<%BANNER%>

Patient Falls in Acute Care Hospitals

Permanent Link: http://ufdc.ufl.edu/UFE0044853/00001

Material Information

Title: Patient Falls in Acute Care Hospitals A Longitudinal Assessment of Nurse Staffing and Unit-Level Characteristics
Physical Description: 1 online resource (228 p.)
Language: english
Creator: Everhart, Damian M
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2012

Subjects

Subjects / Keywords: falls -- hospitals -- nurse -- staffing
Health Services Research, Management, and Policy -- Dissertations, Academic -- UF
Genre: Health Services Research thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Patient falls in acute care hospitals represent a significant and growing public health concern in the United States. Inpatient falls adversely impact patients, health care professionals, organizations and policymakers through increased health care expenditures and decreased quality of patient care. However, prior empirical studies in hospitals have provided inconclusive evidence on interventions that influence patient falls. Given the significant variation in fall rates in hospitals and federal policy regulations that no longer reimburse hospitals for injuries from patient falls, our research examines acute care hospital factors that are associated with patient fall rates, particularly assessing the organizational and unit-level nurse staffing factors that predict fall rates in hospitals over time. This is a retrospective longitudinal study design from July 2006 through December of 2010 at the unit-level using monthly data from the National Database for Nursing Quality Indicators (NDNQI). Latent class growth modeling (LCGM) was employed to determine latent class trajectories of hospital and unit-level fall rates over time. Findings indicated that hospitals were categorized into three linear trajectory groups of ‘consistently high, medium and low’ based on their fall rates over 54 months. Additionally, there was significant variation in fall rates among hospital units. Generalized estimating equations (GEE) was then conducted to understand nurse staffing and organizational factors that predict membership into the ‘consistently high’ fall rate trajectory group, as well as nurse staffing and organizational factors that influence overall fall rates over time. Findings from GEE analyses revealed that greater levels of registered nurse (RN) staffing and hospital Magnet status were significantly associated with reductions in fall rates and a decreased likelihood of membership in the ‘consistently high’ fall rate trajectory group. Unexpected positive associations between licensed practical nurse (LPN) staffing and nursing assistant (UAP) staffing were found with patient falls. Policymakers and administrators may be able to reduce fall rates by maintaining greater RN staffing ratios as well as fostering an environment consistent with that of Magnet hospitals. Future research is also needed to understand contextual factors in Magnet hospitals that impact patient fall rates.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Damian M Everhart.
Thesis: Thesis (Ph.D.)--University of Florida, 2012.
Local: Adviser: Duncan, R. P.
Local: Co-adviser: Schumacher, Jessica Rachael.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2012
System ID: UFE0044853:00001

Permanent Link: http://ufdc.ufl.edu/UFE0044853/00001

Material Information

Title: Patient Falls in Acute Care Hospitals A Longitudinal Assessment of Nurse Staffing and Unit-Level Characteristics
Physical Description: 1 online resource (228 p.)
Language: english
Creator: Everhart, Damian M
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2012

Subjects

Subjects / Keywords: falls -- hospitals -- nurse -- staffing
Health Services Research, Management, and Policy -- Dissertations, Academic -- UF
Genre: Health Services Research thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Patient falls in acute care hospitals represent a significant and growing public health concern in the United States. Inpatient falls adversely impact patients, health care professionals, organizations and policymakers through increased health care expenditures and decreased quality of patient care. However, prior empirical studies in hospitals have provided inconclusive evidence on interventions that influence patient falls. Given the significant variation in fall rates in hospitals and federal policy regulations that no longer reimburse hospitals for injuries from patient falls, our research examines acute care hospital factors that are associated with patient fall rates, particularly assessing the organizational and unit-level nurse staffing factors that predict fall rates in hospitals over time. This is a retrospective longitudinal study design from July 2006 through December of 2010 at the unit-level using monthly data from the National Database for Nursing Quality Indicators (NDNQI). Latent class growth modeling (LCGM) was employed to determine latent class trajectories of hospital and unit-level fall rates over time. Findings indicated that hospitals were categorized into three linear trajectory groups of ‘consistently high, medium and low’ based on their fall rates over 54 months. Additionally, there was significant variation in fall rates among hospital units. Generalized estimating equations (GEE) was then conducted to understand nurse staffing and organizational factors that predict membership into the ‘consistently high’ fall rate trajectory group, as well as nurse staffing and organizational factors that influence overall fall rates over time. Findings from GEE analyses revealed that greater levels of registered nurse (RN) staffing and hospital Magnet status were significantly associated with reductions in fall rates and a decreased likelihood of membership in the ‘consistently high’ fall rate trajectory group. Unexpected positive associations between licensed practical nurse (LPN) staffing and nursing assistant (UAP) staffing were found with patient falls. Policymakers and administrators may be able to reduce fall rates by maintaining greater RN staffing ratios as well as fostering an environment consistent with that of Magnet hospitals. Future research is also needed to understand contextual factors in Magnet hospitals that impact patient fall rates.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Damian M Everhart.
Thesis: Thesis (Ph.D.)--University of Florida, 2012.
Local: Adviser: Duncan, R. P.
Local: Co-adviser: Schumacher, Jessica Rachael.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2012
System ID: UFE0044853:00001


This item has the following downloads:


Full Text

PAGE 1

1 PATIENT FALLS IN ACUTE CARE HOSPITALS: A LONGITUDINAL ASSESSMENT OF NURSE STAFFING AND UNIT LEVEL CHARACTERISTICS By DAMIAN MICHAEL EVERHART A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PAR TIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2012

PAGE 2

2 2012 Damian M ichael Everhart

PAGE 3

3 I dedicate this dissertation to my mother and father for their love, support and guidance throughout my life I love you both and none of this would be possible with you

PAGE 4

4 ACKNOWLEDGMENTS I would like to extend my sincere gratitude to all of my committee members for their mentorship during the dissertation process. Dr. Schumacher, I cannot thank you e nough for being my co chair and meeting with me every we ek to plan, implement and revise my dissertation Your guidance and encouragement during this time has been critical to my academic success. I would also like to ex tend my sincere thanks to my co chai r, Dr. Duncan who has been a tremendous source of guidance and support during the entire PhD program My thanks also go to Dr. Hall Dr. Shorr and Dr. Neff, who all have been a critical part of my doctoral studies over the past three years. I look forward to working with all of you in my future professional academic career. Since it takes a village to complete a dissertation, I would also like to thank my colleagues and classmates in the Health Services Research PhD program. Throughout the pro gram, you a ll have been integral in preparing for courses and proj ects, especially studying for the preliminary examination. I w ill never forget the study sessions, our spring break trip to Key West and your words of encouragement during times of hardship. I am al so forever grateful for my family and friends for their love, support and patience with me during these difficult years. Thank you all very much for putting up with me during this process. A s pecial thanks to my parents and my brother, N ick, for their love and encouragement I also want to thank my girlfriend Meghan for her love and support, as well as always being a significant source of love and strength I am so thankful t hat you were there for me every day during this time.

PAGE 5

5 Finally, I want to thank th e faculty and staff of the Department of Health Services Research, Management and Policy at the University of Florida for all that they do to make this a special program.

PAGE 6

6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 9 LIST OF FIGURES ................................ ................................ ................................ ........ 11 LIST OF ABBREVIATIONS ................................ ................................ ........................... 13 ABSTRACT ................................ ................................ ................................ ................... 15 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 17 2 BACKGROUND AND SIGNIFICANCE ................................ ................................ ... 23 Falls in Acute Care Hos pitals ................................ ................................ .................. 24 Impact of Hospital Falls on Quality of Patient Care ................................ .......... 25 Impact of Hospital Falls on the Elderly ................................ ............................. 26 Impact of Hospital Falls on Patient Well Being, Family Members and Caregivers ................................ ................................ ................................ ..... 27 Impact of Hospital Falls on Health Care Expenditures ................................ ..... 28 Government Policies and Regulations on Inpatient Falls ................................ 29 Risk Factors and Interventions to Reduce Inpatient Falls ................................ ....... 30 Risk Factors Related to Inpatient Falls ................................ ............................. 31 Interventions to Reduce Falls in Hospitals ................................ ........................ 32 Conclusion ................................ ................................ ................................ ........ 35 The Role of Nurses in Preventing Falls in Acute Care Hospitals ............................ 35 Factors Influencing Nursing Practice ................................ ................................ 37 Federal and State Level Nursing Policies ................................ ........................ 38 The National Quality Forum Falls as a Nurse Sensitive Outcome ................. 40 Hospital Organizational Background ................................ ................................ ....... 41 Organizational Characteristics and Nursing Staffing ................................ ........ 41 Organizational Characteristics and Inpatient Falls ................................ ........... 42 3 LITERATURE REVIEW ................................ ................................ .......................... 45 Nurse Staffing and Quality of Patient Care ................................ ............................. 45 Nurse Staffing and Patient Falls ................................ ................................ ............. 46 Limitations in Previous Research ................................ ................................ ............ 48 Variation in Methodology and Study Design ................................ ..................... 49 Variation in Data Sources ................................ ................................ ................. 51 Measurement and Operational Definition of N urse Staffing .............................. 54 Lack of Consistent Theoretical Frameworks ................................ ..................... 57

PAGE 7

7 Conclusion ................................ ................................ ................................ ........ 58 4 CONCEPTUAL FRAMEWORK AND HYPOTHESES ................................ ............. 60 ................................ ................................ ....................... 60 NSPOM) ............................... 61 Aiken, Clarke & Sloane (2002) Theoretical Model ................................ .................. 62 Nurse Staffing and Patient Fall Conceptual (NSPF) Model Development ............... 62 Structure Factors ................................ ................................ .............................. 63 Process Factors ................................ ................................ ............................... 64 Outcome Factors ................................ ................................ .............................. 65 Patient and Policy Contextual Factors ................................ .............................. 65 Specific Aims ................................ ................................ ................................ .......... 66 Study H ypotheses ................................ ................................ ................................ ... 66 5 DATA AND METHODS ................................ ................................ ........................... 73 The National Dataset for Nursing Quality Indicators (NDNQI) ................................ 73 Measures for Data ................................ ................................ ................................ .. 74 Fall Outcome Variables ................................ ................................ .................... 74 Nurse Staffing Variables ................................ ................................ ................... 75 Hospital Organizational Variables ................................ ................................ .... 76 Unit Level Organizational Variables ................................ ................................ 77 Process of Care Variables ................................ ................................ ................ 78 Study Design and Statistical Methods ................................ ................................ ..... 78 Analysis by Specific Aim ................................ ................................ ......................... 79 Latent Class Growth Modeling (LCGM) ................................ ............................ 79 SAS PROC TRAJ Procedure ................................ ................................ .......... 83 Determining Probability Distributio ns and Trajectory Shape ............................. 84 Determining Number of Latent Class Categories and Final Model Selection ... 85 Approaches to Address Miss ing Longitudinal Data ................................ .......... 88 Steps to Address Specific Aim 1 and 1a ................................ ........................... 89 Step #1: Unadjusted monthly hospital fall model ................................ ....... 90 Step #2: Adjusted monthly hospital fall model ................................ ............ 91 Step #3: Unadjusted monthly hospital unit fall models ............................... 92 Step #4: Adjusted monthly hospital unit fall models ................................ ... 92 Step #5: Missing data models and additional sensitivity analyses ............. 93 Generalized Estimating Equations (GEE) ................................ ......................... 93 GEE analysis #1: Consistently high trajectory group predictors at the hospital unit level ................................ ................................ .................... 96 GEE analysis #2: Total monthly falls at the hospital unit level ................... 99 6 RESULTS ................................ ................................ ................................ ............. 110 Results from Descriptive NDNQI Analysis ................................ ............................ 110 Descriptive Characteristics of Patients Who Fell in the NDNQI ...................... 110 Descriptive Characteristics of Nurse Staffing in the NDNQI ........................... 111

PAGE 8

8 Descriptive Characteristics of Hospital Organizations in the NDNQI .............. 111 Results from Specific Aim 1: Trajectories of Hospital and Hospital Unit Fall Rates ................................ ................................ ................................ ................. 112 Step #1: Unadjusted Monthly Hospital Fall Model ................................ .......... 112 Step #2: Adjus ted Monthly Hospital Fall Model ................................ .............. 115 Step #3: Unadjusted Monthly Hospital Unit Fall Models ................................ 116 Unadjusted medical unit model ................................ ................................ 117 Unadjusted surgical unit model ................................ ................................ 119 Unadjusted medical surgical unit model ................................ .................. 120 Step #4: Adjusted Monthly Hospital Unit Fall Models ................................ ..... 122 Step #5: Multiple Imputation Models and Unadjusted Hospital Model by Quarter ................................ ................................ ................................ ........ 123 Specific Aims 2 & 3: Predictors of Patient Falls ................................ .................... 124 Descriptive Results of Latent Class Groups for GEE Analysis #1 .................. 125 Trajectory Group ................................ ................................ ......................... 125 GEE Analysis #2 Results: Impact of Nurse Staffing and Organizational Factors on Pa tient Falls ................................ ................................ .............. 128 7 DISCUSSION AND CONCLUSION ................................ ................................ ...... 152 Descriptive NDNQI Discussion ................................ ................................ ............. 152 Specific Aim 1 Discussion ................................ ................................ ..................... 153 Specific Aims 2 and 3 Discussion ................................ ................................ ......... 156 Study Limitations & Strengths ................................ ................................ ............... 161 Policy Implications and Future Research ................................ .............................. 165 APPENDIX A ADDITIONAL MAGNET HOSPITAL INFORMATION ................................ ........... 170 B SPECIFIC AIM 1 METHODOLOGICAL AND ANALYTIC CONSIDERATIONS .... 173 C SENSITIVITY ANALYSIS MODELS FOR SPECIFIC AIMS 2 AND 3 ................... 206 D SAS PROCEDURE CODES ................................ ................................ ............... 208 LIST OF REFERENCES ................................ ................................ ............................. 211 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 228

PAGE 9

9 LIST OF TABLES Table page 5 1 Variable descriptions ................................ ................................ ........................ 106 5 2 Total fall observation sample size after exclusion c riteria ................................ 108 5 3 Log Bayes approximation factor ................................ ................................ ....... 109 6 1 Fall rates per 1,000 patient days by unit type and total fall rates ...................... 135 6 2 Descriptive characteristics of patients who fell by unit type .............................. 136 6 3 Descriptive characteristics of nurse staffing by unit t ype ................................ .. 137 6 4 Descriptive characteristics of hospital organizations ................................ ........ 138 6 5 Descriptive statistics of hospital observations ................................ ................... 139 6 6 Unadjusted hospital model selection results ................................ ..................... 141 6 7 Unadjusted hospital fall rate model with Poisson maximum likelihood estima tes of three linear trajectories ................................ ................................ 142 6 8 Unadjusted medical unit model selection results ................................ .............. 143 6 9 Unadjusted medical unit fall ra te model with Poisson maximum likelihood estimates of four linear trajectories ................................ ................................ ... 144 6 10 Unadjusted surgical unit model selection results ................................ .............. 145 6 11 Unadjusted surgical unit fall rate model with Poisson maximum likelihood estimates of four linear trajectories ................................ ................................ ... 146 6 12 Unadjusted medical surgical unit model selection resul ts ................................ 147 6 13 Unadjusted medical surgical unit fall rate model with Poisson maximum likelihood estimates of four linear trajectories ................................ ................... 148 6 14 Descriptive characteristics of hospital latent class trajectory groups ................ 149 6 15 Generalized estimating equations predicting membership into the class trajectory group at the hospital unit level ................................ ................................ ................................ .................. 150 6 16 Generalized estimating equations determining the impact of nurse staffing and organizational characteristics on patient falls at the hospital unit level ...... 151

PAGE 10

10 B 1 Descriptive statistics of medical unit observations ................................ ............ 186 B 2 Descriptive statistics of surgical unit observations ................................ ............ 188 B 3 Descriptive statistics of medical surgical unit observations .............................. 190 B 4 Adjusted hospital fall rate model wit h Poisson maximum likelihood estimates of three linear trajectories ................................ ................................ ................. 192 B 5 Unadjusted surgical unit fall rate model with Poisson maximum likelihood estimates of three linear trajectories ................................ ................................ 194 B 6 Adjusted medical unit fall rate model with Poisson maximum likelihood estimates of four linear trajectories ................................ ................................ ... 195 B 7 Adjusted surgical unit fall rate model with Poisson maximum likelihood estimates of two linear trajectories ................................ ................................ ... 197 B 8 Adjusted medical surgical unit fall rate model with Poisson maximum likelihood estim ates of four linear trajectories ................................ ................... 198 B 9 Unadjusted multiple imputation model 1: Poisson maximum likelihood estimates of three hospital fall rate linear trajectories ................................ ....... 200 B 10 Unadjusted multiple imputation model 2: Poisson maximum likelihood estimates of three hospital fall rate linear trajectories ................................ ....... 201 B 11 Unadjusted multiple imputation model 3: Poisson maximum likelihood estimates of three hospital fall rate linear trajectories ................................ ....... 202 B 12 Unadjusted multiple imputation model 4: Poisson maximum likelihood es timates of three hospital fall rate linear trajectories ................................ ....... 203 B 13 Unadjusted multiple imputation model 5: Poisson maximum likelihood estimates of three hospital fall rate linear trajectories ................................ ....... 204 B 14 Unadjusted quarterly hospital fall rate model with Poisson maximum likelihood estimates of three linear trajectories ................................ ................. 205 C 1 Generalized estimating equations predicting membership into the at the hospital level ................. 206 C 2 Ge neralized estimating equations determining the impact of nurse staffing and organizational characteristics on patient falls at the hospital level ............. 207

PAGE 11

11 LIST OF FIGURES Figure page 4 1 Nurse staffing and patient outcomes model ................................ ....................... 70 4 2 Hospital organization, nurse organization and patient outcomes model ............. 71 4 3 Integrate d nurse staffing and patient fall conceptual model ................................ 72 5 1 Probability function and maximum likelihood estimation of the zero inflated Poisson model ................................ ................................ ................................ .. 101 5 2 Unadjusted growth model estimating the average pattern of patient falls among acute care hospitals ................................ ................................ .............. 102 5 3 Adjusted growth model estimating the average pattern of p atient falls among acute care hospitals ................................ ................................ .......................... 103 5 4 Unadjusted growth model estimating the average pattern of patient falls among hospital units ................................ ................................ ......................... 104 5 5 Unadjusted growth model estimating the average pattern of patient falls among hospital units ................................ ................................ ......................... 105 6 1 Unadjusted hospital monthly fall rate trajectory model ................................ ..... 131 6 2 Unadjusted medical unit monthly fall rate trajectory model ............................... 132 6 3 Unadjusted surgical unit monthly fall rate trajectory model ............................... 133 6 4 Unadjusted medical surgical unit monthly fall rate trajectory model ................. 134 B 1 Distribution of fall rates over 54 months in the NDN QI ................................ ..... 173 B 2 A djusted hospital monthly fall rate trajectory model ................................ .......... 174 B 3 Unadjusted medical unit monthly fall rate trajectory mod el with three linear trajectory groups ................................ ................................ ............................... 175 B 4 Unadjusted surgical unit monthly fall rate trajectory model with three linear trajectory groups ................................ ................................ ............................... 176 B 5 Adjusted medical unit monthly fall rate trajectory model with four linear trajectory groups ................................ ................................ ............................... 177 B 6 Adjusted surgical unit monthly fall rate trajectory model with two li near trajectory groups ................................ ................................ ............................... 178

PAGE 12

12 B 7 Adjusted medical surgical unit monthly fall rate trajectory model with four linear trajectory groups ................................ ................................ ..................... 179 B 8 Unadjusted multiple imputation hospital trajectory model 1 .............................. 180 B 9 Unadjusted multiple imputation hospital trajectory model 2 .............................. 181 B 10 Unadjusted multiple imputation hospital trajectory model 3 .............................. 182 B 11 Unadjusted multiple imputation hospital trajectory model 4 .............................. 183 B 12 Unadjusted multiple imputation hospital trajectory model 5 .............................. 184 B 13 Unadjusted quarterly hospital model results with three linear trajectories ........ 185

PAGE 13

13 LIST OF ABBREVIATION S AIC Akaike Information Criteria ANA American Nurses Association ANCC American Nurses Credentialing Center BIC Bayesian Information Criteria CALNOC Collaborative Alliance for Nursing Outcomes CDC Centers for Disease C ontrol and Prevention CDHS California State Department of Health Services CMS Centers for Medicare and Medicaid Services DRG Diagnosis Related Group GEE Generalized Estimating Equation GLM General Linear Modeling HPPD Hours per Patient Day IOM Institute of Medicine LCGM Latent Class Growth Model LGCM Latent Growth Curve Model LPN Licensed Practical Nurse MAR Missing at Random MCAR Missing Completely at Random MNAR Missing Not at Random MSA Metropolitan Statistical Area NDNQI National Database for Nursing Quality Indicators NQF National Quality Forum

PAGE 14

14 NR A Nurse Reinvestment Act OLS Ordinary Least Squares SEM Structural Equation Modeling SPO Structure Process Outcomes RN Registered Nurse UAP Nursing Assistant ZIP Zero Inflated Poisson

PAGE 15

15 Abstract of D issertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy PATIENT FALLS IN ACUTE CARE HOSPITALS: A LONGITUDINAL ASSESSMENT OF NURSE STAFFING AND UNIT LEVEL CHARACTERISTICS By Damian Michael Everhart December 2012 Chair: R. Paul Duncan Cochair: Jessica R. Schumacher Major: Health Services Research P atient falls in acute care hospitals represent a significant and growing public health conce rn in the United States Inpatient falls adversely impact patients health care professionals, organizations and policymakers through increased health care expenditures and decreased quality of patient care However, p rior emp irical studies in hospitals ha ve provided inconclusive evidence on interventions that influen ce patient falls Given the significant variation in fall rates in hospitals and federal policy regulations that no longer reimburse hospitals for injuries from patient falls, our research exam ines acute care hospital factors that are associated with patient fall rates particularly assessing the organizational and u nit level nurse staffing factors that predict fall rates in hospitals over time. This is a retrospective longitudinal study design from July 2006 through December of 2010 at the unit level using monthly data from the National Database for Nursing Quality Indicat ors (NDNQI) Latent class growth modeling (LCGM) was employed to determine latent class trajectories of hospital and unit le vel fall rates over time. Findings indicated that hospitals were categorized into three linear trajectory

PAGE 16

16 groups of ed on their fall rates over 54 months Additionally, there was significant variation in fall rates am ong hospital units. G eneralized estimating equati ons (GEE) was then conducted to understand nurse staffing and organ izational factors that predict as well as nurse staffing and organizati onal factors that influence overall fall rates over time. Findings from GEE an alyse s revealed that greater levels of registered nurse (RN) staffing and hospital Magnet status were significantly associated with reductions in fall rates and a decreased likel ihood of membership in the Unexpected positive associations between l icensed practical nurse (LPN) staffing and nursing assistant (UAP) staffing were found with patient falls P olicymakers and administrators may be able to reduce fall rates by maintaining greater RN staffing ratios as well as fostering an environment consistent with that of Magnet hospitals. Future research is also needed to understand contextual factors in Magnet hospitals that impact patien t fall rates.

PAGE 17

17 CHAPTER 1 INTRODUCTION P atient falls in acute care hospitals represent a significant and growing public health concern in the United States (U.S.) constituting an estimated one million falls per year (Currie, 2008; Oliver et al., 2010; AH RQ, 2009). Falls are the most commonly reported adverse event in hospitals and falls among patients over age 65 are the most frequent cause of non fatal injuries (Morse, 2002; Currie, 2008). Consequences of inpatient falls include increased length of hospi tal stay, greater health care expenditures and increased risk of injury or death for patients (Schwendimann, Buhler, De Geest & Milisen, 2006). Additionally, secondary injuries related to falls account for 15% of re hospitalizations one month after dischar ge, and the total estimated costs of injuries from falls are between $16 to $19 billion (Shekelle et al., 2003; Stevens, Corso, Finkelstein, & Miller, 2006). Other deleterious consequences of falls for patients include psychological distress, activity rest riction and reductions in independent activities of daily living (Tideiksaar, 1993). Moreover, given the rapidly increasing elderly population in the United States, two Institute of Medicine (IOM) reports specifically focused attention on patient safety (I OM, 1999; IOM, 2001) and concluded that hospitals have been unable to sufficiently reduce fall rates in hospitals (Abraham, 2011), underscoring falls as a significant concern for patients, providers and policymakers. Despite the large number of falls in a cute care settings, research on preventing falls has been conducted primarily in rehabilitation units and long term care facilities, and there is a paucity of organizational studies that examine how the acute care structural environment impacts the process es of care and/or the risk of falls (Fischer et al., 2005; Krauss et al., 2007). The current evidence on hospital falls has primarily

PAGE 18

18 focused on identifying patient risk factors and evaluating preventative fall interventions in single facilities ( i.e. fall risk assessment tools, modifications to the hospital environment, patient/provider education interventions) (Currie, 2008). In addition, despite existing patient level interventions, fall rates in hospitals remain high and previous research has failed to identify interventions that influence the overall fall rates in acute care facilities, underscoring the need to conduct further research to identify factors that may contribute to elevated fall rates in order to target and tailor interventions to reduce ho spital fall rates. Previous studies examining patient falls in hospitals have either focused on unit level ( e.g. unit level nurse staffing) or hospital level ( e.g. fall risk assessment tools) predictors of falls separately and have yielded inconsistent r esults. This precludes an understanding of the comprehensive set of hospital factors that predict falls. A few hospital level studies have determined that organizational characteristics, such as hospital Magnet status, have been shown to be associated with lower fall rates (Lake, Shang, Klaus & Dunton, 2010), whereas others have shown no relationship between organizational characteristics and their impact on falls (Mark et al., 2008). Similarly, unit level factors, such as nurse staffing, have been shown in some studies to be associated with lower fall rates (Blegen & Vaughn, 1998; Stovie & Jawad, 2001; Whitman, Kim, Davidson, Wolf, & Wang, 2002; Dunton, Gajewski, Taunton, & Moore, 2004; Dunton, Gajewski, Klaus, & Pierson, 2007; Lake et al., 2010; Patrician et al., 2011), while other studies have shown no association between these measures (Blegen, Goode & Reed, 1998; McGillis, Hall, Doran, & Pink, 2004; Donaldson et al., 2005; Burnes Bolton et al., 2007; Shuldham et al., 2009). However, these divergent find ings in the patient falls

PAGE 19

19 literature can likely be attributed to limitations in research methodology, including cross sectional study designs, small sample sizes, non representative data sources, inconsistent measures of nurse staffing variables and a lac k of consistent theoretical frameworks to guide the research (Lake & Cheung, 2006). In order to guide policy efforts, it is critical to have an understanding of a comprehensive set of organizational and nurse staffing predictors and their influence on fall s given the importance of nurse staffing in the quality of patient care, as well as our limited understanding of hospital characteristics that contribute to patient fall rates over time. To our knowledge, no study to date has examined a comprehensive set o f organizational and nurse staffing predictors on patient falls lo ngitudinally In addition to previo us inconsistent findings in hospital level predictors of patient fall s, studies conducted on fall rates have documented sig nificant variation betwe en hosp itals and hospital units though to date, research has not attempted to explain wha t accounts for this variation. Depending on the clinical unit for example, average fall rates of between 1.7 to 25 falls per 1,000 patient days have been documented (Hernand ez & Miller, 1986; Nyberg et al., 1997; Mahoney, 1998; Schwendimann, 1998; Halfon, Eggli, Melle & Vagnair, 2001, Hitcho et al., 2004; Currie, 2008), though what accounts for this variation is not known. To the extent a common set of predictors of low fall rate facilities can be identified, clinical and state policy can be tailored and targeted to reduce these costly, avoidable adverse outcomes for patients. The purpose of this study is to assess acute care hospital factors that influence patient fall rat es over time. Our research will add to the existing body of knowledge by characterizing hospital fall rates over time in terms of latent class categories and

PAGE 20

20 assessing the organizational and unit level factors that influence patient falls over time. This i nformation can potentially be used to target and tailor interventions to hospitals and cli nical units to reduce patient fall s The specific aims of this current study are to do the following over a 54 month period in t he U.S. from 2006 2010: 1. Assess traject ories of fall rates within acute care hospitals and hospital units to identify latent class categories of consistently high, low or increasing/decreasing fall rates over time. 2. Examine the association between changes in unit level nurse staffing and changes in patient falls over time. 3. Determine the relationship between acute care hospital characteristics and fall rates over time. As the U.S. health care system strives to provide patient centered care where safety and quality are tailored to individual pati ents, there is increasing recognition that interventions to improve health care quality must take into consideration characteristics of the organizations within which they are implemented to maximize effectiveness. However, current recommendations are limi ted to fall interventions targeted at the patient level, which have done little to reduce fall rates in the U.S., suggesting additional information is needed to target and tailor interventions to organizations and patients with the highest need. By underst anding a comprehensive set of both organizational and unit level characteristics associated with consistently high, low and increasing/decreasing fall rates in hospitals over time, specific quality interventions can be utilized to reduce falls among patien ts most at risk. The aims above will be assessed using the 2006 2010 National Dataset for Nursing Quality Indicators (NDNQI), a monthly survey of a nationwide sample of hospita ls from across the U.S The advantages of this study include its longitudinal n ature, national sample, and collection of a comprehensive set of organizational factors.

PAGE 21

21 The longitudinal nature of the survey that contains structure, process and outcome measures, allows for an unprecedented opportunity to explore the relationship betwe en organizational characteristics, nurse staffing and patient falls over time Further, the dataset contains over 1,500 nationwide hospitals, with information collected at both the hospital and unit level, allowing for a comprehensive assessment of these c onstructs and their infl uence on patient falls (Montalvo, 2007; Dunton, Gajewski, Klaus, & Pierson, 2007; Lake et al., 2010) This study constitutes an important contribution to the literature as prior studies have yielded inconsistent results between nurs e staffing and falls, and little information is available to determine the impact of organizational structural characteristics on the risk of patients falling in hospitals, a critical gap as there is currently no information to guide the targeting and tail oring of fall interventions in acute care facilities at the hospital level. In addition, we guide our study with an integrated and comprehensive conceptual framework that incorporates an organizational perspective. Therefore, this study addresses gaps in t he current literature by examining relationships between organizations and patient falls longitudinally using unit level data and advanced statistical techniques to analyze change over time, while accounting for clustering of falls within hospitals and cli nical units within hospitals. Findings from our study are relevant and timely given the recent 2008 reimbursement changes from the Centers of Medicare and Medicaid (CMS) where providers are no longer reimbursed for preventable conditions (CMS, 2008). Add itionally, our results are important given prior mixed evidence from the nurse staffing literature, the ineffective interventions to reduce inpatient falls, the high prevalence rate of falls and rising health care expenditures (Lake & Cheung, 2006; Rosenth al, 2007; Shi

PAGE 22

22 & Singh, 2008; Abraham, 2011). In this study, creating latent class categories of hospitals based on patient fall measure s allowed for an assessment of the characteristics of these hospitals that influence fall rates overall and within indivi dual hospital units. Results of this study have important implications for managers, clinicians, researchers and policymakers, who can use this information to create guidelines and interventions that target hospitals and clinical units with disproportionat ely more falls, potentially reducing these preventable and costly outcomes for patients.

PAGE 23

23 CHAPTER 2 BACKGROUND AND SIGNI FICANCE O ver the previous decade, patient safety in acute care settings has become an increasingly critical nationwide issue with the r elease of several Institute of Medicine (IOM) reports highlighting deficiencies in the quality of ca re received in the U.S. (IOM, 1999; IOM, 2001). The 1999 IOM report To Err is Human: Building a Safer Health System brought national attention to patient sa fety and quality of care by concluding that as many as 44,000 to 98 ,000 deaths in the U.S. are attributed to errors from medical care, and preventable medical error death rates could potentially exceed those related to breast can cer, motor vehicle accident s or AIDS (IOM, 1999). The recommendations outlined by the 1999 IOM report were a call to action for health care providers, policymakers, health services researchers and administrators to reduce the number of adverse events in health care settings by esta blishing national patient safety goals, creating a public reporting system for tracking adverse events, incorporating stringent standards of practice and encouraging additional evidence based research to understand why errors occur in health care (IOM, 199 9; Clancy, 2009). Furthermore, the 1999 report found that most preventable errors were not the result of one individual health care provider, but rather of deficiencies in organizational systems and processes (IOM, 1999). Among the most frequently cited ad verse events, falls are specifically described as preventable occurrences that contribute to decreased quality of patient care in the IOM report (IOM, 1999). Despite these recommendations, interventions to reduce falls have not fully considered both the ho spital and clinical unit to target policy interventions, instead mainly focusing fall prevention efforts at identifying individual predictors of patient falls.

PAGE 24

24 A subsequent report from the Institute of Medicine released in 2001, Crossing the Quality Chas m: A New Health System for the 21 st Century outlined additional recommendations to improve the quality of care provided to patients and prevent further adverse medical events, including a focus on applying scientific evidence to health care delivery, alig ning payment with quality and preparing the health care workforce to meet the demands of a changing health care industry (IOM, 2001). These Institute of Medicine reports shifted the national focus in health care toward building a safer and more efficient health system by using evidence based approaches to reduce the number of adverse medical events, including patient falls. Although extensive research has been conducted on falls since the IOM reports were first published, falls rates among hospitalized pat ients have not subsided in recent decades overall and falls continue to represent a significant concern for patients, providers and policymakers (Unruh, 2002; Currie, 2008; Abraham, 2011). Therefore, it is imperative to conduct further analysis on why hosp italized patient falls occur and explore actionable research on how to prevent falls in acute care settings. Falls in Acute Care Hospitals Hospital falls have been conceptualized in different ways in the literature. Agostini, Baker and Borardus (2001) st on the ground, floor or other lower level, but not as a result of syncope or o verwhelming 11). Although previous definitions have attempted to classify falls based on their etiology, the operational definition of falls by the American Nurses Association (ANA) and National Database for Nursing Quality Indicators (NDNQI) will be used to assess the relationship between organizations, nurse staffing and falls due to the comprehensive nature o f the definition and the inclusions of hospital organizational

PAGE 25

25 unplanned descent to the floor (or extension of the floor, e.g. trash can or other equipment) with or witho ut injury to the patient, and occurs on an eligible reporting nursing unit. All types of falls are to be included whether they result from physiological Given this definition, the consequen ces of inp atient falls in the U.S. health care system are far reaching, impacting patients, family members, health care professionals, health care organizations, policymakers and payers. Several reasons why inpatient falls have become a major safety priori ty in the health care include: (1) increasing incidence rates of falls, (2) a growing focus on falls and quality of patient care, (3) increasing health care expenditures due to falls and, (4) policy regulations that focus on safety and no longer reimburse hospitals for injuries from patient falls. Impact of Hospital Falls on Quality of Patient Care For over 50 years, prevention of falls has been a concern in acute care settings in terms of quality of care (Thurston, 1957; Grubel, 1959; Morgan, Mathison, R ice & Clemmer, 1985; Currie, 2008). However, patient falls continue to represent the most commonly reported adverse event in hospitals and falls among patients over age 65 are the most frequent cause of non fatal injuries (Morse, 2002; Currie, 2008). Appr oximately 1 to 12% of patients in acute care settings will experience a fall during their stay (Vlahov, Myers & Al Ibrahim, 1990; Mahoney, 1998; Halfon et al., 2001; Coussement et al., 2007; Oliver et al., 2010). Given the nearly 40 million hospitalizatio ns each year, the total number of falls could potentially exceed more than one million annually (Currie, 2008; Oliver et al., 2010; AHRQ, 2009).

