02), respectively. Thus current leve ls of Emotional Responses may have reduced effects of current levels of Perceived Stress on BehSx. However, the effect was only slightly reduced. The BehSx intercepts resulting from this model remain highly variable across dyads (estimate 20.42, p= 0.00), and within-person va riability is also significant (estimate 15.13, p= 0.00). Summary of Findings in relation to Proposed Model The m odel was supported to some extent, as se en in the effects presented in Figures 4-6 through 4-9. BehSx and Perceived Stress models were improved by adding treatment, but Emotional Responses model was not. Only BehS x revealed a Treatment*Time interaction that was supportive of the model. Perceived Stress did not change over time in interaction with treatment status. To summarize, BehSx did app ear to diverge over time according to treatment
114 status. Yet BehSx may not have changed as a result of changes in caregivers stress process, since Perceived Stress and Emotional Responses did not diverge significantly according to Treatment status. Regarding the model relationships, Path C (a dding Perceived Stress as a predictor of BehSx) was significant with 33% betweenand 14% w ithin-dyad variance explained. Path B (adding Emotional Responses as a predictor of BehSx) was significant with 27% betweenand 6% within-dyad variance explained. While it was expected that when both predictors were modeled (combined model), the effects of Percei ved Stress would be reduced in the presence Emotional Responses, this model was only partia lly supportive of the hypotheses. Effects of Perceived Stress, on average, were slightly reduc ed, and caregivers current levels of Emotional Responses may have been responsible for the reduction of the reduction. However, the mean levels of Emotional Responses overall were not si gnificant in the combined model. Furthermore, the current levels of Emotional Responses were al so reduced from their previous model. These results are, therefore, inconclusive regard ing whether a mediation effect occurred. Nonetheless, there is strong evidence for the e ffect of caregivers st ress process variables on BehSx separately, and this provides support for th e underlying model premis e, that caregivers stress process may affect BehSx in PWD.
115Table 4-1. Means, standard deviations, and t -tests by treatment status for observed scores across 9 data point s for model variables. Month of study (M) BaseLine M2 M3 M4 M5 M6 M8 M10 M12 All Months Perceived Stress (Short Zarit Burden InterviewBedard) Mean (SD) 21.86 (8.83) 21.43 (7.70) 20.65 (8.97) 20.00 (8.15) 20.15 (9.04) 19.92 (8.88) 19.09 (8.61) 21.10 (9.55) 20.70 (10.48) 20.74 (8.62) Diff. t-test df (p) 5.14 2.11 47 (.04) 5.44 2.53 44 (.02) 5.16 2.06 46 (.05) 3.94 1.55 38 (.13) 4.32 1.52 37 (.11) 3.76 1.31 36 (.14) 5.60 1.91 30 (.07) 6.87 2.08 28 (.05) 7.40 1.92 25 (.07) 5.13 5.81 333.38 (0.000) Emotional Responses (CES-D & NegPANAS combined*) Mean (SD) .102 (.906) -.011 (.942)-.086 (.977) .058 (.908) .054 (.879)-.048 (.866)-.147 (.742) .069 (.932)-.021 (.935)-.002 (.90) Diff. t-test df (p) 0.34 1.33 47 (.19) 0.48 1.87 41.99 (.07) 0.27 0.98 47 (.33) 0.56 2.01 38 (.06) 0.36 1.31 38 (.20) 0.23 .82 37 (.42) 0.54 2.14 23.05 (.04) 0.52 1.84 29 (.08) 0.55 1.42 31 (.17) 0.415 4.51 354.47 (0.000) Behavioral Symptoms (NeuroPsychiatric InventoryQuestionnaireBehaviors) Mean (SD) 11.58 (7.12) 9.49 (7.17) 11.00 (7.53) 11.22 (7.14) 9.04 (6.80) 9.70 (7.00) 10.61 (6.98) 11.07 (7.95) 11.44 (8.44) 10.55 (7.28) Diff. t-test df (p) 2.92 1.18 32 (.25) 3.34 1.36 31 (.19) 4.60 2.12 43 (.04) 2.35 .98 34 (.33) 0.063 .03 34 (.98) 1.89 .82 35 (.42) 2.56 1.04 30 (.31) 2.89 1.53 27 (.14) 3.14 1.65 20.5 (.11) 3.01 3.72 293.98 (0.000) *com bined z-scores from Center for Epidemiologic Study-Depression and negative po rtion of the Positive & Negative Affect Schedule. Note: SD= standard deviation; Diff.= difference in means between control and experimental subjects.
116 Table 4-2. Pearson correlations between model va riables at Baseline (below the diagonal) and Month 12 (above the diagonal & underlined with dashes). r N N Perceived Stress Emotional Response Behavioral Symptoms Perceived Stress 1 .725**(n=30) .786**(n=27) Emotional Response .688**(n=49) 1 726**(n=27) Behavioral Symptoms .442** (n=34) .490** (n=34) 1 **Significant at the 0.01 level (2-tailed).
117 Table 4-3. Pearson correlations (and list-wise means) between model variables across all dyads and data points (list-wise N=308). r Perceived Stress Emotional Response Behavioral Symptoms Perceived Stress 1 .680**.527** Emotional Response -1 .475** Behavioral Symptoms --1 Mean (SD) 20.61 (8.91) -0.005 (.93) 10.49 (7.21) **Significant at the 0.01 level (2-tailed).
118Table 4-4. Model construction criteria for Perceived Stress Model -2LL' AIC BIC Residual (within) variance Intercept (between) variance -2LL' From Bas # P A R df ( p ) r2btw r2wth A) Unconditional means model 2053.47 2059.47 2071.04 12.99 58.33 3 B1) Unconditional linear change Random intercept + slopes 2042.96 2052.96 2072.20 12.65 58.45 10.51 5 2 (<.0005 ) -0.002 0.0261 B2) Unconditional quadratic orthotic chg. Random intercept + slopes 2022.58 2036.58 2063.52 10.79 58.85 30.89 (+20.38) 7 4 (<.0004 ) -0.009 0.1695 Based on above modeling, accepted quadratic orthotic change model with random intercept + slopes (model B2) as "baseline" for t reatment models. All unconditional models used variance components (VC) random structure. B2) Unconditional quadratic orthotic chg. Random intercept + slopes 2022.58 2036.58 2063.52 10.79 58.85 7 -0.009 0.1695 C) Conditional quadratic orth. chg. With Treatment (VC) 2015.80 2035.80 2074.30 10.83 52.70 6.78 10 3 ( 0.08 ) 0.1046 -0.004 D1) Conditional quadratic orth. chg. With Treatment, Month repeated 2009.41 2045.41 2114.70 By Month 53.63 13.17 18 11 (0.28) 0.0984 n/a D2) Conditional quadratic orth. chg. With treatment, month, AR1 2013.32 2035.32 2077.67 11.62 52.57 9.249 11 4 ( 0.06 ) 0.1103 -0.217 D3) Other structures for covariance did not converge or improve model fit (UN, ARH1, AD1, CS, CSH, TP, & TPH covariance struct ures) Based on above modeling, accepted the conditional quadratic orthotic change model with treatment status as the final model for change (C); approached significance and explained 11% of previously unexplained variance between persons (D1-3 did not converge or improve) Note: -2LL' = -2 Log Likelihood; AIC= Akaikes Information Criteria; BIC= Schwarzs Bayesian Criterion; -2LL = change in -2LL from previous baseline model; # PAR= number of parameters in model, used to establish df and p to determine whether model is a significant improvement over baseline model or last accepted model (using chi-square); r2 btw = between-subjects pseudo R-squared, estimate of the portion of variance explained ; r2 wth = withinsubjects pseudo R-squared, estimate of the portion of variance explained; VC= Variance Components (default); UN= unstructured; AR1= Auto-Regressive, First Order; ARH1= Auto-Regressive, Heterogeneous; AD1= Ante-D ependence, First Order; CS= Compound Symmetry; CSH= Compound Symm etry, Heterogeneous; TP= Toeplitz; TPH= Toeplitz, Heterogeneous.
119Table 4-5. Model construction cr iteria for Emotional Responses. Model -2LL' AIC BIC Residual (within) variance Intercept (between) variance -2LL From Bas # P A R df ( p ) r2 btw r2 wth A) Unconditional means model Baseline for change models 626.19 632.19 643.87 0.22 0.60 3 B1) Unconditional linear change Random intercept + slopes 604.98 614.98 634.37 0.19 0.62 21.21 5 2 (<.0001 ) -0.021 0.1458 B2) Unconditional quadratic orth. chg. Random intercept + slopes 595.57 609.57 636.71 0.17 0.62 30.63 (+9.41) 7 2 (.009) -0.034 0.2224 Based on above modeling, accepted quadratic orthotic change model with random intercept & slopes (model B2) as "baseline" for t reatment models. All unconditional models used variance components (VC) random structure. B2) Unconditional quadratic orth. chg. Random intercept + slopes 595.57 609.57 636.71 0.17 0.62 7 -0.034 0.2224 C) Conditional quadratic orth. chg. With Treatment (VC) 593.63 613.63 652.41 0.17 0.60 1.93 9 2 (0.381) 0.0379 0.0002 D1) Conditional quadratic orth. chg. With Treatment, Month repeated 584.99 620.99 690.79 By Month 0.62 10.58 18 11 (0.497) 0.0065 n/a D2) Conditional quadratic orth. chg. With treatment, month, AR1 588.50 610.50 653.15 0.21 0.58 7.07 11 4 (0.132) 0.0772 -0.214 D3) Other structures for covariance did not converge or improve model fit (UN, ARH1, AD1, CS, CSH, TP, & TPH covariance struc tures). Based on above modeling, accepted the unconditional quadratic orthotic change model (B2) as the final model for change (treatme nt did not improve model). Note: -2LL' = -2 Log Likelihood; AIC= Akaikes Information Criteria; BIC= Schwarzs Bayesian Criterion; -2LL = change in -2LL from previous baseline model; # PAR= number of parameters in model, used to establish df and p to determine whether model is a significant improvement over baseline model or last accepted model (using chi-square); r2 btw = between-subjects pseudo R-squared, estimate of the portion of variance explained ; r2 wth = within-subjects pseudo Rsquared, estimate of the portion of variance explained; VC= Variance Components (default); UN= unstructured; AR1= Auto-Regressi ve, First Order; ARH1= Auto-Regressive, Heterogeneous; AD1= Ante-Dependence, First Orde r; CS= Compound Symmetry; CSH= Compound Symmetry, Heterogeneous ; TP= Toeplitz; TPH= Toeplitz, Heterogeneous.
120Table 4-6. Model construction criteria for BehSx. Model -2LL' AIC BIC Residual (within) variance Intercept (between) variance -2LL From Bas # P A R df ( p ) r2 btw r2 wth A) Unconditional means 1879.41 1885.41 1896.61 17.29 32.27 3B1) Unconditional linear change Random intercept + slopes 1853.73 1863.73 1882.35 16.83 31.55 25.67 5 2 (<.0001 ) 0.0223 0.0264 B2) Unconditional linear change Random intercept only 1853.73 1861.73 1876.63 16.83 31.55 25.67 4 1 (<.0001 ) 0.0223 0.0264 B3) Unconditional quadratic change Random intercept + slopes 1850.90 1862.90 1885.24 16.45 31.18 28.50 (+2.83) 6 2 (.243) 0.0339 0.0487 B4) Unconditional quadratic orthotic chg. Random intercept + slopes 1850.80 1862.80 1885.14 16.38 31.35 28.60 (+2.93) 6 2 (.231) 0.0284 0.0527 Based on above modeling, accepted linear change model with random intercept only (model B2) as "baseline" for treatment models (B1 did not converge). All unconditional models used variance components (VC) random structure. B2) Unconditional linear change Random intercept only (VC) 1853.73 1861.73 1876.63 16.84 31.55 4 0.0223 0.0264 C) Conditional linear change with Treatment (VC) 1846.91 1858.91 1181.26 16.61 29.29 6.82 6 2 ( 0.033) 0.0715 0.0134 D) Providing Structure for covariance did not improve model fit (Month as repeated; UN, AR1, ARH1, AD1, CS, CSH, TP, & TPH s tructures) Based on above modeling, accepted the conditional linear change model with treatment status (C) as the final model for change. Note: -2LL' = -2 Log Likelihood; AIC= Akaikes Information Criteria; BIC= Schwarzs Bayesian Criterion; -2LL = change in -2LL from previous baseline model; # PAR= number of parameters in model, used to establish df and p to determine whether model is a significant improvement over baseline model or last accepted model (using chi-square); r2 btw = between-subjects pseudo R-squared, estimate of the portion of variance explained ; r2 wth = withinsubjects pseudo R-squared, estimate of the portion of variance explained; VC= Variance Components (default); UN= unstructured; AR1= Auto-Regressive, First Order; ARH1= Auto-Regressive, Heterogeneous; AD1= Ante-D ependence, First Order; CS= Compound Symmetry; CSH= Compound Symm etry, Heterogeneous; TP= Toeplitz; TPH= Toeplitz, Heterogeneous.
121Table 4-7. Final Model Parameters for each Variable as modeled over Time and by Treatment. Fixed Effects (Population) Estimates Type III sig. tests Estimates of Covariance Parameters Estimate SE df t p f p EstimateSE Wald Z P PERCEIVED STRESS Intercept 18.56 1.5448.6312.070.00395.670.00 Residual (within) 10.831.0510.330.00 Linear time 0 0 Redundant1.660.21 Quadratic time 0.00004 0.0026.961.670.118.520.01 Intercept (between) 52.7011.034.780.00 Treatment (con) 5.03 2.1249.322.380.025.650.02 Linear*Treatment (con) 0.005 0.0048.301.630.11 1.480.23 Linear time 0.000020.000.600.55 (exp) 0.0002 0.0028.880.060.96 Quadratic*Treat. (con) 0.00003 0.0043.380.780.44 0.610.44 Quadratic time 0.000.001.850.07 (exp) 0 0 Redundant EMOTIONAL RESPONSES Intercept 0.0002 0.1248.970.001.000.001.00 Residual (within) 0.170.0210.450.00 Linear time 0.00004 0.0036.960.130.900.020.90 Quadratic time 0.000005 2.2122.442.120.054.510.05 Intercept (between) 0.620.134.690.00 Linear time 0.0000030.002.360.02 Quadratic time 0.000.001.180.24 BEHAVIORAL SYMPTOMS Intercept 9.08 1.1950.237.660.00170.640.00 Residual (within) 16.611.4611.370.00 Linear time 0 0 Redundant0.570.45 Treatment (con) 3.19 1.6450.981.950.063.810.06 Intercept (between) 29.296.454.540.00 Linear*Treatment (con) 0.006 0.00271.361.750.08 3.200.08 (exp) -0.003 0.00271.49-0.760.45 Note: Bold= Significant; Bold Italics= a pproached significance; Intercepts reflect mean number of days from baseline.
122 Table 4-8. Model construction criteria for effects of Caregivers Stress Process Variables on PWD BehSx. Model -2LL' AIC BIC Residual (within) variance Intercept (between) variance -2LL' From Bas # P A R df ( p ) r2 btw r2 wth Based on modeling from Table 4-4, BehSx m odel with random intercept only (model B4) as "baseline" for further modeling of careg iver variables effects. B4) Unconditional linear change Random intercept only (VC) 1853.73 1861.73 1876.63 16.83 31.55 4 0.0223 0.0264 C) Conditional linear change with Perceived Stress (VC) 1810.87 1824.87 1850.94 14.69 21.14 42.86 7 3 (<.0001) 0.3298 0.1271 D) Conditional linear change with Emotional Responses (VC) 1826.85 1840.85 1866.91 15.78 23.10 26.89 7 3 (<.0001) 0.2678 0.0626 Based on above modeling, both Perceived Stress and Emotional Responses have influence on BehSx. Model C (Perceived StressBehSx) will serve as baseline for combined models to examine the relationships in the theoretical model. E1) Conditional linear change (VC) (centered vars. fixed & random) 1806.67 1826.67 1863.90 15.13 20.91 4.205 10 3 (0.24) 0.0344 -0.03 E2) Model E1 with random effects removed (centered fixed only) 1806.67 1822.67 1852.45 15.13 20.42 4.205 8 1 (0.04) 0.0344 -0.03 E3) Model E2 with month added as repeated variable 1804.35 1836.35 1895.93 By Month 20.77 6.518 (+2.313) 16 8 (0.9699) 0.0176 n/a E4) Model E3 with covariance structured using AR1 1806.49 1824.49 1858.00 15.30 20.27 4.381 (+0.176) 9 1 (0.6748) 0.0414 -0.041 E5) Other structures for covariance did not converge or improve model fit (UN, ARH1, AD1, CS, CSH, TP, & TPH covariance struc tures) Based on above modeling, the combin ed model with fixed effects only was accepted over Perceived StressBehSx alone (model E1 did not converge). Note: -2LL' = -2 Log Likelihood; AIC= Akaikes Information Criteria; BIC= Schwarzs Bayesian Criterion; -2LL = change in -2LL from previous baseline model; # PAR= number of parameters in model, used to establish df and p to determine whether model is a significant improvement over baseline model or last accepted model (using chi-square); r2 btw = between-subjects pseudo R-squared, estimate of the portion of variance explained ; r2 wth = within-subjects pseudo R-squared, estimate of the portion of variance explaine d; VC= Variance Components (defau lt); UN= unstructured; AR1= Auto -Regressive, First Order; ARH1= Auto-Regressive, Heterogeneous; AD1= Ante-Dependence, First Order; CS= Compound Symmetry; CSH= Compound Symmetry, Heterogeneous; TP= Toeplitz; TPH= Toeplitz, Heterogeneous.