PAGE 26

26 Among those who fall in hospitals, approximately 6 to 42% of these patients will incur some type of injury a nd 4 to 9% will result in serious injury (Morse, Prowse, Morrow & Federspeil, 1985; Lauritzen, 1996; Morse, 1997; Hitcho et al., 2004; Currie, 2008; Wong et al., 2011). Injuries from falls are wide ranging, and they include skin bruises, soft tissue injur ies, joint injuries, fractures, traumatic brain injuries and possibly death (Bates, Pruess, Souney & Platt, 1995; Mahoney, 1998; Hitcho et al., 2004; Coussement et al., 2008; Wong et al., 2011). For instance, hip fractures are a major complication from in patient falls that lead to long term immobility in 20% of patients, and 25% to 36% of these patients will die within one year of the fracture (Zuckerman, 1996; Rubenstein & Josephson, 2002). Although deaths due primarily to falls are rare in acute care set tings, constituting less than 1% of all falls, they account for approximately 11,000 fatal falls on a yearly basis (Currie, 2008). These complications from falls have serious long term effects in terms of overall quality of life for patients, especially fo r vulnerable hospitalized patients who are at highest risk of falling. Impact of Hospital Falls on the Elderly Research indicates that falls impact patients of all ages in acute care settings, but the elderly population (over age 65) has been shown to di sproportionately fall more than any other age group and injuries among this segment of the population are often magnified (Mahoney, 1998; Rubenstein & Josephson, 2002; Hitcho et al., 2004; Clyburn & Heydemann, 2011). For instance, Inouye, Brown and Tinetti (2009) found that more than 30% of individuals over age 65 will fall at least once per year, and this is increased to 50% for individuals over age 80. In hospitals, incidence rates for the elderly population range from 0.5 to 2.7 falls per bed year (Ruben stein and Josephson, 2002)

PAGE 27

27 and mortality rates due to inpatient falls are substantially higher among the elderly, representing three quarters of all deaths by falls (Hogue, 1982). From a public hea lth standpoint, these figures on elderly inpatient fall ra tes are of of the baby boomer generation, there will be a greater demand for more intensive hospital services, especially in hospital units with a greater proportion o f elderly patients (Shi & Singh, 2008). According to Shi and Singh (2008), the over 85 age group has been the most rapidly growing sector of the U.S. population and by the year 2030, the elderly are expected to constitute 20% of the population. In addition the incidence and prevalence of chronic disease and disability is anticipated to rise in conjunction with the growing elderly population, which increases the risk of patient falls. Therefore, it is imperative to understand contributing characteristics of inpatient falls to reduce the effects of falls in this population. Impact of Hospital Falls on Patient Well Being, Family Members and Caregivers Inpatient falls also have long term effects on patient well being and quality of life through loss of indepen dence, fear of future falls, anxiety, depression and immobility (Vellas et al., 1997; Coussement et al., 2008; Currie, 2008; Faes et al., 2010). Patients who experience a fall are often fearful of future falls and therefore reduce their mobility (Parry et cascade effect where patients lose mobility and independence, exhibit signs of depression, and are more likely to be admitted to a long term rehabilitation care facility (Tideiksaar, 1993; Vellas et al., 1997; Oliver, Hopper & Seed, 2000; Parry et al., 2001; Faes et al., 2010). Another deleterious consequence of falls is the emotional and psychological stress on family members and caregivers who feel they may have failed

PAGE 28

28 the patient by not preventing the fall, which may lead to future anxiety and depression (Liddle & Gilleard, 1994). As a result, inpatient falls significantly contribute to decreased health related quality of life of hospitalized patients, as well as have a profound negative impact on family members and health care providers. Impact of Hospital Falls on Health Care Expenditures In addition to the detrimental impact on quality of care and patient well being, inpatient falls result in significant financial consequences for the health care industry. The economic impact of falls is often seen through increased lengths of stay (Bates et al., 1995), greater utilization of health care resources (Hill et al., 2007; Clyburn & Hyedemann, 2011), higher direct and in direct costs to the patient (Heinrich et al., 2010), potential litigation costs (Fiesta, 1998) and increased total charges for inpatient care (Stevens et al., 2006; Oliver et al., 2010; Wong et al., 2011). For instance, total estimated costs range from $16 to $19 billion for fall related injuries without death and the cost of falls resulting in death were approximately $170 million (Shekelle et al., 2003; Stevens et al., 2006). A study conducted by the Centers for Disease Control and Prevention (CDC) (2011) estimated that the annual direct and indirect health expenditures from injurious falls could reach more than $50 billion by the year 2020 (Clyburn & Heydemann, 2011). Secondary injuries related to falls lead to similar adverse outcomes, accounting for 15 % of re hospitalizations one month after initial discharge, and average length of stay in the hospital was found to increase by 7.5 to 12.3 days on average for patients who fell (Bates et al., 1995; Shekelle et al., 2003; Stevens et al., 2006). Bates and c olleagues (1995) conducted a case control study of hospital falls and found that charges were $4,233 greater in patients with serious fall related injuries than those who

PAGE 29

29 did not fall, representing approximately 60% higher charges for hospital services. Tw o recent studies found that the average patient cost per fall ranged from $1,059 to $13,316, depending on the severity, and patients who experience a fall are hospitalized 6.3 days longer than non fallers (Heinrich, Rapp, Becker & Konig, 2010; Wong et al. 2011). These current and projected figures on total health care expenditures from falls necessitate further research to introduce modifiable interventions for falls that have the potential to decrease the financial constraints on the health care system. Government Policies and Regulations on Inpatient Falls In light of the severe burden of falls on hospitalized patients and the increased financial strain on the health care system, federal policies have been enacted to prevent inpatient falls. The federa l government set goals through the Healthy People 2010 fall prevention to reduce the number of deaths and serious adverse event caused by falls (US DHHS, 2000). However, recent trends in falls data conclude that incidence rates have remained consistent, th ough there were an increased number of deaths caused by falls over this time frame (Currie, 2008; Abraham, 2011; CDC, 2012). Specifically, Currie (2008) determined that there was an upward trend in fall related death rates from 1999 to 2004, increasing to 41 deaths per 100,000 individuals, which exceeds the goals set by Healthy People 2010 of less than 34 deaths per 100,000 population (US DHHS, 2000). Despite the significant attention and resources allocated by the federal government through the Healthy Peo ple initiative, it remains unclear how best to reduce fall rates in acute care hospitals, particularly given the unsuccessful attempts at reducing mortality and morbidity from falls. On October 1 st 2008, the Centers for Medicare and Medicaid Services (CMS) implemented a policy that eliminated reimbursement to hospitals for treatment of injuries

PAGE 30

30 resulting from falls that occur during hospitalization (CMS, 2008). The intent of this policy was to eliminate the previous payment system where hospitals were reimbursed for preventable conditions, such as falls and other avoidable adverse events, as well as to reduce costs, increase quality and to encourage hospitals to take an active role in preventing patient falls (Mattie & Webster, 2008). However, since thi s was enacted, it is unclear how hospitals have responded or whether the policy has been effective in reducing patient fall rates in hospitals or units with particular characteristics. To our knowledge, no recent studies have been conducted longitudinally to determine how hospitals have responded to this policy before and after its implementation. Since the Centers for Medicare and Medicaid (CMS) consider falls to be avoidable events, and e & Webster, 2008, p. 338), identification of underlying causes of falls in hospitals could allow for the possibility of targeting specific interventions to improve quality of care, reduce the number of injurious falls and decrease the overall cost of pati ent care. This is particularly important given potential variability in fall rates among hospital units over time. Additionally, understanding how hospitals respond to changes in reimbursement for patient falls has critical implications for shaping future health p olicies on patient safety Risk Factors and Interventions to Reduce Inpatient Falls To date, a majority of the research on falls has been conducted in rehabilitation and long term care settings due to the co morbid conditions of the patient popul ation, with less research on falls in acute care settings (Fischer et al., 2005; Krauss et al., 2007), despite the fact that an estimated one million inpatient falls occur yearly (Currie, 2008; Oliver et al., 2010; AHRQ, 2009). The current limited evidence on hospitals has primarily focused on cross sectional studies that identify patient risk factors associated

PAGE 31

31 with falls and evaluating individual preventative fall interventions (Currie, 2008), precluding the ability to sufficiently target and tailor exist ing interventions at an organizational level. T o date research on the prevention of inpatient falls has la rgely been focused on two main areas of inquiry: (1) risk factors related to inpatient falls, and (2) interventions to reduce falls in hospitals (pre dominately fall risk assessment tools), including patient/provider educational approaches and modifications to the patient or organizational environment. Risk Factors Related to Inpatient Falls Nearly 300 risk factors for falls have been identified in th e literature, but these risk factors have not consistently been shown to reduce overall fall rates and there is a lack of agreement on which patient level factors should be used in fall prevention interventions (Oliver et al., 2000). In a systematic review Evans and colleagues (2001) reported 28 patient level risk factors consistently associated with inpatient falls, including altered mental status, immobility, age, certain types of medications ( e.g. sedatives and hypnotic medications), and a history of pr evious falls. Findings from this study indicate significant variability in risk factors among studies, which may be the ans, Hodgkinson, Lambert & Wood, 2001, p.43). In 2004, Oliver and colleagues identified five consistent patient level risk factors across multiple falls studies, including unsteady gait, toileting needs ( i.e. urinary incontinence/frequency), confusion, me dications with sedative hypnotic properties and a history of prior falls. However, among the 13 studies that evaluated risk factors of falling in acute care settings, a number of these studies lacked adequate validation

PAGE 32

32 methodology and there were a limited number of studies incorporating environmental risk factors for falling (Oliver et al., 2004). Other factors found to be associated with falls in hospitals include neurological impairment, musculoskeletal conditions ( e.g. osteoporosis, degenerative joint disease), memory loss/dementia, behavioral dysfunction, vision loss, syncope and environmental hazards (Bates et al., 1995; Evans et al., 2001; Oliver et al., 2010). However, prevention of falls has been challenging due to the multidimensional nature of ho spital care where identification and treatment of at risk patients requires coordinated efforts between the organization, staff, family members and patients. Overall, these systematic literature reviews focused on patient level factors, but care provided t o patients is delivered at the hospital unit level, and it is imperative to consider targeting policies and interventions at both the hospital and unit level to effectively reduce falls. Interventions to Reduce Falls in Hospitals Interventions to reduce f alls in hospitals have incorporated a limited number of organizational level approaches, and these interventions have yielded mixed results. At the organizational level, fall risk assessment tools are one of the most widely utilized methods to detect patie nts at risk for falls in acute care settings (Morse, 1997; Oliver, Daly, Martin & McMurdo 2004; Currie, Mellino, Cimino, & Bakken 2004). Creation of risk assessment tools have been guided by patient risk factors found to be associated with a greater risk of falling (Evans et al., 2001). However, risk assessment tools have been shown to inaccurately identify patien ts at high and low risk of falling in clinical settings, and there is significant variation in sensitivity and specificity in detecting falls amo ng at risk groups (Moore, Martin & Stonehouse, 1996; McCollam, 1995; McAvoy, Skinner & Hones, 1996; Oliver et al., 2000; Evans et al., 2001; Oliver at al., 2004).

PAGE 33

33 Based on prior literature studies, this may be due to methodological flaws in research desi gns and disagreement about which patient level factors should be used to prevent falls. As a result, these tools may have unintended consequences of immobilizing patients who are not at high risk of falling or misidentifying patients who are at high risk o f falling (Evans et al., 2001). Other approaches to reducing falls, such as educational interventions for patients/providers (Krishna & Van Cleave, 1983) and environmental modifications (Cumming et al., 2008), have been ineffective in reducing the nationwi de incidences of falls due to an inability to replicate findings across hospitals, lack of generalizability and inadequate sample sizes to detect meaningful differences between study groups. At the patient level, fall prevention strategies have incorporat ed patient bed alarms (Tideiksaar, Feiner & Maby, 1993), armband identification bracelets (Mayo, Gloutney & Levy, 1994), toileting needs assessments (Bakarich, McMillan & Prosser, 1997), medication reviews (Oliver et al., 2000; Cumming et al., 2008) and use of physical restraints (Shorr et al., 2002). Similar to the organization level literature, a majority of these studies have been conducted in only a few hospital settings with small sample sizes, which limits the ability to generalize the findings to other settings. For example, Tideiksaar and researchers (1993) conducted a case control study on the effectiveness of bed alarms in reducing falls among geriatric patients. The authors were unable to determine a significant relationship between the use of bed alarms and reductions in falls, and this may be due to a small, non representative sample (i.e. 70 total patients with 60 women, 10 men and a mean age of 84). In addition, similar sample size problems were found in a randomized control trial of colored armband identification

PAGE 34

34 bracelets, which found that high risk patients with blue colored identification bracelets had higher fall rates than those in the control group (Mayo et al., 1994). Other patient level studies on falls have been conducted using ra ndomized control trials, which are often considered the gold standard in scientific research, but findings from these studies may be difficult to implement and replicate in real world hospital settings. While performing a randomized trial of 24 elderly car e wards in 12 hospitals, Cumming and colleagues (2008) explored how a targeted multi factorial intervention of risk assessment tools, patient/provider education, medication review, hospital environment modifications, patient bed alarms and an exercise regi men impacted inpatient falls. Findings from this study showed that the intervention program did not have a significant effect in reducing patient falls (Cumming et al., 2008). Given these findings, our research may provide insight into why it is important to allocate resources for interventions in certain hospital units with particular characteristics, as interventions may work in some hospital settings and not others. Although there have been multiple proposed interventions to reduce falls in hospitals, fa lls continue to be a growing issue that is not resolved by any one measure and it is still unclear how specific risk factors interact in the acute care setting to modify the risk of falls, which is a critical gap in the literature (Oliver et al., 2000 ; Eva ns et al., 2001 ). A recent systematic review of risk factors and risk assessment tools concluded hospitalized patients, although many of the published fall prevention studies were underpowered and methodologically f 128). Therefore, it is imperative to continue investigating innovative methods to understand why falls occur in

PAGE 35

35 hospital settings and ways to target interventions to improve bot h the safety and efficiency of health care delivery. Conclusion Despite the significant amount of resources used to detect at risk patients and prevent falls, studies on hospital falls have demonstrated significant variability between the highest and lowe st performing hospitals and hospital units. Depending on the clinical unit, there are between 1.7 to 25 falls per 1,000 patient days on average (Hernandez & Miller, 1986; Nyberg et al., 1997; Mahoney, 1998; Schwendimann, 1998; Halfon et al., 2001, Hitcho e t al., 2004; Currie, 2008). As a result, much of the variation in falls in hospitals is still unexplained, and recommendations from the literature do not conclude with modifiable interventions to improve patient fall outcomes at the hospital or unit level. However, to our knowledge, no study to date has comprehensively assessed the relationship between hospital organizational characteristics, nurse staffing and the risk of inpatient falls (Dunton et al., 2004; Lake et al., 2010). These areas of inquiry repr esent a significant gap in the literature given the importance of nurses in preventing adverse events, as well as the need to understand characteristics of hospital organizations with consistently high and low patient fall measures and how these characteri stics influence patient falls. In this study, the impact of both nurse staffing and hospital organizational characteristics on inpatient falls will be examined to analyze differences in fall rates over time and to develop modifiable interventions to reduce fall rates. The Role of Nurses in Preventing Falls in Acute Care Hospitals Often considered a front line defense against adverse events, nurses constitute the largest group of health care professionals in the United States and adequate nurse

PAGE 36

36 staffing has been linked to measures of patient satisfaction and quality of care (Shi & Singh, 2008; Kane, Shamliyan, Mueller, Duval & Wilt, 2007; Unruh, 2008). Nurses play a multifaceted role in the delivery of health services, using their knowledge and skills to mon itor, assess and diagnose patients in a variety of health care settings, including acute care, long term care, assisted living, private agencies and many others (Shi & Singh, 2008). In particular, nurses provide a majority of direct patient care in acute c are settings (nearly 54%), and their responsibilities have a profound impact on patient outcomes and perceptions of care, including evaluation and prevention of patients at risk for falls (Irvine, Sidani & McGillis Hall, 1998; Lake & Cheung, 2008). Sever al types of nursing professionals operate in acute care hospitals, including registered nurses (RNs), licensed practical nurses (LPNs), and nursing assistants (UAPs). ompared t o other nurses (Unruh, 2003, p. 143). RNs are responsible for supervising unlicensed nursing staff, administering medications, advanced patient assessment skills, and communication with physicians and other health care providers (Unruh, 2003). LPNs typically require one year of a licensed state approved program, and they work together with RNs to provide patient care through assessment and evaluation of patient conditions (Unruh, 2003; Shi & Singh, 2008). UAPs work under the supervision of both RNs and LPNs, and they provide limited direct patient care activities, such as bathing, assessing vital signs and other basic patient care duties (Bishop et al., 2008). However, detection and prevention of patient falls requires an adequate balance of staffing needs among all types of nurses (Shuldham, 2004; Clarke & Donaldson, 2008).

PAGE 37

37 Factors Influencing Nursing Practice Although there are estimated to be over 3 million nurses in the health care workforce, ongoing national nursing shortages continue to plagu e the U.S. health care system (Shi & Singh, 2008; Fox & Abrahamson, 2009; Aiken et al., 2011). Other factors influencing the urgent need for research on nurse staffing and patient outcomes include the rapidly aging U.S. population, changes in health care f inancing and public health policy with increased national attention surrounding the quality of care provided in acute care settings (Wunderlich, Sloane & Davis, 1996; Clark & Donaldson, 2008). Specifically, changes in reimbursement to hospitals by prospec tive payment in the form of diagnosis related groups (DRGs), advanced diagnostic technologies and managed care regulations have brought about changes in the structure and processes of hospitals over previous decades (Aiken, Sochalski & Lake, 1997; Blegen & Vaughn, 1998; Wunderlich et al., 2006; Mark & Harless, 2007; Kane et al., 2007). Since the late and total number of inpatient days, which in turn has lead hospitals to restructure their nurse staffing levels to reduce costs and remain economically viable (Wunderlich et al., 1996; Blegen & Vaughn, 1998; Mark & Harless, 2007). Reductions in hospital nurse staffing have occurred through two different approaches: (1) uti lizing various proportions of nursing professional skill mixes (RNs, LPNs and unlicensed nursing assistants) or (2) reducing the total number of hospital costs (Wu nderlich et al., 1996; Blegen & Vaughn, 1998; Kovner, Jones, Chunliu, Gergen, & Basu, 2002; Rivers Tsai, & Munchus 2005; Mark & Harless, 2007). These

PAGE 38

38 conditions, cont inue to represent a growing problem for the health care industry in terms of quality of patient care, especially given the long term shortage of nurses and cost cutting measures from hospitals (Shi & Singh, 2008; Fox & Abrahamson, 2009). In addition, chang es in hospital organizational structure and nurse staffing may have a profound impact on the quality of care provided to patients in hospitals, particularly in terms of patient falls where assessment and monitoring of patients by nurses play s an integral r ole in preventing falls (Lake & Cheung, 2008). For instance, fall risk assessments are implemented by nurses and reduced staffing could potentially have an impact on adequately assessing at risk fall patients (Dempsey, 2004; Lake & Cheung, 2008). Units wi th reductions in nurse staffing may be risk patients in a timely manner, which in turn could result in increased risk of patient falls (Oliver et al., 2010). Additionally, adequate nurse staffing no t only enhances activation of fall prevention protocols, but nurses also educate family members and other staff members of patients at risk of falling, which could lead to continuous monitoring by other hospital staff members. As such, there is an urgent n eed for research on nurse staffing and patient falls to determine an appropriate balance of nurses that maximizes scarce resources and improve quality of patient care in acute care settings, as well as to determine if reductions in nurse staffing have an i mpact on patient falls in hospitals. Federal and State Level Nursing Policies Due to the importance of nurses in delivering high quality health services and the continued nursing shortages, several federal and state policies have been implemented in an at tempt to increase the supply of nurses (Donley et al., 2003; West, Griffith, & Iphofen, 2007). In 2002, President Bush si gned the Nurse Reinvestment Act (NRA) into

PAGE 39

39 law (Donley et al., 2003) The first provision of the NRA included recruiting nurses by incr easing loans and scholarships to nursing students for an agreement to work in critically underserved areas for two years after graduation (Donley et al., 2003). Additionally, the NRA sought to address retention of nurses through continuing education, cross ing training programs and improving communication and collaborations with other health care professionals (Donley et al., 2003). However, after passing the NRA, Congress did not approve an appropriations bill to fund the pro visions (Donley et al., 2003). A s a result, the effectiveness of this policy is still undetermined due to continued underfunding and limited policy evaluation research (Donley et al., 2003; Shi & Singh, 2008). At the state level, a controversial approach to maintaining an adequate sup ply of nurses and ensuring patient safety is through mandated minimum nurse to patient ratios. In 1999, California passed a landmark nursing legislation requiring minimum unit based nurse to patient ratios i n acute care hospitals ( Burnes Bolton et al., 200 1; Donaldson et al., 2005; Burnes Bolton et al., 2007). At that time, no other state had Department of Health Services (CDHS) to specify how these regulations would be enacted (Burnes Bolton et al., 2007). While establishing the regulations in the rulemaking process, the CDHS reported that there was insufficient empirical evidence on standardized nurse to patient ratios that would ensure patient safety and quality of ca re (California DHS, 2003). For this reason, the CDHS gathered information from public hearings, written letters and conducted random hospital surveys to develop thresholds for nurse staffing ratios (Donaldson et al., 2005). The initial nurse to patient

PAGE 40

40 rat ios in 2004 for medical surgical, step down and telemetry units were 1:6, 1:4 and 1:5 respectively, and these were further reduced to 1:5, 1:4 and 1:5 in 2005 (Donaldson et al., 2005). To date, only a few studies have analyzed the impact of the California nurse mandate on patient outcomes, and the results of these studies have concluded with mixed results (Dunton & Schumann, 2005; Donaldson et al., 2005; Burnes Bolton et al., 2007; Aiken et al., 2010). Despite the attempts from federal and state policies to increase the supply of nurses, it is still unclear the impact of nurse staffing on certain metrics of patient care, such as patient falls. Research conducted on nurse staffing and patient outcomes have provided inconclusive results and standardized nurs ing ratios have not been established in the scientific literature (Lang, Hodge, Olson, Romano & Kravitz, 2004). Given resource constraints and the ongoing national nursing shortages, the practice of adding more nurses may not be feasible or cost effective However, using a theoretically based research strategy to target interventions to hospitals and clinical units will ensure the most effective approach to reduce fall rates. The National Quality Forum Falls as a Nurse Sensitive Outcome As a result of t he multiple IOM reports on patient safety and quality of care, significant attention has been paid by policymakers and health care researchers to standardize measures of quality in health care, specifically linking quality measures associated with nursing care (Clark and Donaldson, 2008). These nurse sensitive quality indicators are defined as measures that vary depending on the structure and processes of nursing care, and are conceptually related to the practice of nursing (Clark & Donaldson, 2008). Most r ecently, the National Quality Forum (NQF) has standardized nurse staffing and outcome measures for quality of care research studies (NQF, 2004).

PAGE 41

41 The NQF established an expert panel that rigorously created nursing sensitive measures for public reporting, an d falls were first included as indicators of nursing quality by the American Nurses Association (ANA) in 1997. In addition, falls were added by the National Quality Forum in 2004 as one of the first measurements of hospital quality, demonstrating that fall s are important measures for maintaining high quality of care in acute care setting s and nursing care has a direct influence on whether or not a patient will fall in hospitals (NQF, 2004; Lake & Cheung, 2006; Burnes Bolton et al., 2007; Clark & Donaldson, 2008). Hospital Organizational Background Organizational Characteristics and Nursing Staffing A second important area of inquiry is to examine organizational characteristics of the hospital and their influence on both nurse staffing and patient falls (Aiken & Patrician, 2000). Nurse staffing levels are often affected by the level of patient acuity on associated with greater nurse staffing levels (Blegen & Vaughn, 1998, Hodge et al., 2004). Other factors that exert an influence on nurse staffing levels in hospitals and hospital units include ownership status (Becker & Foster, 1988; Mark, Salyer & Wan, 2000; Seago, Spetz & Mitchell, 2004; Jiang, Stocks & Wong., 20 06; Mark & Harless, 2007), hospital bed size (Becker & Foster, 1988), geographic location (Blegen, Vaughn & Vojir, 2008, Mark & Harless, 2007) and hospital teaching status (Bloom, Alexander & Nichols, 1997). For instance, Mark and colleagues (2000) found that hospital characterist ics significantly impacted unit level nursing skill mix, including the volume of services offered in a particular hospital. Blegen and researchers (2008) also found that nurse staffing (RN hours per patient day and RN skill mix) was positively associated

PAGE 42

42 with hospital complexity (a variable combining bed size, patient acuity, teaching status and hospital technology) and supply of registered nurses in certain geographic areas. Nevertheless, the link between hospital characteristics nurse staffing and patient level falls outcomes is still unknown. Organizational Characteristics and Inpatient Falls Research has shown that hospital organizational characteristics can differentially impact nurse staffing levels (Blegen et al., 2008), but the question of how these organizational characteristics impact patient outcomes is not well understood in the scientific literature. Furthermore, research on hospital structure and patient outcome measures has been inconclusive (Mark, Sayler & Wan, 2 003; Cramer, Jones & Hertzog, 2011). For instance, Mark and colleagues (2008) conducted a study at 143 hospitals with a total of 278 medical surgical units to examine the association between hospital organizational structure, patient level characteristics and safety climate on the risk of inpatient falls. Results from this study did not find a significant effect of hospital structural components on patient falls, but hospital units with greater patient capacity resulted in higher fall rates. The authors su ggest that increased capacity may inhibit the ambulation (Mark et al., 2008). Several other hospital organizational characteristics have been linked to patien t falls, such as Magnet status, type of clinical unit, and nursing process measures (Mick & Mark, 2005; Jiang et al., 2006: Dunton et al., 2007; Mark & Harless, 2007). For example, M agnet status, a designation given to hospitals in recognition of their com mitment to standards of nursing excellence and clinical practice (Rivers et al., 2005; Wishall, 2008; American N urses Credentialing Center (ANCC), 2008 ), was found to be

PAGE 43

43 associated with a 5 10% reduction in the number of falls when compared to non M agnet h ospitals (Dunton et al., 2007; Lake et al., 2010). In 1993, the American Nurses Credentialing Center (ANCC) establishe d an accreditation process for M agnet status and the goals for each th recognize nursing services that use p rofessional standards and research to build programs of nursing excellence for the delivery of care to patients, (2) promote quality in a milieu that support professional nursing practice, (3) provide a vehicle for the dissemination of successful nursing p ractices and strategies among healthcare organizations using the services of vens & Johnston, 2004, p.580). To become M agnet certified, hospitals apply for accreditation and they are evaluated on site by ANCC evaluators based on their demonstrated nursing leadership characteristics and management style, their commitment to new knowledge through research publications, their organizational structure, personnel policies and profession al development, and their commitment to education and quality of care through quality pubic reporting and use of information technology (Havens & Johnston, 2004; Aiken, Havens & Sloane, 2009; Kelly, McHugh & Aiken, 2011). See Appendix A for additional in formation regarding the accreditation process for hospitals to become Magnet certified in the U.S. As previously mentioned, fall rates vary considerably between clinical unit and service type within the hospital, and two articles found some this variation may be due to patient characteristics (Fischer et al., 2005; Krauss et al., 2007; Schwendimann, Buhler, De Geest & Millisen, 2008). As a result, it is important to consider the structural components of clinical units to target tailored interventions for t hese patient populations.

PAGE 44

44 Finally, hospital organizational factors affecting nurse care processes, such as patient fall risk assessments and nursing interventions, have an impact on overall fall rates (Mick & Mark, 2005; Dunton et al., 2007). However, ther e is still a lack of comprehensive understanding between nurse staffing and hospital organizational characteristics on patient falls in acute care settings, and this lack of understanding represents a significant gap in knowledge.

PAGE 45

45 CHAPTER 3 LITERATURE RE VIEW Nurse Staffing and Quality of Patient Care In 1996, the Institute of Medicine (IOM) convened the Committee on the Adequacy of Nurse Staffing in Hospitals and Nursing Homes. The committee concluded that nurse staffing represents an integral part of the delivery of quality patient care in th e U.S. and future research would be necessary to understand the relationship between nurse sensitive outcomes (Wunderlich et al., 1996). Prior to the 1996 IOM study, researchers had not established a consistent associ ation between nurse staffing and patient outcomes (Blegen & Vaughn, 1998). In addition, methodological challenges and ill defined conceptualization of the relationships often limited the ability of researchers to make meaningful comparisons between these v ariables, as did the lack of available national datasets (Blegen & Vaughn, 1998; Clark & Donaldson, 2008). Consequently, the importance of nurse staffing was often agreed upon on a theoretical basis. For example, nurses were considered an important compone nt for safety and quality of care generally, but there were disagreements on established measures of staffing in the literature and how research should influence p olicymaking decisions (Clark & Donaldson, 2008). Following the 1996 IOM report on the adequa cy of nurse staffing, the study of nurse staffing and quality of patient care saw immense growth in the peer reviewed literature (IOM, 2004). In 2002, two seminal contributions to the nurse staffing literature were published, including a publication in JAM A by Aiken and colleagues, and another by Needleman and colleagues in the New England Journal of Medicine Aiken and colleagues determined that there was a 7% increase in the odds of patient mortality and

PAGE 46

46 failure to rescue in surgical patients for every on e patient increase in nurse workload (Aiken et al., 2002). Furthermore, researchers found that a higher proportion of nurses were associated with lower rates of patient mortality (Aiken et al., 2003; Aiken et al., 2011; Needleman et al., 2011), pressure ul cers (Blegen et al., 1998; Unruh, 2003), pneumonia (Kovner et al., 2002), medication errors and nosocomial infections (Lucero, Lake, & Aiken, 2010), urinary tract infections (Needleman et al., 2002), as well as lower rates of other adverse patient outcomes (Lang et al., 2004; Lankshear, Sheldon, & Maynard, 2005; Kane et al., 2007; Unruh, 2008). Despite the increase in research over the past decade, few studies have examined the relationship between nurse staffing and falls, and studies have produced inconsi stent results regarding the nature of this relationship. This may be due to the fact that falls differ significantly from other types of outcomes (Mark, Hughes & Jones, 2004). However, this is a critical omission in the literature and will be the focus of our study. Nurse Staffing and Patient Falls Prior to the 1996 IOM report on the adequacy of nurse staffing, research conducted on nurse staffing and inpatient falls was limited and findings were often inconclusive, with studies finding both positive and ne gative relationships between nurse staffing and falls. Results from a study implemented by Fine in 1959 concluded that there was an increase in patient fall rates with fewer nurses on a particular shift. Several studies prior to 1996 showed insignificant r esults of nurse staffing and patient falls (Kustaborder & Rigner, 1983; Morse, Tylko & Dixon, 1987; Tutuarini, de Haan & Limburg, 1993). Conversely, Sehested and Severin Nielson (1977) found a positive relationship between falls and the number of nurses and a study by Grillo Peck and Risner (1995) suggested that restructuring of RN staffing from 80% to 60% skill mix

PAGE 47

47 reduced the number of falls. However, studies before 1996 are subject to limitations, including less sophisticated technology to track and calculate nursing staffing hours and other critical control variables, such as hospital unit type, teaching status, geographic location and bed size. Additionally, the majority of articles were cross sectional, which inhibits the ability to make causal inf erences. Although these articles laid the framework for future studies, they lacked the ability to detect meaningful relationships. Since 1996, the scientific evidence examining the relationship between nurse staffing and inpatient falls has grown, but the results have remained inconsistent because of variations in methodological designs, variations in data sources, differences in measurement of nurse staffing variables and a lack of theoretical frameworks to guide the research. Three systematic litera ture reviews examining the research between 1990 and 2006 have found insufficient evidence to support the causal association between nurse staffing and falls in acute care settings, and findings from the literature have not provided modifiable unit based r ecommendations to reduce patient falls (Lang et al., 2004; Lake & Cheung, 2006; Kane et al., 2007). In particular, a number of researchers have demonstrated a significant association between reductions of inpatient falls and total nursing hours per patient day (Sovie & Jawad, 2001; Whitman et al., 2002; Dunton et al., 2004; Dunton et al., 2007; Patrician et al., 2011), RN skill mix (Blegen & Vaughn, 1998; Dunton et al., 2004; Dunton et al., 2007; Patrician et al., 2011), RN hours per patient day (Sovie & Ja wad, 2001; Lake et al., 2010), licensed nursing hours per patient day (Unruh, 2003), and patient to nurse ratios (Krauss et al., 2005). On the other hand, studies have found insignificant and unexpected associations between falls and nurse staffing. For e xample, insignificant relationships were found

PAGE 48

48 between inpatient fall rates and measures of total nursing hours per patient day (Blegen & Vaughn, 1998; Blegen et al., 1998; Cho, Ketefian, Barkauskas, & Smith, 2003; Shuldham, Parkin, Firouzi, Roughton, & La u Walker, 2009; Breckenridge Spr oat, Johantgen & Patrician, 2012 ), RN skill mix (Blegen et al., 1998; Cho et al., 2003: Breckenridge Sproat et al., 2012 ) and licensed nursing skill mix (McGillis Hall, Doran, & Pink, 2004). In addition, two articles examin ing the effect of the 1999 California nurse mandate reported insignificant relationships between inpatient falls and multiple nurse staffing measures (Donaldson et al., 2005; Burnes Bolton et al., 2007). Interestingly, a few studies have concluded with una nticipated findings; a greater risk of falls was found to be ass ociated with a higher number of licensed nurses (Unruh, 2003), increased LPN hours per patient day in non ICUs (Lak e et al., 2010) and increased nursing assistant hours per patient day in non ICUs (Lake et al., 2010). These mixed findings preclude the ability of researchers to make meaningful comparisons between fall outcomes, and the question of how nurse staffing contributes to patient falls in acute care settings remains unanswered. Limitat ions in Previous Research Given the previous findings in the literature, there is reason to believe that nurse staffing influences inpatient falls, but the mechanisms through which this occurs and how fall rates are influenced by policy changes over time a re unclear. Further investigation may provide insight as to why prior studies have yielded equivocal results and how this study will address these deficiencies in the literature. As reported in the systematic reviews of the nurse staffing literature, prev ious inconsistent findings are subject to limitations due to: (1) variation in methodology and study design differences,

PAGE 49

49 (2) variations in data sources, (4) measurement and operationalization of nurse staffing variables and, (5) a lack of consistent theore tical frameworks to guide the research. Variation in Methodology and Study Design Methodological and study design differences may account for the significant variation in findings in the nurse staffing and falls literature. Researchers have used a mult itude of different techniques to determine a relationship between these variables, including generalized estimating equations (Blegen & Vaughn, 1998; Lake et al., 2010), generalized linear mixed model (Dunton et al., 2004; Dunton et al., 2007; Breckenridge Sproat et al., 2012 ), logistic regression in a case control study (Krauss et al., 2005), multilevel logistic regression (Cho et al., 2003), Bayesian hierarchical logistic regression (Patrician et al., 2011), repeated measures ANOVA (Donaldson et al., 2005 ; Burnes Bolton et al., 2007), Poisson regression (Unruh, 2003; Manojlovich, Sidani, Covell & Antonakos, 2011), and linear regression models (Blegen et al., 1998; Duffield et al., 2011). However, many of these approaches are limited in their ability to as sess nesting of clinical units within hospitals, as well as to determine relationships between constructs longitudinally. Therefore, we used generalized estimating equations to account for these concerns in our study. Additionally, there was significant va riation found in study designs between articles. For instance, Krauss and colleagues (2005) conducted a case control study design of hospitalized patients at a large 1,300 bed teaching hospital. Findings from this study showed that when compared to nurses with three or fewer patients, the chance of falling for patients was increased three fold for nurses with four to six patients and seven fold for nurses with seven or more patients (Krauss et al., 2005).

PAGE 50

50 Given the variation in methodology and study desig n, Lake and Cheung (2006) recommend implementing statistical techniques that encompass and account for the multilevel dimensions of nurse staffing within hospital organizations, as well as utilizing statistical models that assess change over time. A recent article by Unruh and Zhang (2012) used a latent growth curve model (LGCM) to analyze the association between changes in nurse staffing and patient safety events in Florida from 1994 to 2004. Although this study did not use falls as an outcome measure, the LGCM analyzes longitudinal changes over a period of time to determine causal relationships between measures, which strengthens the methodological design and allows for determination of a temporal precedence (Duncan & Duncan, 2004; Unruh & Zhang, 2012). Ou r study uses latent class growth modeling (LCGM) to assess hospital fa ll rates over time and identify latent class trajectory groups of patient falls over time in hospitals. We then analyzed these categories separately to determine how nurse staffing and o rganizational characteristics impacted fall ra tes in these hospitals In addition to study design and analytic approaches, results of previous studies varied based on the unit of analysis, as well as the use of aggregated patient and nurse level variables (Lake & Cheung, 2006). Some studies included clinical nursing units as control variables (Blegen & Vaughn, 1998), whereas others conducted separate analyses for each unit type (Dunton et al., 2004; Dunton et al., 2007; Lake et al. 2010). Many large datase ts incorporated variables at different levels of analysis, such as hospital level (Cho et al., 2001; Unruh, 2003), individual nursing unit level (Blegen et al., 1998; Blegen & Vaughn, 1998; Sovie & Jawad, 2001; Dunton et al., 2004; Krauss et al., 2005; Dun ton et al., 2007; Shuldham et al., 2009; Lake et al., 2010; Duffield et al., 2011)

PAGE 51

51 and nursing shift level (Patrician et al., 2011). In addition, previous studies have aggregated patient level and nurse level variables to the hospital level (Cho et al., 20 03). These aggregat ed measures at the hospital level likely results in lost variation that takes place at the unit level, which makes it difficult to commun icate study findings to shape hospital or state policy (Whitman et al., 2002; Unruh & Zhang, 2012). Previous literature also found that fall rates ranged significantly between studies depending on study d esigns at the hospital and unit levels. For instance, considerable variation was found in hospital level studies by Cho and colleagues (2003) and Unruh (2003), in which the fall rate in Unruh (2003) was twice that found in Cho and colleagues (2003). Likewise, there was significant variation in fall rates at the nursing unit level (Blegen et al., 1998; Dunton et al., 2004). Dunton and colleagues (2004) ex plored unit level hospital differences in patient falls, and significant differences were found across units; medical surgical units exhibited higher rates of falling when compared to surgical units, and total nurse staffing hours had a significant impact on falls (Dunton et al., 2004). Similar findings were reported in ot her studies at the nursing unit level (Sovie and Jawad, 2001; Lake et al., 2010). Overall, studies were strengthened by using datasets that capture both hospital and unit level data to all ow for analysis of individual nursing units nested within organizations, such as the NDNQI dataset used in our study (Lake & Cheung, 2006). Variation in Data Sources Differences between study results may be attributed to limitations in study design and data sources, including the use of cross sectional studies and inadequate sample sizes to detect meaningful relationships. A number of articles investigating the relationship between nurse staffing and patient falls have been conducted using cross

PAGE 52

52 sectiona l data sources, which limits the ability to make causal inferences (Lake & Cheung, 2006). For instance, some investigators have analyzed data from a single year or single point in time (Cho et al., 2003; Dunton et al., 2004; McGillis Hall et al., 2004; Lak e et al., 2010; Manojlovich et al. 2010), whereas others have analyzed data longitudinally (Blegen and Vaughn, 1998; Sovie & Jawad, 2001; Unruh, 2003; Donaldson et al., 2005; Burnes Bolton et al., 2007; Duffield et al., 2010; Breckenridge Sproat et al., 20 12 ). However, there were differences found between studies, even among those with longitudinal study designs. For example, Sovie and Jawad (2001) found a significant relationship between total nursing hours per patient day (HPPD) and lower fall rates, wher eas this relationship was insignificant when using other longitudinal databases ( Breckenridge Sproat et al., 2012 ). Despite the mixed results between studies, longitudinal data allows for researchers to establish temporal precedence and make causal inferen ces about the relationship between nurse staffing and patient falls, as do statistical techniques that account for change in variation over time, such as latent growth curve models and latent class growth models (Duncan & Duncan, 2004; Nagin, 2005). T o da te, t he majority of published articles have focused on data sources that may not be generalizable to other populations due to non representative study groups and inadequate sample sizes. Prior studies examining the relationship between nurse staffing and p atient falls have been published using data from California (Cho et al., 2003; Donaldson et al., 2005; Burnes Bolton et al., 2007), Pennsylvania (Unruh, 2003), England (Shuldham et al., 2009), Australia (Duffield et al., 2011), Canada (McGillis et al., 200 4), Canada and Michigan (Manojlovich et al., 2010), military hospital databases

PAGE 53

53 (Patrician et al., 2011; Breckenridge Spr oat, Johantgen & Patrician, 2012 ), a teaching hospital in Missouri (Krauss et al., 2005), an unspecified large university teaching hos pital (Blegen et al., 1998) and nationwide U.S. datasets (Sovie & Jawad, 2001; Dunton et al., 2004; Dunton et al., 2007; Lake et al., 2010). In particular, using the CALNOC dataset, two studies analyzed the impact of the California minimum nurse staffing r atio mandate policy on patient falls using a pre post study design at the unit level (Donaldson et al., 2005; Burnes Bolton et al., 2007). Results indicated no significant changes in fall rates after implementation of mandated nurse to patient ratios, alth ough nurse staffing ratios increased over this period of time. These studies provide some insight into the relationship between nurse staffing and patient falls within a certain hospital or geographic region, but findings may only be generalizable to hospi tals used in that particular data source. Nationwide datasets provide researchers the ability to generalize findings to a larger subset of the population, such as those used by Dunton and colleagues (2004) in the National Dataset for Nursing Quality Indica tors (NDNQI), who found that higher total nurse staffing and percentages of registered nurses (RNs) on a unit were associated with reductions in fall rates. Inadequate sample sizes and lack of consistent control variables may have precluded researchers from having sufficient power to detect differences between nurse staffing measures and patient falls. Sovie and Jawad (2001) conducted 16 separate analyses from a dataset that included inpatient records for 29 hospitals. However, only 3 analyses resulted i n significant findings, which may be due to inadequate statistical power to detect a relationship between nurse staffing and falls. Likewise, Shuldham and colleagues (2009) examined two hospitals with cardio pulmonary patients, and they

PAGE 54

54 cited a limited abi lity to detect a relationship between nurse staffing and patient falls due to inadequate statistical power, which may be consistent with other small sample size studies (Blegen et al., 1998). Additionally the empirical evidence is inconsistent in the use of control variables, and a majority of these studies were conducted using limited or no organizational control variables. For instance, only four hospital characteristics were included in an analysis by Cho and colleagues (2003). Thus, e xclusion of organi zational variables co uld have resulted in an omitted variable that is related to both nurse sta ffing and patient falls, leading to further problems of endogenei ty and omitted variable bias that threaten the internal validity of the se studies Measurement a nd Operational Definition of Nurse Staffing In addition to limitations in data sources and sampling, comparisons between studies in the literature are difficult because of the variation in approaches used to measure and operationalize nurse staffing (Lake & Cheung, 2006). These inconsistencies are likely due to a lack of an established conceptual measurement of nurse staffing in the literature (Mark, Hughes & Jones, 2004). The various empirical use of nurse staffing includes: total nursing hours per patie nt day (Blegen et al., 1998; Stovie & Jawad, 2001; Dunton et al., 2004; Donaldson et al., 2005; Burnes Bolton et al., 2007; Dunton et al., 2007; Shuldham et al., 2009; Duffield et al., 2011; Patrician et al., 2011; Breckenridge Sproat et al., 2012 ), propor tion of RNs to total nursing hours (Blegen et al., 1998; Blegen & Vaughn, 1998; Cho et al., 2003; Dunton et al., 2004; Donaldson et al., 2005; Burnes Bolton et al., 2007; Dunton et al., 2007; Duffield et al., 2011; Patrician et al., 2011), RN hours per pat ient day (Stovie & Jawad, 2001; Cho et al., 2003; Donaldson et al., 2005; Bolton et al., 2007; Lake et al., 2010), unlicensed nurses hours per patient day (Donaldson et al., 2005; Bolton et al., 2007; Lake et al.,

PAGE 55

55 2010), proportion of licensed nurses hours (RN and LPN) to total nursing hours (Unruh, 2003; McGillis Hall et al., 2004; Donaldson et al., 2005; Burnes Bolton et al., 2007), proportion of unlicensed nurses hours (UAPs) to total nursing hours per patient day (Lake et al., 2010), patient to nurse r atios (Krauss et al., 2005), proportion of contract nurses hours to total nursing hours (Dunton et al., 2004; Donaldson et al., 2005; Burnes Bolton et al., 2007; Lake et al., 2010) and other nurse skill mix variables (Manojlovich et al., 2010). As a resul t, divergent findings between multiple studies examining nurse staffing and falls outcomes could be explained by the different ways in which nurse staffing variables are operationalized (Manojlovich et al., 2010). Another measurement concern which may p artially explain the mixed findings across studies, is the calculation of nurse staffing hours, which has commonly incorporated flawed indirect and aggregated measures. In a study by Whitman and colleagues (2001), the authors incorporated indirect nursing care provider hours, such as those from unit secretaries and nurse managers, in their calculation of direct care nurse staffing hours, which has the potential to overestimate coefficients and bias the study results. In addition, another concern for measure ment error is the aggregated nurse staffing variable over time. Previous studies have used quarterly or yearly aggregate nurse staffing variables, which does not account for variability in nurse staffing over shorter time periods (Cho et al., 2003; Unruh, 2003; Patrician et al., 2011). Aggregated data provides averages over time, but this information often only captures chronically understaffed hospitals and does not account for variation in nursing hours within the hospital or hospital units. Furthermore, these measures can result in measurement error and may underestimate the impact of nurse staffing on adverse

PAGE 56

56 patient outcomes. Measurement error represents a significant limitation in the nursing literature, as many of the researched outcomes occur as disc rete events on a specific unit, which would be masked by yearly aggregated data (Patrician et al., 2011). To address limitations in the measurement and operationalization of nurse staffing variables, Patrician and researchers (2011) performed a longitudi nal study in multiple military hospitals to examine the relationship between nurse staffing and analysis, the authors found that higher proportions of RNs resulted i n few falls in medical surgical and critical care units, but not in step down units, and a strong relationship was found between total staffing (total nursing care hours per patient shift) and injurious patient falls (Patrician et al., 2011). Results among unit types were different than in previous studies, in that there was a 15% to 51% increase in falls with injury with a one hour decrease of nursing care per shift. These effects may have been undetected in other studies that aggregat ed monthly or yearly staffing. However, there is limited generalizability in this study due to the study sample of military hospitals with substantially different rules, regulations and study populations than those in private sector hospitals. Comparisons between studies on nu rse staffing and falls are strengthened by using evidence based nurse staffing measures that are recommended by researchers with extensive expertise in the literature (Lake & Cheung, 2006). In their systematic review of the nurse staffing and falls literat ure, Lake and Cheung (2006) indicated that both skill mix and skilled nursing hours represent the most widely supported measures

PAGE 57

57 and skill mix endorsed by the Nation addition, measures of nursing hours per patient day (HPPD) and skill mix of RNs to total nursing staff were found to be the most accepted and useful measures for nurse staffing from a panel of internation al experts (Van den Heede, Clarke, Sermeus, Vleugels & Aiken, 2007). Therefore, this study will address these gaps by using standardized metrics recommended by the American Nurses Association (ANA) the National Quality Forum (NQF) and an international pan el of nurse staffing experts. Lack of Consistent Theoretical Frameworks A majority of articles in the nurse staffing and patient falls literature lack a conceptual or theoretical framework to explain the relationship between staffing and inpatient fal ls. A proposed explanation for the inconsistent findings include unarticulated impact on patient falls in hospitals (Mark et al., 2004; Lake & Cheung, 2006). This lack of conceptual clarity limits the ability of researchers to determine how nurses interact within hospital units to impact patient falls outcome, in addition to leading to measures of nurse staffing and/or falls that do not adequately reflect the intended const ruct. In the nurse staffing and patient falls literature, there have only been a few conceptual and theoretical models used that link organizations and patient outcomes, including Process Outcomes (SPO) conceptual framework (1966), C (2001) Nurse Staffing and Patient Outcomes Model (NSPOM), and a theoretical model proposed by Aiken, Sochalski and Lake (1997) and subsequently revised by Aiken, Clarke and Sloane (2002) to include nurse staffing and nurse practice environment. A cont ribution of the current study is the generation and testing of an integrated conceptual model that draws on prior theory and empirical evidence, addressing gaps in

PAGE 58

58 prior theoretical models and articulating a clear pathway between nurse staffing and patient falls in the hospital setting. Conclusion In summary, prior studies examining nurse staffing and patient falls have produced inconsistent findings due to limitations in study design and methodology, variations in d ata sources, operationalization s of nur se staffing variables and a lack of consistent theoretical frameworks to guide the research. In the current study, we will address these critical gaps in the literature by using the National Dataset for Nursing Quality Indicators (NDNQI) over a four year p eriod of time. Additionally, we will utilize advanced statistical methods that allow for examination of change in organizational characteristics, nurse staffing and patient falls at the unit level over time. We will also expand on prior theoretical models to include an integrated and comprehensive conceptual framework linking hospital organizations, nurse staffing and patient falls. This study represents an unprecedented opportunity to examine these relationships longitudinally using unit level data. Prio r studies would be strengthened by incorporating a longitudinal dataset with comprehensive nurse staffing an d organizational characteristics collected at the unit level, such as the NDNQI in our study. In the limited studies that have examined the relation ship between nurse staffing and patient falls using the NDNQI measures of nurse staffing have been associated with decreased fall rates at the unit level (Dunton exception of hospitals, as well as the ability to conduct longitudinal analysis with both hospital and unit level data. However, previous studies using the NDNQI were not conducted

PAGE 59

59 longitud inally with statistical methods that assess changes in fall rates over time which represents a significant strength of our study design. The identification of unit specific nurse staffing and organizational characteristics associated with fall rates over time will allow for targeted interventions to reduce the number of falls in acute care settings.

PAGE 60

60 CHAPTER 4 CONCEPTUAL FRAMEWORK AND HYPOTHESES A limited number of theoretical explanations for why nurse staffing impacts patient falls may account for the si gnificant variability in findings in the literature (Mark et al., 2004). To address the limitations in empirical studies, a conceptual framework is derived from prior models to identify organizational and unit level characteristics that influence patient f alls in hospitals. Currently, only a few theoretical models exist to analyze the nurse staffing evidence at the organizational level and these models Process Outcomes (SPO) conceptual framework ffing and Patient Outcomes Model (NSPOM), and a theoretical model proposed by Aiken, Clarke and Sloane (2002). These models are integrated in the current study to allow for an assessment of a comprehensive set of factors associated with patient falls in ho spitals. Process Outcomes (SPO) framework (1966), health care quality is a function of the components that relate to the organizational characteristics of health care providers (structure), the clini cal and administrative procedures involved in the delivery of health care to patients (process) and the effect of care on the health status of hospital patients (outcomes). Specifically, structure refers to all of the attributes of the hospital setting, s uch as physical attributes, equipment and human labor, including nurse staffing (Donabedian, 1966; Donabedian, 1988; Sovie & Jawad, 2001). The processes are referred to as the manner in which care is delivered, and this is typically achieved through staff activities and functions, such as fall risk assessments by nurses (Donabedian, 1966; Donabedian, 1988; Lake et al., 2010).

PAGE 61

61 Finally, outcomes consist of measures of patient health and wellness, including patient outcome measures and satisfaction with care ( Donabedian, 1966; Donabedian, 1988; Donabedian, 1992). Therefore, the model suggests that organizational structure influences the processes of patient care, which in turn affects patient outcomes. In 2001, Cho developed the Nurse Staffing and Patient Outcomes Model organizational accidents from the industr generic modeling system (Reason, 1995), the NSPOM distinguishes between latent and active errors, where latent errors are the consequences of delayed actions in organizational decision making and active errors direct contact with the human 2001, p.78). Based on these definitions, Reason (1995) proposed a theoretical understanding of organizational accidents. These organiz ational accidents originate after undetected latent failures result in unsafe working environments, which lead to active failures and adverse events (Reason, 1995). For instance, the causal pathway for organizational accidents is a function of failed organ izational structure (e.g. scheduling and planning of nurse staffing, lack of appropriate standards and fall prevention processes) which results in latent failures. In turn, these latent failures are transmitted to the workplace, where understaffed units cr eate an environment with a higher probability of active failures due to inadequate time to care for patients, miscommunications between providers and burnout/fatigue of overworked providers (Cho, 2001). As a result, there is a greater risk of adverse event s, such as falls, on units that are

PAGE 62

62 understaffed due to active failures (errors or violations in nurse care processes), and the consequences of these are decreased health outcomes and increased patient costs (Cho, 2001). In other words, organizational syst ems in place result in understaffed nursing units, and this leads to a higher probability of adverse events, such as falls (See Figure 4 1). Aiken, Clarke & Sloane (2002) Theoretical Model In 2002, Aiken, Clarke and Sloane expanded on a prior theoretical framework based on the association between organizational characteristics and patient outcomes, particularly describing how the organization of nurses in a hospital and the delivery of nursing care influence patient outcomes (Aiken et al., 1997). The revi sed 2002 model incorporated nurse staffing as an important contributor to early detection and surveillance of adverse events and complications when caring for patients (Aiken et al., 2002). This theoretical model indicates that organizations with adequate nurse to patient ratios and skill mixes will result in better detection of complications and processes of care that positively affect patient outcomes (Aiken et al., 2002). However, as opposed to the 1997 model, organizational support for nursing care dire ctly impacts the processes of patient care, and these influenced patient outcomes. Figure 4 2 illustrates the revised 2002 theoretical model by Aiken and colleagues. Nurse Staffing and Patient Fall Conceptual (NSPF) Model Development Despite the theoretica l models that have been developed, a model based on the relationship between organizational contextual factors, nurse staffing and patient falls provides a more detailed description of the relationships between these concepts (Mark et al., 2004). A large n umber of empirical articles do not explicitly state a theoretical link between constructs, and these theoretical links often include outcome variables that are

PAGE 63

63 inherently different from one another. For instance, falls occur at a temporally distinct point in time and require detection and evaluation of at risk patients, whereas other outcomes, such as pressure ulcers, occur over a longer period of time and may require a different balance of nursing hours to monitor and reduce skin breakdown (Cho et al., 200 3). Also, these outcome measures may require different organizational and nursing processes. Therefore, differences in outcomes necessitate the need for a theory focused on organizational characteristics, nurse staffing and patient falls. In our study, a n characteristics and nurse staffing on patient fall outcomes. This int egrated conceptual model allow s for a focused assessment of the philosophical relationships between these constructs, as well as address gaps in pri or theoretical models and guides the current research project ( See Figure 4 3). The Nurse Staffing and Patient Falls (NSPF) conceptual m odel illustrates how the organizational structure of hospitals determines levels of nurse staffing, which in assessment processes. Finally, the organizational stru cture and processes of care influence whether a patient will fall in an acute care setting. Figure 4 3 illustrates the derived conceptual framework and illustrates the hypothesized mechanisms through which the structural characteristics of the organization impact unit level nurse staffing, which in turn influences patient care processes and patient falls in hospitals. Structure Factors The derived NSPF model posits that organizational characteristics determine adequate or inadequate levels of nurse staf fing at the unit level. This interaction

PAGE 64

64 between organizational characteristics and unit level nurse staffing creates an environment in which there are two possible outcomes: 1) chronic understaffed nursing units give rise to latent/active errors and ineff ective care processes or 2) adequately staffed clinical units lead to improved detection of patients at risk for falls. In addition, several studies described in the literature review (see Chapter 3 ) concluded that hospital organizational characteristics, such as hospital bed size, hospital census region, metropolitan location, teaching status and magnet status are associated with variations in levels of nurse staffing and patient outcomes. For example, M agnet status is a certification for hospitals in term s of nursing excellence that has been shown to be associated with reductions in adverse patient outcomes (Lake et al., 2010), as well as increased nursing satisfaction and better nursing work environment s (Kelly et al., 2011). These organizational factors are incorporated into the NSPF model. Furthermore, both nurse staffing and organizational characteristics will play an important role in the processes of patient care, particularly patient risk assessment for patient falls. Process Factors Processes of ca re reflect the manner in which care is delivered to patients within hospital and hospital units (Donabedian, 1966; Donabedian, 1988). In terms of falls, these processes are directly attributed to the ability of nurses to assess and monitor patients (Lake & Cheung, 2008). Fall risk assessments are implemented by nurses and staffing significantly contributes to adequately assessing at risk fall patients (Dempsey, 2004; Lake & Cheung, 2008). As previously mentioned, units with inadequate nurse staffing may be less likely to evaluate at risk patients in a timely manner, which in turn could result in increased risk of patient falls, whereas units with adequate nurse staffing will be better able to monitor and detect patients at risk of falling (Oliver, 2010). The se

PAGE 65

65 processes of care at the unit level are performed by nurses and contribute to awareness, evaluation and detection of patients at risk of falling. Therefore, nursing fall risk assessments and risk assessment scales are conceptually incorporated into the NSPF model as process of care factors. Outcome Factors In the NSPF model depicted in Figure 4 3, outcome measures are the total fall rates and fall rates with injury. These measures are consistent with prior empirical studies that examined the influence of nurse staffing on patient falls (Lake & Cheung, 2006). According to the NSPF model, these negative outcome measures will be higher in units with lower rates of nurse staffing, and conversely, these rates will be lower on units with adequate nurse staffi ng and organizational characteristics that contribute to high rates of staffing levels. Another consideration in this model is the different types of falls, including injurious falls in hospital units. Patient and Policy Contextual Factors Another componen t in this model is the impact of patient characteristics and policy contextual factors, such as the role of the 2008 CMS non reimbursement policy, on hospital structure, process and outcomes factors. First, it is important to consider how patient factors p lay a role in determining adequate nurse staffing at the organizational level, as well as how patient factors can determine whether an inpatient fall will result in injury. This is based on the fact that hospital structure is often determined by community needs assessments and certificates of need in a geographic region (Shi & Singh, 2008). Patient demographic characteristics, such as age and gender play an important role in whether a patient will fall in an acute care setting. In addition, unit level pati ent severity also influences whether a patient will fall on a unit.

PAGE 66

66 However, our dataset lacks several measures of patient level data and the focus of the NSPF is on organizational characteristics that contribute and predict both nurse staffing levels and patient fall outcomes. As a result we will use statistical methodologies, such as sensitivity analyses and proxy measures, to adequately contro l for these factors in our model Specific Aims To address the limitations and gaps in the literature, the spe cific aims are the following for acute care facilities in the U.S.: Specific Aim 1 Assess trajectories of fall rates within acute care hospitals to identify latent class categories of consistently high, low or increasing/decreasing fall rates over time. S pecific Aim 2 Examine the association between changes in unit level nurse staffing and changes in patient falls over time. Specific Aim 3 Determine the relationship between acute care hospital characteristics and patient fall rates over time. Study Hyp otheses Based on prior empir ical evidence and the derived NS PF conceptual framework, hypotheses corresponding to each specific aim have been formulated to evaluate the impact of both organizational characteristics and nurse staffing on patient falls. Beca use of the significant variation in fall rates among hospitals and clinical units in the literature, as well as the 2008 CMS non reimbursement policy, hospitals and hospital units are expected to vary significantly in their fall rates in this study. Theref ore, the goal of Specific Aim 1 was to assess trajectories of acute care hospital patient fall outcomes over a 54 month period of time to identify latent class categories of high, low, or

PAGE 67

67 increasing/decreasing fall rates. Furthermore, a sub aim of Specific Aim 1 was to assess trajectories of acute care hospital clinical unit (i.e. medical, surgical and medical surgical units) patient fall outcomes over a 54 month period of time to identify latent class categories of high, low, or increasing/decreasing fall rates and ascertain which clinical units are significantly contributing to fall rates at the hospital level. Hypothesis 1a Hospital fall rates will vary over time with observed fall rates reflecting latent trajectory groups of high, low, increasing or de creasing fall rates over time. Hypothesis 1b Hospital unit fall rates will vary over time with observed fall rates reflecting latent trajectory groups of high, low, increasing or decreasing fall rates over time. We anticipated that nurse staffing would h ave a positive impact on reducing the fall rates in hospitals given their importance in assessing and monitoring patients in acute care settings. Additionally, given the different types of nurses in hospitals, licensed and unlicensed nursing personnel were anticipated to impact inpatient falls, but increased RN hours per patient day would have a greater impact on falls given their advanced clinical judgment and additional educational requirements. Therefore, the goal of Specific Aim 2 was to examine the ass ociation between unit level nurse staffing and their impact on changes in patient falls over time. It was postulated from this aim that: Hypothesis 2a A higher percentage of RN nursing hours to total number of nursing hours will be associated with members hip into the low patient fall rate trajectory group and a decrease in total patient fall rates over time.

PAGE 68

68 Hypothesis 2b An increase in RN hours per patient day will be associated with membership into the low patient fall rate trajectory group and a decre ase in total patient fall rates over time. Hypothesis 2c An increase in LPN hours per patient day will be associated with membership into the low patient fall rate trajectory group and a decrease in total patient fall rates over time. Hypothesis 2d An in crease in nursing assistant (UAP) hours per patient day will be associated with membership into the low patient fall rate trajectory group and a decrease in total patient fall rates over time. Given our conclusion following the literature review that hospi tal organizational characteristics, such as hospital bed size, metropolitan status, teaching status and magnet status are associated with variations in levels of nurse staffing and patient fall outcomes, the goal of Specific Aim 3 was to determine the rela tionship between organizational characteristics and hospital fall rates over time. Therefore, it was postulated from this aim that: Hypothesis 3a Hospitals with Magnet Status designation will be associated with lower patient fall rates when compared to non magnet hospitals, and they will have a greater likelihood of membership into the low patient fall rate trajectory group Hypothesis 3b Hospitals located in metropolitan areas will have lower patient fall rates when compared to those in non metropolita n areas, and they will have a greater likelihood of membership into the low patient fall rate trajectory group.

PAGE 69

69 Hypothesis 3c Hospitals that are larger in size (greater than 500 beds) will have lower fall rates when compared to hospitals of smaller size, and they will have a greater likelihood of membership into the low patient fall rate trajectory group. Hypothesis 3d Hospitals that are teaching status will have lower fall rates when compared to hospitals that are non teaching, and they will have a gre ater likelihood of membership into the low patient fall rate trajectory group. Although these hospital organizational characteristics are not mutable, it was anticipated that gaining an understanding of how these different types of organizations vary in t heir fall rates over time would enable policymakers and researchers to target resources and develop strategies to specifically reduce fall rates in hospitals with different characteristics. This approach could potentially achieve a higher success rate of r educing falls by tailoring these interventions to fit the needs of various hospital organizations.

PAGE 70

70 Figure 4 1. Nurse staffing and patient outcomes m odel (Cho, 2001)

PAGE 71

71 Figure 4 2. Hospital organization, nurse organization and patient outcomes model (Aiken, Clarke & Sloane, 2002)

PAGE 72

72 Structure Process Outcomes Figure 4 3. Integrated nurse staffing and patient fall conceptual m odel Patient Factors and Policy Context 1. Unit level Patient Demographics & Acuity 2. 2008 CMS Non Reimbursement Policy Organizational Factors Magnet Status, Hospital Size, Census Region, Metropolitan Status, Teaching Status Unit Level Organizational Characteristics 1. Nurse Staffing 2. Unit Type Organizational Processe s 1. Fall Risk Assessment Scale 2. Fall Prevention Protocol in Place Patient Falls 1. Tot al Patient Fall Rate 2. Injurious Fall Rate

PAGE 73

73 CHAPTER 5 DATA AND METHODS This chap ter discusses the data source and variables used in this study, as well as the methods for analyzing each specific aim. This study was conducted using a retrospective, l ongitudinal study design with data collected over a 54 mon th period of time from July 2 006 through December 2010. The National Dataset for Nursing Quality Indicators (NDNQI) In 1996, the American Nurses Association (ANA) conducted pilot tests in seven states and collected information on nurse sensitive indicators that would be used to fo rm a nationwide database (ANA, 1997). Using information from 36 hospitals, the National Database for Nursing Quality Indicators (NDNQI) was created by the ANA in 1998 for the purposes of enhancing research of nurse sensitive indicators at the unit level an d providing a tool for hospitals to compare quality metrics (Montalvo, 2007). The conceptual framework, and the collection and management of information for this database oc curs at the University of Kan sas Medical Center (KUMC) School of Nursing under contract from the ANA (Dunton et al., 2004). Since 1998, the number hospitals participating in the database has significantly grown, and nearly one quarter of all general hospi tals in the United States provided information for the NDNQI in 2009 (Lake et al., 2010). Participation in the NDNQI is voluntary for hospitals, although it does provide benchmark quality measures, as well as satisfies state and federal reporting requirem ents for hospital accreditation (Montalvo, 2007). Hospitals who participate submit unit level information quarterly to the NDNQI, and as part of a contractual

PAGE 74

74 agreement between the NDNQI and the hospitals, all identifying information is excluded from the d ata, such as name of facility, hospital ID number and address (Montalvo, 2007; Lake et al., 2010). Hospitals are provided a secure website to submit their information, and the NDNQI conducts extensive data audits to ensure accuracy, including data review t ools, error reporting on outlying information and feedback on common data error entry mistakes (Dunton et al., 2004; Dunton et al., 2007). In addition, the NDNQI performs reliability studies on collected indicators, such as total number of patient days (Du nton et al., 2007; Simon, Yankovskyy, Klaus, Gajewski & Dunton, 2010; Simon Yankovskyy & Dunton, 2010). These studies have demonstrated high levels of inter rater reliability and rater to standard reliability, as well as standards of compliance with NDN QI guidelines (Hart et al., 2006; Dunton et al., 2007). Measures for Data The following section outlines the operationalization of variables incorporated in the study model, including the outcome measures, nurse staffing measures, organizational characte ristics of hospitals, and process of care variables. Table 5 1 summarizes the measurement and operationalization of these variables in ou r study, including whether the measures are reported o ver time on a monthly, quarterly, or yearly basis. Fall Outcome Variables The outcome of interest in the current study was the rate of patient falls per 1 000 patient days, with the rate of injurious falls per 1 000 patient days used in a descriptive analysis. These are commonly used measures to determine the rate at w hich patients fall within hospitals (NDNQI, 2007; Dunton et al., 2007; Lake et al., 2010). In addition, these fall indicators are reported to NDNQI on a monthly basis and falls are identified

PAGE 75

75 through incident reports. A patient day was defined as a total o f 24 hours beginning the day of admission and not including the day of discharge. For this study, the total fall rate variable was calculated by dividing the total number of patient falls in a month by the total patient days and multiplying by 1 000 as sh own in Equation 5 1 below Monthly Fall Rate = (Monthly fall count/total patient days) x 1 000 (5 1) In the NDNQI, injurious falls are reported as a categorical variable based on the level of severity of the patient injury from falling: 1. None indicates that the patient did not sustain an injury secondary to the fall 2. Minor indicates that those injuries requiring a simple intervention 3. Moderate indicates injuries requiring sutures or splints 4. Major injuries are those that require surgery, casting, further examination ( e.g. a neurological injury). 5. Deaths refers to those that result from injuries sustained from the fall. To determine the injurious fall rate, the categorical injury variable in the NDNQI was dichotomized to measure falls with (including all types of injury) and without injury. This dichotomized variable was then used to calculate the injurious fall rate by dividing the total number of patient falls with injury in a month by the total patient days and multipl ying by 1 000 as shown in Equation 5 2 below Injurious Fall Rate = (Fall Count with Injury/Total Patient Days) x 1 000 (5 2) Nurse Staffing Variables Nurse staffing was measured according to the recommendations from the NQF (2004) and previous expert recommendations on the most effective and useful nurse staffing measurements (Dunton et al., 2004; Dunton et al., 2007; Van den Heede et al., 2007; Lake et al., 2010) Therefore, percent RN skill mix, RN hours per patient day,

PAGE 76

76 nursing assis tant (UAP) hours per patient day, and licensed practical nurse (LPN) hours per patient day were used as primary predictor variables in our study. The measures of nurse staffing were operationalized in the following ways: RN skill mix percentage was calcul ated by dividing the total RN patient care hours by the total nursing care hours on a unit and multiplying by 100. RN hours per patient day were calculated by dividing total RN patient care hours on a specific unit by the total patient days in that month. Nursing assistant (UAP) hours per patient day were calculated by dividing the total UAP patient care hours on a given unit by the total patient days in that month. Finally, LPN hours per patient day were calculated by dividing the total LPN patient care ho urs on a given unit by the total patient days in that month. All nurse staffing variables varied by time on a monthly basis, which means that the NDNQI captures information at the unit level on a monthly basis for these measures. Also, given previous nurse staffing shortages and high rates of turnover in certain units, hospitals have used temporary agency nurses to assist in providing coverage for understaffed units. As a result, nursing hours provided by contract nurses on each unit were incorpora ted in ou r calculation of nursing hours. Hospital Organizational Variables Findings from previous studies have shown that nurse staffing and patient falls differ by organizational characteristics of hospitals, including metropolitan status, census region, teachi ng status, total bed size and Magnet certification status (Dunton et al., 2007; Blegen et al., 2008; Mark & Harless, 2007; Lake et al., 2010). These structural elements that impact quality of patient care were considered in the model to assess their influe nce on patient fall rates over time. In addition, hospital level organizational variables included in our study were not time varying, as they were only reported to

PAGE 77

77 NDNQI at one point in time. Metropolitan status was dichotomized as hospitals located in me tropolitan versus non metropolitan locations. NDNQI reports metropolitan status as either metropolitan, micropolitan or neither. For the purpose of this study, a metropolitan location was reported as a single county or group of adjacent counties with a pop ulation of 50,000 in an urban area (NDNQI, 2007). Next, census region was operationalized as a categorical variable that reflected the loc ation of the hospital in the Northeast, Midwest, South or West region of the United States. Teaching status in the ND NQI reflected a categorical variable that distinguished between an Academic Medical Center (a hospital serving as a clinical site for a college of medicine), teaching hospital and a non teaching hospital. For the purpose of this study, we dichotomized this variable as either teaching (teaching and academic medical center) or non teaching. Total bed size was categorized as less than 100 beds, 100 199 beds, 200 299 beds, 300 399 beds, 400 499 b eds and greater than 500 beds. Finally, Magnet status was a dichot omized variabl e used to differentiate Magnet certified hospitals from non Magnet hospitals. Unit Level Organizational Variables At the unit level, there were three types of adult units that were used for our study: 1) medical, 2) surgical and 3) medical/s urgical units. In practice, these units differ based on their staffing, patient population and services offered. For instance, medical nursing units typically treat higher acuity patients and these patients are at higher risk of falls when compared to surg ical units (Dunton et al., 2004). Additionally, to determine and proportion of male and female patients by using the NDNQI reported characteristics of patients who f ell over this timeframe.