123Table 4-9. Final Model Parameters es timated for BehSx as predicted by Per ceived Stress and Emotional Responses. Fixed Effects (Population) Estimates Type III sig. tests Estimates of Covariance Parameters EstimateSE df t p f p EstimateSE Wald Z p PERCEIVED STRESS ALONE Intercept 1.862.0148.480.930.360.860.36 Residual (within) 14.691.3810.620.00 Linear time 0.000.00277.030.170.870.030.87 Mean Perceived Stress 0.420.0948.674.670.0021.800.00 Intercept (between) 21.144.774.430.00 Centered Perceived Stress 0.330.0831.834.100.0016.790.00 Centered Stress 0.060.051.060.29 EMOTIONAL REPSONSES ALONE Intercept 10.670.7349.2914.630.00213.990.00 Residual (within) 15.781.4510.890.00 Linear time 0.000.00276.940.170.870.030.87 Mean Emotional Responses 3.570.9148.803.920.0015.380.00 Intercept (between) 184.108.40.2060.00 Centered Emotional Resp. 2.140.6118.193.540.0012.500.00 Centered Emotion 1.242.390.520.60 COMBINED MODEL with random intercept only Intercept 4.012.9048.131.380.171.910.17 Residual (within) 15.131.3311.360.00 Linear time 0.000.00277.740.060.950.000.95 Mean Perceived Stress 0.320.1348.712.360.025.570.02 Intercept (between) 20.424.654.390.00 Mean Emotional Responses 1.291.3049.500.990.330.980.33 Centered Perceived Stress 0.290.07259.663.930.0015.460.00 Centered Emotional Resp. 1.290.56261.182.290.025.250.02 Note: Intercepts reflect mean number days from baseline.
124 0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 40.00 45.00 Month of Study Zarit Burden InterviewBedard CON 24.2723.9223.1222.1722.3722.0022.0724.5324.54 EXP 19.1318.4817.9618.2318.0518.2416.4717.6717.14 02345681012 Figure 4-1. Caregivers Perceive d Stress as observed over Time by Treatment Status (means and standard error bars).
125 -1.00 -0.80 -0.60 -0.40 -0.20 0.00 0.20 0.40 0.60 0.80 1.00 Month of Study Combined Z-scores from CES-D and PANAS CON 0.210.190.090.290.220.070.180.470.12 EXP -0.08-0.26-0.23-0.19-0.13-0.15-0.39-0.22-0.26 02345681012 Figure 4-2. Caregivers Emotional Responses as observed over Time by Treatment Status (means and standard error bars).
126 0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 Month of Study NPI-Q Behavior Scale CON 12.7810.9313.6012.059.0911.4111.0413.2514.23 EXP 9.867.778.4410.119.018.839.418.788.90 02345681012 Figure 4-3. PWD Behavioral Symptoms as obser ved over Time by Treatment Status (means and standard error bars).
127 0.00 10.00 20.00 30.00 40.00 50.00 60.00 Month of Study CES-D CON 14.4115.2913.3914.1114.7112.6013.2517.2516.64 EXP 12.8410.8911.0810.5011.5611.769.6110.4410.59 02345681012 Figure 4-4. Caregivers Depre ssive Symptoms as observed ov er Time by Treatment Status (means and standard error bars).
128 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 Month of Study Average Negative PANAS CON 0.750.680.680.800.740.690.770.860.66 EXP 0.530.440.470.530.550.510.370.510.40 02345681012 Figure 4-5. Caregivers Negativ e Affect/Mood as observed over Time by Treatment Status (means and standard error bars).
129 Figure 4-6. Perceived Stress modeled over Time by Treatment Status; Demons trates no effect of Treatment over Time.
130 Figure 4-7. Emotional Responses modeled over Time ; Demonstrates no effect of Treatment over Time.
131 Figure 4-8. Behavioral Symptoms modeled over Time by Treat ment Status; Treatment improved model and interaction of Treatment with Time approached significance.
132 Figure 4-9. Behavioral Symptoms modeled as predicted by Perceived Stress (C -top) and Emotional Responses (B-bottom); Combined model (center) inconclusive regarding mediation of Perceived Stress effects by Emotional Responses.
133 CHAPTER 5 DISCUSSION AND IMPLICATIONS In this study, relationships between BehSx in PWD and care givers Perceived Stress and Em otional Responses were investigated to evalua te the tenability of newly developed model. These variables have been considered in previous studies with BehSx as the predictor; however, in this study BehSx were posited as the outcome Caregivers Perceived Stress and Emotional Responses were modeled as predictors for PW D BehSx. Additionally, these variables were modeled over time. While there were expected improvements in Perceived Stress and Emotional Responses from the parent study intervention, no impact on BehSx was expected. Therefore, any improvements in BehSx were more likely an outcome of changes that occurred in caregiver variables (Perceived Stress and Emotional Responses), lending s upport to the premise that PWD BehSx may be a response to care givers stress process. Support of Proposed Theoretical Model When PWD BehSx were estim ated over Time the trajectories modeled with Treatment status added were a significan t improvement over Time alone, and both Treatment (at mid-study) and Treatment*Time interactions (linear time) appr oached significance. Particularly after the mid-point of the study, PWD in the control group had increasingly worse BehSx, while those in the experimental group improved. Furthermore, the Caregivers Perceived Stress and Emotional Responses each diverged in their measurement acc ording to Treatment stat us after the mid-point in the study, although the Treatment*Time interac tion was non-significant for Perceived Stress and the Emotional Responses model with Treatmen t status did not have improved criteria (did not better represent the populat ions trajectories). The parent study intervention was expected to change the caregi vers stress process variables (Perceived Stress and Emotional Responses), but did not have components to address
134 BehSx. Therefore, these findings of divergence based on Treatment status partially support the underlying premise of the model, that caregivers stress process may negatively affect BehSx. Nonetheless, the lack of significant differences over time according to treatment status in caregiver variables is counter-intuitive when a ttempting to reconcile the models with such a premise. It is unclear how BehSx could ch ange significantly without corresponding significant changes in caregivers stress process variables. The models reflecting the relationships betw een caregivers stress process variables and BehSx also provide some limited evidence to suppor t the model. When caregiver stress process variables were modeled separately, and were, on average, higher than usual either as a group (centered variables averaged across dyads) or in comparison to the group (means considered across dyads), BehSx in the PWD also increased. Additionally, in the combined model with both Perceived Stress and Emotional Responses, there is some limited reduction in effects of average Perceived Stress on BehSx, with current levels of Emotional Responses on average, possibly responsible. It was hypothesized that Emotional Responses would ameliorate the influe nce of Perceived Stress on BehSx both within and across persons. The random, within-dyad effect s of current levels could not be determined because they needed to be removed to allow the model to converge. However, the effect of caregivers current levels Perceived Stress, on average, on BehSx was significant in the combined model, and slightly reduced. The mean Emotional Responses was no longer significant, indicating that average Emotional Re sponse did not influence BehSx. Support of the proposed model would have been improved by the opposite finding, with Emotional Responses remaining significant and Perceive d Stress losing both magnitude of effect and significance.
135 Findings in Relation to Past Research The findings from this study are primarily in agreement with the published literature. In the discussion that follows, previous research will be compared to findings in three areas. First, the general descriptive findings regarding model variables will be discussed. Second, the relationships investigated will be covered. Lastly the directionality of the relationships will be discussed, along with pot ential reasons for th is directionality. Model Variables As in previous studies, BehSx were a substantia l issue in the care dyads in this study. In both experim ental and control status partic ipants, BehSx, on average, remained over 4 throughout the study; this is cons idered the cut-off level for cl inical significance (Lyketsos, 2007). In previous research, multilevel modeling was used to establish that over a 3-month period caregivers stress and PWD BehSx were, on av erage, stable, but that there was significant intra-individual variability, indicating that not ev eryone displayed this shor t-term stability (Fauth et al., 2006). This studys subjec ts also displayed substantial in tra-individual variability, and time alone did not influence seve rity of BehSx in this study. Caregivers in this study, on average, also had consistently high scores on variables reflecting the stress process. The abridged version of the Zarit Burden Interview utilized to measure Perceived Stress had a scale 0-48, and prev ious studies have suggested either 10 (based on sensitivity vs. specificity) or 17 (based on uppe r quartile of sample) as potential cut-offs for excessive burden (Bedard et al., 2001; O' Rourke & Tuokko, 2003); neither of these are considered normative for the population without furthe r research. Nonetheless, caregivers in this study consistently averaged approximately 20, a bove these initial estimates of high Perceived Stress. Although there are no cut-off scores av ailable for the composite measure reflecting Emotional Responses, average depressive symp tom scores were high, and in some months,
136 approached 16, the cut-off for clinical depression risk (Radloff, 1977). In contrast, the mean scores on negative affect were, on average, under 1 on a scale of 0-5. These items were rated upon arising during a one-week period for each meas urement occasion, in contrast to the burden and depressive symptoms that were scored at the beginning of the measurement period during the day, often when caregiving duties for th e day were well under way. Despite these inconsistencies, the over-all picture in this sample is one of highly stressed caregivers. Relationships between Model Variables The rela tionship between the two caregiver variables, Pe rceived Stress and Emotional Responses to stress, has been very well-developed in the literature, and wa s not a concern in this study. This relationship is embedded within the Stress-Health model that was adapted for this study. The Stress-Health model emerged from previous well-accepted theory and a body of research, and has extensive support from reviews of the research litera ture (Goode et al., 1998; Pearlin et al., 1990; Pinquart & Sorensen, 2003a, 2003b, 2007; Schulz & Beach, 1999; Schulz & Martire, 2004; Schulz et al., 1997; Schulz et al., 1995; V italiano et al., 2003). Furthermore, each of the caregivers stress process variables explained approximately one third of the variance in BehSx that was unexplained with change over ti me alone. Therefore, this study confirmed a well-developed body of research th at establishes relati onships between variables representing the caregivers stress process and BehSx. While mo st of the reports and reviews discuss these relationships in the opposite di rection as hypothesized in this study, experts acknowledge that previous work has not establishe d directionality of the relations hips (Black & Almeida, 2004). The combined model did not fully support the hypothesis that Emotional Responses would act in a manner similar to a mediator. It is counter-intuitive that th is study did not fully substantiate the part of the adapted model that is most entrenched in previous theory and research findings. Possible explanations include the measures from the parent study that were utilized for
137 Perceived Stress and Emotional Responses. Differe nt measures may have allowed this model to more closely support the model. Perhaps simply using depressive symptoms alone rather than in combination with negative affect would have provided more evidence of reduction in the influence of Perceived Stress. However, the movement of the mean Emo tional Responses to non-significant in the combined model points to a different possibility. It may be that the directionality between these two caregiver stress process variables from this sa mple does not reflect a clear directional path as supported by the Stress-Health model. For ex ample, the Emotional Responses to stress and Perceived Stress may be more enmeshed together in the stress process, rather than the latter directly leading to the former. In this sample, the caregivers high levels of Perceived Stress and Emotional Responses reflect long-term stress rela ted to caregiving. It is plausible that the Emotional Reponses have themselves become a stressor or have decr eased the caregivers perception of their ability to manage or cope with the stressors, thus l eading to higher Perceived Stress. Directionality of Relationships between Caregiver Stress Process Variables and BehSx The findings in this study are in agreem ent with previous works that proposed caregiver influence on PWD outcomes. One review of the literature examined and highlighted caregivers influence on PWD outcomes, in particular, BehSx (Dunkin & Anderson-Hanley, 1998). In a study of over 5000 dementia care dyads in a population-based study, careg iver depression and burden, indicative of the stress process, were among predictors of increased BehSx (Sink et al., 2006). In a prospective study of 96 dementia care dyads, caregiver role st ress (negative attitude toward the care recipient) predicted worse social behaviors in PWD. Overal l, 32% of variance in care recipient quality of life (psychological well-being, produc tive behaviors, & functional ability) was explained by caregiver factors (Burgener & Twigg, 2002b).
138 While not a direct reflection of the relationships in this study, there is a small body of previous research that proposes that caregiver coping and manage ment strategies may influence BehSx and other outcomes in PWD. This incl udes studies regarding1) caregiver managements strategies and relations hip with BehSx (de Vugt et al., 2004 ), 2) caregivers cognitive decline and related increased BehSx, possibly linked to the inability to provide an optimum care environment (de Vugt et al., 2006), 3) caregiv ers training in management of BehSx and subsequent improvements in caregivers stress process variables and BehSx (Teri et al., 2005), and lastly, 4) caregiver coping st rategies and the relationship be tween ineffective coping and low survival rates in PWD (McClendon et al., 2004). It is reasonab le that coping and management strategies would indirectly resu lt in changes in BehSx associat ed with a more intense stress process. Several researchers have presented ev idence that might help to explain why caregivers who have a more intense stress process care for PW D who have more severe BehSx. For example, previous quality of relationship (i.e., how c ommunal or reciprocal the relationship was), predicted both depressive sympto ms in caregivers and frequency of harmful treatment of the care recipient (Williamson & Shaffer, 2001). Additiona lly, caregivers with higher strain and distress were also higher in expressed emotion, or crit icism toward the care recipient, and those more critical were more likely to attr ibute BehSx to willfulness in the PWD, or as controllable by the PWD (Tarrier et al., 2002). Anot her study showed that longitudina lly, caregivers high expressed emotion was predictive of increased negative behaviors over time (Vitaliano et al., 1993). Caregivers that are burdened, stra ined, distressed, or depressed ma y treat the care recipient in ways that increase stress or leave emotional n eeds unmet in PWD, which may in turn lead to increased BehSx (Algase et al., 1996; Hall & Buckwalter, 1987; Smith et al., 2004).
139 In summary, the findings in this study were in general alignment with findings in previous studies. Despite the different approach in the model tested herein, there is support from the literature for this approach. The finding in this study that is in contrast to the current science is the lack of a clear mediation or reduction in the influence of Perceived Stress on BehSx when Emotional Responses is modeled simultaneously. While current levels of Emotional Responses may have had some effect as expected, mean Emotional Responses become non-significant, calling the reason for the slight reduction of Perceived Stress BehSx effects into question. It is not clear why this finding occurred in the current sample, and these relationships bear investigation in further analyses. Strengths, Limitations, and Directions for Future Research The use of multilev el modeling allowed change w ithin dyads to be considered in the study analyses. These analyses provided more precise estimates than would be seen in traditional analyses. For example, population estimates al low for individual traj ectories to borrow strength from the population averages when there is higher occasion-to-occasion variability within the persons measurements than exists in the sample. Furthermore, missing data points are accounted for efficiently. This reduces meas urement error and adds precision to estimates. This emphasis on precision over unbiasedness is the primary motivation to utilize multilevel modeling to estimate change trajectories (Schafer, 2007; Singer & Willett, 2003). While modeling likely strengthened estimates in this study, it is important to recall that estimates may be biased toward the sample, partic ularly if there is substantial missing data. Of concern is the missingness due to attrition (70% completed study) or missed data collection points in this study; power and precision, particularly in later measurement points, may have been compromised (Schafer, 2007). Since results in this study pointed to more divergence in groups at the end of the study, fu ture research might prudently use a piecewise approach to
140 modeling, allowing the traj ectories to change after month 6. As previously discussed, the systematically missing data may also have influenced analyses regarding the primary outcome in this study, BehSx; however, attempts were made to reduce that possibility. One limitation of this study is its secondary nature, and the lack of control over data collection. However, this study was conceptualized shortly after the beginning of data collection in the original study, and specific data were coll ected with the intent of this study in mind. Measures collected in the origin al study can arguably be considered appropriate measures for stress process variables consid ered in the dissertation study. However, it is not known whether these are the best measures for these concepts. This issu e may need to be clarified in future research. Additionally, a convenience purposive samp le was used rather than random selection. Descriptive statistics will allow for readers to make conclusions about the degree to which findings are generalizable, since it is not realistic to collect a population-based sample for a study such as the one designed here. A concern regarding the data available from the parent study was that MMSE measures were the only measure of cognition in persons with dementia, and that this measure was not repeated at collection points; onl y one score was obtained, within th e first 2 months of the study. Furthermore, there is a high rate of missingness of MMSE data for persons with dementia, with approximately half of dementia subjects comple ting an MMSE. In the subjects who did have this data collected, there were no significant differences according to treatment status. Nonetheless, it would be valuable to have repeated measures of this variable, at least three times during the study period, for better co ntrol of dementia severity in within-individua l analyses over time.
141 Some note concern in the litera ture about using caregiver repo rt of behaviors in the person with dementia, rather than direct observation. Ot hers acknowledge that ther e is intuitive logic in using caregivers as raters, noting that caregive rs know better than anyone else what behaviors are exhibited, and can capture a wider time frame than direct observation would allow (Davis et al., 1997). The Cohen-Mansfield Agitation Inve ntory, a common caregiver-report measure of BehSx, was significantly correlated to direct obs ervation of BehSx, using informants who were formal caregivers who likely had heavy patient loads (Cohen-Mansfield & Libin, 2004). These findings support the appropriatene ss of using caregiver report when direct observation is not a realistic option, perhaps particul arly appropriate in community-b ased dyads where the caregiver and person with dementia are in close proximity much of the time and caregivers are intimately aware of BehSx patterns. Additionally, in rese arching dementia care dyads in the community, the use of direct observation could be perceived as intrusive and disrup tive to PWD, and the study of BehSx in the natural setting is key to th e model being tested. It would be difficult to maintain such a natural environmental setting using direct observation or videotaping, with researcher presence. As tec hnology improves, careful considerat ion of the most appropriate methods of measurement are in order, with c onsideration of subjects privacy in decisions regarding implanted videotaping vs. usi ng caregiver report instruments. Regarding the analyses, weaknesses included the lack of covariates in these basic initial assessments of the model. For example, base d on previous research, gender, relationship, and race could have possibly served as factors in the changes and relationships. In this study, these were equal in the two groups, and were controlled for in that manner. Furthermore, sample descriptives were assessed fo r correlation with comp letion of the study. Nonetheless, further analyses are in order to assess their impact more directly. An example of a potential covariate
142 that was not available in the orig inal study data is that of the qu ality of the relationship prior to onset of the care dyad relationship. Covariates such as these are easily included in modeling and should be considered in future research. Lastly, the causal pathways were assessed with data that was collected simultaneously, and the proposed relationships are imbedded within a dyadic environment that is influential for both members of the dyad. For example, BehSx might have caused changes in Perceived Stress and Emotional Responses as well as vice versa. While the modeling of mean and centered predictors and change over time in variables theoretically supported direction of causality, analyses which more specifically assess the direct ionality of the relationships, such as multivariate mediation (simultaneous assessment of model pathways within dyads) and lagged comparisons to determine which variables change first, are in order in future studies. Summary This study has provided initial confirm ation of the proposed model, and enough findings that meet expectations in order to move forward with further test ing. It is reasonable that BehSx act as a stressor among many in caregivers stress pr ocess, particularly when the upset, bother, or distress they cause leads to their perception as a pr oblem. However, it is al so plausible that they may emerge as an outcome of that process, when caregivers Perceived Stress and Emotional Responses impact their interactions and ability to provide an optimal care environment. Further research to evaluate this model may lead to improved intervention st rategies that target caregivers stress process but also improve BehSx, interrupti ng this cycle and positively impacting more distal outcomes for both members of these vulnerable dyads Such research has the potential to improve the quality of life of millions of community-based dementia care dyads, reducing the social, emotional, and fi nancial costs of this condition.
143 APPENDIX A MISSING DATA Systematically Missing Items in the NPI-Q This study had system atically missing data in the NPI-Q behavior scale items related to a computer glitch. Several methods were considered to address this missingness. First, the following were investigated as corrective options. Mean Substitution. Simple averaging methods were considered, often referred to as mean substitution. All other subject s answers for that item during that data collection point were averaged together and the resulting mean was imputed where systematically missing items occurred. This resulted in values that were representa tive of the sample. Substitution of the Mean of Po tential Item Scores. Another simple method involved consideration of the potential scor es had the item been answered. In this case, the possible severity scores ranged from 1-3, since 0 was not an option (in the firs t question, caregivers were asked whether the behavior domain occurre d, with 0=no and 1=yes, but in this item, severity was rated only if the behavior occurr ed). The mean of this range is 2, indicating moderate severity. This resulted in values th at were indicative of the potential range in the population The sensitivity of the mean substitution method was evaluated by comparing it to the method of using the mean of Potential Item Scores The average of the im puted values for mean substitution was 1.84, calculated by summing all imputed values and dividing by the total number of imputations. This would provide similar results to simply using the average potential score. Using either of these methods could have inflated (or deflated) the sum score by as much as 3 points, however. In this study, the interest was in change within subjects, as well as acr oss subjects to yield population estimates. Sums using these imputed values could have distorted or muted those effects over time. Therefore, these methods of imputation of data from sources other than the subject ultimately seemed unrealistic. To addr ess this issue, methods of substitution were considered that utilized the pe rsons own data, such as:
144 Last value carried forward. This method of us ing the value from the previous occasion for that person was unrealistic for 2 reasons. First, the beha vior domain may not have occurred previously and the attempt was to re place the item, not the sum score. Secondly, since this was a study of change, the lack of variability from one occasion to the next would be problematic. Complex computations from related information. In this study, there was information available at that time point for that individual in the form of the distress caused by the behavior. Complex computation of that caregi vers usual ranking of the severity of the behavior in relation to the distress it caused could have been undertaken. Multiple imputation. Use of multiple item resp onse sets in an iterative comparative manner could have resulted in converging on the best possible estimation of the missing response using both the subjects and the samples data. This method is superior to those previously considered in pr ecision of the imputed value, but is effort-intense and technically complex. According to an expert on missing data, such intensive methods may not be necessary when the range of the missing item is narrow, an d the volume of missingness is low. The effort and time needed to accomplish such methods is counter-productive in such simple cases (Schafer, 2007). Solution for Systematically Missing Data in this Study Based on these considerations, a sim ple met hod that capitalized on the subjects own data was utilized. When items were missing, the items that were answered for the subject, for that data point, were averaged together, and then mu ltiplied by 12 to return the summary score to the familiar scale. This method utilized what was known about the subjects BehSx severity, without imputing based on data from other sources. Similar techniques to the one chosen for the NP I-Q missing data were used in the original study for other variables representing the model. The CES-D was averaged if at least 15 of the 20 items were scored, and then returned to scale. The Short ZBI-Bedard was scored as a mean if 9 items of the 12 were scored, and was multiplied by 12 for use in this study to reflect the
145 familiar scale. The Negative PANAS was scored each morning during a 1-week data collection period, and means were reported if at least 3 values were present. This simple method provided the most realistic option for the systematically missing data in this sample. In addition to the systematically missing data within the NPI-Q instrument, there were missing data points for approximately one third of the subjects due to missed appointments or drop-out. These have been considered missing at random (see Chapter 4) and were addressed through the use of multilevel modeling to accommodate such missingness.
146 APPENDIX B STATISTICAL ANALYSES Multilevel Analyses Preparation of the Data Prior to estim ation, the longitudinal data were prepared by restructuring the data to a univariate format. This meant that each subjec t had one entry (case) for each data point from which any data were available, sometimes referre d to a long or stacked rather than wide data structure. This allowed the computer program to establish parame ters over time based on the data point connected to each case. In multilevel models for change, a sensible tim e metric is chosen based on the study design and hypotheses. This allows for data that is not precisely time-structured to be included in analyses, with variable timing and spacing of waves easily accommodated. Additionally, while balanced data with equal numbers of measurements may be appealing, this occurrence is usually not realistic in dementia caregiv ing research. Failure to obtain precisely balanced data does not limit the multilevel analyses, wh ich can accommodate missingness within individuals waves of data. In this study, several considerations were ma de to accommodate the data collection scheme in the parent study. While there were 9 data collection occasions over 12 months, the spacing was not time-structured. The experimental and control group differed in length between Baseline and Month 2 due to a reliability pha se in subjects who received the system. Additionally, while data was collected each mont h during the first half of the study, spacing was every two months in the last half. In order to better reflect data collect ion patterns in individual dyads, days from baseline was utilized as the time metric in analyses. Four subjects from the parent study were eliminated; 1 had no outcome data, and three had only one data point.
147 Subjects who had between 2 and 9 data points colle cted were included in analyses. Lastly, in restructuring the data, on those occasions when time-varying variables that represented model concepts were not collected, the occasion was e liminated. This allowed estimation of population parameters based on all available data, without pr ediction of those occasions that were missing. Since only 67 % of subject completed the study, th is allowed 33% of subj ects to be included in analyses of change with partial data. In this study, covariates were not modeled for simplicitys sake; the random assignment with equal groups at baseline for potential c ovariates was appealed to for reducing the potential for spurious findings. Estimation Techniques Maxim um likelihood estimation was used in an alyses to add precision to estimation of trajectories, determining the maximum likeli hood that the samples data came from the population. Random or within -person trajectories were estimate d, as well as fixed, or populationbased trajectories. Such model estimates are created by balancing the individuals variability over time against the groups variab ility. If within-subject variability is greater than that of the group, more weight is given to the average, and if between-subject variability is greater than the individuals, the persons values are more heav ily weighted. The number of data points present for each individual is also considered in weightin g the influence of the subjects data, with fewer observations resulting in more shrinkage toward the mean for that indivi duals trajectory and lessened contribution to the popul ation trajectory (Fitzmaurice et al., 2004; Singer & Willett, 2003). According to Singer & Willett (2003), thes e techniques make model-based estimates considerably more precise than observed data because measurement error is reduced, and thus estimates more closely mirror the true populatio n characteristics as compared to sample averages. However, the estimates may be biased toward the sample if there is a large amount of
148 missing data or a small sample. Restricted maximum likelihood is an alternative method for estimating multi-level models that offers less bi as and handles high correlation more effectively, but is only useful for evaluating random eff ects (Fitzmaurice et al., 2004; Garson, 2007; Singer & Willett, 2003). In this study, the fixed effects were also of interest, so the maximum likelihood method was the best option. The modele d fixed effects are considered estimated, and random effects are said to be p redicted; the latter can be used to graph individuals and also average trajectories. In this study, 67% of the subjects completed th e study. The dropout of 33% of subjects as the study progressed means that model-based estimates toward the end of the study may be less precise than those during the first pa rt of the study. Modeling Nested Data The centr al feature of multilevel models is th at the lower level units are clustered or nested within upper level units and the covariance structure am ong level one units is modeled to fit the data, rather than requiring that data me et assumptions about the structure. Longitudinal or repeated measures data repres ent a clustering of measurements w ithin subjects. When data is collected in this manner, use of statistical techniques that do not accommodate the design may lead to erroneous inferences. For example, in traditional analyses, intra-individual variability is considered measurement error or noise, and incomplete data may exclude the subject from analyses; individual change trajec tories are not assessed. In c ontrast, multilevel models allow intra-individual change, as well as deviation from ones own change trajectory to be included in analyses. Additionally, the ability of multile vel modeling to accommodate unbalanced (timeunstructured or incomplete) data improves precision of estimates at both levels of data collection (Fitzmaurice et al., 2004; Si nger & Willett, 2003). Assumptions underlying multi-level modeling in clude a) random groupings (assumed to be a random sample of all such groupsin this study, dyads were the grouping units); b) non-
149 independence of lower lever units (measures with in persons are correlated); c) independent blocks (subjects-or persons in this study) that are assumed to have similar covariance, but are independent of each other; d) properly specifie d covariance structure; e) similar group sizes (estimation handles unbalanced designs, but gr ossly unbalanced data due to missingness may inflate Type I errors); f) estimates that ar e produced with asymptotic (approximate) normal distribution in larger samples; and g) adequate sample size. The number of subjects that constitutes an adequate sample size is contr oversial, with estimates ranging from 20-150. An important issue is that the numbe r of observations per unit of th e sample increases the overall power, and power is also determined at each level according to the number of units at that level. Severely unbalanced data at either level may comp romise power and require a larger sample size (Garson, 2007). Modeling Change over Time Variables representing each concep t in the theo retical model were estimated in a multilevel model, including CG Perceived Stress, CG Emotional Responses, and PWD Behavioral Symptoms. The Short ZBI-Bedard represented Pe rceived Stress, the composite of Depressive Symptoms and Negative Mood scores represen ted Emotional Responses, and the NPI-Q was used to measure PWD Behavioral Symptoms. Random coefficient models were used to asse ss within-person and between person change over time. In a multilevel model for change, level 1 analyses evaluate the time-varying parameters over the course of the study within i ndividuals. Level 2 analys es utilize predictors posited at the dyad/individual level that are not time-varying; to estimate differences in averaged change trajectories. Level one units are consid ered non-independent or correlated within level grouping units (grouping units follow the command subject = in SPSS). Level two categorical predictors are called f actors (factors follow the command by in SPSS), and interval or higher
150 predictors are considered cova riates (covariates follow the co mmand with in SPSS). Level two factors can be used to estimate average trajectories across levels of the factor (Singer, 1998). In this study, each variable was modeled with Days and by Treatment Status. Treatment Status by Days interaction was also included to dete rmine if trajectories differed over time by treatment status. Preliminary models were used to prepare fo r model fit analyses. The outcome was first modeled without any predictors. This so-calle d unconditional means model established the samples estimated mean, across all persons and data points, and allowe d intra-class correlation to be calculated, to determine how much of the total variance was at each level, within persons vs. across persons, to substantiate further an alyses. An unconditional growth model, with persons as the subject (or blocking variable for covariance matrix), and with the default covariance structure, was estimated to determine the effect of time within the sample. Across persons (fixed effects), the inte rcept reflected the average outcome value at time 0, and the time coefficient (Days) reflected the average change in the outcome for each unit of time. Within persons (random effects), inter cepts and slopes were assessed fo r significant variance to model with a person-level covariate (Singer, 1998). Comparisons over Time by Treatment Status In this study, Treatm ent Status was consider ed a person-level (lev el 2) factor with influence on slopes over time within persons, and was thus modeled as a fixed effect. Results included the following fixed effects (or eff ects across persons for population estimates): Averaged intercepts and slopes (time) while controlling for Treatment Status Treatment Status (impact on averaged initia l status, at time 0, or intercept) Treatment Status in interaction with time (f or each unit of time, averaged slope changed X amount)
151 In this study, linear time was centered at mid-study (average # of days from baseline) in order to separate out the quadratic effects. Quadratic change (centered linear change*centered linear change) was regressed on linear change (cen tered), and the residuals from the regression represented orthogonal quadratic change, or the qu adratic effect that was independent of linear change. Therefore, references to time 0 or inte rcept in this study repres ent the mid-point of the study. In addition to fixed effects, random (with in-person) estimates for intercepts and slopes were assessed for reduction in variance and if this occurred, the percentage of reduction was calculated (Singer, 1998). Predic ted trajectories were displayed ove r time in a graphical format according to treatment status. Modeling the Covariance Once the fin al was selected, the random erro rs were structures to allow appropriate conclusions from the results. Unless the covari ances are modeled, erroneous inferences may be made. In longitudinal data, the structure must c onsider within-person corr elation over time (all of subject As measures are correlated), and between-person correla tion (all subjects measurements at a particular time may display correlation). A dding a repeated statement with occasion or, in this case, month, reflects the time structure of the data (days from baseline remained a covariate). Unstructured is a complex structure that allows variances and correlations among elements to be unpredictable, often results in numerous parameters in the modeling, and in this study, had difficulty converging. Variance components is the default structure in SPSS and was used for initial models in this studyother structures were modeled to assess for improvement of the criteria from the variance components model. In SPSS, variance components uses a diagonal structure for repeated statements. Diagonal has heterogeneous variances, with no correlati on between elements. Ante -Dependence, First Order has heterogeneous variances and heterogeneous correl ations across elements, with the correlation
152 between two non-adjacent elements determ ined by those between the two elements. Compound Symmetry-Heterogeneous has heterogeneous variances and eq ual correlation between elements. Auto-regressive, First Order has homogenous variances, with correlation between elements lagged such that data from points furthe r away in time are less correlated. Auto-regressive Heterogeneous is similar, but with heterogeneous variances. Toeplitz is also similar, except that lagged correlations may vary according to unique f eatures of the time point (rather than just distance) (Fitzmaurice et al., 2004; Garson, 2007; Singer & Willett, 2003). These co-variance structures were modeled, and the one with the most improvement in information criteria was chosen for reporting. Building Models using Information Criteria and PsuedoR2 Models were built using the change in the -2 Log Likelihood (-2LL) information criteria. The current model was compared to the previous one by subtracting the current -2LL from that of the previous one. The resulting chi square units were assessed for significant change in the model by comparing the value to standard chi squa re tables. Only those models that converged appropriately and also had significant change in the -2 LL were accepted as improved. In building the models, each models ability to explain variance identified at both Level 2 (fixed effects; between-persons variance) and Le vel 1 (random effects; w ithin-persons variance) were considered. This reducti on in variance through m odeling is somewhat comparable to the use of R2 in traditional models. However, this so-called pseudoR2 is a bit different. This figure represents the reduction of explainable variance in the models, so a fairly small change can result in a large percentage of change (Singer, 1998). In comparing models to structure the covarian ce, the information crite ria that automatically adjust for degrees of freedom were considered. These included the AIC, or Akaike Information Criteria, and BIC, or Schwarz Bayesian Criteria. These criteria are flexible and more amenable
153 to use with non-nested models, where decisi ons do not necessarily center on removing or retaining a covariate. The BIC more strongly penalizes the addition of covariance parameters, and thus may result in choosing a model that is too parsimonious. The AIC strikes a balance between complexity and parsimony (Fitzmauric e et al., 2004). The latter was therefore considered more relevant in this study. Howe ver, they often agreed, and most often did not improved upon the criteria in baseline models. Additional Considerations There are th ree considerations regarding the manner in which these analyses were conducted that bear mentioning. First, a repeated statement was not used in the initial models. While adding a repeated statement may add to the specification of the covariance structure of the measurements within persons, this is often not necessary in long itudinal modeling, since choosing the covariant structure e ffectively describes how the errors act over time (Singer, 1998; Singer & Willett, 2003). Nonetheless, once the fina l model was chosen and attention turned to structuring covariances, this was accomplished usi ng month as a repeated statement. Secondly, the choice of using Treatment Status as an influen tial person-level factor rather than specifying a three-level model with persons nested within gr oups was made for ease in interpretation. Lastly, the centering of time for the be ginning, middle, and end of the study, as recommended by Singer & Willett, to assess the influence of Treatment Status on the intercept and slopes was not included. Time was centered at mid-study to addre ss collinearity with qua dratic time, and model intercepts reflect mid-study. The univariate test of differences in the estimated marginal means was also not used, since the change over time by treatment status was the interest rather than differences at a particular point in the study. In future analyses, additional models or analyses may lend more insight to differenc es at various points in the study, questions that were not posed in this study.