PAGE 78

78 Process of Care Variables Process variables incorporated in our study included the presence of a fall prevention protocol on each unit and the type of fall risk assessment scale used by each unit (Shorr et al., 2002). The presen ce of a fall prevention protocol was measured as the proportion of units with a fall prevention protocol. In the NDNQI, the presence of a fall prevention protocol was only assessed when a patient fell on a particular unit, and observations are missing when there were no falls on a particular unit during a given month. Therefore, if a hospital unit had a fall prevention protocol in place before and after a zero fall month observation, we assigned that fall protocol value (yes or no) for the month without a f all. This process resulted in less than 10% of missing values for this variable. Lastly, the type of fall risk assessment scale was operationalized as a categorical variable that reflected the risk assessment instrument used by each unit, including the Mor Study Design and Statistical Methods Our research study was a retrospective longitudinal study that used latent class growth modeling (LCGM) to determine latent class trajectory groups of hospital fall rates and g eneralized estimating equations (GEE) to understand external factors that predicted membership into latent class trajectory groups with the highest fall rates, as well as to test the effects of nurse staffing and organizational characteristics on fall rate s in hospitals over time. We took advantage of the key features from the NDNQI, analyzing structure, process and outcome measures at the unit level. For the purpose of this study, NDNQI data were examined from July 1 st 2006 through December 31 st 2010, wh ich represents 54 months of data. We excluded hospitals unit observations without complete monthly patient fall count data and nursing hours from our study.

PAGE 79

79 Table 5 2 offers a comparison of the original sample of fall observations and the study sample of f all observations after excluding hospital units without complete information. In addition, adult medical ( i.e. general medical or medical specialty care including cardiac, infectious disease, bone marrow transplant), surgical ( i.e. general surgical or surg ical specialty care including bariatric, transplant, trauma and cardiothoracic), and medical/surgical units ( i.e. provide care to both types of patients) were included in this analysis, as these represent a majority of all patient care units in hospitals. Our final sample size was 1,592 hospitals and 7,871 hospital units. All statistical analyses were conducted using Statistical Analysis Software ( SAS ) version 9.1 ( SAS 2004). Analysis by Specific Aim Latent Class Growth Modeling (LCGM) To analyze Specif ic Aim 1, latent class growth modeling (LCGM) was selected to categorize the change in hospital fall rates (average aggregated hospital fall rate per month) over the 54 month time period, as well as to evaluate the change in unit level fall rates over this timeframe. The unit of analysis for this method was the hospital level for Specific Aim 1. In addition, we also conducted LCGM on aggregated medical, surgical and medical surgical unit fall rates to assess variability in fall rates at the unit level over time, as well as to determine if there was an average causal effect when aggregating fall rates to the hospital level. This method of analysis was used to develop common latent class categories among hospitals and hospital units based on their fall rates o ver time (Duncan & Duncan, 2004). For instance, if latent categories produce similar straight lines in a given sample, the trajectory classes would reflect the average differences in the intercept and slopes for those given hospitals, and these could be us ed for further analysis in our study. LCGM was used to summarize growth at the

PAGE 80

80 hospital and hospital unit level, and findings were applied to determine which predictors exert an impact on hospital and unit level latent class trajectory membership and fall rates over time (Muthn, 1997; Duncan & Duncan, 2004). Traditional latent growth curve models (LGCMs) are developed using methodology from structural equation modeling (SEM), and they require many of the same statistical assumptions, such as the inclusio n of large sample sizes and data collection occurring at the same time as when the subjects are observed (Duncan & Duncan, 2004). This SEM framework in LGCMs can be used on repeated measures of data to model hospital fall rate growth trajectories. In addit ion, this statistical technique assumes that changes in hospital falls would be systematically related to the passage of time, which is consistent with our dataset that reports 27 months before and after the 2008 CMS non reimbursement policy (Burchinal & A ppelbaum, 1991). LGCMs can include both fixed and time to study within hospital effects (the effect of changes in i ndependent variables on dependent variables within the hospital) or between hospital effects (the effect of the mean of the independent variable on the dependent variable) (Duncan et al., 2006; growth curve methodology consists of two stages: (1) a regression curve, not necessarily linear, is fit to the repeated measures of each individual in the sample and (2) the parameters for n the original create a latent construct (Unruh & Zhang, 2012).

PAGE 81

81 Given prior assumptions from traditional growth curve modeling that a single growth model or single avera ge growth estimate can adequately estimate the fall performance in the entire population of hospitals in the NDNQI, we incorporated latent class growth modeling (LCGM) to determine categories of multiple change patterns among hospitals longitudinally (Nagi n, 1999; Raudenbush, 2001; Nagin, 2005; Jung & Wickrama, 2008). The rationale for this is that subpopulations of the dataset may emerge from the data whose growth trajectories are inherently different than the single growth estimate, such as variations in hospital fall rates in our model (Jung & Wickrama, 2008). These latent class growth models (LCGM) have been employed extensively in the fields of sociology, psychology and criminology (Jones et al., 2001; Nagin, 2005; Jung & Wickrama, 2008), where modeling trajectories of individual behavior patterns, such as juvenile physical aggression, opposition and hyperactivity (Nagin & Tremblay, 1999), have provided researchers with novel information concerning behavioral development over time for a given population. In addition, these methods have been utilized in previous health care studies (Laptook, Klein & Dougherty, 2006; Umberson, Williams & Powers; Liu & Needham, 2006; Fergus, Zimmerman & Caldwell, 2007; Schmiege, Meek, Bryan & Petersen, 2012), such as ass essing distress trajectories of newly diagnosed breast cancer patients (Henselmans et al., 2010). However, to our knowledge, these methods have not been utilized in the in the patient safety and hospital falls literature. LCGMs create group based traject ories by assessing change patterns in given outcomes across multiple units of time (Nagin, 1999; Nagin, 2005; Andruff et al., 2009). This semi parametric method allows researchers to classify subgroups of a population

PAGE 82

82 following a consistent pattern of chan ge in a given outcome by modeling probability distributions that are best specified to fit the data (Jones et al., 2001; Andruff et al., 2009). For example, although each hospital and hospital unit has a distinct course of fall rates over 54 months, the ch ange patterns between individual hospitals are summarized by an intercept and slope that corresponds equally to a given trajectory of hospitals (Nagin, 1999; Nagin, 2005; A ndruff et al., 2009). Figure 5 2 illustrates our latent class growth model study des ign diagram, representing repeated measures of aggregated hospital fall rates per month. The intercept of this time varying latent variable can be interpreted as the initial mean hospital fall performance measure of a given hospital trajectory, the slope i s the average change in hospital fall performance rates per month for a hospital trajectory, and the categorical variable is the latent class type. unobserved populat ama, 2008, p. 304). LCGMs are suited to identify unobserved heterogeneity in a given population, which means that membership in a particular group is unknown prior to analysis of the data (Jones et al., 2001). These groups are therefore called latent classe s because their membership in a trajectory group is unobserved or latent (Lubke and Muthen, 2005). Due to the fact that hospitals and hospital units are a mixture of distinct heterogeneous groups with variations in organizational and unit level characteris tics (e.g. size), group based modeling is appropriate to allow for analysis of changes in monthly fall rates over time using both unadjusted and adjusted methodology (Jones et al., 2001; Nagin, 2005). The LCGM accomplishes this by creating latent trajector y classes, or latent categorical variables, which encompass different classes of growth trajectories around various means (Jung &

PAGE 83

83 Wickrama, 2008). A maximum likelihood function is used for a specific probability distribution to generate model parameters ba sed on a general quasi Newton procedure (Dennis, Gay & We lsch, 1981). Figure 5 1 demonstrates the probability function and maximum likelihood e stimation of the zero inflated Poisson (ZIP) model that will be used in our study (Jones et al., 2001), where: Pr(Y i = y i ) = Probability of observing the data at trajectory y i ijk = Extra Poisson probability of a zero When conducting LCGM analys is, Jung and Wickrama (2008, p. 307 316) suggest using the following steps to generate these latent class categories: 1) specify a single class latent growth curve model, (2) specify an unconditional latent class model without covariates, (3) determine the number of classes, (4) address convergence issues, and (5) specify a conditional latent class model with covariates. The product of this approach is the creation of separate categories of growth models for each latent class, which were individually analyzed in Specific Aims 2 and 3. These separate categories were anticipated to indicate whether a hospital and/or hospital unit is high performing, low performing or increasing/decreasing i n terms of fall rates over this period of time. SAS PROC TRAJ Procedure Latent class growth analysis was employed in this study using PROC TRAJ in SAS 9.1 ( SAS Institute, 2004). The PROC TRAJ procedure is a group based mixture model procedure created b y Nagin (1999; 2005) for the estimation of trajectories, allowing for flexibility in the specification of probability distributions, the number of latent class categories generated, and both time varying and time invariant covariates (Jones et al., 2001). Overall, determining the best model fit is an iterative process that requires

PAGE 84

84 both subjective knowledge of the dataset and interpretation of measures of statistical fit (Nagin, 1999). The sections that follow outline the methodology for model selection usi ng PROC TRAJ, including specifying a distribution of the dependent variable in the model, determining the number of latent classes to be used, accounting for missing data, as well as providing a detailed outline of each step in the LCGM analysis process fo r this study. In addition, all SAS PROC TRAJ coding and relevant SAS state ments can be found in Appendix D Determining Probability Distributions and Trajectory Shape Using PROC TRAJ, LCGMs require the specification of one of three probability distrib utions of the dependent variable that can used to estimate parameters in the model: (1) cens ored normal, (2) zero inflated Poisson and (3) logistic models (Jones et al., 2001). The censored normal (CNORM) model is typically used for psychometric scale data and it is helpful when modeling data that clusters at the minimum and maximum points of the scale, but can also be used for normally distributed continuous data with specified maximum and minimum values outside of the range of the dataset (Jones et al., 2001). The zero inflated Poisson (ZIP) model is useful for count data with a greater number of zeroes th an would be expected under the Poisson assumption (Lambert, 1992). Lastly, the logistic (LOGIT) model is used to model binary or dichotomous data (Jones et al., 2001). For the purposes of this study, both the cen sored normal and zero inflated Poisson could be used to model fall rates over ti me. However, the zero inflated Poisson (ZIP) was most appropriate given the non normal distribution of falls and a h igher number zer oes than expected (see Figure B 1 in Appendix B for a graphic representation of the fall rate distribution), as well as a lack of

PAGE 85

85 clustering around maximum and minimum values in our dataset that is typically found in the censored normal dis tribution. Additionally, PROC TRAJ allows for specification of the shape of each trajectory group, which is a polynomial function of the independent variable as a linear, quadratic or cubed term (Jones et al., 2001). The linear trend may gradually increa se or decrease at varying points in time or remain constant, a quadratic trend may change (increase, decrease or remain constant) up to a certain point before changing in magnitude or direction, and a cubic trend typically has at least two changes in magni tude or direction over time (Andruff et al., 2009). Preliminary descriptive statistics guided model selection in addition to empirical evidence from the literature for specifications on optimal model fit (Jones et al., 2001). For each model, we tested the linear, quadratic and cubic functions of each trajectory up to a maximum of five trajectories. Following the recommendations of Helgeson, Snyder and Seltman (2004), we eliminated non significant (p>0.05) quadratic and cubic terms, but retained non signifi cant linear terms. Once completed, we retested the models to obtain values of Bayesian Information Criteria (BIC) and percent membership in trajectory groups to ascertain model fit. Determining Number of Latent Class Categories and Final Model Selection Determining the number of latent class categories and final model selection is such as the Bayesian Information Criterion (BIC) and the log Bayes approximation factor (Na gin, 1999; Nagin, 2005; Andruff et al., 2009). Jones and colleagues (2001) likelihood evaluated at the maximum likelihood estimate less one half the number of parameters in the model times the log of the sample (p. 390). The BIC is consistently a negative number and empirical evidence suggests

PAGE 86

86 selecting the least negative value of the BIC for model selection (Kass & Raftery, 1995; Nagin, 2005). The change in BIC from one model to another can then used to estimate the BIC log Bayes approximation factor, which is represented by Equation 5 3 below : 2log e (B 10 (5 3 ) g the more complex model (i.e. the model with more latent class categories) from the less complex model (i.e. the model with fewer latent class categories), and findings from this are interpreted as evidence favoring or against the alternative mode l (Jones et al., 2001). Table 5 3 displays the ranges and interpretation of the log Bayes approximation factor. The Bayes factors (B 10 prior probability that the alternative hypothesis is correct equals one 2001, p. 390). Following the guidance summarized in Table 5 3, values over 6 are considered strong evidence in support of the more complex model. Finally, another measure of model fit is the Akaike Information Cri teria (AIC) (Akaike, 1974), which is similar to the BIC and does not vary based on sample size (Nagin, 2005). However, the AIC measure has been found to be less reliable than the BIC in research studies because it is more likely to select unnecessary addit ional model groups (Nagin, 2005). Therefore, for the purpose of this study, we used the BIC and log Bayes approximation factor, the percentage of hospital trajectory membership and guidance from published studies to choose the most parsimonious model with the appropriate number of latent class groups. PROC TRAJ allows researchers to specify the number of latent class categories to include in the model using the NGROUP statement and the shape of the trajectory

PAGE 87

87 using the ORDER statement (Andruff et al., 200 9). To determine an adequate number of latent classes, we followed the recommendations of Nagin (2005) by utilizing an iterative stepwise approach by first running a one group linear model (ORDER=1), then fitting a two group linear model (ORDER=1,1), then a three group linear model (ORDER=1,1,1) until a maximum number of five groups h ave been modeled (see Appendix D for PROC TRAJ codes). We then compared the BIC and log Bayes approximation values of these groups after each step, continually analyzing the pa rameter estimates for significant p values less than 0.05 and percentages of group membership in each latent class trajectory, which should be at least five percent (Andruff et al., 2009). Next, we repeated this process with more complex polynomial shapes of quadratic and cubic forms, continually changing latent classes and ordering based upon statistical significance and judgment from the literature. As previously mentioned, we excluded statistically insignificant quadratic and cubic functions, as well as models with less than five percent group membership (Andruff et al., 2009). This iterative process of model selection was conducted on trajectory models for both monthly hospital fall rates and monthly hospital unit (medical, surgical and medical surgical units) fall rates. Finally, after selecting the number of latent class categories and checking model fit, we analyzed the posterior probabilities, which represent the probability a given hospital is in a particular trajectory group, where a maximum probab ility was to assign hospitals membership into a category for which they had the highest probability of membership (Jones et al., 2001; Andruff et al., 2009). For instance, if three latent class trajectories emerged from a model and a given hospital was ass igned posterior

PAGE 88

88 probabilities of Trajectory 1=0.15, Trajectory 2=0.05, and Trajectory 3=0.8, the probability assignment wou ld be 0.8 from Trajectory 3. These posterior probabilities can then be used to calculate the average posterior pro bability of membership for hospitals and hospital units in particular trajectories as a test of reliability that distinguishes between hospitals with different change patterns (Andruff et al., 2009). Overall, average posterior probabilities should be great er than 0.7 and these were calculated for each model in our study (Nagin, 2005; Andruff et al., 2009). Approaches to Address Missing Longitudinal Data When analyzing longitudinal datasets, a common methodological issue pertains to missing data and attr ition due to dropout of subjects with repeated measures over time (Hedeker & Gibbons, 1997; Twisk & deVente, 2002). Since the number of hospitals in the NDNQI increased steadily over the 54 month timefra me of our study (see Table 6 1 in Chapter 6 for total hospital fall observations per month), it is important to consider attrition and incomplete data from hospitals that began reporting to the NDNQI at different time points and hospitals dropping out of the dataset, which represents only a small percentage of hospitals (less than 10%) who started in July of 2006 and dropped out of the sample by December of 2010. According to Little and Rubin (1987), there are three distinct types of missing data: (1) Missing completely at random (MCAR), in which missingness of data does not depend on characteristics of participants in the dataset, observed responses or characteristics and, (3) Missing not at random (MNAR), in which missingnes s depends on unobserved or some observed characteristics. In addition, Laird (1988) further defined missing data as ignorable versus non ignorable, where

PAGE 89

89 MAR is an ignorable pattern is missingness, and MNAR is non ignorable and may lead to biased coefficie nts. The SAS PROC TRAJ procedure assumes missing data is MAR and subjects are included in the analysis if they have at least one observation of valid data on the dependent variable, and any observation with completely missing trajectory data is removed f rom the analysis (Nagin, 1999; Nagin, 2005; Dodge, Shen & Ganguli, 2008). Therefore, it was critical in the current investigation to consider how missing values could have impacted the overall analysis. We accomplished this by conducting a sensitivity ana lysis using PROC MI to account for both monotone and non monotone patterns arise when missing data points occur at the beginning and end of datasets (i.e. when hospitals ente r or drop out of the dataset), whereas non monotone missing data patterns exist when participating hospitals have missing values in between continuous SAS PROC MI allows for imputation of both monotone and non monoto ne missing values using the Monte Carlo Markow Chain (MCMC) methodology, which uses algorithms to sample various probability distributions assumptions, we conducted multiple imputation using PROC MI to impute hospital fall rate values in our dataset to determine how missing values affected our results. The findings from sensitivity analyses are summarized in the results section, with addition al detail provided in Appendix B Steps to Address Specific Aim 1 and 1a The following section outlines the steps in the analyses for Specific Aim 1 and Specific Aim 1a. Model selection methodology is provided in the results section

PAGE 90

90 (Chapter 6), including a table of the BIC, log Bayes app roximation factor, and the percentages of hospitals and hospital units that reside within certain latent class trajectories. In addition, we provided a graphical representation of the models in each step and calculated average posterior probabilities for r eliability and model adequacy. Specific Aim 1 : Assess trajectories of fall rates within acute care hospitals to identify latent class categories of consistently high, low or increasing/decreasing fall rates over time. In order to conduct the analysis us ing PROC TRAJ, all data were reshaped into a month by creating separate monthly fall rate variables. This was done by using PROC TRANSPOSE in SAS Next, for the adju sted fall rate models, binary dummy variables were created for all categorical predictors, as PROC TRAJ does not allow for differentiation between continuous and categorical predictors (Nagin, 2005). Therefore, dummy variables were created for time invaria nt categorical organizational variables of teaching status, M agnet status, hospital bed size, metropolitan status and census region. In addition, time varying nurse staffing and process of care covariates (e.g. fall prevention protocol in place) were also created by using PROC TRANSPOSE for the adjusted models. In sum, multiple datasets were created to accommodate each of the models in our study. Step #1: Unadjusted monthly hospital fall model Step #1 was conducted to determine trajectories of hospital fa ll rates over 54 months without adjusting for covariates in the model. Fall rates were aggregated to the hospital PROC TRAJ was used to model hospital fall rates monthly and dete rmine the average

PAGE 91

91 tr ajectories over time. Figure 5 2 provides a depiction of the generated intercept and slope for each latent class category from month 1 through month 54. The results of this model are presented in the results section in the next chapter. Step #2: Adjusted monthly hospital fall model Step #2 was conducted to determine trajectories of hospital fall rates over 54 months while controlling for time varying nurse staffing covariates and time invariant organizational characteristics in the mode l separately as a form of sensitivity analysis. According to Jones and colleagues (2001), time invariant covariates influence the probability of trajectory membership, whereas time varying covariates affect the levels of the observed trajectory. Therefore, fall rates, nurse staffing variables and organizational characteristics were aggregated to the hospital was created to analyze fall rate trajectories. PROC TRAJ was used to model hospital fall rates monthly and determine the ave rage trajectories over time. Time varying nurse staffing measures were incorporated in the model by using the TCOV statement and time invariant organizational factors were incorporated in the model by using the RISK statement in PROC TRAJ. Unit level fall risk assessment scale measures were not incorporated in this analysis, as these measures cannot be aggregated to the hospital level because units within the same hospital may use different fall ri sk assessment scales. Figure 5 3 provides a depiction of the generated intercept and slope for each latent class category from month 1 through month 54. The results of this model are presented in the next chapter, and tables and fi gures can be found in Appendix B Specific Aim 1a : Assess trajectories of fall rates within acute care hospital clinical units (medical, surgical and medical surgical units) to identify latent class categories of fall rates over time.

PAGE 92

92 Step #3: Unadjusted monthly hospital unit fall models Step #3 was conducted to address Specific Aim 1a and determine trajectories of hospital unit (medical, surgical and medical surgical unit) fall rates over 54 months without adjusting for covariates in the model. In addition, Steps #3 and #4 were conducted to better understand the variation in hospital un its and to ascertain whether aggregated hospital fall rate trajectory models remove the variation seen at the unit level. Therefore, fall rates were aggregated to the unit created to model these fall rate trajectories. PROC T RAJ was used to model unit level fall rates monthly and determine the average trajectories over time. A total of three separate analyses were conducted on medical, surgical and medical surg ical unit fall rates. Figure 5 4 provides a depiction of the genera ted intercept and slope for each latent class category from month 1 through month 54. The results of these models are pre sented in the next chapter. Step #4: Adjusted monthly hospital unit fall models Step #4 was conducted to determine trajectories of hos pital unit (medical, surgical and medical surgical) fall rates over 54 months while controlling for time varying nurse staffing covariates and time invariant process of care measures as a form of sensitivity analysis. Therefore, fall rates, nurse staffing variables and process measures were aggregated to the unit trajectories. PROC TRAJ was used to analyze unit level fall rates monthly and determine the average trajectories over time. Time varying nu rse staffing measures and time varying monthly unit protocol measures were incorporated in the model by using the TCOV statement and time invariant process factors (e.g. risk assessment scales) were incorporated in the model by using the RISK statement in PROC TRAJ. Hospital

PAGE 93

93 level organizational characteristics were not incorporated in this model because this was strictly a unit level analysis to determine trajectories of units based on whether they were reported as medical, surgical or medical su rgical (Na gin, 2005). Figure 5 5 provides a depiction of the generated intercept and slope for each latent class category from month 1 through month 54. The results of these models are presented in the next chapter, and tables and figures can be found in Appendix B Step #5: Missing data models and additional sensitivity analyses In order to ensure the previous LCGM results accurately describe trajectories of hospital falls over time, we conducted multiple imputation using both monotone and non monotone missing stat ements in PROC MI to account for missing data and attrition in the NDNQI. We imputed five different datasets to make comparisons between imputed values and non imputed trajectory model analyses, with final results calculated according to n rules (Rubin, 1987 ). In addition, we conducted a LCGM at the hospital quarter level to assess any average causal effects in the models between monthly and quarterly hospital fall rates that could have been produced by averaging fall rates across quarters instead of months. The results of these sensitivity analyses are summarized in the results sec tion and presented in Appendix B Generalized Estimating Equations (GEE) In the following section, we provide a detailed description of our analysis plan for Sp ecific Aims 2 and 3: Specific Aim 2 : Examine the association between changes in unit level nurse staffing and changes in patient falls over time Specific Aim 3 : Determine the relationship between acute care hospital characteristics and patient fall rates over time

PAGE 94

94 To test Specific Aims 2 and 3, a generalized estimating equation (GEE) model approach was selected to estimate coefficients and determine the impact of both nurse staffing and organizational characteristics on hospital fall rates over time, as w ell as to group. Created by Liang and Zeger (1986), the GEE method was formulated to produce dinal or repeated measure research designs with non 2004, p. 128). When compared to ordinary least squares regression (OLS), GEEs are a form of general linear modeling (GLM) that are able to best estimate probabilities of depen dent variables using data longitudinally (Gardner, Mulvey & Shaw, 1995). General linear models allow for modeling of non normally distributed dependent variables, are fitted using maximum likelihood estimation, and require the specification of a link func tion that describes the relationship between the systematic component (linear predictor) and the expected value of the dependent variable (McCallagh & Nelder, 1989; Ballinger, 2004; Dobson & Barnett, 2008). GEE incorporates a correlation matrix that encomp asses the within subject correlation of responses of dependent variables with many different distributions (e.g. normal, binomial and negative binomial distributions), and produce coefficients that are similar to OLS regression models with a normal depende nt variable distribution (Ballinger, 2004, p.130). As a result, GEEs are able to calculate appropriate standard errors that account for within hospital clustering and repeated measures (Lake et al., likelihood estimation of and the variance calculated using a link function, which is a transformation function that

PAGE 95

95 allows the dependent variable to be expressed as a vector of parameter estimates ( y = 0 1 X 1 2 X 2 3 X 3 .) in the form of an Additionally, GEEs incorporate a variance function in the form of a transformation matrix that calculates the variance of the parameters (McCullagh & Nelder, 1989). The results of these calculations form a matrix o variance (Ballinger, 2004). The output from these equations is then utilized in starting involves minimizing the exten t of change in the parameter estimates from a perfectly Nelder, 1989; Ballinger, 2004, p. 131). This allows for stabilization and interpretation of and standard errors) in the GEE model. Overall, using GEE requires specification of (1) the link function to be used, (2) the distribution of the dependent variable and, (3) the correlation structure of the dependent variable (Ballinger, 2004, p.131), wh ich are described below in the analytic approach for Specific Aims 2 and 3. Other considerations for utilizing GEE include the possibility of interaction terms in the model, which were formally tested in this study. The commonly used correlation structures in GEE are independence, exchangeable, autoregressive and unstructured (Ballinger, 2004). Incorrectly specified correlation structures have the potential to bias According to Ballinger (2004), autoregressive correlation struc tures are useful when modeling data that are repeated clusters longitudinally, exchangeable correlation structures are used when the observations are equally correlated, unstructured correlation structures are modeled when researchers want to obtain estima tes of all correlations between within

PAGE 96

96 hospital responses, and the ind ependence correlation structures assume that observations are independent of each other, which does not account for the within subject correlation (Fitzmaurice, Laird & Rotnitzky, 1993). For the purpose of this study, we ran multiple models with each correlation structure and dependent variable distribution to ascertain the best fit based on the nature of the relationship between independent and dependent variables, as well as used compa risons between standard errors and the lowest Pans quasi likelihood under the independence model information criteria (QIC) score, which is a useful measure when choosing the appropriate correlation structure and overall model fit (Ballinger, 2004). Additi onally, log likelihood estimates were also used to determine adequate model fit. All analyses were conducted in SAS version 9.1 using the PROC GENMOD procedure ( SAS Institute, 2004) and relevant SAS codes can be found in Appendix D Final model selectio n was also determined by using bivariate analysis for significant predictors of falls, as well as correlations between predictors to reduce the threat of multicollinearity. The following section outlines the two steps of analyses that were conducted for th is study and the specifications used for each procedure: GEE a nalysis #1: Consistently high trajectory group predictors at the hospital unit level For this analysis, we used a GEE approach to account for repeated measures in the NDNQI over time and cluste ring of units within hospitals. We examined the impact of nurse staffing and organizational characteristics on the probability of latent class that arose from the results of Specif ic A im 1, controlling for all other variables in the model (i.e. proportion of a fall prevention protocol in place and unit type). All measures were

PAGE 97

97 aggregated to the hospital unit level, and consistent with methodology from Hyer and colleagues (2011), a v among months of measurement in the same hospital unit over time. The dependent variable in this model was a dichotomous variable that distinguished between hospital our sample (i.e. a combination of h rate trajectory groups). Independent variables in this model included nurse staffi ng (RN skill mix percentage, RN HPPD, LPN HPPD and UAP HPPD) and organizational characteristics (metropolitan status, census region, teaching status, total bed size, and Magnet certification status) and controlled for unit level process of care measures. Since unit type is known to be associated with high and low fall rates in hospitals (i.e. medical units have higher fall rates when compared to surgical and medical surgical units) and nurse staffing measures vary by unit (Dunton et al., 2004; Lake et al 2010), we examined the hospital unit level analysis by accounting for clustering of units within hospitals. Therefore, using unit level nurse staffing, as well as incorporating unit type (i.e. medical, surgical and medical surgical units) in the model as a form of risk adjustment, we modeled the GEE using a binomial distribution with a logit link and an unstructured correlation structure, which incorporates a correlation matrix that estimates every possible correlation between within subject responses and inputs these correlations into the calculation of the variances (Fitzmaurice, Laird & Rotnitzky, 1993; Ballinger, 2004). These are commonly used distributions and link functions when conducting logistic regression using GEE (Ballinger, 2004). From this an alysis, we

PAGE 98

98 hypothesized (see Chapter 4 for hypotheses) that greater levels of nurse staffing and org anizational characteristics of M agnet status, teaching status, metropolitan locations and larger hospital bed size would be associated with a greater odds o f membership in the consistently lower group trajectories (representing the trajectories at or below the In addition, calculations from the GEE model provided odds ratio estimates, standard errors and p values, where statistical significance is determined at the p<0.05 level. The GEE latent class category analysis is represe nted by E quation 5 4 below : logit {E(Y \ roup Membership)=B 0 + B 1 Nurse_Staffing + B 2 Hospital_Factors + B 3 Process_Factors + u (5 4 ) Due to concerns that collapsing measures across unit types in hospitals could potentially impact or bias the a ssociation between nurse staffing and trajectory group membership, we used GEE methodology to predict nurse staffing and organizational characteristics at the hospital level as a form of sensitivity analysis. This was done to examine the potential for an a verage causal effect resulting from the aggregation of measures to the hospital level and to determine the impact of accounting for correlations of units clustered within hospitals from GEE Analysis #1. An unstructured correlation was selected as the most appropriate correlation structure for our hospital level model. Therefore, using aggregated hospital nurse staffing and falls measures we modeled the GEE using a binomial distribution with a logit link at the hospital level. Findings from this sensitivity analysis are summarized in the results section, with addition al detail provided in Appendix C, Table C 1

PAGE 99

99 GEE a nalysis #2: Total monthly falls at the hospital unit level For the analysis predicting total monthly falls, GEE was employed to predict the imp act of nurse staffing and organizational characteristics on total monthly patient fall rates. All measures were analyzed at the hospital unit used to adjust for the correlation among months of measurement in the same hosp itals and units over time (Hyer et al., 2011). The dependent variable in this model was the monthly hospital unit fall count per month. The independent variables in this model included nurse staffing (RN skill mix percentage, RN HPPD, LPN HPPD and UAP HPPD ) and organizational characteristics (metropolitan status, census region, teaching status, total bed size, and Magnet status) and controlled for process of care mea sures and unit type. GEE Analysis #2 examined hospital unit level fall predictors and accoun ted for clustering of units within hospitals. We incorporated unit type (i.e. medical, surgical and medical surgical units) in the model as a form of risk adjustment to account for differences in patient populations known to be associated with differential fall risk. In addition, we modeled the GEE using a negative binomial distribution with a log link function because the distribution of fall counts were over dispersed (i.e. variance of 5.05 was greater than the mean of 2.49). Following methodology from La ke and colleagues (2010), the dependent variable was the monthly unit level fall count, and we incorporated patient days as an exposure variable on the right side of the equation, which is equivalent to modeling fall rates as the dependent variable. This w as done by including patient days in the OFFSET statement in PROC GENMOD ( SAS Institute, 2004). The correlation structure used for the total fall rate models was an autoregressive structure which accounted for correlations within clusters (unit fall rates