154 Modeling Mediation To test the hypotheses related to the relationshi ps in the m odel, a specialized technique was used to evaluate the effects of caregiv er variables on care recipient variables within dyads and across dyads (Marsiske, 2008; McCrae et al., 2008). Variables were created to reflect both between-persons effects and within-p ersons effects of both predictors: Within-person averaged Perceived Stress and Emotional Responses were assessed for effects between persons at Level 2, with each caregiver contributing the same value at each data point collected to the fi xed effects estimates (across dyads in the population) of BehSx. The interpretation of this effect was that, on average, caregiv ers general level of Perceived Stress and Emotional Responses infl uenced BehSx in general. For illustration, when caregivers in the population had, say, higher Perceived Stress, PWD might have more severe BehSx across all data points. Within-person centered Perceived Stress and Emotional Responses reflected effects both at Level 1 and at Level 2, w ith each subject contributing to the fixed effects estimates based on the data points collected for that dya d. Individual caregive rs contribution at a given data point reflected their values devi ation from their own mean. These variables were interpreted at both levels: At Level 1, the current level of centered Pe rceived Stress and Emotional Responses were assessed for impact on BehSx within the dyad. Fo r illustration, when a caregivers level of Perceived Stress and Emotional Responses were higher than average, BehSx in the PWD in their care might be more severe. At Level 2, the current levels of centered Perceived Stress and Emotional Responses, on average in the population, were assessed for influence on current BehSx across dyads in the population. This effect reflected how th e caregivers changing stress process might influence changes in the BehSx over tim e, within the population as a whole. In this study, these variables allowed assessments regarding mean and cen tered influence of both variables on BehSx, but there were not formal test s of mediation. Nonetheless, the effect of caregivers Perceived stress on PWD BehSx was assessed with an expectation that this relationship would at least partia lly dissipate when Emotional Res ponses were also considered. In general, formal tests of mediation models require that a) the variables co-vary with each other, b) causes occur before th e presumed outcomes, and c) sour ces of spurious co-variation are eliminated. The temporal featur e of predictors, mediators, and outcomes in any mediation model
155 is difficult to ensure, particularly when the m easures are collected contemporaneously, as they occurred in this study, at the same data points. Experimental manipulatio n of X and M, as was done in this study using within-p erson centered values, lends strength to a mediation models specification. However, as some experts suggest, this may require separate analyses or subsequent studies to establish de finitive evidence of mediation, pa rticularly relate d to direction of causality (Bauer et al ., 2006; Kenny et al., 2003). In this study, in addition to manipulating the variables to reflect centered values, the change over time may lend support to direction of causality. In the pare nt study, X (Perceived Stress) and M (Emotional Responses) were hypothesi zed to change according to treatment status, but the outcome (BehSx) was not ex pected to be influenced by the experimental system. All variables were modeled over time according to ra ndomly assigned treatment status in separate analyses. Therefore, the diverging change over time in the BehSx according to treatment status may lend some support to directional ity in the causal hypotheses.
156 APPENDIX C PRELIMINARY TREND ANALYSES Multileve l models for change assume intra-in dividual individual linea r trends in change, and deviations from this assumption must be co nsidered in modeling. Singer & Willett (2003) recommend examining intra-individual Level 1 em pirical change data an d also evaluation of trends according to Level 2 predictors, for assess ing tenability of this assumption (page 128) and to become familiar with subjects trends. Initially, the general form or shape of th e trend over time was assessed (e.g., linear, quadratic, cubic). Subjects BehS x were first plotted against time in months and are displayed in Figure C-1, according to treatment status. This plot was assessed for initial appearance of general trends over time. A regression for each treatment group is plotted among the measurements, and demonstrates a flat or somewhat linear trend. Next, preliminary assessment of individual subjects BehSx was accomplished by graphing them over data points and examining them for evidence of non-linear patterns. Figure C-2 displays PWD BehSx for all dyads paneled by tr eatment status. Individuals BehSx patterns from occasion to occasion display a great deal of vari ability. Due to this variability, it is difficult to identify linear or non-linear trends as a usual occurrence for individuals in either treatment group. The variability of this data within persons emphasizes one of the strengths of multilevel modeling; that is to model intra-individual traj ectories while also cons idering a subjects own variation around that trajectory. To further assess the within-individual tre nds, scatter plots for each individual were developed. Figure C-3 and C-4 re flect these plots with regular regressions of BehSx on months for control and experimental subj ects, respectively. Examination of these descriptive plots lends some support to the use of linear modeling for Be hSx. However, while many of the subjects had
157 fairly stable trends, there was variation in both direction and magnitude of change over time within both treatment status groups, again underscoring the need to allow individuals trajectories to differ within person to adequately represent the data. Analyses for caregivers stress process variab les revealed similar va riability when withinperson change was assessed. The variation both between subjects (i ncluding direction and magnitude), and within subjects over time (months in the study), remains the most notable feature of the trends (see figures C-5 through C-12). These assessm ents of prominent variability lend support to using analyses that will identify the form of th e change trend over time. This became the first step in modeling each these variables change over time.
158 03581 0Month of Study 0 10 20 30 con exp 03581 0Month of Study BehSx over time by Treatment Figure C-1. Outcome PWD BehSx plotted against Months paneled by Treatment Status with Regressions.
159 12 10 8 6 5 4 3 2 0 Month of Study 35.00 30.00 25.00 20.00 15.00 10.00 5.00 0.00 Mean NPI-Q Behavior 35.00 30.00 25.00 20.00 15.00 10.00 5.00 0.00 con exp Experimental / Control PWD Behavioral Symptoms Figure C-2. Observed variation in PWD Behavi oral Symptoms over Time within subjects, paneled by Treatment Status.
160 -25 0 25 50 901.00 902.00 903.00 904.00 906.00 907.00908.00909.00910.00911.00 920.00921.00922.00923.00924.00 925.00926.00927.00928.00941.00 942.00943.00944.00945.00946.00 947.00 -25 0 25 50 -25 0 25 50 -25 0 25 50 -25 0 25 50 0510Month of Study -25 0 25 50 Figure C-3. Behavioral Symptoms over Months paneled by Dyad with RegressionsControl dyads only.
161 0 10 20 30 101.00 102.00 104.00 105.00 106.00 107.00 108.00 109.00 110.00 111.00 120.00 121.00 123.00 125.00 126.00 127.00 142.00 144.00 145.00 146.00 147.00 148.00 151.00 0 10 20 30 0 10 20 30 0 10 20 30 0510Month of Study 0 10 20 30 0510Month of Study 0510Month of Study Figure C-4. Behavioral Symptoms over Months pane led by Dyad with regressions Experimental dyads only.
162 03581 0Month of Study 20 30 40 50 60 con exp 03581 0Month of Study Perceived Stress over time by Treatment Figure C-5. Predictor Caregiver Perceived Stress plotted agains t Months paneled by Treatment Status with Regressions.
163 03581 0Month of Study -1 0 1 2 3 con exp 03581 0Month of Study Emotional Responses over time by Treatment Figure C-6. Predictor Caregive r Emotional Responses plotte d against Months paneled by Treatment Status with Regressions.
164 12 10 8 6 5 4 3 2 0 Month of Study 60 40 20 Mean Short Zarit Burden Interview-Bedard Sum Scores 60 40 20 con exp experimental / control Figure C-7. Observed variation in caregivers Perceived Stress over Time within subjects, paneled by Treatment Status.
165 12 10 8 6 5 4 3 2 0 Month of Study 2 0 -2 Mean Combined Depressive Symptoms (CES-D) and Negative Affect (PANAS) 2 0 -2 con exp experimental / control Figure C-8. Observed variation in caregivers Emotional Respons es over Time within subjects, paneled by Treatment Status.
166 901.00 902.00 903.00 904.00 906.00 907.00 908.00 909.00 910.00 911.00 920.00 921.00 922.00 923.00 924.00 925.00 926.00 927.00 928.00 941.00 942.00 943.00 944.00 945.00 946.00 947.00 10 20 30 40 50 60 10 20 30 40 50 60 10 20 30 40 50 60 10 20 30 40 50 60 0510 10 20 30 40 50 60 0510 Figure C-9. Perceived Stress over Months paneled by Dyad w ith RegressionsControl dyads only.
167 101.00 102.00 104.00 105.00 106.00 107.00 108.00 109.00 110.00 111.00 120.00 121.00 123.00 125.00 126.00 127.00 142.00 144.00 145.00 146.00 147.00 148.00 151.00 20 30 40 50 60 20 30 40 50 60 20 30 40 50 60 0510 20 30 40 50 60 0510 0510 0510 0510 Figure C-10. Perceived Stress plotted over Months paneled by Dyad with Regressions Experimental dyads only.
168 901.00 902.00 903.00 904.00 906.00 907.00 908.00909.00910.00911.00920.00921.00 922.00923.00924.00925.00926.00927.00 928.00941.00942.00943.00944.00945.00 946.00947.00 -4 -2 0 2 -4 -2 0 2 -4 -2 0 2 -4 -2 0 2 0510 -4 -2 0 2 0510 Figure C-11. Emotional Responses plotted ove r Months paneled by Dyad with regressions Control dyads only.
169 101.00 102.00 104.00 105.00 106.00 107.00 108.00 109.00 110.00 111.00 120.00 121.00 123.00 125.00 126.00 127.00 142.00 144.00 145.00 146.00 147.00 148.00 151.00 -1 0 1 2 3 4 -1 0 1 2 3 4 -1 0 1 2 3 4 0510 -1 0 1 2 3 4 0510 0510 0510 0510 Figure C-12. Emotional Responses plotted ove r Months paneled by Dyad with regressions Experimental dyads only.
170 LIST OF REFERENCES Aalten, P., de Vugt, M. E., Lousberg, R., Korten, E., Jaspers, N., Senden, B., et al. (2003). Behavioral problem s in dementia: A factor an alysis of the neurops ychiatric inventory. Dementia and Geriatric Cognitive Disorders, 15 (2), 99-105. Acton, G. J., & Kang, J. (2001). Interventions to reduce the burden of caregiving for an adult with dementia: A meta-analysis. Research in Nursing and Health, 24 (5), 349-360. Algase, D. L., Beck, C., Kolanowski, A. M., Wha ll, A. L., Berent, S. K., Richards, K., et al. (1996). Need-driven dementia-compromised behavior: An alternative view of disruptive behavior. American Journal of Alzheimer's Disease, 11 (6), 10-19. Almberg, B., Grafstrom, M., & Winblad, B. ( 1997). Caring for a demented elderly person-burden and burnout among caregiving relatives. Journal of Advanced Nursing, 25 (1), 109-116. Alzheimer's association: Sta tistics about Alzheimer's diseas e. (2006). Retrieved Sep. 9, 2006, from http://www.alz.org/AboutAD/statistics.asp Am irkhanyan, A. A., & Wolf, D. A. (2003). Caregi ver stress and noncaregiv er stress: Exploring the pathways of psychiatric morbidity. Gerontologist, 43 (6), 817-827. Ankri, J., Andrieu, S., Beaufils, B., Grand, A., & Henrard, J. C. (2005) Beyond the global score of the Zarit burden interview: Useful dimensions for clinicians. International Journal of Geriatric Psychiatry, 20 (3), 254-260. Annerstedt, L., Elmstahl, S., Ingvad, B., & Sa muelsson, S. M. (2000). Family caregiving in dementia--an analysis of the caregiver's burden and the "bre aking-point" when home care becomes inadequate. Scandinavian Journal of Public Health, 28 (1), 23-31. Archbold, P. G., Stewart, B. J., Greenlick, M. R., & Harvath, T. (1990). Mutuality and preparedness as predictors of caregiver role strain. Research in Nursing and Health, 13(6), 375-384. Ashford, J. W., Kolm, P., Colliver, J. A., Beki an, C., & Hsu, L. N. (1989). Alzheimer patient evaluation and the mini-mental state: Item characteristic curve analysis. Journals of Gerontology, 44(5), P139-146. Atienza, A. A., Collins, R., & King, A. C. (2001). The mediating effects of situational control on social support and mood following a stressor: A prospective study of dementia caregivers in their natural environments. Journals of Gerontology B Se ries: Psychological Sciences and Social Sciences, 56 (3), S129-139. Ayalon, L., Gum, A. M., Feliciano, L., & Arean, P. A. (2006). Effectiveness of nonpharmacological interventions for the manage ment of neuropsychiatric symptoms in patients with dementia: A systematic review. Archives of Internal Medicine, 166 (20), 2182-2188.