PAGE 100

100 within hospitals) over time (Ballinger, 2004). From this analysis, we hypothesized (see Chapter 4 for hypotheses) that greater levels of hospital nurse staffing and organizational characteristics of magnet status, teaching status, metropolitan location an d greater hospital bed size would inversely impact total hospital fall rates. In addition, incidence rate ratios we re calculated from this model. These incidence rate ratios reflect the anticipated change in the incidence of the patient fall rate that acco mpanies a one unit change in the independent variables (nurse staffing and organizational characteristics), holding all other covariates constant in the model (Lake et al., 2010). Calculations from the regression models for GEE provide incidence rate ratio estimates, standard errors and p values, where statistical significance is determined at the p<0.05 level. The GEE total hospital fall analysis used in the current investigation is repre sented by Equation 5 5 below : ) = B 0 + B 1 Nurse_Staffing + B 2 Hospital_Factors + B 3 Process_Factors + u (5 5 ) Similar to GEE Analysis #1, we repeated the total monthly falls analysis at the hosp ital level to determine if collapsing measures across unit types in hospitals may impact or bias the association between nurse staffing, organizational characteristics and total falls and whether there is an average causal effect of collapsing these measur es. Therefore, using a negative binomial distribution with a log link function and an autoregressive correlation structure, we modeled a GEE at the hospital level. Findings from this sensitivity analysis are summarized in the results section (Chapter 6), w ith additional detail provided in Appendix C, Table C 2

PAGE 101

101 Figure 5 1. P robability function and maximum likelihood estimation of the zero in flated Poisson model

PAGE 102

102 Figure 5 2 Unadjusted growth model estimating the average pattern of patie nt falls among acute care hospitals over 54 months

PAGE 103

103 Figure 5 3 Adjusted growth model estimating the average pattern of patient falls among acute care hospitals over 54 months controlling for nursing staffing and organizational factors

PAGE 104

104 Figure 5 4 Unadjusted growth model estimating the average pattern of patient falls among hospital units over 54 months

PAGE 105

105 Figure 5 5 Unadjusted growth model estimating the average pattern of patient falls among hospital units over 54 months

PAGE 106

106 Table 5 1. Variable d escriptions Variable Explanation Time Varying Dependent Variables 1. Total Fall Rate Patient fall rate per 1 000 Patient Days Monthly 2. Injurious Fall Rate Injurious fall rate pe r 1 000 Patient Days Monthly Nurse Staffing Variables Nurse Staffing (1) RN Skill Mix Percentage Monthly (2) RN HPPD* (3) UAP HPPD* (4) LPN HPPD* Hospital Organizational Variables Hospital Bed Size T otal number of hospital beds Not time varying (<100, 100 199, 200 299, 300 399, 400 499, >500) Metropolitan Status MSA=1, non MSA=0 Not time varying Census Region Northeast, Midwest, South, West Regions Not time varying Teaching Status Teaching=1, non Teaching=0 Not time varying Magnet Status Magnet=1, non Magnet=0 Not time varying Unit Level Organizational Variables Hospital Unit Type (1) Medical Unit Monthly (2) Surgical Unit (3) Medical/Surgical Unit Patient Age Average patient age of patients who Quarterly fell on a unit Patient Gender Proportion of patient gender on unit Quarterly

PAGE 107

107 Table 5 1 Continued Variable De scription Time Varying Process of Care Variables Fall Risk Assessment Scale Type of risk assessment scale used by unit Monthly (Morse, Schmid, Hendrich II, Other) Fall Prevention Protocol Proportion of units with a fall protoco l Monthly *HPPD=hours per patient day

PAGE 108

1 08 Table 5 2 Total fall observation sample size after exclusion criteria Original Sample Sample with Exclusion Criteria Percentage of Missing Fall Observations (%) Total Fall Count Observations 746,801 746,710 0.01

PAGE 109

109 Table 5 3. Log Bayes approximation factor 2log e (B 10 ) (B 10 ) Evidence Against H 0 0 to 2 1 to 3 Not Significant 2 to 6 3 to 20 Positive 6 to 10 20 to 150 Strong >10 >150 Very Strong Reference: Jones e t al., 2001

PAGE 110

110 CHAPTER 6 RESULTS The results of this study are presented in the following sections, including the descriptive analysis of the NDNQI dataset and the r esults for each specific aim: Results from Descriptive NDNQI Analysis The total fall rate f or all observations in the NDNQI from July 2006 through December of 2010 was 3.65 falls per 1 000 patient days (standard deviation (SD)=3.2). Of tho se patients who fell, the injurious fall rate over the entire sample was 0.84 injurious falls per 1 000 pati ent days. Both the total fall rate and injurious fall rate were highest in medical units (fall rate=4.13 falls per 1 000 patient days; injurious fall rate=0.97 injurious falls per 1 000 patient days). The second highest fall rates were found in medical sur gical units (fall rate=3.70 falls per 1 000 patient days; injurious fall rate=0.87 injurious falls per 1 000 patient days), followed lastly by surgical units with a fall rate of 2.91 falls per 1 000 patient days and an injurious fall rate of 0.62 injuriou s falls per 1 000 patient days. Table 6 1 shows the rate of falls and injurious fa ll rate by unit type for the entire NDNQI sample. Descriptive Characteristics of Patients Who F ell in the NDNQI Among all patients who sustained a fall in the NDNQI dataset o ver the 54 month timeframe of the study, the average age of these patients was 66 years old. When stratifying these by clinical unit type, the average ages of patients who fell on medical units were higher than those on both medical surgical and surgical u nits (66.28 on medical units 64.75 on surgical units and 66 on medical surgical units). Only slight differences in gender were found overall (50.03% females versus 49.97% males), but medical surgical and surgical units were found to have a higher percent age of female

PAGE 111

111 fallers when compared to males (50.46% females and 49.54% males in medical surgical units; 52.44% females and 47.56% males in surgical units; 2 =255, p<0.001). Medical units had a greater percentage of male fallers overall (47.98% female and 52.02% male). Table 6 2 displays the age and gender of patients who fell by units in the entire NDNQI sample. Descriptive Characteristics of Nurse S taffing in the NDNQI Table 6 3 displays the descriptive measures of total nurse staffing and nurse st affing based on unit type. The majority of nursing hours were provided by RNs (overall mean=5.73 RN hours per patient day) and the least amount of nursing hours in hospitals were provided by LPNs (overall mean=0.45 hours per patient day). RN hours per pati ent day (HPPD) were highest in surgical units (mean=5.9 RN HPPD) and lowest in medic al units (mean=5.7 RN HPPD). Nursing assistant HPPD were highest in medical units (2.53 UAP HPPD) and LPN HPPD were highest on medical surgical units (0.49 LPN HPPD). The p ercentage of RN skill mix did not vary significantly, but the highest percentage of RNs hours to total nursing hours was on surgical units (mean=67%; SD=10.64 %). Descriptive Characteristics of Hospital O rganizations in the NDNQI Table 6 4 summarizes the organizational characteristics of hospitals in the NDNQI sample. Overall, the largest percentage of hospitals based on bed size was 100 to 199 beds (31.33%) and 200 to 299 beds (21.02%). The lowest percentage of hospitals based on hospital bed size in the NDNQI was 400 to 499 (7.07%) and more than 500 beds (8.28%). However, when compared to general acute care hospitals in the U.S., the NDNQI has a higher percentage of hospitals with bed sizes over 300 (Lake et al., 2010). In our sample, 32.06% of hospitals were reported to be teaching status and

PAGE 112

112 87.90% were located in a metropolitan location. Additionally, 21% of hospitals reported Magnet status, which is higher than the average U.S. hospital population of approximately 7% (Kelly et al., 2011). NDNQI hospita ls were mainly located in the South (52.30%) and Northeast (22.67%), and the West and Midwest regions only accounted for approximately 25% of all NDNQI hospitals. Results from Specific Aim 1: Trajectories of Hospital and Hospital Unit Fall Rates All traje ctory models were fit with linear parameters because quadratic and cubic function parameters were not statistically significant and parameter estimates did not converge in these models. However, given the descriptive statistics of hospital fall rates over time presented in Table 6 5, we would not expect hospital fall rates to follow a quadratic or cubic trend because the means and standard deviations of these fall rates have remained constant over time. Also, because the distribution of hospital fall rates include a larger number of zeroes than would be expected under normal conditions, we used the zero inflated Poisson (ZIP) option in PROC T RAJ for each model (see Figure B 1 in Appendix B for distribution of fall rates). The total number of hospitals that r eported fall information to the NDNQI was 1,592. Within these hospitals, clinical units were analyzed according to whether they were reported in NDNQI as medical, surgical and medical surgical units. Hospital unit samples included 2,558 medical units, 1,86 5 surgical units and 3,448 medical surgical units. The following section summarizes model results found in this study for all five steps, including descriptive statistics: Step #1: Unadjusted Monthly Hospital Fall Model Table 6 5 summarizes descriptive inf ormation on the number of hospitals, mean hospital fall rates, standard deviations and minimum and maximum values over a 54 month period of time from July 2006 through December of 2010 in the NDNQI. The

PAGE 113

113 overall mean fall rate for hospitals across the entir e 54 month sample was 3.65 falls per 1 000 patient days. The mean fall rates were stable across this time period, ranging from a low of 3.48 in September of 2010 to a high of 4.04 in December of 2006. As previously mentioned, model selection for the unad justed monthly hospital fall rate model was an iterative process that incorporated both statistical analysis and subjective determination of the best number of latent class trajectory groups. Although BIC and log Bayes approximation factor values revealed the four trajectory group model to be an optimal fit when compared to the three group model (see Table 6 6), trajectory #1 in the four trajectory model contained only 2.2% of hospitals, which is below the recommended minimum of 5% of the sample within a gi ven latent trajectory group (Nagin, 2005; Andruff et al., 2009). Therefore, because the four group trajectory did not fit all model requirements, we chose the three group linear trajectory model as the best fit for the unadjusted monthly hospital fall mode l. After model selection, the unadjusted monthly hospital fall model revealed three trajectories of fall rates, and this is represented graphically in Figure 6 1 Furthermore, Table 6 7 provides the output of the Poisson maximum likelihood estimates, stan dard errors and p values for the unadjusted monthly hospital fall model. Trajectory #1 at the bottom of Figure 6 1 revealed a trajectory of hospitals that had consistently lower fall rates over the 54 month time frame as compared to trajectories #2 and #3. This trajectory had a slightly decreasing statistically significa nt linear parameter estimate (p< 0.0001) with a baseline fall rate around 2.5 falls per 1 000 patient days. For the

PAGE 114

114 fall rate trajectory and 22.8% of all hospitals were estimated to be in this trajectory group. Next, trajectory #2 revealed hospitals that have had consistently lower fall rates than trajectory #3, but higher fall rates than trajectory #1. This trajectory had a statistically significant decreasing linear parameter estimate (p=0.001) with a baseline fall rate around 4 falls per 1,000 patient days Trajectory #2 consists of 54.9% of hospitals and this trajectory will be referred to as the rate trajectory. Trajectory #3 included hospitals that have had consistently higher fall rates over time when compared to trajectories #1 and #2. Similar to trajectory #1 and #2, trajectory #3 had a statistically signifi cant decreasing linear parameter estimate (p<0.001) with a baseline fall rate over 5 falls per 1,000 patient days and this group accounted for 22.3% of hospitals in the sample. Therefore, traje ctory #3, with fall rates above the overall NDNQI mean fall ra te (i.e. 3.65 falls per 1,000 patient days) will be group Finally, to ensure reliability and model accuracy across trajectories, average posterior probability results indicated that a ll three trajectories were above the recommended minimum value of 0.7 (trajectory #1=0.92, trajectory #2=0.89 and, trajectory #3=0.91). Findings from Step #1 partially supported Hypothesis 1 a which stated that h ospital fall rates will vary over time with observed fall rates reflecting latent trajectory groups of high, low, increasing or decreasing fall rates over time. Although we did not find evi dence of latent class trajectory groups with consistently increasing or decreasing patient fall rates over tim e, we found consistently high (22.3%), medium (54.9%) and low (22.8%) linear latent class trajectory groups of hospital fall rates from the LCGM

PAGE 115

115 analysis in Step #1. These unadjusted high, medium and low hospital trajectory groups were used in the assessme nt of Specific Aims 2 and 3 to better understand the impact Step #2: Adjusted Monthly Hospital Fall Model As a sensitivity analysis, we investigated whether both time varying and time invariant external factors impacted hospital membership into a particular trajectory group. We specified the same number of latent classes as the un adjusted hospital falls model in Step #1 and controlled for both time varying nurse staffing characteristics (RN HPPD, LPN HPPD, UAP HPPD and percent RN skill mix) and time invariant organizational characteristics (metropolitan status, census region, teach ing status, total bed size, and Magnet status). Findings revealed no significant change in hospital group membership between the adjusted and unadjusted models in Steps #1 and #2. Therefore, for the purpose of this study, we utilized trajectory membership information from the unadjusted hospital fall rate model in Step #1 for Specific Aims 2 and 3 to conduct generalized estimating equation (GEE) methodology and predict factors that te trajector y group Figure B 2 in Appendix B illustrates the graphical representation of the three trajectories in the adjusted aggregated monthly h ospital fall model and Table B 4 in Appendix B provides the output of the Poisson maximum likelihood estimates, standar d errors and p values for the adjusted monthly hospital fall model. Similar to the unadjusted hospital fall rate model, the adjusted hospital fall rate rates over t ime. Adjusted trajectory #1 2 has a

PAGE 116

116 decreasing statistically significant linear parameter estimate (p<0.0001) with a baseline fall rate around 2.5 falls per 1,000 patient days and contains 24.1% of hospitals in the populati on, which is only slightly different from the unadjusted model of 22.8% of hospitals in Step #1. Trajectory #2 had a decreasing statistically significant linear parameter estimate (p<0.0001) and it represented 52.7% of hospitals in the sample, which was si milar to the unadjusted model of 52.7% of hospitals. Finally, trajectory #3 had a decreasing statistically significant linear parameter estimate (p<0.0001) and this trajectory consists of 23.2% of hospitals, which was only slightly changed from the unadjus ted model percentage of 22.3%. Average posterior probabilities for the adjusted hospital fall rate model were identical to the unadjusted hospital model in Step #1 with trajectory #1 at 0.92, trajectory #2 at 0.89 and trajectory #3 at 0.91. Therefore, when controlling for time varying and time invariant covariates in the model, membership probabilities in the adjusted models were not significantly different from the unadjusted hospital fall rate trajectory model. For the purpose of this study, we utilized t rajectory membership information from the unadjusted hospital fall rate model in Step #1 for Specific Aims 2 and 3 to conduct generalized estimating equation (GEE) methodology l fall rate trajectory group Step #3: Unadjusted Monthly Hospital Unit Fall Models Steps #3 and #4 were conducted to assess fall rates within acute care hospital clinical units to identify latent class trajectories of fall rates over time. In contrast to the hospital trajectory models in Steps #1 and #2, unit level trajectory models provide an in depth understanding of fall rates separately according to whether hospitals are reported

PAGE 117

117 as medical, surgical or medical surgical units. The results of Step # 3 are provided below: Unadjusted medical unit model The overall mean fall rate over 54 months for medical units was 4.13 falls per 1 000 patient days with an average sta nda rd deviation of 3.25. Table B 1 in Appendix B displays the fall rate means, stand ard deviations and minimum/maximum values per month for all medical units in our study. The results of these descriptive statistics reveal that medical unit fall rates have only decreased slightly over the study period, and these medical units have higher fall rates when compared to surgical and medical surgical units over time. Model selection for the medical unit model was conducted using the BIC values he latent class groups (Henselmans et al., 2010, p.163). A maximum number of five groups were pre determined and used to assess optimal model fit. The BIC, log Bayes approximation factor values and percent membership probability results summarized in Figur e 6 8 revealed the four trajectory group model to be an optimal fit when compared to other group models. The five trajectory group model incorporated two latent class trajectories with less than 5% of all medical units, which is lower than the recommended number of 5% probability of membership in a particular group, and the log Bayes approximation factor favored the four group model over the three group model (see Figure B 3 in Appendix B for the unadjusted three group medical unit trajectory model). Theref ore, because the four group trajectory fits all of model requirements, we chose this trajectory model as the best fit for the unadjusted aggregated monthly medical unit fall model.

PAGE 118

118 The unadjusted monthly medical unit fall model revealed four trajectories of fall rates, and this is represented graphically by Figure 6 2. Table 6 9 summarizes the Poisson maximum likelihood estimates, standard errors and p values for the unadjusted monthly medical unit fall model. Findings from this model indicated that there were four slightly decreasing linear trajectories with estimated membership probabilities of 19%, 42.3%, 32.8% and 6%, respectively. Trajectory #1 at the bottom of Figure 6 2 reveals categories of hospitals that consistently had the lowest fall rates over 54 months when compared to trajectories #2, #3 and #4. This trajectory had a slightly decreasing statistically significant linear parameter estimate (p < 0.0001) with a baseline fall rate around 2.9 falls per 1,000 patient days For the purpose of this stu dy, trajectory #1 will hospitals were estimated to be in this trajectory group. Trajectory #2 revealed a statistically significant decreasing linear parameter estim ate (p=0.001) with a baseline fall rate around 3.75 falls per 1,000 patient days and this group incorporated the largest percentage of medical units (42.3%). As a result, y. Trajectory #3 consists of 32.8% of medical units with a baseline fall rate of approximately 5 falls per 1,000 patient days high % of hospitals in our sample and has a baseline fall rate over 7 falls per 1,000 patient days which is significantly higher than other trajectory groups in the model. However, the linear parameter for both trajectory #3 and trajectory #4 were not statisti cally significant. Average posterior probabilities for the unadjusted medical unit model were 0.85 for

PAGE 119

119 trajectory #1, 0.82 for trajectory #2, 0.90 for trajectory #3 and 0.91 for trajectory #4, which exceeds the recommended minimum of 0.70. Unadjusted sur gical unit model The overall mean fall rate for surgical units over 54 months was 2.93 falls per 1 000 patient days with a sta nda rd deviation of 2.93. Table B 2 in Appendix B shows the fall rate means, standard deviations and minimum/maximum values per mo nth for all surgical units in our study. The results of these descriptive statistics revealed that surgical fall rates have remained consistently lower than medical and medical surgical units over the study period, which may be the result of lower acuity p atients on these units when compared to medical and medical surgical units (Dunton et al., 2004). Model selection for the surgical unit model revealed the two trajectory group model to be an optimal fit when compared to other latent class trajectory mode ls (see Table 6 10). This was due to the fact that although the three group trajectory model demonstrated better model fit statistics, trajectory #3 was influenced by outlier observations and did not adequately categorize surgical units when compared to th e tw o trajectory mod el (see Figure B 4 and Table B 5 in Appendix B for the three group surgical unit model comparison). Therefore, based on subjective determination and conceptual meaningfulness of the surgical fall rate model, the two group trajectory mod el was the best fit for the unadjusted aggregated monthly surgical fall model. The unadjusted monthly surgical unit fall model revealed two trajectories of fall rates for the NDNQI surgical unit population, and this is illustrated graphically by Figure 6 3. Tabl e 6 11 shows the output of the Poisson maximum likelihood estimates, standard errors and p values for the unadjusted monthly surgical unit fall model. Findings from this model indicated that there were two relatively unchanged linear

PAGE 120

120 trajectories ov er time with estimated surgical unit membership probabilities of 70.5% in trajectory #1 and 29.5% in trajectory #2. Trajectory #1 at the bottom of Figure 6 3 has a baseline fall rate around 2.5 falls per 1,000 patient days and trajectory #2 has a baseline fall rate around 4.5 falls per 1,000 patient days Although these parameter estimates provided the best model fit, they were not statistically significant at the p<0.05 ently surgical unit model was 0.95 for trajectory #1 and 0.94 for tra jectory #2, which is above the recommended average of 0.7. Unadjusted medical surgical unit model The overall mean medical surgical unit fall rate over 54 months was 3.70 falls per 1 000 patient days with a stan dard deviation of 3.22. Table B 3 in Append ix B illustrates the fall rate means, standard deviations and minimum/maximum values per month for all medical surgical units in our study. Findings from these preliminary statistics reveal that medical surgical fall rates are consistently higher than surg ical unit fall rates, but lower than medical unit fall rates. Model selection for the medical surgical unit model used a maximum number of five groups to calculate the best model fit. The BIC, log Bayes approximation factor values and percent membership pr obability results show the four trajectory group model to be an optimal fit when compared to other group models (see Table 6 12). In comparison to the four group model, the five group trajectory model had a trajectory group with only a 3.4% estimated proba bility of group membership, which is below the

PAGE 121

121 5% recommended cutoff. As a result, we selected the four group model as the most parsimonious LCGM for medical surgical units in our study. After determining model selection, Figure 6 4 illustrates the graphic al representation of medical surgical units, which can be divided into four separate linear trajectory groups with unit membership probabilities of 26%, 45.5%, 22.6% and 5.9%, respectively. Table 6 13 provides the output of the Poisson maximum likelihood e stimates, standard errors and p values for the unadjusted monthly medical surgical unit fall model. Trajectory #1 at the bottom of Figure 6 4 reveals hospitals in this category consistently had the lowest fall rates over 54 months when compared to trajecto ries #2, #3 and #4. This trajectory has a slightly decreasing statistically significant linear parameter estimate (p=0.002) with a baseline fall rate of approximately 2.5 falls per 1,000 patient days medical surgical unit fall rate trajectory and 26% of all hospitals were estimated to be in this trajectory group. Trajectory #2 revealed a statistically significant decreasing linear parameter estimate (p<0.0001) with a baseline fall rate around 4 falls per 1,000 patient days and this group incorporated the largest percentage of medical units (45.5%). As a result, trajectory Trajectory #3 had a statistically significant decr easing linear parameter estimate (p<0.0001) and consisted of 22.6% of medical surgical units with a baseline fall rate of approximately 5 falls per 1,000 patient days Trajectory #3 is classified as the jectory. Finally, trajectory #4 had a statistically significant decreasing linear parameter estimate (p=0.002) and had a

PAGE 122

122 baseline fall rate over 7 falls per 1,000 patient days which was significantly higher than the previous trajectory groups. This group included 5.9% of medical surgical units in our surgical fall rate trajectory. Average posterior probabilities for the unadjusted medical surgical unit model were 0.87 for trajectory #1, 0.82 for t rajectory #2, 0.88 for trajectory #3 and 0.92 for trajectory #4, which all exceeded the recommended minimum of 0.70. Findings from the medical, surgical and medical surgical unit trajectory models in Step #3 partially supported Hypothesis 1a, which state d that h ospital unit fall rates will vary over time with observed fall rates reflecting latent class trajectory groups of high, low, increasing or decreasing fall rates over time. Although we did not find evi dence of latent class trajectory groups with con sistently increasing or decreasing patient fall rates over time, we found that latent class categories emerged from three separate LCGM analyses. Specifically, medical unit trajectories were classified as either low, low medium, high medium and high, surgi cal unit trajectories were classified as either high or low, and medical surgical unit trajectories were classified as either low, low medium, high medium and high in terms of unit level fall rates over time. Step #4: Adjusted Monthly Hospital Unit Fall M odels To ensure model adequacy with the medical, surgical and medical surgical LCGMs in our study, we analyzed adjusted models by controlling for both time varying and time invariant external factors as a form of sensitivity analysis. This was conducted t o compare the probability of unit membership in each trajectory group while adjusting for factors that influence fall rates in hospital medical, surgical and medical surgical units. We specified the same number of latent classes as the unadjusted unit leve l models in Step #3 and controlled for both time varying characteristics (i.e. RN HPPD,

PAGE 123

123 LPN HPPD, UAP HPPD, percent RN skill mix and unit level proportion of fall protocols used) and time invariant process characteristics (i.e. type of fall risk assessment scale used on each unit). Findings revealed only relatively small changes in clinical unit trajectory membership between the adjusted and unadjusted models in Steps #3 and #4. Figures B 5 through B 7 in Appendix B depict the graphical representation of t he adjusted medical, surgical and medical surgical uni t tr ajectory models, and Table s B 6 through B 8 in Appendix B provides the output of the Poisson maximum likelihood estimates, standard errors and p values for the adjusted monthly medical, surgical and medical surgical unit fall models. Given only minor differences were found between adjusted and unadjusted hospital unit level models, we further confirmed that the unadjusted models in Step #3 accurately represented the variability in hospital trajectory Step #5: Multiple Imputation Models and Unadjusted Hospital Model by Quarter Step #5 tested whether missing data affected the results of prior LCGM analyses, as well as tested the average causal effect of aggregated hospital falls on a quarterly basis. After creating five separate monthly unadjusted hospital datasets using MCMR multiple imputation with both monotone and non monotone missing data patterns, we analyzed these datas ets using PROC TRAJ. When comparing the non imputed unadjusted hospital fall rate model to the five imputed models, there were only minor fluctuations in group membership percentages between these models, suggesting findings are robust for the non imputed unadjusted hospit al fall rate model (see Figure s B 8 through B 12 and Table s B 9 through B 13 in Appendix B for imputed results ). Therefore, we will use the non imputed dataset models for Specific Aims 2 and 3.

PAGE 124

124 Furthermore, Figure B 13 and Tabl e B 14 in Appendix B represent the unadjusted hospital fall rate model by quarter in the NDNQI. Findings from this model demonstrated three distinct categories of falls rates that are relatively stable, which is consistent with monthly unit level trajectories. Howe ver, trajectory #2 incorporates more than 67% of all hospitals in the sample, and this may indicate an average causal effect of aggregating hospital fall rates each quarter when compared to the monthly model. By aggregating hospital fall rates quarterly, w e removed some of the variation in falls over time, which strengthens the evidence that monthly fall rates should be used when modeling latent cla ss trajectories. Specific Aims 2 & 3: Predictors of Patient Falls The results of GEE analysis for Specific Aim s 2 and 3 are outlined in three separate sections below. The first section provides descriptive results of the latent class groups for GEE Analysis #1. The second section describes the results of the GEE Analysis #1, which predicts factors that impact memb trajectory group, reflecting hospital membership in the trajectory group of fall rates at or above the mean in the sample. Finally, the third section displays results of the GEE Analysis #2, which estimates nursing and o rganizational factors that influence patient falls over the 54 months included in the study sample. After conducting bivariate analysis and analyzing correlations between measures of predictor variables, we excluded only the fall risk assessment scale vari able because it was not significantly associated with patient falls in bivariate analysis. All other measures of nurse staffing and organizational characteristics were incorporated into the final multivariate GEE models. In addition, we tested interaction terms among nurse staffing and organizational predictors, and these interaction terms were not found to be statistically significant.

PAGE 125

125 Descriptive Results of Latent Class Groups for GEE Analysis #1 Table 6 14 summarizes descriptive findings of hospitals in each latent class mean fall rate was 4.96 falls per 1 000 patient days, which is compared to 3.63 falls per 1 rate trajectory group and 2.50 falls per 1 15% of hospitals were designated a had less RN and UAP nursing hours when compared to the lower fall rate trajectory average hours of LPN staffing (mean=0.61 LPN HPPD) when compared to the lower fall PPD). Trajectory Group Table 6 15 displays the results of GEE Analysis #1, which provides adjusted odds ratio (OR) estimates of nurse staffing and organizational characteristics that for unit type and unit level fall risk protocol. As previously mentioned, we specified a binomial distribution with a logit link and an unstructured correlat ion matrix, which demonstrated the best model fit based on QIC and log likelihood fit statistics (QIC=215,427). From our nurse st affing hypotheses (Hypothesis 2a through 2d in Chapter 4), we posited that hospitals with higher levels of RN, LPN and UAP hour s per

PAGE 126

126 patient day (HPPD) and a higher percentage of RN skill mix would be less likely to be in lysis #1 supported Hypothesis 2a ; a higher number of RN HPPD was found to be a ssociated with a greater odds of membership into the lower fall rate groups and less likely to be in words, a one unit increase in RN HPPD was associated with 6.2% gre ater odds of membership in the lower fall rate trajectory groups. group increased as LPN HPPD staffing and the percentage of RN skill mix increased. A one unit increase i n LPN HPPD staffing was associated with 32% greater odds of being widely depending on the type of nursing care and when comparing nurse staffing hours from Table 6 14, a o ne hour increase in LPN HPPD staffing is equivalent to over one and a half of the LPN standard deviation (LPN HPPD SD=0.7). In comparison, a one unit increase in RN HPPD is only 0.625 of the SD (overall RN HPPD SD=1.6). In addition, although the percentage of RN skill mix was associated with a greater skill mix percentage was associated with only a 1.4% greater odds of being in the which is statistically significant (p=0.039) but may not be clinically relevant. fall rate trajectory and acute care hospital organizational characteristics were as follows:

PAGE 127

127 group, (2) hospitals in metropolitan statistical areas are less likely to be in the ter than 500 from GEE Analysis #1 indicate that hospitals with Magnet stat us and larger hospital bed trajectory over ti me, which supports Hypotheses #3a and #3c Interestingly, hospitals with Magnet status had 52.7% greater odds of being in the lower fall rate trajectories hospital bed size was associated with greater odds of being in the consistently lower fall rate trajectory group. Hospitals with 100 to 200 beds (OR=2.15, p<0.001), 200 to 299 (OR=1.61, p=0.002), and 400 to 499 beds (OR=2.44, p<0.001) were significantly more over 500 be ds, which supports Hypoth esis #3c Conversely, the hypothesized relationships that teaching status hospitals and hospitals in a metropolitan area are less Analysis #1. Lastly, hospitals loc ated in the Midwest had 48% greater odds of being in ompared to the West (p=0.017). Finally, findings from our sensitivity analysis, summarized in Table C 1 of Appendix C confirm that aggregating unit lev el nurse staffing measures to the hospital level resulted in a loss of variation among coefficients, though it did not affect the trajectory and the independent variable s. Odds ratios and statistical significance

PAGE 128

128 between these analyses did not substantially change for organizational characteristics, such as hospital bed size and Magnet status (OR=0.661, p<0.001). We did find that collapsing measures across unit types to t he hospital level impacted the statistical significance among nurse staffing variables. For instance, the only statistically significant nursing factor in the hospital level sensitivity analysis was LPN HPPD (OR=1.5, p<0.001), whereas measures of RN HPPD a nd percent RN skill mix were both statistically significant in our previous hospital unit level analysis. Therefore, this sensitivity analysis demonstrated the importance of conducting GEE analysis that accounts for clustering of units within hospitals ins tead of using aggregated hospital measures. GEE Analysis #2 Results: Impact of Nurse Staffing and Organizational Factors on Patient Falls Table 6 16 displays the results of GEE Analysis #2, including incidence rate ratio (IRR) estimates, standard errors and p values. As previously mentioned, we modeled the GEE with a negative binomial distribution and a log link function, which provided the best model fit based on QIC and log likelihood ratio fit statistics (QIC= 38,834). In this study, the IRR is the e xpected change in the dependent variable (patient falls) for a one unit increase in the independent variables of nurse staffing and organizational characteristics while controlling for unit type and unit level fall risk protocol. In terms of the relationsh ip between nurse staffing and patient falls, all nurse staffing variables were found to be statistically significant, but in different directions. Our nurse st affing hypotheses (Hypothesis #2a through 2d in Chapter 4) posited that hospitals with higher lev els of RN, LPN and UAP hours per patient day (HPPD) and a higher percentage of RN skill mix would be associated with reductions in patient falls over time. Similar to

PAGE 129

129 GEE Analysis #1, RN HPPD was found to be inversely associated with patient falls (IRR= 0. 985, p=0.049), and for every one unit change in RN HPPD, the likelihood of patient falls decreased by 1.5%, which is equal to a 1.5% reduction in the fall rate (Lake et al., 2010). Other nurse staffing variables yielded significant positive results; LPN HP PD were associated with an 8.3% increase in patient falls (p<0.001), UAP HPPD were associated with a 6.1% increase in patient falls (p<0.001) and RN skill mix percentage was associated with a 0.6% increase in patient falls (p=0.01). In regard to the relat ionship between acute care hospital organizational characteristics and patient falls in GEE Analysis #2, we hypothesized that hospitals with Magnet designation, location in a metropolitan area, larger bed size and teaching status would be associated with r eductions in overall fall rates over time. Magnet status was again found to be statistically significant and inversely associated with patient falls (IRR=0.943; p<0.001). This association was also elicited if we excluded nurse staffing variables from the m odel, strengthening the evidence of an association between Magnet status and patient fall rates. Furthermore, if Magnet status was excluded from the m odel, measures of RN HPPD and percent RN skill mix were not statistically significant in GEE Analysis #1, but remained significant in GEE Analysis #2 Metropolitan area and bed size were not found to be statistically significant, which did not support our hypotheses. However, hospital teaching status was unexpectedly found to be associated with a 4.6% increase in patient fall rates (IRR=1.046; p<0.001) and we will discuss potential explanations for this in Chapter 7. In addition, similar to GEE Analysis #1, hospitals located in the Midwest were more likely to have greater patient falls when compared to hospital s in the West region (IRR=1.115, p<0.001).

PAGE 130

130 Furthermore, our sensitivity analysis for GEE Analysis #2 demonstrated that aggregating unit level measures to the hospital level resulted in a loss of variation among nurse staffing measures, providing additional evidence that researchers should account for clustering of units within hospitals over time (see Table C 2 in Appendix C ). These findings are consistent with the sensitivity analysis in GEE Analysis #1. Also, these results suggest that further research ma y need to consider both the hospital and unit level, as this has implications for policymakers and managers when implementing policies in acute care settings.