171 Ballard, C., Lowery, K., Powell, I., O'Brien, J., & James, I. (2000). Impact of behavioral and psychological symptoms of dementia on caregivers. International Psychogeriatrics, 12 93. Banaszak-Holl, J., Fendrick, A. M., Foster, N. L ., Herzog, A. R., Kabeto, M. U., Kent, D. M., et al. (2004). Predicting nursing home admission: Estimates from a 7-year follow-up of a nationally representative sa mple of older Americans. Alzheimer Disease and Associated Disorders, 18 (2), 83-89. Bauer, D. J., Preacher, K. J., & Gil, K. M. (2006). Conceptualizing and testing random indirect effects and moderated mediation in multilevel models: New procedures and recommendations. Psychological Methods, 11 (2), 142-163. Beck, C. (1998). Psychosocial and behavioral interventions for Al zheimer's disease patients and their families. American Journal of Ge riatric Psychiatry, 6 (2 Suppl 1), S41-48. Beck, J. G., Novy, D. M., Diefenbach, G. J., St anley, M. A., Averill, P. M., & Swann, A. C. (2003). Differentiating anxiety a nd depression in older adults with generalized anxiety disorder. Psychological Assessment, 15 (2), 184-192. Bedard, M., Molloy, D. W., Squire, L., Dubois, S., Lever, J. A., & O'Donnell, M. (2001). The Zarit burden interview: A new s hort version and screening version. Gerontologist, 41 (5), 652-657. Beekman, A. T., Deeg, D. J., Van Limbeek, J., Br aam, A. W., De Vries, M. Z., & Van Tilburg, W. (1997). Criterion validity of the center for epidemiologic studies depression scale (CES-D): Results from a community-based samp le of older subjects in the Netherlands. Psychological Medicine, 27 (1), 231-235. Berger, G., Bernhardt, T., Weimer, E., Peters, J ., Kratzsch, T., & Frolich, L. (2005). Longitudinal study on the relationship between symptomatol ogy of dementia and levels of subjective burden and depression among family careg ivers in memory clinic patients. Journal of Geriatric Psychiatry and Neurology, 18 (3), 119-128. Bergman-Evans, B. (1994). A health profile of spousal Alzheimer's caregivers. Depression and physical health characteristics. Journal of Psychosocial Nursing and Mental Health Services, 32 (9), 25-30. Bertrand, R. M., Fredman, L., & Saczynski, J. (2006) Are all caregivers created equal? Stress in caregivers to adults with and without dementia. Journal of Aging and Health, 18(4), 534551. Black, W., & Almeida, O. P. (2004). A systema tic review of the association between the behavioral and psychological symptoms of dementia and burden of care. International Psychogeriatrics, 16 (3), 295-315. Blake, H., Lincoln, N. B., & Clarke, D. D. (2003). Caregiver strain in spouses of stroke patients. Clinical Rehabilitation, 17 (3), 312-317.
172 Blane, D., Netuveli, G., & Montgomery, S. M. (2 008). Quality of life, health and physiological status and change at older ages. Social Science and Medicine, 66 (7), 1579-1587. Bookwala, J., & Schulz, R. (1998). The role of neuroticism and mastery in spouse caregivers' assessment of and response to a contextual stressor. Journals of Gerontology B Series: Psychological Sciences and Social Sciences, 53 (3), P155-164. Bourgeois, M. S. (1993). Effects of memory aids on the dyadic conversations of individuals with dementia. Journal of Applied Behavior Analysis, 26 (1), 77-87. Brummett, B. H., Babyak, M. A., Siegler, I. C., V italiano, P. P., Ballard, E. L., Gwyther, L. P., et al. (2006). Associations among perceptions of social support, negativ e affect, and quality of sleep in caregivers and noncaregivers. Health Psychology, 25 (2), 220-225. Bucks, R. S., & Radford, S. A. (2004). Em otion processing in Alzheimer's disease. Aging and Mental Health, 8 (3), 222-232. Buckwalter, K. C., Gerdner, L., Kohout, F., Hall, G. R., Kelly, A., Richards, B., et al. (1999). A nursing intervention to decrea se depression in family caregivers of persons with dementia. Archives of Psychiatric Nursing, 13 (2), 80-88. Buhr, G. T., Kuchibhatla, M., & Clipp, E. C. (2006). Caregivers' reasons for nursing home placement: Clues for improving discussions with families prior to the transition. Gerontologist, 46 (1), 52-61. Burgener, S., & Twigg, P. (2002a). Interventions for persons with irreversible dementia. Annual Review of Nursing Research, 20, 89-124. Burgener, S., & Twigg, P. (2002b). Relationships among caregiver factors and quality of life in care recipients with irreversible dementia. Alzheimer Disease and Associated Disorders, 16(2), 88-102. Burns, R., Nichols, L. O., Martindale-Adams, J., Graney, M. J., & Lummus, A. (2003). Primary care interventions for dementia caregiver s: 2-year outcomes from the REACH study. Gerontologist, 43 (4), 547-555. Butcher, H. K., Holkup, P. A., Park, M., & Maas, M. (2001). Thematic analysis of the experience of making a decision to place a family member with Alzheimer's disease in a special care unit. Research in Nursing and Health, 24 (6), 470-480. Callahan, C. M., Boustani, M. A., Unverzagt, F. W., Austrom, M. G., Damush, T. M., Perkins, A. J., et al. (2006). Effectiveness of collabo rative care for older adults with Alzheimer disease in primary care: A randomized controlled trial. Journal of the American Medical Association, 295(18), 2148-2157. Callahan, C. M., Hui, S. L., Nienaber, N. A., Musick, B. S., & Tierney, W. M. (1994). Longitudinal study of depression and health services use among elderly primary care patients. Journal of the American Geriatric Society, 42 (8), 833-838.
173 Caron, C. D., & Bowers, B. J. (2003). Deciding whether to continue, share, or relinquish caregiving: Caregiver views. Qualitative Health Research, 13 (9), 1252-1271. Carruth, A. K., Tate, U. S., Moffett, B. S., & Hill, K. (1997). Reciprocity, emotional well-being, and family functioning as determinants of fa mily satisfaction in caregivers of elderly parents. Nursing Research, 46(2), 93-100. Caswell, L. W., Vitaliano, P. P., Croyle, K. L ., Scanlan, J. M., Zhang, J., & Daruwala, A. (2003). Negative associations of chronic stress and cognitive performance in older adult spouse caregivers. Experimental Aging Research, 29 (3), 303-318. Cavanaugh, J. C., Dunn, N. J., Mowery, D., Feller, C., Niederehe, G., Fruge, E., et al. (1989). Problem-solving strategies in de mentia patient-caregiver dyads. Gerontologist, 29 (2), 156-158. Chang, B. L. (1999). Cognitive-behavioral in tervention for homebound caregivers of persons with dementia. Nursing Research, 48 (3), 173-182. Cho, S.-H. (2003). Using multilevel analysis in patient and organizational outcomes research. Nursing Research, 52 (1), 61-65. Chou, K. R., LaMontagne, L. L., & Hepworth, J. T. (1999). Burden experienced by caregivers of relatives with dementia in Taiwan. Nursing Research, 48 (4), 206-214. Clyburn, L. D., Stones, M. J., Hadjistavropoulo s, T., & Tuokko, H. (2000). Predicting caregiver burden and depression in Alzheimer's disease. Journals of Gerontology, B series: Psychological Sciences and Social Sciences, 55 (1), S2-13. Coehlo, D. P., Hooker, K., & Bowman, S. ( 2007). Institutional placem ent of persons with dementia: What predicts occurrence and timing? Journal of Family Nursing, 13 (2), 253277. Cohen-Mansfield, J. (2000a). Non-pharmacologi cal management of behavioral problems in persons with dementia : The TREA model. Alzheimer's Care Quarterly, 1 (4), 22-34. Cohen-Mansfield, J. (2000b). Theoretical framew ork for behavioral problems in dementia. Alzheimer's Care Quarterly, 1(4), 8-21. Cohen-Mansfield, J. (2001a). Managing agit ation in elderly patients with dementia. Geriatric Times Retrieved December 2, 2004, from http://www.geriatrictimes.com/g010533.html Cohen-Mansfield, J. (2001b). Nonpharm acologic interventions for inappropriate behaviors in dementia: A review, summary, and critique. American Journal of Geriatric Psychiatry, 9(4), 361-381. Cohen-Mansfield, J., & Libin, A. (2004). Assess ment of agitation in elderly patients with dementia: Correlations between inform ant rating and direct observation. International Journal of Geriatric Psychiatry, 19 (9), 881-891.
174 Cohen, C. A., Colantonio, A., & Vernich, L. (20 02). Positive aspects of caregiving: Rounding out the caregiver experience. International Journal of Geriatric Psychiatry, 17 (2), 184188. Cohen, S., Kamarck, T., & Mermelstein, R. (198 3). A global measure of perceived stress. Journal of Health and Social Behavior, 24 385-396. Colling, K. B. (2004). Caregiver interventions for passive behaviors in dementia: Links to the NDB model. Aging and Mental Health, 8 (2), 117-125. Compton, S. A., Flanagan, P., & Gregg, W. (1997). Elder abuse in people with dementia in Northern Ireland: Prevalence and predictors in cases referred to a psychiatry of old age service. International Journal of Geriatric Psychiatry, 12 (6), 632-635. Cooper, C., Balamurali, T. B., & Livingston, G. (2007). A systematic review of the prevalence and covariates of anxiety in caregivers of people with dementia. International Psychogeriatrics, 19 (2), 175-195. Covinsky, K. E., Newcomer, R., Fox, P., Wood, J., Sands, L., Dane, K., et al. (2003). Patient and caregiver characteristics associated with de pression in caregivers of patients with dementia. Journal of General Internal Medicine, 18(12), 1006-1014. Coyne, A. C., Reichman, W. E., & Berbig, L. J. (1993). The relationship between dementia and elder abuse. American Journal of Psychiatry, 150 (4), 643-646. Crawford, J. R., & Henry, J. D. (2004). The positive and negative affect schedule (PANAS): Construct validity, measurement properties a nd normative data in a large non-clinical sample. British Journal of Clinical Psychology, 43(Pt 3), 245-265. Croog, S. H., Burleson, J. A., Sudilovsky, A., & Ba ume, R. M. (2006). Spouse caregivers of Alzheimer patients: Problem responses to caregiver burden. Aging and Mental Health, 10(2), 87-100. Croog, S. H., Sudilovsky, A., Burleson, J. A., & Baume, R. M. (2001). Vulnerability of husband and wife caregivers of Alzheimer disease patients to caregiving stressors. Alzheimer Disease and Associated Disorders, 15 (4), 201-210. Cuijpers, P. (2005). Depressive disorders in caregivers of dementia patients: A systematic review. Aging & Mental Health, 9 (4), 325-330. Cummings, J. L. (1997). The neuropsychiatr ic inventory: Assessing psychopathology in dementia patients. Neurology, 48 (5 Suppl 6), S10-16. Cummings, J. L., Mega, M., Gray, K., Rosenberg -Thompson, S., Carusi, D. A., & Gornbein, J. (1994). The neuropsychiatric i nventory: Comprehensive assessment of psychopathology in dementia. Neurology, 44 (12), 2308-2314.
175 Czaja, S. J., Eisdorfer, C., & Sc hulz, R. (2000). Future directions in caregiving: Implications for intervention research (pp. 283). In R. Schulz (Ed.), Handbook of dementia caregiving intervention research. New York: Springer Publishing Company. Davey, A., & Eggebeen, D. J. (1998). Patterns of intergenerational exchange and mental health. Journals of Gerontology B Series: Psychol ogical Sciences and Social Sciences, 53 (2), P86-95. Davis, L. L., Buckwalter, K., & Burgio, L. D. (1997). Measuring problem behaviors in dementia: Developing a methodological agenda. Advances in Nursing Science, 20 (1), 40-55. de Vugt, M. E., Jolles, J., van Osch, L., Steven s, F., Aalten, P., Lousberg, R., et al. (2006). Cognitive functioning in spousal caregivers of dementia patients: Findings from the prospective MAASBED study. Age and Ageing, 35 (2), 160-166. de Vugt, M. E., Stevens, F., Aalten, P., Lousbe rg, R., Jaspers, N., & Verhey, F. R. (2005). A prospective study of the effect s of behavioral symptoms on the institutionalization of patients with dementia. International Psychogeriatrics, 17 (4), 577-589. de Vugt, M. E., Stevens, F., Aalten, P., Lousber g, R., Jaspers, N., Winke ns, I., et al. (2003). Behavioural disturbances in dementia patien ts and quality of the marital relationship. International Journal of Geriatric Psychiatry, 18 (2), 149-154. de Vugt, M. E., Stevens, F., Aalten, P., Lousberg, R., Jaspers, N., Winkens, I., et al. (2004). Do caregiver management strategies influence patient behavior in dementia? International Journal of Geriatric Psychiatry, 19 (1), 85-92. Deichert, N. T., Fekete, E. M., Boarts, J. M., Druley, J. A., & De lahanty, D. L. (2008). Emotional support and affect: Associations with health behaviors and active coping efforts in men living with HIV. AIDS and Behavior, 12 (1), 139-145. Dellasega, C. (1991). Caregiving stress among community caregivers for the elderly: Does institutionalization make a difference? Journal of Community Health Nursing, 8 (4), 197205. Denollet, J., & De Vries, J. (2006). Positive and negative affect within the realm of depression, stress and fatigue: The two-factor distress model of the global mood scale (GMS). Journal of Affective Disorders, 91 (2-3), 171-180. Diehl, M. (2005). Personal convers ation with Dr. Manfred Diehl. Donaldson, C., Tarrier, N., & Burns, A. (1997). The impact of the symptoms of dementia on caregivers. British Journal of Psychiatry, 170, 62-68. Donaldson, C., Tarrier, N., & Burns, A. (1998). Determinants of carer stress in Alzheimer's disease. International Journal of Geriatric Psychiatry, 13 (4), 248-256.
176 Doody, R., Stevens, J. C., Beck, C., Dubinsky, R. M., Kaye, J. A., Gwyther, L., et al. (2001). Practice parameter: Management of dementia (an evidence-based review): Report of the quality standards committee of th e American academy of neurology. Neurology, 56 (9), 1154-1166. Dunkin, J. J., & Anderson-Hanley, C. (1998). Dementia caregiver burden: A review of the literature and guidelines for assessment and intervention. Neurology, 51 (1 Suppl 1), S5360; discussion S65-57. Dyer, C. B., Pavlik, V. N., Murphy, K. P., & Hyman, D. J. (2000). The high prevalence of depression and dementia in elder abuse or neglect. Journal of the American Geriatric Society, 48 (2), 205-208. Eloniemi-Sulkava, U., Notkola, I. L., Hamalain en, K., Rahkonen, T., Viramo, P., Hentinen, M., et al. (2002). Spouse caregivers' perceptions of influence of dementia on marriage. International Psychogeriatrics, 14 (1), 47-58. Ernst, R. L., Hay, J. W., Fenn, C., Tinklenberg, J., & Yesavage, J. A. (1997). Cognitive function and the costs of Alzheimer's disease. An exploratory study. Archives of Neurology, 54 (6), 687-693. Farran, C. J., Loukissa, D., Perraud, S., & Paun, O. (2003). Alzheimer's disease caregiving information and skills. Part i: Care recipient issues and concerns. Research in Nursing and Health, 26 (5), 366-375. Farran, C. J., Loukissa, D., Perraud, S., & Paun, O. (2004). Alzheimer's disease caregiving information and skills. Part ii: Family caregiver issues and concerns. Research in Nursing and Health, 27 (1), 40-51. Fauth, E. B., Zarit, S. H., Femia, E. E., Hofer, S. M., & Stephens, M. A. (2006). Behavioral and psychological symptoms of dementia and careg ivers' stress appraisals: Intra-individual stability and change over short-term observations. Aging and Mental Health, 10 (6), 563573. Fisher, P. A., & Laschinger, H. S. (2001). A rela xation training program to increase self-efficacy for anxiety control in Alzheimer family caregivers. Holistic Nursing Practice, 15 (2), 4758. Fitzmaurice, G. M., Laird, N. M., & Ware, J. H. (2004). Applied longitudinal analysis Hoboken, New Jersey: John Wiley & Sons. Folstein, M., Folstein, S., & McHugh, P. (1975) Mini-mental state: A practical method for grading the cognitive state of patients for the clinician. Journal of Psychiatric Research, 12, 189-198. Forester, B. P., & Oxman, T. E. (2003). Measures to assess the noncognitive symptoms of dementia in the primary care setting. Primary Care Companion to the Journal of Clinical Psychiatry, 5 (4), 158-163.