PAGE 131

131 Figure 6 1. Unadjusted hospit al monthly fall rate trajectory model

PAGE 132

132 Figure 6 2. Un adjusted medical unit monthly fall rate trajectory model

PAGE 133

133 Figure 6 3. Unadjusted surgical unit monthly fall rate trajectory model

PAGE 134

134 Figure 6 4. Unadjusted medical surgical unit monthly fall rate trajectory model

PAGE 135

135 Table 6 1. Fall rates per 1 000 patient days by unit type and total fall rates (N=7 871) Unit Type N Mean Std Dev Min Max Medical 2558 4.13 3.25 0 58.83 Surgical 1865 2.91 2.93 0 62.50 Medical/Surgical 3448 3.70 3.22 0 69.37 Total 7871 3.65 3.19 0 69.37

PAGE 136

136 Table 6 2. Descriptive characteristics (age and gender) of patients who fell by unit type (N=212,684) Unit Type Observations (N) Mean Age (SD) Male (%) Female (%) Medical 75,417 66.28(16.6) 52.02 47.98 Surgical 47,8 59 64.75 (16.9) 47.56 52.44 Medical/Surgical 89,408 66.00(17 .0 ) 49.54 50.46 Total 212,684 65.82(16.8) 49.97 50.03

PAGE 137

137 Table 6 3. Descriptive characteristics of nurse staffing by unit type RN HPPD* LPN HPPD* UAP HPPD* Skill mix% Unit Type N Mean(SD) Mean(SD) Mean(SD) Mean(SD) Medical 2558 5.66(1.5 3 ) 0.42(0.68 ) 2.53(0.98 ) 66 (10.95 ) Surgical 1865 5.90(1.55 ) 0.44(0.69 ) 2.48(1.0 2 ) 67(10.64 ) Medical/Surgical 3448 5.68(1.6 4 ) 0.49(0.7 3 ) 2.50(1.0 4 ) 66(10 .89 ) Total 7871 5.73(1.57 ) 0.45(0.7 0 ) 2.5 0 (1.0 2 ) 66(10.86 ) *HPPD=Hours per patient day

PAGE 138

138 Table 6 4. Descriptive characteristics of hospital organizations (N=1 592) Organizational NDNQI Acute Care Hospitals (%) Characteristics Hospital Bed Size <100 19.62 100 199 31.33 200 299 21.02 300 399 12.68 400 499 7.07 500+ 8.28 Teaching Status 32.06 Magnet Status 21.02 Metropolitan Status 87.90 Region N ortheast 22.67 Midwest 12.59 West 12.41 South 52.30

PAGE 139

139 Table 6 5. Descriptive statistics of hospital observations: Fall rate means, standard deviations, minimums and maximums by month and year (N=1 592) Variable Year Month Obser vations Mean Std Dev Min Max Fall1 2006 7 837 3.74 2.26 0 34.88 Fall2 2006 8 842 3.79 2.22 0 20.83 Fall3 2006 9 843 3.73 2.06 0 22.94 Fall4 2006 10 855 3.95 2.13 0 26.00 Fall5 2006 11 857 3.83 2.13 0 22.94 Fall6 2006 12 859 4.04 2.79 0 35.56 Fall7 2007 1 904 3.83 1.96 0 15.63 Fall8 2007 2 902 3.82 2.08 0 21.86 Fall9 2007 3 901 3.67 2.01 0 21.51 Fall10 2007 4 936 3.66 1.95 0 18.18 Fall11 2007 5 936 3.73 2.09 0 2 1.51 Fall12 2007 6 935 3.55 1.94 0 16.69 Fall13 2007 7 968 3.69 2.19 0 26.47 Fall14 2007 8 968 3.64 2.03 0 23.26 Fall15 2007 9 968 3.65 2.05 0 20.76 Fall16 2007 10 1019 3.70 2.31 0 37.37 Fall17 2007 11 1018 3.83 2.18 0 25.48 Fall18 2007 12 1016 3.73 2.40 0 42.81 Fall19 2008 1 1063 3.66 2.04 0 18.52 Fall20 2008 2 1062 3.68 2.04 0 27.03 Fall21 2008 3 1060 3.62 1.88 0 15.34 Fall22 2008 4 1088 3.78 2.23 0 32.26 Fall23 2008 5 1086 3.60 2.07 0 15.87 Fall24 2008 6 1085 3.51 1.97 0 18.32 Fall25 2008 7 1108 3.61 2.05 0 27.62 Fall26 2008 8 1107 3.71 2.30 0 45.23 Fall27 2008 9 1106 3.64 2.14 0 31.09 Fall28 2008 10 1111 3.84 2.07 0 18.7 7 Fall29 2008 11 1109 3.79 1.96 0 13.18 Fall30 2008 12 1109 3.75 1.96 0 13.07 Fall31 2009 1 1155 3.78 2.02 0 18.43 Fall32 2009 2 1154 3.68 2.04 0 16.30 Fall33 2009 3 1152 3.63 1.83 0 12.03 Fall34 2009 4 1170 3.6 1 1.91 0 13.33 Fall35 2009 5 1167 3.57 1.91 0 16.71 Fall36 2009 6 1168 3.51 1.93 0 14.24 Fall37 2009 7 1199 3.63 1.98 0 14.71 Fall38 2009 8 1199 3.65 2.13 0 33.33 Fall39 2009 9 1197 3.65 2.20 0 23.03 Fall40 2009 10 1224 3.71 1.97 0 17.86

PAGE 140

140 Table 6 5. Continued Variable Year Month Observations Mean Std Dev Min Max Fall41 2009 11 1224 3.62 1.96 0 13.40 Fall42 2009 12 1222 3.63 2.03 0 20.27 Fall43 2010 1 1274 3.60 1.96 0 16.76 Fall44 2010 2 1274 3.52 1.97 0 16.70 Fall45 2010 3 1271 3.55 1.92 0 19.48 Fall46 2010 4 1322 3.65 2.13 0 21.65 Fall47 2010 5 1322 3.57 1.97 0 18.87 Fall48 2010 6 1320 3.56 2.11 0 26.32 Fall49 2010 7 1351 3.51 2.30 0 47.81 Fall50 2010 8 1350 3.48 2.16 0 34.03 Fall51 2010 9 1349 3.48 2.40 0 57.97 Fall52 2010 10 1380 3.61 2.10 0 18.60 Fall53 2010 11 1378 3.58 2.02 0 15.63 Fall54 2010 12 1372 3.60 2.09 0 20. 83

PAGE 141

141 Table 6 6. Unadjusted hospital model selection results (N=1 592) Number of Groups Estimated probabilities (% in each group) BIC 2log e (B 10 ) 1 2 3 4 1 124180 100 .0 2 120532 7296 52.2 47.2 3 119435 2194 22.8 54.9 22. 3 4 118946 978 20.5 51.8 22.5 2.2

PAGE 142

142 Table 6 7. Unadjusted hospital fall rate model with Poisson maximum likelihood estimates of three linear trajectories (N=1,592) Group Parameter Estimate Standard Error Prob>|T| 1 Intercep t 2.317 0.241 <0.0001** Linear 0.0001 0 0.00001 <0.0001** 2 Intercept 1.703 0.126 <0.0001** Linear 0.00002 0.00001 0.001 0 ** 3 Intercept 2.334 0.174 <0.0001** Linear 0.00004 0.00001 <0.0001** Group Membership 1 22.80% 1.34 <0.0001** 2 54.90% 1.53 <0.0001** 3 22.30% 1.36 <0.0001** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 143

143 Table 6 8. Unadjusted medical unit model selection results (N=2,558) Number of Groups Estimated probabilities (% in each group) BIC 2log e (B 10 ) 1 2 3 4 5 3 215669 36 .0 49.0 15 .0 4 214822 1,694 19 .0 4 2.2 32.8 6 .0 5 214040 1,564 2.6 30.3 44.7 20.8 1.6 *Group numbers 1 and 2 resulted in false convergence

PAGE 144

144 Table 6 9. Unadjusted medical unit fall rate model with Poisson maximum likelihood estimates of four linear trajectories (N=2,558) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 2.313 0.259 <0.0001** Linear 0.00007 0.00001 <0.0001** 2 Intercept 2.222 0.145 <0.0001** Linear 0.00004 0.00001 0.001 0 ** 3 Intercept 1.939 0.141 <0.0001** Linear 0.00001 0.00001 0.160 0 4 Intercept 1.701 0.260 <0.0001** Linear 0.0002 0 0.00001 0.098 0 Group Membership 1 19.00% 1.239 <0.0001** 2 42.30% 1.322 <0.0001** 3 32.80% 1.419 <0.0001** 4 6.00% 0.662 <0.0001** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 145

145 Table 6 10. Unadjusted surgical unit model selection results (N=1,865) Number of Groups Estimated probabilities (% in each group) BIC 2log e (B 10 ) 1 2 3 4 1 150400 100 .0 2 145016 10,768 70.5 29.5 3 143014 4,004 50.5 43.2 6.4 4 142136 1,756 45.8 30 .0 20 .0 4.1

PAGE 146

146 Table 6 11. Unadjusted surgical unit fall rate model with Poisson maximum likelihood estimates of four linear trajectories (N=1,865) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 1.203 0.123 <0.001 ** Linear 0.000 1 0 0.00001 0.752 2 Intercept 1.6334 0.161 <0.001 ** Linear 0.00001 0.00001 0.554 Group Membership 1 70.50% 1.241 <0.001 ** 2 29.50% 1.241 <0.001 ** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 147

147 Table 6 12. Unadjusted medical surgical unit model selection results (N= 3,448 ) Number of Groups Estimated probabilities (% in each group) BIC 2log e (B 10 ) 1 2 3 4 5 1 275493 100 .0 2 265352 20,282 68.3 31.7 3 262612 5,480 3 5.2 47.8 17 .0 4 261539 2,146 26 .0 45.5 22.6 5.9 5 260803 1,466 3.4 27.8 42.9 20.6 5.2

PAGE 148

148 Table 6 13. Unadjusted medical surgical unit fall rate model with Poisson maximum likelihood estimates of four linear trajectories (N=3,448) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 1.650 0.196 <0.0001 ** Linear 0.00003 0.00001 0.002 0 ** 2 Intercept 1.946 0.106 <0.0001** Linear 0.00003 0.00001 <0.0001** 3 Intercept 2.275 0.133 <0.0001** Linear 0.00003 0.00001 <0.0001** 4 Intercept 2.918 0.243 <0.0001** Linear 0.00004 0.00001 0.002 0 ** Group Membership 1 26.00% 1.025 <0.0001** 2 45.50% 1.104 <0.00 01** 3 22.60% 0.915 <0.0001** 4 5.90% 0.507 <0.0001** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 149

149 Table 6 14. Descriptive characteristics of hospital latent class trajectory groups (N=1,592) Fall R ates RN HPPD* LPN HPPD* UAP HPPD* RN Skill Mix% Latent Class Group Hospital N Mean(SD) Mean(SD) Mean(SD) Mean(SD) Mean(SD) Consistently High 344 4.96(3.8) 5.57(1.6) 0.61(0.8) 2.40(1.0) 65(12) Consistently Medium 891 3.63(3.0 ) 5.73(1.5) 0.42(0.7) 2.50(1.0) 66(10) Consistently Low 357 2.50(2.5) 5.79(1.8) 0.39(0.7) 2.63(1.2) 66(11) *HPPD=Hours per patient day

PAGE 150

150 Table 6 15. Generalized estimating equations (GEE) predicting membership into the level (N=234,057) Variable Odds Ratios Std Error Confidence Interval Pr>|Z| Intercept 0.077 0.539 (0.027 0.222) <0.001** Nursing Factors RN HPPD 0 .942 0.028 (0.892 0.995) 0.032* LPN HPPD 1.323 0.062 (1.171 1.495) <0.001** UAP HPPD 1.006 0.053 (0.907 1.115) 0.914 RN Skill Mix 1.014 0.007 (1.001 1.027) 0.039* Organizational Factors Magnet 0. 659 0.086 (0.557 0.780) <0.001** Teaching 1.154 0.102 (0.945 1.409) 0.160 Metropolitan 0.736 0.178 (0.519 1.043) 0.085 Northeast a 1.062 0.162 (0.774 1.459) 0.708 South a 0.978 0.145 (0.736 1.300) 0.877 Midwes t a 1.481 0.165 (1.072 2.046) 0.017* Bed100 b 1.452 0.224 (0.937 2.250) 0.095 Bed100to199 b 2.145 0.162 (1.561 2.948) <0.001** Bed200to299 b 1.610 0.156 (1.186 2.185) 0.002** Bed300to399 b 1.362 0.158 (0.999 1.855) 0. 050 Bed400to499 b 2.444 0.155 (1.804 3.312) <0.001** surgical), unit fall protocol use and date link *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level a = Reference group for Region is West b = Reference group for Bed Size is >500 beds

PAGE 151

151 Table 6 16. Generalized estimating equations (GEE) determining the impact of n urse staffing and organizational characteristics on patient falls at the hospital unit level (N=230,939) Variable Incidence Rate Ratios Std Error Confidence Interval Pr>|Z| Intercept 0.002 0.128 (0.001 0.002) <0.001** Nur sing Factors RN HPPD 0.985 0.008 (0.970 0 .999 ) 0.049* LPN HPPD 1.083 0.015 (1.052 1.116) <0.001** UAP HPPD 1.061 0.014 (1.033 1.091) <0.001** RN Skill Mix 1.006 0.002 (1.002 1.009) 0.010* Organizational Fact ors Magnet 0.943 0.013 (0.919 0.969) <0.001** Teaching 1.046 0.014 (1.017 1.075) 0.001** Metropolitan 0.981 0.026 (0.933 1.032) 0.467 Northeast a 1.020 0.024 (0.974 1.068) 0.402 South a 1.033 0.021 (0.992 1 .075) 0.114 Midwest a 1.115 0.023 (1.065 1.168) <0.001** Bed100 b 0.962 0.029 (0.909 1.018) 0.183 Bed100to199 b 1.011 0.022 (0.968 1.055) 0.622 Bed200to299 b 0.994 0.021 (0.954 1.037) 0.786 Bed300to399 b 0.976 0.230 (0.933 1.020) 0.272 Bed400to499 b 1.032 0.024 (0.985 1.082) 0.185 surgical), unit fall protocol use and date ve binomial distribution and a log link *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level a = Reference group for Region is West b = Reference group for Bed Size is >500 beds

PAGE 152

152 CHAPTER 7 DISCUSSION AND CONCLUSION Descript ive NDNQI Discussion Descriptive NDNQI findings revealed an overall fall rate across our sample of 3.65 falls per 1 000 patient days. At the unit level, there was variability among fall rates by unit type, and medical units on average had higher fall rate s (mean=4.13 falls per 1 000 patient days) than medical surgical (mean=3.70 falls per 1 000 patient days) and surgical units (mean=2.91 falls per 1 000 patient days). These findings were similar to other research studies using the NDNQI dataset (Dunton et al., 2004, Dunton et al., 2007, Lake et al., 2010). Additionally, the average age of patients who fell on medical units (mean=66.3 years old) was higher when compared to medical surgical (mean=66 years old) and surgical units (mean=64.8 years old), which i s consistent with prior research that a greater incidence of fall rates are typically found as patient age increases (Rubenstein & Josephson, 2002; Clyburn & Heydemann, 2011). Nurse staffing measures varied across unit types and these differences were see n in RN, LPN and UAP staffing. For instance, surgical units had higher levels of RN HPPD (mean=5.9 RN HPPD) when compared to medical (mean=5.7 RN HPPD) and medical surgical units (mean=5.7 RN HPPD). In addition, average UAP HPPD were highest on medical uni ts (mean=2.53 UAP HPPD) and lowest on surgical units (mean=2.48 UAP HPPD). Finally, average LPN HPPD was the highest on medical surgical units (mean=0.49 LPN HPPD). Overall, these findings suggest that there is variation in nurse staffing hours based on th e unit type in hospitals, which was found in prior empirical studies (Dunton et al., 2004; Dunton et al., 2007; Lake et al., 2010).

PAGE 153

153 Descriptive hospital characteristics revealed the largest percentage of NDNQI hospitals based on bed size were 100 to 19 9 beds (31.33%) and 200 to 299 beds (21.02%). However, when compared to general acute care hospitals in the U.S., the NDNQI has a higher percentage of hospitals with bed sizes over 300 (Lake et al., 2010). In addition, 21% of hospitals in our sample had Ma gnet accreditation, which exceeds the U.S. average of 7% of hospitals with Magnet status (Kelly et al., 2011). In addition, 32.06% of hospitals were reported to be teaching status and 87.90% were located in a metropolitan location. Finally, a majority of N DNQI hospitals were located in the South (52.30%) and Northeast (22.67%), and the West and Midwest regions only accounted for approximately 25% of all NDNQI hospitals. These results demonstrated that hospitals in the NDNQI were more often located in a metr opolitan area, had a higher percentage of teaching and Magnet status, and were larger in size when compared to other hospitals in the U.S. Specific Aim 1 Discussion The results of Specific Aims 1 and 1a identified distinct latent class trajectories of ho spitals and hospital units based on their fall rates over time. When conducting LCGM on hospital fall rates, three linear trajectory classes emerged and these were hig from the data. Because of the unique study design, these findings shed new light on classifying hospitals longitudinally based on their fall rates over time, and to our kno wledge, no other studies have used this type of methodology to categorize and predict hospital falls. In addition, we found that there were differences in trajectory membership between hospital units; medical units were categorized into four distinct

PAGE 154

154 laten t class trajectories, surgical units were classified into two distinct latent class trajectories and medical surgical units were classified into four distinct latent class trajectories with differences in baseline fall rate measures across units. Medical u nits had consistently higher fall rates (average 4.2 falls per 1 000 patient days) over time when compared to medical surgical and surgical units, and baseline fall rate measures were consistently higher than other units. Surgical units had the lowest fall rate with an average of 2.91 falls per 1 000 patient days. Based on these descriptive fall rate findings among units, we expected to find differences in the number of latent classes in each of the groups. In addition, these unit level trajectories reveale d that much of the variation in fall rates at the hospital level were the result of medical and medical surgical units, which has important policy implications for targeting interventions to medical and medical surgical units within hospitals. Interesting ly, although we expected that the 2008 CMS non reimbursement policy would have an impact on fall rate trajectories after implementation, trajectories of hospital and hospital unit fall rates indicate only slightly decreasing linear trends that do not devia te significantly from their baseline intercept. These findings indicate that if hospitals were in a high or low performing trajectory before the CMS policy, they remained in this group after policy implementation, suggesting that the non reimbursement poli cy may not be as effective in reducing patient falls as policymakers anticipated after 2008. Furthermore, our findings are consistent with a quasi experimental study by Lee and colleagues (2012) who found that the 2008 CMS non reimbursement policy has not been effective in significantly reducing both catheter associated bloodstream infections and catheter associated urinary tract infections,

PAGE 155

155 which may strength en the argument that financial penalties from the 2008 CMS non reimbursement policy are not enoug h of an incentive for hospital administrators to systematic reduce these adverse events in hospitals However, dat a from our study do es not capture al l hospitals in the U.S. and only considers 27 months after the policy, so there may be improvements in fall rates when considering a longer span of time, but this is unlikely given the relatively small changes in fall rates over the study period. Perhaps the reason hospitals and hospital units remain in a particular fall rate trajectory despite continued policy intervention is due to resource constraints, reimbursement considerations or organizational processes that could not be measured in this study. Nevertheless, prior attempts at reducing falls have not been substantially successful and innovative strategies will be required to target interventions and create effective fall prevention policy. Given the significant amount of attention and resources used to reduce falls over the past few decades, including two IOM reports (IOM, 1999, IOM, 2001) and both state a nd federal policy interventions, it is clear that scholars must consider alternative ways to combat this epidemic. As previously mentioned in the Background section of this study (Chapter 2) previous fall risk assessment interv entions have focused on pati ent level interventions, but they have not been successful at reducing fall rates uniformly over time in hospitals or hospital units. This study provides additional evidence that hospitals have been unsuccessful in systematically reducing falls after the 2 008 CMS policy by showing hospital and hospital unit trajectories with only slightly decreasing parameter estimates over a 54 month period of time. These findings may also support the need for further qualitative inquiry of hospitals and hospital units in consistently high and low latent class trajectory groups to

PAGE 156

156 investigate subjective causal factors for hig her fall rates on these units. However, the latent class trajectory group analysis allow s policymakers to target hospitals with consistently high fall rates for policies or interventions. Specific Aims 2 and 3 Discussion The results of Specific Aims 2 and 3 identified several nurse staffing and rate trajectory group and were associated with patient fall rates longitudinally. From a nurse staffing perspective, there were mixed findings in the GEE analyses: RN HPPD were found to be significantly associated with lower fall rates and membership in the low fall rate trajec tory group, whereas LPN HPPD were associated with higher fall rates and RN skill mix had a weak positive association with f alls over time, and nursing assistant (UAP) HPPD had a statistically signific ant positive association in GEE Analysis #2, but was not significant in GEE Analysis #1. However, these findings may be due to differences in outcome measures between GEE Analysis #1 and GEE Analysis #2. Despite these unexp ected findings, both LPN and nurs ing assistant (UAP) staffing were found in a previous study using GEE to be associated with an increase in falls (Lake et al., 2010). The authors suggest ed that th is could be due to hospitals increasing the number of nursing assistant and LPN staffing on u nits with higher fall rates to reduce expenses, and that higher proportions of nursing assistant and LPN HPPD do not actually cause an increase in fall rates (Lake et al., 2010). Another potential reason for these unexpected findings may be due to the fact that hours of nursing care from LPNs and UAPs were significantly lower when compared to RNs, as well as the fact that LPN staffing is not a critical component of staffing in hospitals when compared to other health care settings, such as nursing

PAGE 157

157 homes and assisted living facilities (Shi & Singh, 2008). These results may also be due to potential unobserved confounders that are related to nurse staffing and patient falls, such as nursing work environment and nursing education For instance, Aiken and colleagu es (2003) found that hospitals with a higher proportion of degrees resulted in reductions in rates of surgical mortality and failure to rescue However, our study used four separate monthly nurse staffing measures over 54 months at t he unit level, which provided robust evidence of chronically understaffed units in hospitals and the ir impact on patient falls. As discussed in Chapter 2, RNs require additional formal education and use o o ther nurses (Unruh, 2003, p. 143). RNs are responsible for supervising unlicensed nursing staff, administering medications, advanced patient assessment skills, and communication with physicians and other health care providers (Unruh, 2003). Successful reduc tions in hospital falls in the literature have been demonstrated through proper initi al fall risk screening follow up assessment during the hospital stay, education of family mem bers and individualized patient centered interventions for at risk patients, which are all implemented and assessed by the RN in hospital clinical units (Perell et al., 2001; Rubenstein, 2005). This could help to explain why RN staffing was associated with reductions in falls in both GEE analyses in our study. In addition, RN staff ing has been found to be important in reducing other types of patient outcomes, such as mortality and failure to rescue (Aiken et al., 2002), as well as reductions in rates of urinary tract infections, pneumonia and cardiac arrest (Needleman et al., 2002). However, an alternative explanation for these findings could be related to a higher proportion of RN staffing in surgical units, which

PAGE 158

158 typically have lower fall rates than other units (i.e. medical and medical surgical units). Resear chers may also want to incorporate other nursing factors, such as level of nursing education and years of experience when conducting research studies, as these may be important factors that influence nursing judgment and promote better patient outco mes (Dunton et al., 2007). Al though we found that LPN and UAP staffing was associated with increased fall rates in GEE Analysis #2, our nurse staffing results for Specific Aim 2 do not diminish the importance of these types of nu rses in hospitals. LPNs and nursing assistants work with RNs to provide direct patient care in hospitals through assessment, evaluation and treatment of patient conditions (Unruh, 2003). As a result, it is important for researchers, clinicians and policymakers to determine an adequate balance of nurses in hosp itals that provides the highest quality of care for patients. We recommend continued research be conducted in the nurse staffing literature to provide specific nurse staffing ratios in hospitals, and these studies may need to consider analyzing nurse to pa tient ratios at the nursing shift level or designing a qualitative study to understand the perspectives of nurses on clinical units (Patrician et al 2011). Nevertheless, our findings do highlight the importance of RN staffing in hospitals and hospital un its in terms of pat ient falls, which has significant implications for policymakers and hospital administrators to continue maintaining an adequate supply of RNs in acute care settings. From an organizational perspective, the results from GEE Analyses for Specific Aim 3 found that despite the various types of sensitivity analyses and aggregated measures to the hospital and unit level, Magnet status remained consistent across all

PAGE 159

159 models; hospital Magnet status was associated with lower fall rates and Magnet fall rate trajectory group when compared to hospitals with over 500 bed s, but were not statistically s ignificant in GEE Analysis #2. Similarly, Dunton and colleagues (2004) used 2002 NDNQI data and found that smaller hospitals were more likely to have increased fall rates when compared to larger hospitals. This relationship m ay be the result of fewer resources in smaller hospitals and an inability to maintain adequate nurse staffing levels, which represents an area for future inquiry. Also, teaching status was found to be significantly associated with an increased risk of fall s in GEE Analysis trajectory group, and this may be due to a lack of patient risk adjustment measures. Nevertheless, we did control for unit type, which is a strong predi ctor of patient acuity and commonly used for risk adjustment (Lake & Cheung, 2006). Finally, the Midwest region was found to have significantly higher fall rates when compared to t he West region, but these findings are difficult to interpret given the fact that we do not have sufficient information on hospital location. In sum, these results reaffirm the importance of considering how organizational characteristics impact patient falls over time and can be used to target policy interventions to particul ar types of hospitals. Since Magnet status was found to be a significant predictor of both decreased fall rates over time and membership in the consistently lower fall rate trajectories, understanding the contextual factors that contribute to these find ings might provide insight into why Magnet hospitals have been successful in reducing fall rates. A recent

PAGE 160

160 article by Kelly and colleagues (2011) found that nurses who work in Magnet hospitals reported significantly better work environments and were consis tently more satisfied with their work. This may be due to fact that Magnet hospitals tend to hav e more advanced technology (e.g. advanced capabi lities for clinical procedures) focus more on education for nurse s, have a higher proportion of b ared nurses and have a greater proportion of certified nurses in various specialties (Kelly et al., 2011). In structures, unit based decision making processes, influential nurse executives and & Sloane, 2009, p.26). These organizational culture and leadership factors may play an important role in n may lead to better nurse and patient outcomes. In the nurse staffing and patient falls literature, Magnet status has been shown to be significantly reduce fall rates (Dunton et al., 2007; Lake et al., 2010), and other studies have also found better patie nt outcomes in Magnet hospitals when compared to other hospitals (Scott et al., 1999; Havens & Aiken, 1999). In conclusion, our theoretically based conceptual model posited that hospital organizational characteristics and unit level nurse staffing influen ce the processes of patient car e, which then impact s patient fall rates in hospitals. Using LCGM, we were able to identify latent trajectories of hospitals based on changes in fall rates over time and test our conceptual model to determine how differences in organizational characteristics (e.g. Magnet status and bed size) and nurse staffing (e.g. RN HPPD and LPN HPPD) influence patient fall rates. Our study results supported the derived NSPF conceptual model in Chapter 4; RN HPPD and Magnet status hospitals were both found

PAGE 161

161 to be associated with reductions in patient falls Also, w e found partial support for our hypothesis that hospitals with greater bed size were associated lower fall rates when compared to hospitals with smaller bed size Study Limitati ons & Strengths Despite significant strengths of the current study, which include longitudinal data and an extensive set of structure and process level predictors at the unit level, there are s everal limitations that should be noted These limitations incl ude a self selected sample, a lack of formal risk adjustment measures, fall counts reported thr ough incident reports, uneven sample sizes and missing data, the potential for omitted variables and the potential for response bias when hospitals classify clin ical unit type in their response to the survey. First, the NDNQI sample represents a self selected sample of hospitals. It is likely that given hospitals are required to pay to be included in the survey, smaller hospitals may not be able to participate due to financial barriers (Dunton et al., 2007). Additionally, despite the nationwide scope of this data source, prior research has shown that the NDNQI contains a higher percentage of large r bed size hospitals Magnet s tatus and metropolitan location when co mpared to other nationally representative datasets (Dunton et al., 2007; Lake et al., 2010). As a result, the implications of this self selected sample is that we might expect different relationships between nurse staffing and falls in hospitals with lower levels of nurse staffing (i.e. greater variability) and fewer resources, such as those found in smaller hospitals. For example, some of the associations between nurse staffing and patient falls may only be evident in these smaller hospitals. Therefore, ho spitals that are not included in the NDNQI may have performed worse in terms of fall rates over time and the problem may be magnified nationwide. This is due to the fact that previous studies have shown that hospitals with

PAGE 162

162 smaller bed size and non Magnet s tatus typically have higher fall rates when compared to their counterparts, which was supported in our study (Dunton et al., 2007; Lake et al. 2010). Furthermore, the NDNQI does not allow for adjustment of patient acuity, and formal risk adjustment method s can only be performed by controlling for unit type, which is a common and useful approach that distinguishes between severity of patient illness across clinical units (Dunton et al., 2004; Lake & Cheung, 2006; Dunton et al., 2007). Though not available i n the current dataset, incorporating a validated formal patient level risk adjustment measure, such as the Charlson Comorbidity Index, would have allowed for better statistical control of patient acuity (Charlson, Pompei, Ales & MacKenzie, 1987). In the c urrent context, facilities with higher nurse staffing may have higher acuity patients, who would also be at higher risk of falling. Importantly, Magnet status hospitals tend to have higher acuity patients and many studies have shown that these higher acuit y facilities result in reductions in falls and other adverse outcomes (Scott et al., 1999; Havens & Aiken, 1999; Lake et al., 2010; Kelly et al., 2011). In the current investigation, Magnet status hospitals were associated with lower rates of falls, sugges ting current associations not adjusted for patient acuity may be conservative. Additional limitations include fall counts that are reported through incident reports, which may inaccurately represent the actual number of falls on a unit (Dunton et al., 200 4). However, there is no reason to expect that reporting of fall counts would be different by hospital with different skill mix or staffing, and hospital identity is anonymous, so there is no incentive for hospitals to report i naccurate information. Furthe rmore limitations of missing data and uneven sample sizes each year in the

PAGE 163

163 NDNQI dataset were addressed in our study by use of sensitivity analyses and multiple imputation methodology (Harless & Mark, 2006). Since nurse staffing hours are reported on a mo nthly basis, the aggregated staffing measures do not capture shift by shift and daily variability, and this could introduce measurement error with a bias toward the null hypothesis. As a result, our findings may have produced conservative estimates regardi ng the impact of nurse staffing on hospital falls over time. Other limitations in our study include the potential for omitted variables that are related to both the independent and dependent variables, such as for profit status and nursing level characteri stics of self perceived work environment and nurse education level. These factors could influence both job satisfaction and burnout rates for nurses, which could possibly also have an adverse effect on patient fall ra tes (Kelly et al. 2011). Nevertheless we incorporated four separate monthly nurse staffing measures over 54 months at the unit level, as well as organizational characteristics that were not considered i n prior patient falls studies. Future research should be conducted to account for these mis sing variables by linking the NDNQI with other nationwide hospital datasets to allow for better statistical control and adjustment for additional hospital characteristics, such as those found in the American Hospital Association (AHA) survey data. This wou ld allow for an even more comprehensive set of hospital factors. Also, the NDNQI recently began including measures of nursing work environment, which would provide greater understanding of nursing level characteristics and patient falls in the future. Fina lly, our results could be impacted by the manner in which hospitals classify clinical unit type in their response to the NDNQI survey. However, the NDNQI works extensively with hospitals through the use of data coordinators and reported information

PAGE 164

164 is cont inually tested for reliability and accuracy (Dunton et al., 2004; Simon et al. 2010), so we would not expect this to be different between hospitals in our study, or between hospitals with higher or lower levels of nurse staffing. Finally, our study was li mited by the fact that we were unable to determine if non Magnet status hospitals were utilizing principles of Magnet facilities or actively seeking Magnet certification during the time of the NDNQI survey. However, if non Magnet hospitals were incorporati ng Magnet principles, we would expect there to be fewer organizational differences between facilities with and without Magnet status and by extension, less of a difference between fall rates in Magnet and non Magnet status facilities. Despite this potentia l, facilities with Magnet hospital designation had significantly lower fall rates when compared to non Magnet hospitals Further research should be conducted to better understand the components of Magnet status that lead to reductions in patient falls in h ospitals. Despite these limitations, our study has significant strengths, including the longitudinal nature of the dataset, which allows for determination of temporal precedence and causal inferences about the relationship between nurse staffing and patien t falls (Dunton et al., 2007). Also, the NDNQI is a nationwide, large database with over 1,500 hospitals, and information collected at both the hospital and unit level provides a detailed understanding of the relationships between constructs and concepts ( Montalvo, 2007; Dunton et al., 2007). As a result, the NDNQI is currently the only database that contains structure, process and outcome measures nationally at the hospital unit level. This represented an opportunity to explore the relationship between org anizational characteristics, unit level nurse staffing and patient falls longitudinally

PAGE 165

165 using group based modeling techniques of LCGM, which allowed for unprecedented analysis of hospitals based on their fall rates over time. To our knowledge, no other stu dy has used these methods in the nurse staffing and patient falls literature to categorize and classify hospitals based on their fall rates over time, which represents a significant advancement in the patient safety literature. Also, we incorporated an int egrated conceptual model that draws on prior theory and empirical evidence, addressing gaps in prior theoretical models and articulating a clear pathway between hospital organizations, nurse staffing and patient falls in the hospital setting. Policy Impli cations and Future Research The results of this study have implications for policymakers, managers and future research studies. At the federal policymaking level, we anticipated that the 2008 CMS non reimbursement policy would have encouraged hospitals to reduce their overall fall rates given the fact they are not compensated for fall related injuries (Mattie & Webser, 2008). However, latent class trajectory models illustrated only slightly decreasing fall rate trends over time and these were much lower tha n we initially anticipated, particularly after implementation of the policy in October of 2008. As a result, policymakers will need to consider reasons why fall rates have not been systematically reduc ed and what can be done to effectively decrease falls i n hospitals. One potential reason for the unsuccessful reduction in falls from the 2008 CMS non reimbursement policy could be that hospitals with higher fall rates may not have the financial resources to implement fall risk strategies and hire new nurses, indicating that these hospitals may need additional governmental assistance or other strategies to successfully reduce patient falls. Future research should be employed using a pre post study design to

PAGE 166

166 investigate the effectiveness of the 2008 CMS non reim bursement policy on patient falls and other adverse events From a hospital administration point of view, findings from our study revealed that an effective approach to reducing falls should be to focus research and policy interventions for patient safet y at both the hospital and clinical unit level. Targeting policies solely at the hospital level may not be effective because patient care is provided at the unit level and some studies in healthcare quality have found that targeting interventions to the un it level while providing p atient centered care is an effective approach to improving patient outcomes (Nelson, Batalden & Godfrey, 2007). According to Nelson and colleagues (2007), successful characteristics of clinical units in terms of quality and patie interdependence of the care team, performance results, process improvement, patient hospitals. Additionally, policy strategies may need to focus on some of the key aspects of Magnet accreditation, such as through improving on areas of nursing education, organizational leadership and providing better nursing work environments (Kelly et al., 2011). Due to financial const raints, all hospitals may not be able to achieve Magnet status, but there may be particular aspects of Magnet certification that might assist in reducing fall rates, and hospitals c ould attempt to replicate these principles of Magnet in their organizations Therefore, f uture research should be conducted to investigate some of the key aspects of Magnet status and h ow they impact patient safety. G iven the importance of nurses to the quality of patient care, including reductions of falls and other

PAGE 167

167 important nu rse sensitive indicators in acute care settings, policymakers and managers should continue to assist in maintaining an adequate nursing workforce in the United States through educational programs for nurses and retention programs in health care settings, s uch as those incorporated in the Nurse Reinvestment Act of 2002 (Donl ey et al., 2003). Future research studies should incorporate qualitative methodology in hospitals, particularly in consistently high and low fall rate trajectory groups, to determine how organizational culture and unit level processes influence fall rates in hospitals. Subjective responses from nursing staff members, managers and patients on these units could provide insight as to why fall rates differ between hospitals and hospital units. Additional investigation using qualitative research should also be employed at Magnet facilities to examine the key characteristics of these hospitals and units in reducing patient falls by interviewing nurses, managers and patients. This is due to the fa ct that the mechanism s through which Magnet status is associated with reductions in better nurse staffing, commitment to nursing education or organizational structural components (Kelly et al., 2011). As an example, a focused ethnographic study using participant observation, semi structured interviews and key informant interviews could be used to develop a descriptive understanding of the organizational culture and proc esses within hospital units and to describe the experiences of nur ses in those units (Richards & Morse, 2007). This type of study would provide a clear and descriptive understanding why patient falls are occurring on th ese types of units (Richards & Morse, 2007).

PAGE 168

168 Future quantitative research should be employed using nationally representative data at the hospital and unit level to examine the impact of nurse staffing on patient fall rates over time, such as from the NDNQI. Researchers should also consider m ulti level methods that account for variation in units within hospitals longitudinally. Additiona lly, future research should account for clustering of units within hospitals, as much of the variation in unit level predictors are lost when aggregating these measures to the hospital level. This could be accomplished by conducting multivariate analysis separately for various unit types, such as medical, surgical and medical surgica l units (Dunton et al., 2004). As previously mentioned, research should also con sider accounting for shift level variability in nursing staffing, which could help to explain some of the variation in nurse staffing and fall rates at the unit level (Patrician et al., 2011 ). Lastly, g iven recommendations from empirical studies, as well as pressure from hospita ls to ambulate patients immediately after surgery, such as after hip fracture surgery ( Kamel, Iqbal Moga llapu, Maas & Hoffman, 2003), f uture quantitative studies shoul d compare assisted and unassisted fall rates to understand uni t level predictors of these different types of patient falls. This would be an important contribution to the literature given the fact that assisting patients to ambulate may contribute to increasing patient fall rates in hospitals, while at the same time policies could contribute to declining fall rates, with a net effect of little change over time. The current investigation did not find evidence of increasing or decreasing fall rate trajectory groups in hospit als or hospital units. To explore the potentia l for changes in trends for assisted versus unassisted falls contributing to static fall rates over time, we ran preliminary descriptive models comparing assisted and unassisted fall rates in hospital units. Findings suggest

PAGE 169

169 no significant differences betw een unassisted versus assisted falls in hospital units over time, so it is unlikely changes contributed to observed effects in the current investigation. Overall, t his dissertation represents a significant contribution to public health and patient safety i n hospitals given the importance of patient falls in terms of quality of care, health care expenditures and overall quality of life for patients. Although we found some mixed evidence regarding nurse staffing and organizational characteristics on patient f alls over time, we found that Magnet hospital designation and greater levels of RN staffing consistently resulted in lower fall rates in the NDNQI from 2006 to 2010, suggesting that hospitals may be able to reduce fall rates by maintaining greater RN staff ing ratios as well as fostering an environment consistent with that of Magnet hospitals. We used innovative methodologies that categorize and classify hospitals based on their fall rates over time, which allowed for an in depth assessment of nursing and or ganizational characteristics that influence fall rates. Our results have important implications for managers, clinicians, researchers and policymakers to create safety guidelines and interventions that specifically target hospitals and clinical units, part icularly replicating some of the key findings from this dissertation. In conclusion, the results of this study represent a significant and important contribution to the field of public health and patient safety within acute care setting s

PAGE 170

170 APPENDIX A ADDITIO NAL MAGNET HOSPITAL INFO RMATION The following section describes specific instructions on the application proces s for hospitals to become Magnet status certified. Through the Magnet Recognition Program the American Nurs es Credentialing Center (ANCC) evalu ates hospitals through a rigorous application process and hospitals must meet the following eligibility criteria to be considered (ANCC, 2008 p.5) ): 1. The applicant organization must exist within a healthcare organization. They must be based in the U.S. or the international community. All phases of the application review are conducted in the English language. 2. The applicant organization must designate one person as the Chief Nursing Officer (CNO), who is ultimately responsible for sustaining the stan dards of nursing practice throughout the organization. All areas for which the CNO is responsible for nursing practice must be included in the application, regardless of reporting relationships. The C NO must be an active participant on the application or decision making and strategic planning body. least 75% of nurse managers of individu al units/wards/clinics must have at least a baccalaureate degree in nursing upon submission of the application, effective January 1 2011. By January 1, 2013, 100% of nurse managers of individual units/wards/clinics must have at least a baccalaureate degre e in nursing upon submission of the application. Nurse Administrators must be implemented throughout the nursing system. 5. Applicants for Magnet recognition must collect data reflect ing nursing sensitive outcomes/quality indicators at the unit level, assess changes at least quarterly, and compare benchmark aggregate hospital level performance of that data against a national benchmark database for at least two years prior to submitting written documentation. Once hospitals have met all eligibility requirements, they can then apply to meet the requirements of Magnet status through a review in four separate phases (ANCC, 2008, p.10):

PAGE 171

171 Phase 1 Application The hospital applicant complet es an electronic application specifying a primary contact person and the settings for whi ch the CNO has responsibility. Applications are accepted throughout the year. Phase 2 Submission of Written Documentation The review process begins when the writte n documentation addressing the required sources of evidence for each model component is submitted by the applicant. This written document must reflect the innovative, dynamic, excellence focused features of the organization that are developed, disseminated and enculturated throughout the organization. It also must demonstrate how the management philosophy, as well as how t he standards are incorporated within the nursing service. Phase 3 Site Visit. A site visit occurs if the scores for the sources of evidence fall within a range of excellence. The purpose of the site visit is to verify, clarify and amplify the content o f the written documentation and evaluate the organizational setting in which nursing is practiced. Most site visits are three days in duration. The review of the written documentation and the site visit appraisal are conducted by professional, registered n urses with experience in evaluating quality indicators, nursing services administration, and nursing care. Phase 4 Award Decision The Commission on Magnet reviews the completed appraisal report to determine if Magnet Recog nition status will be awarded. The Commission, who is responsible for monitoring, compliance with program standards, criteria and the sources of evidence, then meet and vote to determine if the hospital will be designated as Magnet status. The award decisions for Magnet status are bas ed on five model components, including (ANCC, 2008. p.7 8): 1. Transformational Leadership Quality of nursing leadership and management style 2. Structural Empowerment Organizational structure, personnel policies and programs, community and the healthcare organization, image of nursing, and professional development 3. Exemplary Professional Practice Professional m odels of care, quality of care through ethics, patient safety and quality infrastructure, quality improvement, consultation and resources, autono my, nurses as teachers and interdisciplinary relationships.

PAGE 172

172 4. New Knowledge, Innovations, and Improvements Quality of care through research and evidence based practice, and quality improvement. 5. Empirical Quality Outcomes Quality of care

PAGE 173

173 APPENDIX B SPECIFIC AIM 1 METHO DOLOGICAL AND ANALYT IC CONSIDERATIONS Figure B 1. Distribution of fall rates over 54 months in the NDNQI

PAGE 174

174 Figure B 2. A djusted hospital monthly fall rate trajectory model controlling for time varying nurse staffing factors and time invariant organizational factors

PAGE 175

175 Figure B 3. Unadjusted medical unit monthly fall rate trajectory model with three linear trajectory groups

PAGE 176

176 Figure B 4. Unadjusted surgical unit monthly fall rate trajectory model with three linear trajectory grou p

PAGE 177

177 Figure B 5. Adjusted medical unit monthly fall rate trajectory model with four linear trajectory groups

PAGE 178

178 Figure B 6. Adjusted surgical unit monthly fall rate trajectory model with two linear trajectory groups

PAGE 179

179 Figure B 7. Adjusted medica l surgical unit monthly fall rate trajectory model with four linear trajectory groups

PAGE 180

180 Figure B 8. Unadjusted multiple imputation h ospita l trajectory m odel 1

PAGE 181

181 Figure B 9 Unadjusted multiple imputation hospital trajectory m odel 2

PAGE 182

182 Figure B 10 Unadjusted multiple imputation hospital trajectory m odel 3

PAGE 183

183 Figure B 11 Unadjusted multiple imputation hospital t raject ory m odel 4

PAGE 184

184 Figure B 12 Unadjusted multiple imputation hospital trajectory m odel 5

PAGE 185

185 Figure B 1 3. Unadjusted quarterly hospital model results with three linear trajectories

PAGE 186

186 Table B 1 Descriptive statistics of medical unit observations: Mean fall rates, standard deviations, minimums and maximums per month Variable Year M onth Observations Mean Std Dev Min Max Medical1 2006 7 1221 4.23 3.33 0 35.21 Medical2 2006 8 1220 4.36 3.22 0 35.88 Medical3 2006 9 1224 4.33 3.34 0 25.86 Medical4 2006 10 1276 4.24 3.12 0 20.41 Medical5 2006 11 1275 4.29 3.24 0 23.70 Medical6 2006 12 1265 4.45 3.32 0 25.50 Medical7 2007 1 1364 4.3 3.18 0 20.92 Medical8 2007 2 1358 4.16 3.31 0 35.09 Medical9 2007 3 1360 4.17 3.11 0 28.38 Medical10 2007 4 1408 4.13 3.14 0 30.67 Medical11 2007 5 1407 4.19 3.18 0 23.17 Medical12 2007 6 1394 4.17 3.32 0 29.70 Medical13 2007 7 1440 4.06 2.99 0 20.85 Medical14 2007 8 1437 4.20 3.10 0 32.65 Medical15 2007 9 1438 4.17 3.23 0 28.02 Medi cal16 2007 10 1505 4.20 3.23 0 24.55 Medical17 2007 11 1500 4.14 3.17 0 27.07 Medical18 2007 12 1512 4.16 3.17 0 19.96 Medical19 2008 1 1588 4.13 3.27 0 26.55 Medical20 2008 2 1591 4.14 3.11 0 22.26 Medical21 2008 3 1589 4.12 3.05 0 25.24 Medical22 2008 4 1624 4.23 3.27 0 29.07 Medical23 2008 5 1622 4.06 3.20 0 36.95 Medical24 2008 6 1620 3.98 3.29 0 45.11 Medical25 2008 7 1631 4.13 3.34 0 42.25 Medical26 2008 8 1626 4.30 3 .34 0 29.68 Medical27 2008 9 1625 4.09 3.28 0 36.29 Medical28 2008 10 1646 4.23 3.24 0 31.03 Medical29 2008 11 1650 4.27 3.29 0 26.39 Medical30 2008 12 1648 4.23 3.29 0 31.18 Medical31 2009 1 1711 4.36 3.36 0 30.53 Medical32 2009 2 1708 4.25 3.40 0 38.07 Medical33 2009 3 1706 4.15 3.15 0 25.67 Medical34 2009 4 1743 4.13 3.23 0 42.27 Medical35 2009 5 1736 4.21 3.32 0 34.29 Medical36 2009 6 1733 4.03 3.16 0 38.42 Medical37 200 9 7 1779 4.13 3.39 0 39.82 Medical38 2009 8 1776 4.09 3.15 0 29.63 Medical39 2009 9 1776 4.11 3.48 0 36.78 Medical40 2009 10 1823 4.18 3.18 0 32.00

PAGE 187

187 Table B 1 Continued Variable Year Month Observations Mean Std Dev Min Max Medical41 2009 11 1826 4.15 3.53 0 47.62 Medical42 2009 12 1830 4.32 3.61 0 50.64 Medical43 2010 1 1893 4.04 3.04 0 22.09 Medical44 2010 2 1887 3.97 3.17 0 23.74 Medical45 2010 3 1889 4.00 2.97 0 18.2 6 Medical46 2010 4 1960 4.06 3.28 0 33.56 Medical47 2010 5 1959 3.96 2.99 0 22.61 Medical48 2010 6 1954 3.96 3.33 0 58.82 Medical49 2010 7 1991 3.98 3.27 0 30.17 Medical50 2010 8 1979 3.96 3.20 0 31.25 Medical51 20 10 9 1979 3.92 3.30 0 55.56 Medical52 2010 10 1995 4.09 3.39 0 32.37 Medical53 2010 11 1994 3.99 3.18 0 30.08 Medical54 2010 12 1980 3.99 3.21 0 31.25

PAGE 188

188 Table B 2. Descriptive statistics of surgical unit observation s: Mean fall rates, standard deviations, minimums and maximums per month Variable Year Month Observations Mean Std Dev Min Max Surgical1 2006 7 931 2.91 2.96 0 21.07 Surgical2 2006 8 920 2.82 2.71 0 24.15 Surgical3 2006 9 924 2.83 2.75 0 18.99 Surgical4 2006 10 966 3.02 2.94 0 27.41 Surgical5 2006 11 966 2.94 2.92 0 27.25 Surgical6 2006 12 958 3.00 2.83 0 17.86 Surgical7 2007 1 1047 2.95 2.86 0 23.53 Surgical8 2007 2 1038 2.96 2 .88 0 27.40 Surgical9 2007 3 1025 2.96 2.56 0 16.15 Surgical10 2007 4 1085 2.84 2.99 0 36.14 Surgical11 2007 5 1083 2.88 2.96 0 33.90 Surgical12 2007 6 1081 2.77 2.93 0 22.06 Surgical13 2007 7 1099 3.04 3.35 0 44.9 4 Surgical14 2007 8 1100 2.76 2.85 0 32.92 Surgical15 2007 9 1096 2.99 3.20 0 32.05 Surgical16 2007 10 1164 2.85 2.93 0 29.63 Surgical17 2007 11 1158 2.95 2.85 0 27.05 Surgical18 2007 12 1166 2.90 2.84 0 24.22 Surgi cal19 2008 1 1221 2.93 2.71 0 16.70 Surgical20 2008 2 1217 3.13 2.95 0 23.70 Surgical21 2008 3 1219 2.95 2.79 0 23.81 Surgical22 2008 4 1252 3.04 3.00 0 25.32 Surgical23 2008 5 1245 2.83 3.00 0 33.90 Surgical24 2008 6 1240 2.86 2.77 0 18.75 Surgical25 2008 7 1240 2.68 2.67 0 16.98 Surgical26 2008 8 1238 2.98 2.93 0 29.41 Surgical27 2008 9 1235 2.97 3.24 0 47.62 Surgical28 2008 10 1234 3.17 3.15 0 25.53 Surgical29 2008 11 1232 3.08 2.96 0 18.43 Surgical30 2008 12 1229 3.03 2.99 0 18.87 Surgical31 2009 1 1278 3.01 2.88 0 30.49 Surgical32 2009 2 1278 2.95 3.11 0 26.79 Surgical33 2009 3 1274 3.07 2.97 0 28.57 Surgical34 2009 4 1300 2.82 2. 80 0 33.33 Surgical35 2009 5 1292 2.81 2.82 0 23.26 Surgical36 2009 6 1286 2.88 3.00 0 30.93 Surgical37 2009 7 1315 2.92 3.02 0 31.58 Surgical38 2009 8 1321 2.75 2.87 0 33.33 Surgical39 2009 9 1322 2.84 2.95 0 48.1 9

PAGE 189

189 Table B 2. Continued Variable Year Month Observations Mean Std Dev Min Max Surgical40 2009 10 1359 2.95 2.96 0 38.46 Surgical41 2009 11 1360 2.89 3.04 0 34.48 Surgical42 2009 12 1364 2.90 2.89 0 21.90 Surgical 43 2010 1 1381 2.85 2.88 0 31.75 Surgical44 2010 2 1382 2.94 2.87 0 22.42 Surgical45 2010 3 1377 2.85 2.76 0 26.02 Surgical46 2010 4 1418 2.88 2.81 0 32.89 Surgical47 2010 5 1418 2.87 3.23 0 62.50 Surgical48 2010 6 1411 2.77 2.87 0 28.17 Surgical49 2010 7 1426 2.74 2.81 0 25.64 Surgical50 2010 8 1425 2.71 2.70 0 17.24 Surgical51 2010 9 1424 2.79 2.69 0 21.98 Surgical52 2010 10 1424 2.96 3.16 0 52.63 Surgical53 2010 11 1429 2 .92 2.96 0 25.32 Surgical54 2010 12 1418 2.99 3.13 0 37.03

PAGE 190

190 Table B 3. Descriptive statistics of medical surgical unit observations: Mean fall rates, standard deviations, minimums and maximums per month Variable Year Month Observatio ns Mean Std Dev Min Max Medsurg1 2006 7 1577 3.80 3.24 0 34.88 Medsurg2 2006 8 1567 3.82 3.35 0 26.25 Medsurg3 2006 9 1574 3.81 3.22 0 20.07 Medsurg4 2006 10 1589 4.12 3.49 0 32.36 Medsurg5 2006 11 1588 4.04 3 .43 0 26.67 Medsurg6 2006 12 1587 4.04 3.77 0 44.03 Medsurg7 2007 1 1693 4.02 3.25 0 25.29 Medsurg8 2007 2 1668 4.03 3.30 0 21.86 Medsurg9 2007 3 1654 3.89 3.28 0 29.22 Medsurg10 2007 4 1747 3.86 3.34 0 40.00 Meds urg11 2007 5 1746 3.79 3.22 0 28.80 Medsurg12 2007 6 1750 3.60 3.19 0 36.14 Medsurg13 2007 7 1767 3.74 3.33 0 29.85 Medsurg14 2007 8 1769 3.74 3.23 0 25.16 Medsurg15 2007 9 1763 3.64 3.12 0 25.23 Medsurg16 2007 10 1900 3.79 3.08 0 22.73 Medsurg17 2007 11 1896 3.92 3.27 0 26.09 Medsurg18 2007 12 1889 3.72 3.18 0 22.73 Medsurg19 2008 1 2004 3.78 3.17 0 28.69 Medsurg20 2008 2 2014 3.75 3.24 0 33.50 Medsurg21 2008 3 2011 3.66 3 .00 0 25.16 Medsurg22 2008 4 2070 3.77 3.25 0 32.26 Medsurg23 2008 5 2069 3.58 3.13 0 27.03 Medsurg24 2008 6 2057 3.53 3.01 0 25.64 Medsurg25 2008 7 2088 3.72 3.25 0 27.62 Medsurg26 2008 8 2083 3.67 3.22 0 45.23 M edsurg27 2008 9 2068 3.72 3.41 0 31.09 Medsurg28 2008 10 2095 3.92 3.38 0 31.91 Medsurg29 2008 11 2091 3.81 3.50 0 46.51 Medsurg30 2008 12 2087 3.81 3.27 0 23.53 Medsurg31 2009 1 2170 3.77 3.26 0 43.48 Medsurg32 200 9 2 2173 3.68 3.17 0 22.14 Medsurg33 2009 3 2159 3.74 3.11 0 31.37 Medsurg34 2009 4 2205 3.74 3.28 0 25.96 Medsurg35 2009 5 2200 3.60 3.18 0 25.86 Medsurg36 2009 6 2185 3.53 3.09 0 35.71 Medsurg37 2009 7 2216 3.66 3.23 0 25.45 Medsurg38 2009 8 2217 3.63 3.15 0 24.28 Medsurg39 2009 9 2213 3.66 3.16 0 22.96

PAGE 191

191 Table B 3. Continued Variable Year Month Observations Mean Std Dev Min Max Medsurg40 2009 10 2261 3.67 3.09 0 27.20 Medsurg41 2009 11 2261 3.71 3.27 0 36.36 Medsurg42 2009 12 2250 3.60 3.14 0 20.27 Medsurg43 2010 1 2355 3.69 3.09 0 28.93 Medsurg44 2010 2 2359 3.53 3.06 0 26.85 Medsurg45 2010 3 2357 3.61 3.09 0 26.10 Medsurg46 2 010 4 2434 3.67 3.28 0 30.30 Medsurg47 2010 5 2436 3.60 3.04 0 24.10 Medsurg48 2010 6 2426 3.52 3.14 0 32.82 Medsurg49 2010 7 2466 3.46 2.93 0 19.99 Medsurg50 2010 8 2462 3.52 3.14 0 26.55 Medsurg51 2010 9 2460 3. 49 3.08 0 32.26 Medsurg52 2010 10 2522 3.60 3.30 0 69.37 Medsurg53 2010 11 2515 3.53 3.31 0 43.48 Medsurg54 2010 12 2500 3.58 3.15 0 22.37

PAGE 192

192 Table B 4 Adjusted hospital fall rate model with Poisson maximum likelihood estimates of three linear trajectories (N=1,592) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 1.907 0.293 <0.0001** Linear 0.00008 0.00001 <0.0001** RN 0.008 0.010 0.410 0 LPN 0.068 0.022 0.002 0 ** UAP 0.039 0.018 0.575 0 Skillmix 0.005 0.249 0.058 0 Protocol 0.044 0.023 0.060 0 2 Intercept 1.060 0.176 <0.0001** Linear 0.00003 0.00001 <0.0001** RN 0.013 0.008 0.125 0 LPN 0. 104 0.013 <0.0001** UAP 0.065 0.014 <0.0001** Skillmix 0.008 0.181 <0.0001** Protocol 0.074 0.014 <0.0001** 3 Intercept 2.189 0.210 <0.0001** Linear 0.00006 0.00001 <0.0001** RN 0.046 0.008 <0. 0001** LPN 0.080 0.015 <0.0001** UAP 0.015 0.014 0.298 0 Skillmix 0.001 0.180 0.503 0 Protocol 0.004 0.019 0.827 0 Time Invariant Organizational Variables 1 Constant (0.00) 2 Constant 2. 563 0.548 0.000 ** bed100 1.251 0.404 0.002 bed100to200 0.781 0.375 0.037 bed200to300 0 766 0.376 0.041 bed300to400 0.827 0.384 0.031 bed400to500 0.747 0.446 0.094 MSA 0.257 0.289 0.374 Magnet 0.353 0.179 0.048* teaching 0.413 0.180 0.021* northeast 0.854 0.332 0.010* south 0.806 0.309 0.009 ** west 1.212 0.352 0.001 ** 3 Constant 1.397 0.619 0.024 bed100 0.396 0.494 0.422

PAGE 193

19 3 Table B 4 Continued Group Parameter Estimate Standard Error Prob>|T| Bed100to200 0.170 0.457 0.710 bed200to300 0.134 0.462 0.772 bed300to400 0.323 0.482 0.503 bed400to500 0.247 0.518 0.633 MSA 0.481 0.29 7 0.105 Magnet 0.735 0.224 0.001 ** teaching 0.642 0.204 0.002** northeast 0.916 0.343 0.008 ** south 1.209 0.321 <0.00 1** west 1.700 0.381 <0.00 1** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 194

194 Table B 5 Unadjusted surgical unit fall rate model with Poisson maximu m likelihood estimates of three linear trajectories (N=1,865) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 1. 130 0.164 <0.0001** Linear 0.000 01 0.00001 0.650 0 2 Intercept 1.579 0.136 <0.0001** Linear 0.000 01 0.00001 0.650 0 3 Intercept 2.607 0.387 <0.0001** Linear 0.00002 0.00002 0.356 0 Group Membership 1 50.40% 1.427 <0.0001** 2 43.20% 1.410 <0.0001** 3 6.40% 0.595 <0.0001** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 195

195 Table B 6 Adjusted medical unit fall rate model with Poisson m aximum likelihood estimates of four linear trajectories (N=2,558) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 1.295 0.379 0.001 0 ** Linear 0.00005 0.00002 0.007 0 ** RN 0.025 0 .008 0.002 0 ** LPN 0.041 0.029 0.158 0 UAP 0.085 0.022 <0.0001** Skillmix 0.002 0.226 0.279 0 Protocol 0.046 0.024 0.052 0 2 Intercept 2.173 0.184 <0.0001** Linear 0.00006 0.00001 <0.0001** RN 0.021 0.0 09 0.015 0 LPN 0.070 0.015 <0.0001** UAP 0.038 0.014 0.008 0 ** Skillmix 0.001 0.188 0.626 0 Protocol 0.070 0.011 <0.0001** 3 Intercept 1.522 0.225 <0.0001** Linear 0.00001 0.00001 0.218 0 RN 0.011 0.0 08 0.178 0 LPN 0.085 0.014 <0.0001** UAP 0.037 0.013 0.003 0 ** Skillmix 0.003 0.172 0.088 0 Protocol 0.058 0.011 <0.0001** 4 Intercept 3.719 0.349 <0.0001** Linear 0.000 0.00002 0.858 0 RN 0.112 0.010 <0 .0001** LPN 0.115 0.024 <0.0001** UAP 0.085 0.017 <0.0001** Skillmix 0.032 0.240 <0.0001** Protocol 0.139 0.024 <0.0001** Time Invariant Organizational Variables 1 Constant (0.00) 2 Constant 0.923 0.182 <0.0001** Morse 0.106 0.192 0.581 0 Schmid 0.498 0.377 0.187 0 Hendrich 0.528 0.295 0.073 0 3 Constant 0.815 0.234 <0.0001** Morse 0.095 0.168 0.570 0 Schmid 0.512 0.373 0.170 0

PAGE 196

196 Table B 6 Cont inued Group Parameter Estimate Standard Error Prob>|T| Hendrich 0.036 0.294 0.902 4 Constant 0.863 0.284 0.002 ** Morse 0.358 0.245 0.144 Schmid 1.391 0.425 0.001 ** Hendrich 0.321 0.464 0.490 *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 197

197 Table B 7 Adjusted surgical unit fall rate model with Poisson maximu m likelihood estimates of two linear trajectories (N=1,865) Group Parameter Estimate Standar d Error Prob>|T| 1 Intercept 1.443 0.191 <0.0001** Linear 0.00007 0.00001 <0.0001** RN 0.090 0.006 <0.0001** LPN 0.112 0.013 <0.0001** UAP 0.075 0.011 <0.0001** Skillmix 0.010 0.157 <0.0001** Protocol 0.08 2 0.013 <0.0001** 2 Intercept 1.021 0.180 <0.0001** Linear 0.00003 0.00001 <0.0001** RN 0.030 0.007 <0.0001** LPN 0.063 0.014 <0.0001** UAP 0.041 0.014 0.004 0 ** Skillmix 0.004 0.177 0.020 0 ** Protocol 0.063 0.009 <0.0001** Time Invariant Organizational Variables 1 Constant ( 0.00 ) 2 Constant 0.956 0.089 <0.0001** Morse 0.353 0.131 0.009 0 ** Schmid 0.250 0.274 0.363 0 Hendrich 0.074 0.1 82 0.683 0 *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 198

198 Table B 8 Adjusted medical surgical unit fall rate model with Poisson m aximum likelihood estimates of four linear trajectories (N=3,448) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 2.65 0.317 <0.0001** Linear 0.00006 0.00002 0.001 0 ** RN 0.094 0.009 <0.0001** LPN 0.109 0.020 <0.0001** UAP 0.004 0.016 0.789 0 Skillmix 0.0 02 0.220 0.453 0 Protocol 0.110 0.022 <0.0001** 2 Intercept 1.672 0.264 <0.0001** Linear 0.00005 0.00001 <0.0001** RN 0.034 0.011 0.002 0 ** LPN 0.048 0.019 0.012 0 UAP 0.018 0.018 0.325 0 Skillmix 0. 006 0.329 0.068 0 Protocol 0.061 0.014 <0.0001** 3 Intercept 1.699 0.152 <0.0001** Linear 0.00004 0.00001 <0.0001** RN 0.030 0.006 <0.0001** LPN 0.068 0.012 <0.0001** UAP 0.027 0.011 0.018 0 Skillmix 0.003 0.146 0.047 0 Protocol 0.052 0.009 <0.0001** 4 Intercept 2.363 0.154 <0.0001** Linear 0.00006 0.00001 <0.0001** RN 0.074 0.005 <0.0001** LPN 0.052 0.010 <0.0001** UAP 0.002 0.010 0.864 0 Sk illmix 0.001 0.122 0.251 0 Protocol 0.049 0.011 <0.0001** Time Invariant Organizational Variables 1 Constant ( 0.00 ) 2 Constant 1.593 0.152 <0.0001** Morse 0.114 0.225 0.611 0 Schmid 1.88 0 0.906 0.038 0 Hendrich 0.120 0.318 0.707 0 3 Constant 2.102 0.146 <0.0001** Morse 0.180 0.217 0.407 0

PAGE 199

199 Table B 8 Continued Group Parameter Estimate Standard Error Prob>|T| Schmid 1.800 0.217 0.045* Hendric h 0.290 0.306 0.343 4 Constant 1.508 0.156 <0.0001** Morse 0.174 0.233 0.454 Schmid 1.49 0 0.929 0.108 Hendrich 0.068 0.335 0.839 *Denotes significance at p<0.05 level **Denotes significance at p<0 .01 level

PAGE 200

200 Table B 9 Unadjusted multiple imputation model 1: Poisson maximum likelihood estimates of three hospital fall rate linear trajectories (N=1,592) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 2.855 0.237 <0.0001** Linear 0.0001 0.00001 <0.0001** 2 Intercept 1.843 0.121 <0.0001** Linear 0.00003 0.00001 <0.0001** 3 Intercept 2.136 0.171 <0.0001** Linear 0.00003 0.00001 0.004 0 ** Group Membership 1 21.95% 1.337 <0.0001** 2 55.60% 1.512 <0.0001** 3 22.45% 1.362 <0.0001** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 201

201 Table B 10 Unadjusted multiple imputation model 2: P oisson maximum likelihood estimates of three hospital fall rate linear trajectories (N=1,592) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 2.854 0.237 <0.0001** Linear 0.0001 0.00001 <0.0001** 2 Intercept 1 .892 0.121 <0.0001** Linear 0.00003 0.00001 <0.0001** 3 Intercept 2.026 0.177 <0.0001** Linear 0.00002 0.00001 0.040 0 Group Membership 1 22.71% 1.404 <0.0001** 2 55 .73% 1.518 <0.0001** 3 20.56% 1.330 <0.0001** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 202

202 Table B 11 Unadjusted multiple imputation model 3: Poisson maximum likelihood estimates of three hospital fall rate linear trajectories (N=1,592) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 2.609 0.228 <0.0001** Linear 0.0001 0.00001 <0.0001** 2 Intercept 1.908 0.121 <0.0001** Linear 0.00003 0.00001 <0.0001** 3 Intercept 2.154 0.175 <0.0001** Linear 0.00003 0.00001 0.040 0 Group Membership 1 22.96% 1.383 <0.0001** 2 55.83% 1.523 <0.0001** 3 21.21% 1.362 <0.0001** *D enotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 203

203 Table B 12 Unadjusted multiple imputation model 4: Poisson maximum likelihood estimates of three hospital fall rate linear trajectories (N=1,592) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 2.942 0.233 <0.0001** Linear 0.0001 0.00001 <0.0001** 2 Intercept 1.780 0.121 <0.0001** Linear 0.00003 0.00001 <0.0001** 3 Intercept 2.158 0.171 <0.0001** Linear 0.00003 0.00001 0.002 0 ** Group Membership 1 22.76% 1.380 <0.0001** 2 55.51% 1.519 <0.0001** 3 21.72% 1.330 <0.0001** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 204

204 Table B 13 Unadjusted multiple imputation model 5: Poisson maximum likelihood estimates of three hospital fall rate linear trajectories (N=1,592) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 2.725 0.230 <0.0001** Linea r 0.0001 0.00001 <0.0001** 2 Intercept 1.852 0.121 <0.0001** Linear 0.00003 0.00001 <0.0001** 3 Intercept 2.133 0.174 <0.0001** Linear 0.00003 0.00001 0.005 0 ** Group Membership 1 22.81% 1.357 <0.0001** 2 55.76% 1.521 <0.0001** 3 21.43% 1.353 <0.0001** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 205

205 Table B 14 Unadjusted quarterly hospital fall rate model with P oisson maximum likelihood estimates of three linear trajectories (N=1,592) Group Parameter Estimate Standard Error Prob>|T| 1 Intercept 1.701 0.925 0.066 0 Linear 0.0006 0.0001 0.261 0 2 Intercept 1.734 0.195 <0.0001** Linear 0.00003 0.00001 0.010 0 3 Intercept 2.433 0.275 0.000 0 ** Linear 0.00005 0.00002 0.002 0 ** Group Membership 1 8.02% 2.100 <0.0001** 2 67.14% 2.106 <0.0001** 3 24.84% 2.900 <0.0001** *Denotes significance at p<0.05 level **Denotes significance at p<0.01 level

PAGE 206

206 APPENDIX C SENSITIVITY ANALYSIS MODELS FOR SPECIFIC AIMS 2 AND 3 Table C 1 Generalized estimating equations (GEE) predicting membership in to the level (N=244,367) Variable Odds Ratios Std Error Confidence Interval Pr>|Z| Intercept 0.037 0.903 (0.006 0.218) <0.001 ** Nursing Factors RN H PPD 0.906 0.070 (0.790 1.038) 0.154 LPN HPPD 1.500 0.103 (1.226 1.835) <0.001 ** UAP HPPD 1.044 0.114 (0.835 1.304) 0.706 RN Skill Mix 1.023 0.013 (0.997 1.050) 0.084 Organizational Factors Magnet 0.661 0.116 (0.527 0.829) <0.001 ** Teaching 1.318 0.153 (0.977 1.777) 0.071 Metropolitan 0.835 0.242 (0.520 1.340) 0.454 Northeast a 0.963 0.214 (0.634 1.485) 0.861 South a 0.905 0.191 (0.623 1.315) 0.601 Midwest a 1.347 0.227 (0.863 2.101) 0.189 Bed100 b 1.865 0.282 (1.074 3.241) 0.027 Bed100to199 b 2.298 0.222 (1.485 3.551) <0.001 ** Bed200to299 b 1.725 0.230 (1.099 2.707) 0.018 Bed300to399 b 1.479 0.216 (0.968 2.259) 0.071 Bed4 00to499 b 1.898 0.293 (1.069 3.370) 0.029 *Denotes significance at p<0.05 level *Denotes significance at p<0.01 level a = Reference group for Region is West b = Reference group for Bed Size is >500 beds

PAGE 207

207 Table C 2 Generalized estimating equations (GEE) determining the impact of nurse staffing and organizational characteristics on patient falls at the hospital level (N=243,284) Variable Incident Rate Ratios Std Error Confidence Interval Pr>|Z| Intercept 0.002 0.298 (0.001 0.003) <0.001 ** Nursing Factors RN HPPD 0.978 0.018 (0.944 1.012) 0.199 LPN HPPD 1.107 0.037 (1.029 1.191) 0.006 ** UAP HPPD 1.061 0.035 (0.991 1.136) 0.089 RN Skill Mix 1.009 0.004 (1.001 1.018 0.034 Organizational Factors Magnet 0.943 0.017 (0.912 0.976) 0.003 Teaching 1.033 0.018 (0.997 1.070) 0.074 Metropolitan 0.97 0 0.033 (0.910 1.034) 0.345 Northeast a 1.050 0.032 (0.986 1.117) 0.130 South a 1.044 0.029 (0.987 1.104) 0.130 Midwest a 1.140 0.030 (1.074 1.209) <0.001 ** Bed100 b 0.961 0.034 (0.900 1.026) 0.238 Bed100to199 b 1.026 0.028 (0.972 1.083) 0.358 Bed200to299 b 0.996 0.026 (0.946 1.049) 0.880 Bed300to399 b 0.978 0.028 (0.926 1.033) 0.429 Bed400to499 b 1.033 0.032 (0.970 1.0 99) 0.313 *Denotes significance at p<0.05 level **Denotes significance at p <0.01 level a = Reference group for Region is West b = Reference group for Bed Size is >500 beds