177 Fredrickson, B. L., & Losada, M. F. (2005). Po sitive affect and the complex dynamics of human flourishing. The American Psychologist, 60 (7), 678-686. Gainey, R. R., & Payne, B. K. (2006). Caregive r burden, elder abuse and Alzheimer's disease: Testing the relationship. Journal of Health and Human Services Administration? 29 (2), 245-259. Gallagher-Thompson, D., Dal Canto, P. G., Jac ob, T., & Thompson, L. W. (2001). A comparison of marital interaction patter ns between couples in which the husband does or does not have Alzheimer's disease. Journals of Gerontology B: Psychological Sciences and Social Sciences, 56 (3), S140-150. Gallicchio, L., Siddiqi, N., Lange nberg, P., & Baumgarten, M. (2002). Gender differences in burden and depression among informal caregiver s of demented elders in the community. International Journal of Geriatric Psychiatry, 17 (2), 154-163. Garand, L., Dew, M. A., Eazor, L. R., DeKosky, S. T., & Reynolds, C. F., 3rd. (2005). Caregiving burden and psychiatric morbidity in spouses of persons with mild cognitive impairment. International Journal of Geriatric Psychiatry, 20 (6), 512-522. Garson, G. D. (2007). Linear mixed models. Statnotes: Topics in Multivariate Analysis Retrieved October 7, 2007, from http://www2.chass.ncsu.edu/garson/pa765/multilevel.htm Gaugler, J. E., Kane, R. L., Kane, R. A., & Ne wcom er, R. (2005). The longitudinal effects of early behavior problems in the dementia caregiving career. Psychology and Aging, 20 (1), 100-116. Gaugler, J. E., Pot, A. M., & Zarit, S. H. (2007) Long-term adaptation to institutionalization in dementia caregivers. Gerontologist, 47 (6), 730-740. Geldmacher, D. S., & Lin, P. J. (2006). Evaluation and management of dementia in primary care: Highlights from the fourth annual dementia congress, Atlanta, Georgia, October, 2005. Clinician Reviews, Self-Study Supplement Gerdner, L. A., Buckwalter, K. C., & Ree d, D. (2002). Impact of a psychoeducational intervention on caregiver respons e to behavioral problems. Nursing Research, 51(6), 363374. Gilliam, C. M., & Steffen, A. M. (2006). The re lationship between caregiving self-efficacy and depressive symptoms in dementia family caregivers. Aging and Mental Health, 10 (2), 7986. Gitlin, L. N., Belle, S. H., Burgio, L. D., C zaja, S. J., Mahoney, D., Gallagher-Thompson, D., et al. (2003a). Effect of multico mponent interventions on careg iver burden and depression: The REACH multisite initiative at 6-month follow-up. Psychology and Aging, 18 (3), 361-374.
178 Gitlin, L. N., Corcoran, M., Winter, L., Boy ce, A., & Hauck, W. W. (2001). A randomized, controlled trial of a home envi ronmental intervention: Effect on efficacy and upset in caregivers and on daily function of persons with dementia. Gerontologist, 41 (1), 4-14. Gitlin, L. N., Hauck, W. W., Dennis, M. P., & Wint er, L. (2005). Maintenance of effects of the home environmental skill-building program fo r family caregivers and individuals with Alzheimer's disease and related disorders. Journals of Gerontology, Series A: Biological Sciences and Medical Sciences, 60 (3), 368-374. Gitlin, L. N., Winter, L., Corcoran, M., Dennis, M. P., Schinfeld, S., & Hauck, W. W. (2003b). Effects of the home environmental skill-build ing program on the caregiver-care recipient dyad: 6-month outcomes from the Philadelphia REACH initiative. Gerontologist, 43 (4), 532-546. Glozman, J. M. (2004). Quality of life of caregivers. Neuropsychology Review, 14 (4), 183-196. Goode, K. T., Haley, W. E., Roth, D. L., & For d, G. R. (1998). Predicting longitudinal changes in caregiver physical and mental health: A stress process model. Health Psychology: official journal of the Division of Health Psychology, American Psychological Association, 17 (2), 190-198. Gort, A. M., Mingot, M., Gomez, X., Soler, T., To rres, G., Sacristan, O., et al. (2007). Use of the Zarit scale for assessing caregiver burden and collapse in caregiving at home in dementias. International Journal of Geriatric Psychiatry, In press Gotell, E., Brown, S., & Ekman, S. L. (2003). Influence of caregiver singing and background music on posture, movement, and sensory awareness in dementia care. International Psychogeriatrics, 15 (4), 411-430. Gottlieb, B. H., & Rooney, J. A. (2004). Coping eff ectiveness: Determinants and relevance to the mental health and affect of family caregivers of persons with dementia. Aging and Mental Health, 8 (4), 364-373. Graff, M. J., Vernooij-Dassen, M. J., Thijssen, M ., Dekker, J., Hoefnagels, W. H., & Rikkert, M. G. (2006). Community based occupational ther apy for patients with dementia and their care givers: Randomised controlled trial. BMJ: Clinical Research ed., 333 (7580), 1196. Grammer, K., Fink, B., Oberzaucher, E., Atzmuller, M., Blantar, I., & Mi tteroecker, P. (2004). The representation of self reported affect in body posture and body posture simulation. Collegium Antropologicum, 28 Suppl 2 159-173. Grov, E. K., Fossa, S. D., Sorebo, O., & Dahl, A. A. (2006). Primary caregivers of cancer patients in the palliative phase: A path analysis of variables influencing their burden. Social Science and Medicine, 63 (9), 2429-2439. Gruffydd, E., & Randle, J. (2006). Alzheimer' s disease and the psychosocial burden for caregivers. Community Practice, 79 (1), 15-18.
179 Haight, B. K., Bachman, D. L., Hendrix, S., Wa gner, M. T., Meeks, A., & Johnson, J. (2003). Life review: Treating the dyadi c family unit with dementia. Clinical Psychology and Psychotherapy, 10 165-174. Haley, W. E. (1997). The family caregiver's role in Alzheimer's disease. Neurology, 48 (Suppl 6), S25-S29. Haley, W. E., LaMonde, L. A., Han, B., Burton, A. M., & Schonwetter, R. (2003). Predictors of depression and life satisfaction among spousal caregivers in hospice: Application of a stress process model. Journal of Palliative Medicine, 6 (2), 215-224. Haley, W. E., Levine, E. G., Brown, S. L., Berry, J. W., & Hughes, G. H. (1987). Psychological, social, and health conseque nces of caring for a relative with senile dementia. Journal of American Geriatric Society, 35 (5), 405-411. Haley, W. E., & Pardo, K. M. (1989). Relationship of severity of dementia to caregiving stressors. Psychology and Aging, 4 (4), 389-392. Haley, W. E., Roth, D. L., Colet on, M. I., Ford, G. R., West, C. A., Collins, R. P., et al. (1996). Appraisal, coping, and social support as mediators of we ll-being in black and white family caregivers of patients with Alzheimer's disease. Journal of Consulting and Clinical Psychology, 64 (1), 121-129. Hall, G. R. (1994). Caring for people with Alzh eimer's disease using the conceptual model of progressively lowered stress thre shold in the c linical setting. Nursing Clinics of North America, 29(1), 129-141. Hall, G. R., & Buckwalter, K. C. (1987). Progres sively lowered stress th reshold: A conceptual model for care of adults w ith Alzheimer's disease. Archives of Psychiatric Nursing, 1 (6), 399-406. Hamel, M., Gold, D. P., Andres, D., Reis, M., Da stoor, D., Grauer, H., et al. (1990). Predictors and consequences of aggressive behavior by community-based dementia patients. Gerontologist, 30 (2), 206-211. Hamilton, M. (1967). Developmen t of a rating scale for primary depressive illness. British Journal of Social and Clinical Psychology, 6 (4), 278-296. Hansberry, M. R., Chen, E., & Gorbien, M. J. (2005). Dementia and elder abuse. Clinical Geriatric Medicine, 21 (2), 315-332. Harrell, L. E., Marson, D., Chatterjee, A., & Parri sh, J. A. (2000). The severe mini-mental state examination: A new neuropsychologic instrument for the bedside assessment of severely impaired patients with Alzheimer disease. Alzheimer Disease and Associated Disorders, 14(3), 168-175. Hart, J. A. (2004). Alzheimer's disease. Retrieved September 23, 2004, from http: www.nlm.nih.gov/medlineplu s/ency/article/000760.htm
180 Harwood, D. G., Barker, W. W., Ownby, R. L., & Duara, R. (2000) Relationship of behavioral and psychological symptoms to cognitive impairment and functional status in Alzheimer's disease. International Journal of Geriatric Psychiatry, 15 (5), 393-400. Hebert, L. E., Scherr, P. A., Bienias, J. L., Be nnett, D. A., & Evans, D. A. (2003a). Alzheimer disease in the US population: Preval ence estimates using the 2000 census. Archives of Neurology, 60 (8), 1119-1122. Hebert, R., Bravo, G., & Preville, M. (2000). Relia bility, validity and refe rence values of the Zarit burden interview for assessing informal caregivers of community-dwelling older persons with dementia. Canadian Journal on Aging, 19 494-507. Hebert, R., Dubois, M. F., Wolfson, C., Chambe rs, L., & Cohen, C. (2001). Factors associated with long-term institutionalization of older people with dementia: Data from the Canadian study of health and aging. Journals of Gerontology, A Series: Biological Sciences and Medical Sciences, 56 (11), M693-699. Hebert, R., Levesque, L., Vezina, J., Lavoie, J. P., Ducharme, F., Gendron, C., et al. (2003b). Efficacy of a psychoeducative group program for caregivers of demented persons living at home: A randomized controlled trial. Journals of Gerontology B Series: Psychological Sciences and Social Sciences, 58 (1), S58-67. Hebert, R. S., Dang, Q., & Schulz, R. (2007). Religi ous beliefs and practices are associated with better mental health in family caregivers of patients with dementia: Findings from the REACH study. American Journal of Geriatric Psychiatry, 15(4), 292-300. Herrmann, N., & Black, S. E. (2000). Behavioral disturbances in dementia: Will the real treatment please stand up? Neurology, 55 (9), 1247-1248. Heru, A. M., Ryan, C. E., & Iqbal, A. (2004). Fa mily functioning in the caregivers of patients with dementia. International Journal of Geriatric Psychiatry, 19 (6), 533-537. Hill, T. D., Ross, C. E., & Angel, R. J. (2005). Neighborhood disorder, psychophysiological distress, and health. Journal of Health and Social Behavior, 46 (2), 170-186. Hooker, K., Bowman, S. R., Coehlo, D. P., Lim, S. R., Kaye, J., Guariglia, R., et al. (2002). Behavioral change in persons with demen tia: Relationships with mental and physical health of caregivers. Journals of Gerontology B Series: Psychological Sciences & Social Sciences, 57 (5), P453-460. Hooker, K., Manoogian-O'Dell, M., Monahan, D. J ., Frazier, L. D., & Shifren, K. (2000). Does type of disease matter? Gender differences among Alzheimer's and Parkinson's disease spouse caregivers. Gerontologist, 40 (5), 568-573. Hoyert, D. L., Kung, H.-C., & Smith, B. L. (2005). Deaths: Preliminary data for 2003. National Vital Statistics Reports, 53 (15).
181 Jansson, W., Nordberg, G., & Grafstrom, M. (20 01). Patterns of elderly spousal caregiving in dementia care: An observational study. Journal of Advanced Nursing, 34 (6), 804-812. Jones, R. N., & Gallo, J. J. (2001). Educati on bias in the mini-mental state examination. International Psychogeriatrics, 13 (3), 299-310. Kallin, K., Gustafson, Y., Sandman, P. O., & Karlsson, S. (2005). Factors associated with falls among older, cognitively impaired people in geriatric care settings : A population-based study. American Journal of Ge riatric Psychiatry, 13 (6), 501-509. Kaufer, D. I., Cummings, J. L., Ketchel, P., Sm ith, V., MacMillan, A., Sh elley, T., et al. (2000). Validation of the NPI-Q, a brief clinical form of the neuropsychiatric inventory. Journal of Neuropsychiatry and Clin ical Neurosciences, 12 (2), 233-239. Kenny, D. A., Kashy, D. A., & Cook, W. L. (2006). Dyadic data analysis New York, New York: The Guilford Press. Kenny, D. A., Korchmaros, J. D., & Bolger, N. (2003). Lower level mediation in multilevel models. Psychological Methods, 8 (2), 115-128. Killgore, W. D. (2000). Evidence for a third factor on the positive and negative affect schedule in a college student sample. Perceptual and Motor Skills, 90 (1), 147-152. Kitwood, T. (1997). Dementia reconsidered: The person comes first Bristol, PA: Open University Press. Klein, D. A., Steinberg, M., Gali k, E., Steele, C., Sheppard, J. M., Warren, A., et al. (1999). Wandering behaviour in community-re siding persons with dementia. International Journal of Geriatric Psychiatry, 14 (4), 272-279. Knight, B. G., Silverstein, M., McCallum, T. J., & Fox, L. S. (2000). A sociocultural stress and coping model for mental health outcomes among African Amer ican caregivers in southern California. Journals of Gerontology, B seri es: Psychological Sciences and Social Sciences, 55 (3), P142-150. Knight, R. G., Williams, S., McGee, R., & Olam an, S. (1997). Psychometric properties of the centre for epidemiologic studies depression scale (CES-D) in a sa mple of women in middle life. Behaviour Research and Therapy, 35 (4), 373-380. Kolanowski, A. M., Fick, D., Waller, J. L., & Ah ern, F. (2006). Outcomes of antipsychotic drug use in community-dwelling elders with dementia. Archives of Psychiatric Nursing, 20 (5), 217-225. Kolanowski, A. M., Litaker, M., & Buettner, L. (2005). Efficacy of theory-based activities for behavioral symptoms of dementia. Nursing Research, 54 (4), 219-228. Kolanowski, A. M., & Whall, A. L. (2000). To ward holistic theory-b ased intervention for dementia behavior. Holistic Nursing Practice, 14 (2), 67-76.
182 Kuzuya, M., Masuda, Y., Hirakawa, Y., Iwata, M., Enoki, H., Hasegawa, J., et al. (2006). Falls of the elderly are associated with bu rden of caregivers in the community. International Journal of Geriatric Psychiatry, 21 (8), 740-745. Langa, K. M., Larson, E. B., Wallace, R. B., Fendrick, A. M., Foster, N. L., Kabeto, M. U., et al. (2004). Out-of-pocket health care expenditu res among older Americans with dementia. Alzheimer's Disease and Associated Disorders, 18 (2), 90-98. Lange, R. T., Hopp, G. A., & Kang, N. (2004). Psyc hometric properties and factor structure of the neuropsychiatric inventor y nursing home version in an elderly neuropsychiatric population. International Journal of Geriatric Psychiatry, 19 (5), 440-448. Lee, H. S., Kim, D. K., & Kim, J. H. (2006). Stre ss in caregivers of demented people in Korea--a modification of Pearlin and colleagues' stress model. International Journal of Geriatric Psychiatry, 21 (8), 784-791. Leinonen, E., Korpisammal, L., Pulkkinen, L. M., & Pukuri, T. (2001). The comparison of burden between caregiving spouses of depressive and demented patients. International Journal of Geriatric Psychiatry, 16 (4), 387-393. Levenstein, S., Prantera, C., Va rvo, V., Scribano, M. L., Berto, E., Luzi, C., et al. (1993). Development of the perceived stress ques tionnaire: A new tool for psychosomatic research. Journal of Psychosomatic Research, 37 19-32. Levesque, L., Ducharme, F., & Lachance, L. (1 999). Is there a diffe rence between family caregiving of institutionalized elde rs with or without dementia? Western Journal of Nursing Research, 21 (4), 472-491; discussion 491-477. Lewis, M. L., Hepburn, K., Narayan, S., & Kirk, L. N. (2005). Relationship matters in dementia caregiving. American Journal of Alzheimers Disease and Other Dementias, 20 (6), 341347. Liang, J., Krause, N. M., & Bennett, J. M. (2001) Social exchange and well-being: Is giving better than receiving? Psychology and Aging, 16 (3), 511-523. Loewenstein, D. A., Arguelles, S., Bravo, M., Freeman, R. Q., Arguelles, T., Acevedo, A., et al. (2001). Caregivers' judgments of the functi onal abilities of the Alzheimer's disease patient: A comparison of proxy reports and objective measures. Journals of Gerontology B Series: Psychological Scienc es and Social Sciences, 56 (2), P78-84. Logan, B. (2004). Family caregiver alliance fa ct sheet: Caregiver's guide to understanding dementia behaviors. Retrieved Sept. 29, 2004, from http://www.caregiver.org/caregiver /jsp/print_friendly.jsp? nodeid=391 Logsdon, R. G., McCurry, S. M., & Teri, L. (2007). Evidence-based psychological treatments for disruptive behaviors in i ndividuals with dementia. Psychology and Aging, 22 (1), 28-36.
183 Logsdon, R. G., Teri, L., McCurry, S. M., Gibbons L. E., Kukull, W. A., & Larson, E. B. (1998). Wandering: A significant problem am ong community-residing individuals with Alzheimer's disease. Journals of Gerontology B Series: Psychological Sciences and Social Sciences, 53 (5), P294-299. Lyketsos, C. G. (2007). Neuropsychiatric symptoms (behavioral and psychological symptoms of dementia) and the developmen t of dementia treatments. International Psychogeriatrics, 19(3), 409-420. Lyketsos, C. G., Lopez, O., Jones, B., Fitzpatrick, A. L., Breitner, J., & DeKosky, S. (2002). Prevalence of neuropsychiatric symptoms in dementia and mild cognitive impairment: Results from the cardiovascular health study. Journal of the American Medical Association, 288(12), 1475-1483. Lyons, K. S., & Sayer, A. G. (2005). Usi ng multilevel modeling in caregiving research. Aging and Mental Health, 9 (3), 189-195. Lyons, K. S., Sayer, A. G., Archbold, P. G., Ho rnbrook, M. C., & Stewart, B. J. (2007). The enduring and contextual effects of physical health and depression on care-dyad mutuality. Research in Nursing and Health, 30 (1), 84-98. Lyons, K. S., Zarit, S. H., Sayer, A. G., & Whitlatch, C. J. (2002). Caregiving as a dyadic process: Perspectives from caregiver and receiver. Journals of Gerontology B Series: Psychological Sciences and Social Sciences, 57 (3), P195-204. Mahoney, R., Regan, C., Katona, C., & Livingston, G. (2005). Anxiety and depression in family caregivers of people with Alzhei mer disease: The LASER-AD study. American Journal of Geriatric Psychiatry, 13 (9), 795-801. Marsiske, M. (2008). Personal co mmunication with Dr. Marsiske. Martire, L. M., & Hall, M. (2002). Dementia caregiving: Recent research on negative health effects and the efficacy of caregiver interventions. CNS Spectrums, 7(11), 791-796. Mausbach, B. T., Patterson, T. L., Rabinowitz, Y. G., Grant, I., & Schulz, R. (2007). Depression and distress predict time to cardiovascul ar disease in dementia caregivers. Health Psychology, 26(5), 539-544. McClendon, M. J., Smyth, K. A., & Neundorfer, M. M. (2004). Survival of persons with Alzheimer's disease: Caregiver coping matters. Gerontologist, 44(4), 508-519. McClendon, M. J., Smyth, K. A., & Neundorfer, M. M. (2006). Long-term-care placement and survival of persons with Alzheimer's disease. Journals of Gerontology, B series: Psychological Sciences and Social Sciences, 61 (4), P220-227. McConaghy, R., & Caltabiano, M. L. (2005). Caring for a person with dementia: Exploring relationships between perceived burd en, depression, coping and well-being. Nursing and Health Sciences, 7 (2), 81-91.
184 McCrae, C. S., McNamara, J. P., Rowe, M. A., Dzie rzewski, J. M., Dirk, J., Marsiske, M., et al. (2008). Sleep and affect in older adults: Using multilevel modeling to examine daily associations. J Sleep Res, 17 (1), 42-53. Mitrani, V. B., Feaster, D. J., McCabe, B. E., Czaja, S. J., & Szapocznik, J. (2005). Adapting the structural family systems rating to assess the patterns of interaction in families of dementia caregivers. Gerontologist, 45 (4), 445-455. Mitrushina, M., & Satz, P. (1991). Reliability an d validity of the mini-mental state exam in neurologically intact elderly. Journal of Clinical Psychology, 47 (4), 537-543. Mittelman, M. S., Haley, W. E., Clay, O. J., & Roth, D. L. (2006). Improving caregiver wellbeing delays nursing home placement of patients with Alzheimer disease. Neurology, 67(9), 1592-1599. Mittelman, M. S., Roth, D. L., Clay, O. J., & Haley, W. E. (2007). Preserving health of Alzheimer caregivers: Impact of a spouse caregiver intervention. American Journal of Geriatric Psychiatry, 15 (9), 780-789. Mittelman, M. S., Roth, D. L., Coon, D. W., & Haley, W. E. (2004a). Sustained benefit of supportive intervention for depression symp toms in caregivers of patient with Alzheimer's disease. American Journal of Psychiatry, 161 1-7. Mittelman, M. S., Roth, D. L., Haley, W. E., & Zarit, S. H. (2004b). Effects of a caregiver intervention on negative caregiver appraisals of behavior problems in patients with Alzheimer's disease: Results of a randomized trial. Journals of Gerontology B Series: Psychological Sciences and Social Sciences, 59 (1), P27-34. Mok, W. Y., Chu, L. W., Chung, C. P., Chan, N. Y., & Hui, S. L. (2004). The relationship between non-cognitive symptoms and functional impairment in Alzheimer's disease. International Journal of Geriatric Psychiatry, 19 (11), 1040-1046. Motohashi, Y., Maeda, A., Wakamatsu, H., Higuc hi, S., & Yuasa, T. (2000). Circadian rhythm abnormalities of wrist activity of institutionalized dependent elderly persons with dementia. Journals of Gerontology A Series: Biol ogical Sciences and Medical Sciences, 55(12), M740-743. Murman, D. L., Chen, Q., Powell, M. C., Kuo, S. B., Bradley, C. J., & Colenda, C. C. (2002). The incremental direct costs associated with behavioral symptoms in AD. Neurology, 59(11), 1721-1729. Musil, C. M., Haug, M. R., & Warner, C. D. (1998) Stress, health, and de pressive symptoms in older adults at three ti me points over 18 months. Issues in Mental Health Nursing, 19 (3), 207-224. Myers, J. K., & Weissman, M. M. (1980). Use of a self-report symptom scale to detect depression in a community sample. American Journal of Psychiatry, 137 (9), 1081-1084.
185 Nagatomo, I., Akasaki, Y., Uchida, M., Tomina ga, M., Hashiguchi, W., & Takigawa, M. (1999). Gender of demented patients and specific fam ily relationship of caregiver to patients influence mental fatigue and burde ns on relatives as caregivers. International Journal of Geriatric Psychiatry, 14 (8), 618-625. Neundorfer, M. M., McClendon, M. J., Smyth, K. A ., Strauss, M. E., & McCallum, T. J. (2006). Does depression prior to caregiving increase vulnerability to depressive symptoms among caregivers of persons with Alzheimer's disease? Aging and Mental Health, 10 (6), 606615. Neundorfer, M. M., McClendon, M. J., Smyth, K. A ., Stuckey, J. C., Strauss, M. E., & Patterson, M. B. (2001). A longitudinal study of the re lationship between levels of depression among persons with Alzheimer's disease and levels of depression among their family caregivers. Journals of Gerontology B Series: Psychological Sciences and Social Sciences, 56 (5), P301-313. Newsom, J. T., Rook, K. S., Nishishiba, M ., Sorkin, D. H., & Mahan, T. L. (2005). Understanding the relative importance of positive and negative social exchanges: Examining specific domains and appraisals. Journals of Gerontology B Series: Psychological Sciences and Social Sciences, 60 (6), P304-P312. Nobili, A., Riva, E., Tettamanti, M., Lucca, U., Liscio, M., Petrucci, B., et al. (2004). The effect of a structured intervention on caregivers of patients with dementia and problem behaviors: A randomized controlled pilot study. Alzheimer Disease and Associated Disorders, 18 (2), 75-82. Normann, H. K., Asplund, K., & Norberg, A. (1998). Episodes of lucidity in people with severe dementia as narrated by formal carers. Journal of Advanced Nursing, 28 (6), 1295-1300. Normann, H. K., Henriksen, N., Norberg, A., & Asplund, K. (2005). Lucidity in a woman with severe dementia related to conversation. A case study. Journal of Clinical Nursing, 14 (7), 891-896. O'Connor, D. W., Pollitt, P. A., Hyde, J. B., Fellow s, J. L., Miller, N. D., Brook, C. P., et al. (1989). The reliability and validity of the mini-mental state in a British community survey. Journal of Psychiatric Research, 23 (1), 87-96. O'Rourke, N., & Tuokko, H. A. (2003). Psychometri c properties of an ab ridged version of the Zarit burden interview within a representative Canadian caregiver sample. Gerontologist, 43(1), 121-127. Oliver, D., Connelly, J. B., Vict or, C. R., Shaw, F. E., Whiteh ead, A., Genc, Y., et al. (2007). Strategies to prevent falls and fractures in hospitals and care homes and effect of cognitive impairment: Systematic review and meta-analyses. BMJ: Clinical Research ed., 334(7584), 82.
186 Onishi, J., Suzuki, Y., Umegaki, H., Nakamura, A., Endo, H., & Iguchi, A. (2005). Influence of behavioral and psychological symptoms of dementia (BPSD) and environment of care on caregivers' burden. Archives of Gerontol ogy and Geriatrics, 41 (2), 159-168. Opie, J., Rosewarne, R., & O'Connor, D. W. ( 1999). The efficacy of psychosocial approaches to behaviour disorders in dementia: A systematic literature review. Australian and New Zealand Journal of Psychiatry, 33 (6), 789-799. Ory, M. G., Hoffman, R. R., 3rd, Yee, J. L., Te nnstedt, S., & Schulz, R. (1999). Prevalence and impact of caregiving: A detailed comparison between dementia and nondementia caregivers. Gerontologist, 39 (2), 177-185. Ostwald, S. K., Hepburn, K. W., Caron, W., Burns, T., & Mantell, R. (19 99). Reducing caregiver burden: A randomized psychoeducational interv ention for caregivers of persons with dementia. Gerontologist, 39 (3), 299-309. Papastavrou, E., Kalokerinou, A., Papacostas, S. S., Tsangari, H., & Sourtzi, P. (2007). Caring for a relative with dementia: Family caregiver burden. Journal of Advanced Nursing, 58(5), 446-457. Parks, S. H., & Pilisuk, M. (1991). Caregiver burden: Gender and the psychological costs of caregiving. American Journal of Orthopsychiatry, 61 (4), 501-509. Pastor, Y., Balaguer, I., Pons, D ., & Garcia-Merita, M. (2003). Testing direct and indirect effects of sports participation on perceived health in Spanish adolescents between 15 and 18 years of age. Journal of Adolescence, 26 (6), 717-730. Pearlin, L. I., Mullan, J. T., Semple, S. J., & Skaff, M. M. (1990). Caregiving and the stress process: An overview of concepts and their measures. Gerontologist, 30 (5), 583-594. Perneczky, R., Wagenpfeil, S., Komossa, K., Gr immer, T., Diehl, J., & Kurz, A. (2006). Mapping scores onto stages: Mini-mental state examination and clinical dementia rating. American Journal of Geriatric Psychiatry, 14 (2), 139-144. Pinquart, M., & Sorensen, S. (2003a ). Associations of stressors and uplifts of caregiving with caregiver burden and depressi ve mood: A meta-analysis. Journals of Gerontology B Series: Psychological Scien ces and Social Sciences, 58 (2), P112-128. Pinquart, M., & Sorensen, S. (2003b). Differences between caregivers and noncaregivers in psychological health and physical health: A meta-analysis. Psychology and Aging, 18 (2), 250-267. Pinquart, M., & Sorensen, S. (2004). Associations of caregiver stressors and uplifts with subjective well-being and depressive mood: A meta-analytic comparison. Aging and Mental Health, 8 (5), 438-449.
187 Pinquart, M., & Sorensen, S. (2006). Helping ca regivers of persons with dementia: Which interventions work and how large are their effects? International Psyc hogeriatrics, 18 (4), 577-595. Pinquart, M., & Sorensen, S. (2007). Correlates of physical health of informal caregivers: A meta-analysis. Journals of Gerontology B Series: Psychological Sciences and Social Sciences, 62 (2), P126-137. Powers, D. V., Gallagher-Thompson, D., & Kraem er, H. C. (2002). Coping and depression in Alzheimer's caregivers: Longit udinal evidence of stability. Journals of Gerontology B Series: Psychological Scien ces and Social Sciences, 57 (3), P205-211. Quayhagen, M. P., & Quayhagen, M. (1996). Disc overing life quality in coping with dementia. Western Journal of Nursing Research, 18(2), 120-135. Quayhagen, M. P., Quayhagen, M., Corbeil, R. R., Hendrix, R. C., Jackson, J. E., Snyder, L., et al. (2000). Coping with dementia: Evaluati on of four nonpharmacologic interventions. International Psychogeriatrics, 12 (2), 249-265. Quayhagen, M. P., Quayhagen, M., Corbeil, R. R., Roth, P. A., & Rodgers, J. A. (1995). A dyadic remediation program for car e recipients with dementia. Nursing Research, 44(3), 153-159. Radloff, L. S. (1977). The CES-D scale: A self-report depression scale for research in the general population. Applied Psychology Measurement, 1 (3), 385-401. Rapp, M. A., Rieckmann, N., Gutzmann, H., & Folstein, M. F. (2002). [micro-mental test a short method of dementia screening]. Der Nervenarzt, 73 (9), 839-844. Reisberg, B., Borenstein, J., Salob, S. P., Ferri s, S. H., Franssen, E., & Georgotas, A. (1987). Behavioral symptoms in Alzheimer's disease: Phenomenology and treatment. Journal of Clinical Psychiatry, 48 Suppl, 9-15. Riggs, J. A. (2001). The health and long-term car e policy challenges of Alzheimer's disease. Aging and Mental Health, 5 (Suppl. 1), S138-S145. Rose-Rego, S. K., Strauss, M. E., & Smyth, K. A. (1998). Differences in the perceived wellbeing of wives and husbands caring for persons with Alzheimer's disease. Gerontologist, 38(2), 224-230. Roth, D. L., Haley, W. E., Owen, J. E., Clay, O. J., & Goode, K. T. (2001). Latent growth models of the longitudinal effects of deme ntia caregiving: A comparison of African American and white family caregivers. Psychology and Aging, 16 (3), 427-436. Rowe, M. (2003a). Night alert prompting system. University of Florida Health Science Center Institutional Review Board Project # 345-2003.
188 Rowe, M. (2003b). People with dementia who become lost. American Journal of Nursing, 103(7), 32-39; quiz 40. Rowe, M., & Bennett, V. (2003). A look at deaths o ccurring in persons with dementia lost in the community. American Journal of Alzheimer's Disease and Other Dementias, 18 (6), 343352. Rowe, M., & Fehrenbach, N. (2004). Injuries su stained by community-dwelling individuals with dementia. Clinical Nursing Research, 13 (2), 98-110, discussion 111-116. Rowe, M., Feinglass, N. G., & Wiss, M. E. (2004) Persons with dementia who become lost in the community: A case study, curr ent research, and recommendations. Mayo Clinic Proceedings, 79(11), 1417-1422. Rowe, M., & Glover, J. C. (2001). Antecedents, descriptions and consequences of wandering in cognitively impaired adults and the Safe Return (SR) program. American Journal of Alzheimer's Disease and Other Dementias, 16 (6), 344-352. Rowe, M., Kelly, A., Horne, C., Lane, S., Ca mpbell, J. M., Lehman, B., et al. (2008). Unpublished manuscript. Reducing dangerous ni ghttime events in persons with dementia using a nighttime monitoring system. Rowe, M., Lane, S., & Phipps, C. (2007). Carewatch: A home monitoring system for use in homes of persons with cognitive impairment. Topics in Geriatric Rehabilitation, 23 (1), 38. Savundranayagam, M. Y., Hummert, M. L., & Montgomery, R. J. (2005). Investigating the effects of communication problems on caregiver burden. Journals of Gerontology B: Psychological Sciences and Social Sciences, 60 (1), S48-55. Schafer, J. L. (2007). Personal co mmunication regarding missing data. Schafer, J. L., & Graham, J. W. (2002). Missi ng data: Our view of th e state of the art. Psychological Methods, 7 (2), 147-177. Schmand, B., Lindeboom, J., Hooijer, C., & Jonk er, C. (1995). Relation between education and dementia: The role of test bias revisited. Journal of Neurology, Neurosurgery, and Psychiatry, 59 (2), 170-174. Schroevers, M. J., Sanderman, R., van Sonderen, E., & Ranchor, A. V. (2000). The evaluation of the center for epidemiologic studies depression (CES-D) scale: Depressed and positive affect in cancer patients and healthy reference subjects. Quality of Life Research, 9 (9), 1015-1029. Schulz, R. (2001). Some critical issues in caregiver intervention research. Aging and Mental Health, 5(Supplement 1), S112-S115.
189 Schulz, R., & Beach, S. R. (1999). Caregiving as a risk fa ctor for mortality: The caregiver health effects study. Journal of the American Medical Association, 282 (23), 2215-2219. Schulz, R., Belle, S. H., Czaja, S. J., McGinni s, K. A., Stevens, A., & Zhang, S. (2004). Longterm care placement of dementia patient s and caregiver health and well-being. Journal of the American Medical Association, 292 (8), 961-967. Schulz, R., Burgio, L., Burns, R., Eisdorfer, C., Gallagher-Thompson, D., Gitlin, L. N., et al. (2003). Resources for enhancing Alzheimer's car egiver health (REACH): Overview, sitespecific outcomes, and future directions. Gerontologist, 43 (4), 514-520. Schulz, R., & Martire, L. M. (2004). Family car egiving of persons with dementia: Prevalence, health effects, and support strategies. American Journal of Geriatric Psychiatry, 12 (3), 240-249. Schulz, R., Newsom, J., Mittelmark, M., Burton, L., Hirsch, C., & Jackson, S. (1997). Health effects of caregiving: The car egiver health effects study: An ancillary study of the cardiovascular health study. Annals of Behavioral Medicine, 19 (2), 110-116. Schulz, R., O'Brien, A. T., Bookwala, J., & Fl eissner, K. (1995). Psychiatric and physical morbidity effects of dementia caregivi ng: Prevalence, correlates, and causes. Gerontologist, 35 (6), 771-791. Sebern, M. (2005). Shared care, elder and fam ily member skills used to manage burden. Journal of Advanced Nursing, 52 (2), 170-179. Shaffer, D. R., Dooley, W. K., & Williams on, G. M. (2007). Endorsement of proactively aggressive caregiving strategies moderates th e relation between caregiver mental health and potentially harmful caregiving behavior. Psychology and Aging, 22 (3), 494-504. Shah, A., & Allen, H. (1999). Is improvement possible in the measurement of behaviour disturbance in dementia? International Journal of Geriatric Psychiatry, 14 (7), 512-519. Shurgot, G. R., & Knight, B. G. (2005). Influence of neuroticism, ethnicit y, familism, and social support on perceived burden in dementia caregive rs: Pilot test of the transactional stress and social support model. Journals of Gerontology B Series: Psychological Sciences and Social Sciences, 60 (6), P331-P334. Silvestri, A., Rosano, G., Zannino, G., Ricca, F., Marigliano, V., & Fini, M. (2004). Behavioral disturbances in Alzheimer's disease: A non-pharmacological therapeutic approach. Archives of Geront ology and Geriatrics (Suppl 1), 379-386. Singer, J. D. (1998). Using SAS proc mixed to fit multilevel models, hierarchical models, and individual growth models. Journal of Educational and Behavioral Statistics, 24 (4), 323355. Singer, J. D., & Willett, J. B. (2003). Applied longitudinal data analysis: Modeling change and event occurrence New York, New York: Oxford University Press.
190 Sink, K. M., Covinsky, K. E., Barnes, D. E., Ne wcomer, R. J., & Yaffe, K. (2006). Caregiver characteristics are associated with ne uropsychiatric symptoms of dementia. Journal of American Geriatric Society, 54 (5), 796-803. Sink, K. M., Holden, K. F., & Yaffe, K. (2005). Pharmacological treatment of neuropsychiatric symptoms of dementia: A review of the evidence. Journal of the American Medical Association, 293(5), 596-608. Small, G. W., Rabins, P. V., Barry, P. P., Buckho ltz, N. S., DeKosky, S. T., Ferris, S. H., et al. (1997). Diagnosis and treatment of Alzheimer disease and related disorders. Consensus statement of the American Association fo r Geriatric psychiatry, the Alzheimer's Association, and the American Geriatrics Society. Journal of the American Medical Association, 278(16), 1363-1371. Smith, M., & Buckwalter, K. (2005). Be haviors associated with dementia. American Journal of Nursing, 105(7), 40-52; quiz 53. Smith, M., Gerdner, L. A., Hall, G. R., & Buck walter, K. C. (2004). History, development, and future of the progressively lowered stress threshold: A conceptual model for dementia care. Journal of the Americ an Geriatric Society, 52 (10), 1755-1760. Smith, M., Hall, G. R., Gerdner, L., & Buckwalter, K. C. (2006). Application of the progressively lowered stress threshold model across the continuum of care. Nursing Clinics of North America, 41 (1), 57-81, vi. Snow, L. A., Hovanec, L., & Brandt, J. (2004). A controlled trial of arom atherapy for agitation in nursing home patients with dementia. Journal of Alternative and Complementary Medicine, 10 (3), 431-437. Snyder, J. R. (2000). Impact of caregiver-receive r relationship quality on burden and satisfaction. Journal of Women and Aging? 12 (1-2), 147-167. Son, J., Erno, A., Shea, D. G., Femia, E. E., Zarit, S. H., & Stephens, M. A. (2007). The caregiver stress process and health outcomes. Journal of Aging and Health, 19(6), 871887. Sorensen, S., Pinquart, M., & Duberstein, P. (2 002). How effective are interventions with caregivers? An updated meta-analysis. Gerontologist, 42 (3), 356-372. Spitznagel, M. B., Tremont, G., Davis, J. D., & Foster, S. M. (2006). Psyc hosocial predictors of dementia caregiver desire to institutionaliz e: Caregiver, care recipient, and family relationship factors. Journal of Geriatric Psychiatry and Neurology, 19 (1), 16-20. Spruytte, N., Van Audenhove, C., Lammertyn, F., & Storms, G. (2002). The quality of the caregiving relationship in informal care for older adults with dementia and chronic psychiatric patients. Psychological Psychotherapeutics, 75 (Pt 3), 295-311.
191 Steele, H., Phibbs, E., & Woods, R. T. (2004). Coherence of mind in daughter caregivers of mothers with dementia: Links with their mothers' joy and relatedness on reunion in a strange situation. Attachment and Human Development, 6 (4), 439-450. Steinberg, M., Sheppard, J., Tschanz, J. T., Nort on, M. C., Steffens, D. C., Breitner, J., et al. (2003). The incidence of mental and behavior al disturbances in dementia: The Cache County study. The Journal of Neuropsychiatry and Clinical Neurosciences, 15 (3), 340345. Steinberg, M., Tschanz, J. T., Norton, M., Breitner J. C. S., & Lyketsos, C. G. (2006). Point and 5-year period prevalence of neuropsychiat ric symptoms in dementia: The Cache County study. Poster, International Conference on Alzheimer's Disease Madrid. Tarrier, N., Barrowclough, C., Ward, J., Dona ldson, C., Burns, A., & Gregg, L. (2002). Expressed emotion and attributions in the ca rers of patients with Alzheimer's disease: The effect on carer burden. Journal of Abnormal Psychology, 111 (2), 340-349. Taylor, D. H., Jr., & Sloan, F. A. (2000). How mu ch do persons with Alzheimer's disease cost Medicare? Journal of the American Geriatric Society, 48 (6), 639-646. Teel, C. S., & Press, A. N. ( 1999). Fatigue among elde rs in caregiving and noncaregiving roles. Western Journal of Nursing Research, 21(4), 498-514; discussion 514-420. Teri, L. (1997). Behavior and caregiver burden: Behavioral problems in patients with Alzheimer disease and its association with caregiver distress. Alzheimer Disease and Associated Disorders, 11 Suppl 4, S35-38. Teri, L. (1999). Training families to provi de care: Effects on people with dementia. International Journal of Geriatric Psychiatry, 14 (2), 110-116; discussion 116-119. Teri, L., Gibbons, L. E., McCurry, S. M., Logsdon, R. G., Buchner, D. M., Barlow, W. E., et al. (2003). Exercise plus behavioral management in patients with Alzheimer's disease. Journal of the American Medical Association, 290 2015-2022. Teri, L., McCurry, S. M., Logsdon, R., & Gibbons, L. E. (2005). Training community consultants to help family members impr ove dementia care: A randomized controlled trial. Gerontologist, 45 (6), 802-811. Thomas, P., Chantoin-Merlet, S., Hazif-Thomas, C., Belmin, J., Mont agne, B., Clement, J. P., et al. (2002). Complaints of informal caregivers providing home care for dementia patients: The PIXEL study. International Journal of Geriatric Psychiatry, 17 (11), 1034-1047. Thomas, P., Lalloue, F., Preux, P. M., Hazif-Thom as, C., Pariel, S., Insc ale, R., et al. (2006). Dementia patients caregivers qua lity of life: The PIXEL study. International Journal of Geriatric Psychiatry, 21 (1), 50-56.
192 Thommessen, B., Aarsland, D., Braekhus, A., Okse ngaard, A. R., Engedal, K., & Laake, K. (2002). The psychosocial burden on spouses of the elderly with stroke, dementia and Parkinson's disease. International Journal of Geriatric Psychiatry, 17 (1), 78-84. Tombaugh, T. N., & McIntyre, N. J. (1992) The mini-mental state examination: A comprehensive review. Journal of the American Geriatric Society, 40 (9), 922-935. Torti, F. M., Jr., Gwyther, L. P., Reed, S. D ., Friedman, J. Y., & Schulman, K. A. (2004). A multinational review of recent trends a nd reports in dementia caregiver burden. Alzheimer Disease and Associated Disorders, 18 (2), 99-109. Touhy, T. (2004). Dementia, personhood, and nursi ng: Learning from a nursing situation. Nursing Science Quarterly, 17 (1), 43-49. Tractenberg, R. E., Weiner, M. F., & Thal, L. J. (2002). Estimating the prevalence of agitation in community-dwelling persons with Alzheimer's disease. Journal of Neuropsychiatry and Clinical Neurosciences, 14 (1), 11-18. Upton, N., & Reed, V. (2006). The influen ce of social support on caregiver coping. International Journal of Psychiatric Nursing Research, 11(2), 1256-1267. van Doorn, C., Gruber-Baldini, A. L., Zimmerman, S ., Hebel, J. R., Port, C. L., Baumgarten, M., et al. (2003). Dementia as a risk factor for falls and fall injuries among nursing home residents. Journal of American Geriatric Society, 51 (9), 1213-1218. Vitaliano, P. P., Young, H. M., Russo, J., Ro mano, J., & Magana-Amato, A. (1993). Does expressed emotion in spouses predict subsequent problems among care recipients with Alzheimer's disease? Journals of Gerontology, 48 (4), P202-209. Vitaliano, P. P., Zhang, J., & Scanlan, J. M. (2003). Is caregiving hazardous to one's physical health? A meta-analysis. Psychological Bulletin, 129 (6), 946-972. Walker, L. O., & Avant, K. C. (2005). Strategies for theory cons truction in nursing, 4th ed. Upper Saddle River, New Jers ey: Pearson Education, Inc. Ward, S., Opie, J., & O'Connor, D. W. (2003). Fa mily carers' responses to behavioural and psychological symptoms of dementia. International Journal of Geriatric Psychiatry, 18(11), 1007-1012. Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology, 54(6), 1063-1070. Weissman, M. M., Sholomskas, D., Pottenger, M ., Prusoff, B. A., & Locke, B. Z. (1977). Assessing depressive symptoms in five psychiatric populations: A validation study. American Journal of Epidemiology, 106 (3), 203-214.
193 Whitlatch, C. J. (2001). Including the person with dementia in family care-giving research. Aging and Mental Health, 5 (Supp 1), S20-S22. Whitlatch, C. J., Greene, R., Judge, K., Femia, E., & Zarit, S. H. (2006a, November). Developing and evaluating a dyadic intervention for family caregivers and persons with early stage dementia. Paper presented at the 58th Annual Scientific Meeting of the Gerontological Society of America, Orlando, FL. Whitlatch, C. J., Judge, K., Zarit, S. H., & Femi a, E. (2006b). Dyadic intervention for family caregivers and care receivers in early-stage dementia. Gerontologist, 46 (5), 688-694. Whitlatch, C. J., Zarit, S. H., & von Eye, A. (1991). Efficacy of interventions with caregivers: A reanalysis. Gerontologist, 31 (1), 9-14. Wijeratne, C. (1997). Review: Pathways to morbidity in carers of dementia sufferers. International Psychogeriatrics, 9 (1), 69-79. Wilkins, S. S., Castle, S., Heck, E., Tanzy, K., & Fahey, J. (1999). Im mune function, mood, and perceived burden among caregivers participa ting in a psychoeducat ional intervention. Psychiatric Services, 50 (6), 747-749. Willette-Murphy, K., Todero, C., & Yeaworth, R. ( 2006). Mental health and sleep of older wife caregivers for spouses with Alzheime r's disease and related disorders. Issues in Mental Health Nursing, 27(8), 837-852. Williams, J. R., Jr. (2005). Depression as a medi ator between spousal be reavement and mortality from cardiovascular disease: Apprecia ting and managing the adverse health consequences of depression in an elderly surviving spouse. Southern Medical Journal, 98(1), 90-95. Williamson, G. M., & Shaffer, D. R. (2001). Relationship quality and potentially harmful behaviors by spousal caregivers: How we were then, how we are now. The family relationships in late life project. Psychology and Aging, 16 (2), 217-226. Williamson, G. M., Shaffer, D. R., & Schulz, R. (1998). Activity restriction and prior relationship history as contributors to mental health outcomes among middle-aged and older spousal caregivers. Health Psychology, 17 (2), 152-162. Wolf, R. S. (1998). Caregiver stress, Al zheimer's disease, and elder abuse. American Journal of Alzheimer's Disease(March/April), 81-83. Woods, R. T. (2001). Discovering the person with Alzheimer's disease: Cognitive, emotional and behavioral aspects. Aging and Mental Health, 5 (Supp 1), S7-S16. Yaffe, K., Fox, P., Newcomer, R., Sands, L., Lindquist, K., Dane, K., et al. (2002). Patient and caregiver characteristics and nursing home placement in patients with dementia. Journal of the American Medical Association, 287 (16), 2090-2097.
194 Zanetti, O., Frisoni, G. B., Bianchetti, A., Ta manza, G., Cigoli, V., & Trabucchi, M. (1998). Depressive symptoms of Alzheimer caregivers are mainly due to personal rather than patient factors. International Journal of Geriatric Psychiatry, 13 (6), 358-367. Zarit, S. H., Femia, E. E., Watson, J., Rice-Oe schger, L., & Kakos, B. (2004). Memory club: A group intervention for people with early-sta ge dementia and their care partners. Gerontologist, 44 (2), 262-269. Zarit, S. H., & Leitsch, S. A. (2001). Developing and evaluating commun ity based intervention programs for Alzheimer's patients and their caregivers. Aging and Mental Health, 5 (Supp 1), S84-S98. Zarit, S. H., Orr, N. K., & Zarit, J. M. (1985). The hidden victims of Alzheimer's disease: Families under stress. New York: New York University Press. Zarit, S. H., Reever, K. E., & Bach-Peterson, J. (1980). Relatives of the impaired elderly: Correlates of feelings of burden. Gerontologist, 20 (6), 649-655. Zarit, S. H., Todd, P. A., & Zarit, J. M. ( 1986). Subjective burden of husbands and wives as caregivers: A longitudinal study. Gerontologist, 26 (3), 260-266.
195 BIOGRAPHICAL SKETCH Judy Ca mpbell was born in Mississippi in 1957, the second child of Mr. and Mrs. Wilbur A. McGee. She graduated from Weir High School in 1975, and attended Holmes Junior College in Goodman before completing her Bachelor of Science in Nursing at the University of Southern Mississippi in 1979. Judy has practiced as a Registered Nurse in medical-surgical, gerontological, and gynecological units in Ne braska (1979-83) and Florida (1983-1994). Judy took the role of Nurse Educator at Brevard Community College in 1990. She received tuition assistance and support to complete her Masters degree, awarded in 1995 at the University of Florida. Judy received the Adva nced Registered Nurse Practitioner Certificate on graduation, as a Gerontological NP. Judy returned to education and taught for 14 years at before pursuing her doctoral education. Judy attended the University of Florida from 2004-2008, and was awarded the UF Alumni Fellowship, the Hartford Building Academic Geriatric Nursing Capacity Scholarship, a position as a UF Gerontology Research Trainee (nonfunded), and the UF College of Nursing Outstanding Research Award. She has been awarded membership in the National Scholars Honor Society, Sigma Phi Omega (gerontology h onor society), and Sigma Theta Tau (nursing honor society). Judy has a long history of professional activity and community service within nursing, primarily through the Florida Nurses Association a nd as a mentor at the local and state level for the Florida Nursing Students Association. She is a member of the Southern Nursing Research Society, the Gerontological Soci ety of America, and the American Nurses Association. Judy has been married to David Campbell sin ce 1979. She also has two adult children, one granddaughter, and two nephews who have blessed her life. Judy has lived in Brevard County, Floridas Space Coast, in Cocoa (Port St. John), since 1983.