PAGE 208

208 APPENDIX D SAS PROCEDURE CODES /* Trajectories of hospitals without covariates */ proc traj data=damian.completehospwide outplot=op outstat=os out=outhos p; id idhosp; var fall1 fall54; indep date1 date54; model zip; ngroups 3 ; order 1 1 1 ; iorder 1 ; run ; % trajplot (op, os, 'Unadjusted Hospital Fall Rates vs. Months' 'Zero Inflated Poisson 'Fall Rates' 'Months' ); /* Trajectories of hospitals over time with covariates */ proc traj data=damian.completehospwide outplot=op outstat=os out=outhospcov; id idhosp; var fall1 fall54; indep date1 date54; model zip; ngroups 3 ; order 1 1 1 ; iorder 1 ; tcov rn1 rn54 lpn1 lpn54 uap1 uap54 skillmix1 skillmix5 4 protocol1 protocol54 ; risk bed100 bed100to199 bed200to299 bed300to399 bed400to499 msa magnet teaching1 northeast south west; run ; % trajplot (op, os, 'Adjusted Hospital Fall Rates vs. Months' 'Zero Inflated Poisson 'Fall Rates' 'Months' ); /* medical fall trajectories without covariates*/ proc traj data=damian.medicalwide outplot=op outstat=os out=outmedical; id idunit; var medical1 medical54; indep date1 date54; model zip; ngroups 3 ; order 1 1 1 ; iorder 1 ; run ; % trajplot (op, os, 'Unadju sted Medical Fall Rates vs. Months' 'Zero Inflated Poisson 'Fall Rates' 'Months' ); /*surgical fall trajectories without covariates*/ proc traj data=damian.surgicalwide outplot=op outstat=os out=outsurgical; id idunit; var surgical1 surgical54; indep date1 date54; model zip; ngroups 2 ; order 1 1 ; iorder 1 ; ; run ; % trajplot (op, os, 'Unadjusted Surgical Fall Rates vs. Months' 'Zero Inflated Poisson 'Fall Rates' 'Months' ); /* medical surgical fall trajectories without covariates*/ proc traj data =damian.medsurgwide outplot=op outstat=os out=outmedsurg; id idunit; var medsurg1 medsurg54; indep date1 date54; model zip; ngroups 4 ; order 1 1 1 1 ; iorder 1 ; run ; % trajplot (op, os, 'Unadjusted Med Surg Fall Rates vs. Months' 'Zero Inflated Poisson Fall Rates' 'Months' );

PAGE 209

209 /*medical falls model with covariates */ proc traj data=damian.medicalwide outplot=op outstat=os out=outsurgicalad; id idunit; var medical1 medical54; indep date1 date54; model zip; ngroups 4 ; order 1 1 1 1 ; iorder 1 ; tcov rn1 rn 54 lpn1 lpn54 uap1 uap54 skillmix1 skillmix54 protocol1 protocol54; risk morse schmid hendrich; run ; % trajplot (op, os, 'Adjusted Medical Unit Fall Rates vs. Months' 'Zero Inflated Poisson 'Fall Rates' 'Months' ); /*surgical falls model with covari ates */ proc traj data=damian.surgicalwide outplot=op outstat=os out=outsurgicalad; id idunit; var surgical1 surgical54; indep date1 date54; model zip; ngroups 2 ; order 1 1 ; iorder 1 ; tcov rn1 rn54 lpn1 lpn54 uap1 uap54 skillmix1 skillmix54 protocol1 pro tocol54; risk morse schmid hendrich; run ; % trajplot (op, os, 'Adjusted Surgical Unit Fall Rates vs. Months' 'Zero Inflated Poisson 'Fall Rates' 'Months' ); /*medical surgical falls model with covariates */ proc traj data=damian.medsurgwide outplot =op outstat=os out=outmedsurgad; id idunit; var medsurg1 medsurg54; indep date1 date54; model zip; ngroups 4 ; order 1 1 1 1 ; iorder 1 ; tcov rn1 rn54 lpn1 lpn54 uap1 uap54 skillmix1 skillmix54 protocol1 protocol54; risk morse schmid hendrich; run ; % trajpl ot (op, os, 'Adjusted Med Surg Unit Fall Rates vs. Months' 'Zero Inflated Poisson 'Fall Rates' 'Months' ); /* Hospital unit group membership*/ proc genmod data =damian.gee3vsall_2 descending or der =data; class grp3 idhosp xdesignationfid datem; model grp3 = rnhppd lpnhppd uaphppd skillmixrn10 preprotocol magnet teaching1 msa northeast south midwest bed100 bed100to199 bed200to299 bed300to399 bed400to499 date medicalsurgical medicalunit / d =binomi al link =logit type3; repeated subject = idhosp(xdesignationfid) / type =un; run ; /*Hospital unit level GEE analysis on total fall rate */ proc genmod data =damian.geeanalysis3 order =data ; class idhosp xdesignationfid datem; model fallscount = rnhppd lpnhppd uaphppd skillmixrn10 preprotocol magnet teaching1 msa northeast south midwest bed100 bed100to199 bed200to299 bed300to399 bed400to499 medicalsurgical medicalunit date

PAGE 210

210 / offset =logdenom d =negbin link =log type3; repeated subject = xdesignationfid(idho sp) / type =ar; run ; /* Hospital group membership */ proc genmod data =damian.gee3vsall_2 descending order =data; class grp3 idhosp xdesignationfid datem; model grp3 = hosprnhppd hosplpnhppd hos puaphppd hospskillmixrn10 hosp_protocol magnet teaching1 msa northeast south midwest bed100 bed100to199 bed200to299 bed300to399 bed400to499 date / d =binomial link =logit; repeated subject = idhosp / type =un; run ; /* Hospital level GEE analysis of impac t on total fall rates */ proc genmod data =damian.geeanalysis4 order =data; class idhosp datem; model sum_fallscount = hosprnhppd hosplpnhppd hospuaphppd hospskillmixrn10 hosp_protocol magnet teaching1 msa northeast south midwest bed100 bed100to199 bed200to2 99 bed300to399 bed400to499 date / offset =logdenom d =negbin link =log; repeated subject =idhosp / type =ar ; run ;

PAGE 211

211 LIST OF REFERENCES Abraham, S. (2011). Fall prevention conceptual framework. The Health Care Manager 30(2), 179 184. Agency for Healthcare Research and Quality (AHRQ) (2009). Healthcare cost and utilization project state inpatient databases. Rockville, MD: Agency for Healthcare Research and Quality. Retrieved from: http://hcupnet.ahrq.gov/HCUPnet.jsp Agostini, J., Baker, D & Bog ardus, S (2001). Chapter 26. Prevention of falls in hospitalized and institutionalized older people. In: Making health care s afer: A c ritical analysis of patient safety p ractices. Washington D.C.: The Agency for Healthcare Resea rch and Quality. Aiken, L., Soch alski, J., & Lake, E. (1997). Studying outcomes of organizational change in health services. Medical Care 35(11), NS6 NS18. Aiken, L. & Patrician, P. (2000). Measuring organizational traits of hospitals: The revised nursing work index. Nursing Resear ch, 49(3), 146 153. Aiken, L., Clarke, S., & Sloane, D. (2002). Hospital staffing, organization, and quality of care: Cross national findings. Nurs ing Outlook 50, 187 194. Aiken, L., Clarke, S., Sloane, D., Sochalski, J., & Silber, J. (2002). Hospital nurse staffing and patient mortality, nurse burnout, and job dissatisfaction. Journal of the American Medical Association 288(16), 1987 1993. Aiken, L., Clarke, S., Cheung, R., Sloane, D., & Silber, J. (2003). Educational levels of hospital nurses and surgical patient mortality. Journal of the American Medical Association, 290(12), 1617 1623. Aiken, L., Havens, D. & Sloane, D. (2009). The Magnet nursing services recognition program: A comparison of two groups of Magnet hospitals Journal of Nursing A dministration 100(3) 26 35. Aiken, L., Sloane, D., Cimiotti, J., Clarke, S., Flynn, L., Seago, J., Spetz, J., & Smith, H. (2010). Implications of the California nurse staffing mandate for other states. Health Services Research, 45(4), 904 921. Aiken, L., Cimiotti, J., Sloane, D., Smith, H., Flynn, L. & Neff, D. (2011). Effects of nurse staffing and nurse education on patient deaths in hospitals with different nurse work environments Medical Care 49(12), 1047 1053. Akaike, H. (1974). A new look at the statistical model indentification. IEEE Transactions on Automatic Control, 19, 716 722

PAGE 212

212 American Nurses Association (1997). RN staffing, length of stay and patient outcomes Washington, D C : America n Nurses Publishing. American Nurses Association (1996). Nursing quality indicators : Definitions and implications Washington, DC: American Nurses Publishing. American Nurses Association (ANA) (2005). National database for nursing quality indicator s: Guidelines for data collection and submi ssion on quarterly indicators V ersion 5.0. American N urses Credentialing Center (20 08 ). Overview of ANCC Magnet Recognition Program. Retrieved from: http://www.nursecredentialing .org/magnet Andruff, H., Carr aro N., Thompson, A., Gaudreau, P., & Louvet, B. (2009). Latent class growth modeling: A tutorial. Tutorials in Quantitative Methods for Psychology, 5(1), 11 24. Bakarish, A., McMillan, V. & Prosser, R. (1997). The effect of a nursing intervention on t he i ncidence of older patient falls. Aust ralian J ournal of Adv anced Nurs ing 15(1), 26 31. Ballinger, G. (2004). Using generalized estimating equations for longitudinal data analysis. Organizational Research Methods, 7 127 150. Bates, D., Pruess, K., S ouney, P. & Platt, R. (1995). Serious falls in hospitalized patients: Correlates and resource utilization. The American Journal of Medicine 99, 137 143. Becker, E., & Foster, R. (1988). Organizational determinants of nurse staffing patterns. Nursing Ec onomic, 6 71 75. Bishop, C., Weinberg, D., Leutz, W., Dossa, A., Pfefferie, S., & Zincavage, R. (2008). fa ctors and impact on resident well being Gerontologist, 48(1), 36 45. Bl egen, M., Goode, C., & Reed, L. (1998). Nurse staffing and patient outcomes. Nursing Research, 47, 43 50. Blegen, M. & Vaughn, T. (1998). A multisite study of nurse staffing and patient occurrences. Nursing Economics 16(4), 196 203. Blegen, M., Vaugh n, T., & Vojir, C. (2008). Nurse staffing levels: Impact of organizational characteristics and registered nurse supply Health Services Research, 43(1), 154 173.

PAGE 213

213 Bloom, J., Alexander, J., & Nichols, B. (1997). Nurse staffing patterns and hospital effic iency in the United States Social Sciences in Medicine, 44, 147 155. Bolton, L., Jones, D., Aydin, C., Donaldson, N., Brown, D., Lowe, M., McFarland, P., & J ournal of Nurs ing Scholars h ip 33 179 184. Breckenridge Sproat, S., Joh antgen, M. & Patrician, P. (2012 ). Influence of unit level staffing on medication errors and falls in military hospitals. West ern J ournal of Nurs ing Res earch 34(4), 455 474 Burchinal, M. & Appelbaum, M (1 991). Estimating individual developmental f unctions: Methods and their a ssumptions Child Development, 62(1), 23 43. Burnes Bolton, L., Aydin, C., Donaldson, N., Brown, D., Sandhu, M., Fridman, M. & Aronow, H. (2007). Mandating nurse staffing ratios in California: A comparison of staffing and nursing sensitive outcomes pre and post regulation. Policy, Politics & Nursing Practice 8(4), 238 250. California State Department of Health Services. (2003). Final statement of reasons, R 37 01. Retrieved fro m : http://www.dhs.ca.gov/Inc/pubnotice/NTPR/R 37 01_FSOR.pdf. Centers for Dise ase Control and Prevention. (2012). National Center for Injury Control and Prevention w eb based injury statistics query and reporting system (WISQRS) Atl anta: CDC Retrieved from : http://www.cdc.gov/injury/wisqars/index.html Centers for Disease Control and Prevention (2012). Costs of falls among older a dults Retrieved from: http://www.cdc.gov/HomeandRe creationalSafety/falls/data/cost estimates.html Centers for Me dicare & Medicaid Services. (2008). Medicare program: Changes to the hospital inpatient prospective payment systems and fiscal year 2009 rates Federal Register, 73(16), 48433 49084. Charlso n, M., Pompei, P., Ales, K., & MacKenzie, C. (1987). A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. Journal of Chronic Disease, 40(5), 373 383. Cho, S. (2001). Nurse staffing and adverse patient ou tcomes: A systems approach. Nurs ing Outlook 49 78 85. Cho, S., Ketefian, S., Barkauskas, V. & Smith, D. (2003). The effects of nurse staffing on adverse events, morbidity, mortality, and medical costs. Nursing Research 52(2), 71 79.

PAGE 214

214 Clancy, C. (2009 ). Ten years after To Err is Human. American Journal of Medical Quality, 24(6), 525 528. Clyburn, T., & Heydemann, J. (2011). Fall prevention in the elderly: Analysis and comprehensive review of methods used in the hospital and in the home. Journal of t he American Academy of Orthopaedic Surgeons, 19 402 419. Clark, S. & Donaldson, N (2008). Nurse staffing and patient care quality and safety. In: Patient safety and quality: An evidence based handbook for nurses. AHRQ Publication No. 08 0043. Rockvill e, MD: Agency for Healthcare Research and Quality. Coussement, J., De Paepe, L., Schwendimann, R., Denhaerynck, K., Dejaeger, E. & Milisen, K. (2008). Interventions for preventing falls in acute and chronic care hospitals: A systematic review and meta analysis. J ournal of the Am erican Geriatr ics Soc iety 56 29 36. Cramer, M., Jones, K., & Hertzog, M. (2011). Nurse staffing in critical access hospitals: Structural factors linked to quality care. J ournal of Nurs ing Care Qual ity 26(4) 335 343. Cummi ng, R., Sherrington, C., Lord, S., Simpson, J., Vogler, C., Camerson, I., & Naganathan, V. (2008). Cluster randomized trial of a targeted multifactorial intervention to prevent falls among older people in hospital. B ritish M edical J ournal 1 6. Currie, L ., Mellino, L., Cimino, J., & Bakken, S. (2004). Development and representation of a fall injury risk assessment instrument in a clinical information system. Stud y of Health Technol ogy Info r m ation 107, 721 725. Currie, L. (2008). Fall an d injury prevent ion. In: Patient safety and quality: An evidence based handbook for nurses. AHRQ Publication No. 08 0043. Rockville, MD: Agency for Healthcare Research and Quality. Dempsey, J. (2004). Falls prevention revisited: A call for a new approach. Journal of C linical Nursing 134, 79 485. Dennis, J., Gay, D. & Welsch, R. (1981). An adaptive nonlinear least square algorithm. ACM Transactions on Mathematical Software, 7(3), 348 368. Dobson, A., & Barnett, A. (2008). Introduction to generalized linear m odels (3 rd ed.). Boca Raton, FL: CRC Press.

PAGE 215

215 Dodge, H., Shen, C. & Ganguli, M. (2008). Application of the pattern mixture patent trajectory model in an epidemiological study with non ignorable missingness. Journal of Data Science, 6, 231 246. Donabedian, A. (1966). Evaluating the quality of medical care. Milban k Memorial Fund Quarterly 44(1), 166 203. Donabedian, A. (1988). The quality of care: How can it be assessed? Journal of the American Medical Association 260 1743 1748. Donabedian, A. (1992). Th e role of outcomes in quality assessment and assurance. Quality Review Bulletin, 11 356 360. Donaldson, N., Bolton, L., Aydin, C., Brown, D., Elashoff, J. & Sandhu, M. (2005). patient ratios on unit level nurse st affing and patient outcomes. Policy, Politics & Nursing Practice 6(3), 198 210. Donley, R., Flaherty, M.J., Sarsfield, E., Taylor, L., Maloni, H., & Flanagan, E. (2003). What does the Nurse Reinvestment Act mean to you? Online Journal of Issues in Nursi ng 8(1). Pallas, L., Aisbett, C., Roche, M., King, M., & Aisbett, K. (2011). Nurse staffing, nursing workload, the work environment and patient outcomes. Applied Nursing Research 24(4), 244 255. Duncan, T., & Duncan, S (2004). An introduction to latent growth curve modeling. Behavioral Therapy, 35, 333 363. Duncan T., Duncan S., & Strycker, L. (2006 ). An introduction to latent variable growth curve modeling: Concepts issues and applications (2nd ed.) Mahwah, NJ: Lawrence Erlbaum Associates Dunton, N., Gajewski, B., Taunton, R. & Moore, J. (2004). Nurse staffing and patient falls on acute care hospital units Nurs ing Outlook 42, 53 59. Dunton, N. & Schumann, M. (2005). Early evidence on California staffing ratios should be interpreted with caution. Policy, Politics & Nursing Practice, 6(4), 354 357. Dunton, N., Gajewski, B., Klaus, S. & Pierson, B. (2007). The relationship of nursing workforce chara cteristics to patient outcomes. The Online Journal of Is sues in Nursing 12(3). Enloe, M., Wells, T., Mahoney, J., Pak, M., Gangnon, R., Pellino, T., Hughes, S. & Leahy Gross, K. (2005). Falls in acute care: An academic medical center six year review J ournal of Patient Saf ety 1(4), 208 214.

PAGE 216

216 Evans, D., Hod gkinson, B., Lambert, L. & Wood, J. (2001). Falls risk factors in the hospital setting: A systematic review International Journal of Nursing Practice, 7, 38 45. Faes, M., Reelich, M., Joosten Weyn Banningh, L., Gier, M., Esselink, R., & Olde Rikkeert, M. (2010). Qualitative study on the impact of falling in frail older persons and family caregivers: Foundations for an intervention to prevent falls. Aging and Ment al Health, 14(7), 834 842. Fergus, S., Zimmerman, M., & Caldwell, C. (2007). Growth trajec tories of sexual risk behavior in adolescents and young adults. Journal of Public Health, 97(6), 1096 1101. Fiesta, J. (1998). Liability for falls. Nurs ing Manage ment 29 24 26. Fine, W. (1959). An analysis of 277 falls in hospital. Gerontological Clin ics 1, 292 300. Fischer, I., Krauss, M., Dunagan, W., Birge, S., Hitcho, E., Johnson, S., Constantinou, E. & Fraser, V. (2005). Patterns and predictors of inpatient falls and fall related injuries in a large academic hospital Infection Control and Ho spital Epidemiology 26, 822 827. Fitzmaurice, G., Laird, N., & Rotnitzky, A. (1993). Regression models for discrete longitudinal responses. Statistical Science, 8, 284 309. Fox, R., & Abrahamson, K. (2009). A critical examination of the US nursing shor tage: Contributing factors, public policy implications. Nursing Forum, 44(4), 235 244. Gardner, W., Mulvey, E. & Shaw, E. (1995). Regression analyses of counts and rates: Poisson overdispersed Poisson and negative binomial models Psychological Bullet in, 118, 392 404. Grillo Peck, A., & Risner, P. (1995). The effect of a partnership model on quality and length of stay. Nursing Economics, 13(6) 367 374. Grubel, F. (1959). Falls: A principal patient incident. Hosp ital Manag ement 88 37 38. Halfon, P., Eggli, Y., Melle, G., & Vagnair, A. (2001). Risk of falls for hospitalized patients: A predictive model based on routinely available data Journal of Clinical Epidemiology, 54, 1258 1266. Hanley, J., Negassa, A., Edwardes, M., & Forrester, J. (2003) Statistical analysis of correlated data using generalized estimating equations: An orientation. American Journal of Epidemiology, 157 364 375.

PAGE 217

217 Hardin, J., & Hil be, J. (2003). Generalized estimating equations Boca Raton, FL: CRC Press. Harless, D., & Mark, B. (2006). Addressing measurement error bias in nurse staffing research. Health Services Research, 41 2006 2024. Hart, S., Bergquist, S., Gajewski, B., & Dunton, N. (2006). Reliability testing of the National Database of Nursing Quality Indic ators pressure ulcer indicator. Journal of Nursing Care Quality, 21, 256 265. Havens, D. & Aiken, L. (1999). Shaping systems to promote desired outcomes: The M agnet hospital model. J ournal of Nurs ing Adm inistration 29(2), 14 20. Havens, D. & Johnsto n, M. (2004). Achieving Magnet hospital recognition: Chief nursing executives and Magnet coordinators tell their stories. JONA, 34(12), 579 588. Hedeker, D. & Gibbons, R. (1997). Applications of random effects pattern mixture models for missing data in longitudinal studies. Psychological Methods, 2 64 78. Heinrich, S., Rapp, K., Rissman, U., Becker, C. & Konig, H. (2009). Cost of falls in old age: A systematic review Osteop o ros is Int ernational 21, 891 902. Helgeson, V., Snyder, P. & Seltman, H. (2004). Psychological and physical adjustment to breast cancer over 4 years: Identifying distinct trajectories of change. Health Psychology, 23 3 15. Henselmans, I., Helgeson, V., Seltman, H., Vries, J., Sanderman, R. & Ranchor, A. (2010). Identificat ion an d prediction of distress trajectories in the first year after a breast cancer diagnosis. Health Psychology, 29(2) 160 168. Hernandez, M. & Miller, J. (1986). How to reduce falls. Geriatr ic Nurs ing 7, 97 102. Hill, K., Vu, M., & Walsh, W. (2007) Falls in the acute hospital setting impact on resource utilization. Aust ralian Health Rev iew 31(3), 471 477. Hitcho, E., Kr auss, M., Birge, S., Dunagon, W ., Fischer, I., Johnson, S., Nast, P., Costantinou, E. & Fraser, V. (2004). Characteristics and circumstances of falls in a hospital setting. J ournal of Gen eral Intern al Med icine 19, 732 739. Hodge, M., Romano, P., Harvey, D., Samuels, S., Olson, V., Sauve, M., & Kravitz, R. (2004). Licensed caregiver characteristics and staffing in California acute care hospital units. Journal of Nursing Administration, 34, 125 133. Hogue, C. (1982). Injury in late life: I. Epidemiology. II. Prevention. J ournal of the Am erican Geriatr ics Soc iety 30, 183 190.

PAGE 218

218 Hyer, K., Thomas, K., Branch, L., Harman, J., Joh nson, C., & Weech Maldonado, R. (2011). The influence of nurse staffing levels on quality of care in nursing homes. The Gerontologist, 1 7. Inouye, S., Brown, C., & Tinetti, M. (2009). Medicare nonpayment, hospital falls, and unintended consequences. N ew Engl and J ournal of Med icine 360(23), 2390 2393. Institute of Medicine (1996). Nursing staff in hospitals and nursing homes: Is it adequate? Washington, D.C.: National Academy Press. Institute of Medicine (1999). To err is human: Building a safer health system Washington, D.C.: National Acade my Press. Institute of Medicine (2001). Crossing the quality chasm Washington, DC: National Academy Press. Institute of Medicine (2004 ). Keeping patients safe: Transforming the work environment of nur ses Washington, DC: National Academy Press. health care. Nursing Economics, 16 58 64, 87. Jiang, H., Stocks, C. & Wong, C. (2006). Disparities between two common data sources on hospital nurse staffing. Journal of Nursing Scholarship 38(2), 187 193. Jones, B., Nagin, D. & Roeder, K. (2001). A SAS procedure based on mixture models for estimating developmental trajectories. Sociological Methods & Research, 29 (3), 374 393. Jung, T., & Wickrama, K. (2008). An introduction to latent class growth analysis and growth mixture modeling. Social and Personality Psychology, 2(1) 302 317. Kamel, H., Iqbal, M., Mogallapu, R., Maas D., & Hoffman R. (2003). Time to a mbulation after hip fracture surgery: Relation to hospitalization outcomes. J Gerontol A Biol Sci Med Sci, 58(11), M1042 M1045. Kane, R., Shamliyan, T., Mueller, C., Duval, S. & Wilt, T. (2007). The association of registered nurse staffin g levels and pa tient outcomes: A systematic review and meta analysis. Medical Care 4 5(12), 1195 1204. Kass, R. & Raftery, A. (1995). Bayes factors. Journal of the American Statistical Association, 90, 773 795.

PAGE 219

219 Kelly, L. McHugh, M. & Aiken, L. (2011). Nurse outcome s in Magnet and Non Magnet hospitals. Journal of Nursing Administration 41(10) 428 433. Kovner, C., Jones, C., Chunliu, Z., Gergen, P., & Basu, J. (2002). Nurse staffing and postsurgical adverse events: An analysis of administrative data from a sample of US hospitals, 1990 1996. Health Services Research 37(3), 611 629. Krauss, M., Evanoff, B., Hitcho, E., Ngugi, K., Dunagan, W., Fischer, I., Birge, S., Johnson, S., Constantinou, E., & Fraser, V. (2005). A case control study of patient, medication a nd care related risk factors for inpatient falls J ournal of Gen eral Intern al Med icine 20, 116 122. Krauss, M., Nguyen, S., Dunagan, C., Birge, S., Constantinou, E., Johnson, S., Caleca, B ., & Fraser, V. (2007). Cir c umstances of patient falls an d injuri es in 9 hospitals in a M idwestern healthcare system. Infection Control and Hospital Epidemiology 28(3), 544 550 Krishna, K., & Van Cleave, R. (1983). Decrease in the incidence of patient falls in a geriatric hospital after educational programs J ournal of Am erican Geriatr ics Soc iety 31 187. Kustaborder, M., & Rigner, M. (1983). Interventions for safety Journal of Gerontological Nursing, 9(3) 159 182. Laird, N. (1988). Missing data in longitudinal studies. Statistics in Medicine, 7, 305 315. Lak e, E & Cheung, R (2006). Are patient falls and pressure ulcers sensitive to nurse staffing? Western Journal of Nursing Research 28(6), 654 677. Lake, E., Shang, J., Klaus, S. & Dunton, N. (2010). Patient falls: Association with hospital magnet stat us and nursing unit staffing. Research in Nurs ing and Health, 33(5), 413 425. Lam bert, D. (1992). Zero inflated Poisson regressions with an application in manufacturing. Technometrics, 34, 1 13. Lankshear, A., Sheldon, T., & Maynard, A. (2005). Nurse st affing and healthcare outcomes: A systematic review of the international research evidence. Advances in Nursing Science, 28, 163 174. Lang, T., Hodge, M., Olson, V., Romano, P. & Kravitz, R. (2004). Nurse patient ratios: A systematic review on the effe cts of nurse staffing on patient, nurse employee, and hospital outcomes. Journal of Nursing Administration 34(7/8) 326 337.

PAGE 220

220 Laptook, R., Klein, D. & Dougherty, L. (2006). Ten year stability of depressive personality disorder in depressed outpatients The American Journal of Psychiatry, 163(5), 865 871. Lauritzen, J (1996). Hip fractures: Incidence, risk factors, energy absorption, and prevention. Bone, 18 65S 75S. Lee, G., Kleinman, K., Soumerai, S., Tse, A., Cole, D., Fridkin, S., Horan, T., Platt, R., Gay, C., Kassler, W., Goldmann, D., Jernigan, J., & Jha, A. (2012). Effect of nonpayment for preventable infections in U.S. hospitals. The New England Journal of Medicine 367(15), 1428 1437. Liang, K., & Zeger, S. (1986). Longitudinal data an alysis using generalized linear models. Biometrika 73 13 22. Liddle, J., & Gilleard, C. (1994). The emotional consequences of falls for older people and their families. Clin ical Rehabil itation 9, 110 114. Little, R. & Rubin, D. (1987). Statistical a nalysis with missing d ata. New York, NY: John Wiley & Sons. Lubke, G. & Muthn, B. (2005). Investigating population heterogeneity with factor mixture models. Psychological Methods, 10(1), 21 39. Lucero, R., Lake, E., & Aiken, L. (2010), Nursing care quality and adverse events in US hospitals. Journal of Clinical Nursing, 19, 2185 2195. Mahoney, J (1998). Immobility and falls. Clin ical Geriatr ic Med icine 14 699 726. Manojlovich, M., Sidani, S., Covell, C., & Antonakos, C. (2011). Nurse Dose: Li nking staffing variables to adverse patient outcomes Nursing Research, 60(4), 214 220. Mark, B., Sayler, J., & Wan, T. (2000). Market, hospital, and nursing unit characteristics as predictors of nursing unit skill mix: A contextual analysis. Journal of Nursing Administration, 30, 553 560. Mark B., Sayler, J., & Wan, T. (2003). Professional nursing practice: impact on organizational and patient outcomes J ournal of Nurs ing Adm inistration 33, 224 233 Mark, B., Hughes, L., & Jones, C. (2004). The role of theory in improving patient safety and quality health care. Nurs ing Outlook 52 11 16. Mark, B., & Harless, D. (2007). Nurse staffing, mortality, and length of stay in for profit and not for profit hospitals. Inquiry, 44, 167 186.

PAGE 221

221 Mark, B., Hughes, L., Belyea, M., Bacon, C., Chang, Y., & Jones, C. (2008). Exploring organizational context and structure as predictors of medication errors and patient falls. Journal of Patient Safety 4 66. Mahoney, J (1998). Immobility and falls. Clin ical Geriatr ic Med icine 14, 699 726. The Health Care Manager, 27(4), 338 349. Mayo, N., Gloutney, L. & Levy, A. (1994). A ra ndomized trial of identification bracelets to prevent falls among patients in rehabilitation hospital. Arch ives of Phys ical Med icine and Rehabil itation 75(12), 1302 1308. McAvoy, S., Skinner, T. & Hines, M. (1996). Clinical methods: Fall risk assessmen t tool. Applied Nursing Research 9 213 218. McCallagh, P., & Nelder, J. (1989 ). Generalized linear models (2 nd ed.). London: Chapman and Hall. McCollam, M (1995). Evaluation and implementation of a research based falls assessment innovation. Nursin g Clinics of North America 30, 507 514. McGillis Hall, L., Doran, D. & Pink, G. (2004). Nurse staffing models, nursing hours and patient safety outcomes. Journal of Nursing Administration, 34(1), 41 45. Mick, S., & Mark, B. (2005). The contribution of organizational theory to nursing health services research. Nurs ing Outlook, 53 317 323. Montalvo, I. (2007). The National Database of Nursi ng Quality Indicators TM (NDNQI ). OJIN: The Online Journal of Issues in Nursing, 12(3), Manuscript 3. Moore, T. Martin, J., & Stonehouse, J. (1996). Predicting falls: Risk assessment tool versus clinical judgment. Perspectives 20 8 11. Morgan, V., Mathison, J., Rice, J., & Cemmer, D. (1985). Hospitals falls: A persistent problem. Am erican J ournal of Public He alth, 75(7), 775 777. Morse, J., Prowse, M., Morrow N., & Federspeil, G. (1985). A retrospective analysis of patient falls. Can adian J ournal of Public Health, 76 116 118. Morse, J., Tylko, S., & Dixon, H. (1987). Characteristics of fall prone patient. The Gerontologist, 27(4 ), 516 522. Morse, J. (1997). Preventing patient falls Thousand Oaks, CA: SAGE Publications, Inc.

PAGE 222

222 Morse, J. (2002). Enhancing the safety of hospitalization by reducing patient falls. Am erican J ournal of Infect ion Control, 30, 37 6 380. Muthn, B. (1997). Latent variable modeling of longitudinal and multilevel data. In : Sociological m ethodology (pp. 453 480). Boston MA : Blackwell. Muthn, L., & Muthn, B. (1998). Los Angeles: Muthn & Muthn. Nagin, D. (1 999). Analyzing developmental trajectories: A semi parametric, group based approach. Psychological Methods, 4 139 177. Nagin, D. (2005). Group based modeling of development Cambridge, MA: Harvard University Press. Nagin, D. & Tremblay, R. (1999). and hyperactivity on the path to physically violent and non violent juvenile delinquency. Child Development, 70, 1181 1196. National Quality Forum. (2004). National voluntary consensus standards for nursing sensitive care: An initial performance measure set Washington D C : National Quality Forum. National Quality Forum. (2009). NQF endorsed standards. Retrieved from : http://www.qualityforum.org/Measures_List.aspx. National Database for Nu rsin g Quality Indicators. (2007). Guidelines for data collection and submission on quarterly indicators. Kan sas City, KS. Needleman, J., Buerhaus, P., Mattke, S., Stewart, M., & Zelevinsky, K. (2002). Nurse staffing levels and the quality of care in hospi tals. New England Journal of Medicine 346, 1715 1722. Needleman, J., Buerhaus, P., Pankratz, V., Leibson, C., Stevens, S. & Harris, M. (2011). Nurse staffing and inpatient hospital mortality. New England Journal of Medicine 364(11), 1037 1045. Nyberg L., Gustafson, Y., Janson, A., Sandman, P., & Eriksson, S. (1997). Incidence of falls in three different types of geriatric care. A Swedish perspective study. Scand inavian J ournal of Soc ial Med icine 25(1) 8 13. Oliver, D., Hopper, A. & Seed, P. (20 00). Do hospital fall prevention programs work? A systematic review J ournal of the Am erican Geriatr ics Soc iety 48 1679 1689.

PAGE 223

223 Oliver, D., Daly, F., Martin, F., & McMurdo, M. (2004). Risk factors and risk assessment tools for falls in hospital in pa tients: A systematic review. Age and Ageing 33 122 130. Oliver, D., Healey, F. & Haines, T. (2010). Preventing falls and fall related injuries in hospitals. Clin ical Geriatr ic Med icine 26 645 692. Parry, S., Steen, N., Galloway, S., Kenny, R., & Bo nd, J. (2001). Falls and confidence related quality of lif e outcome measures in an older B ritish cohort. Postgrad Med ical J ournal 77(904), 103 108. Patrician, P., Loan, L., McCarthy, M., Fridman, M., Donaldson, N., Bingham, M. & Brosch, L. (2011). The association of shift level nurse staffing with adverse patient events. Journal of Nursing Administration 41(2), 64 70. Perell, K., Nelson, A., Goldman, R., Luther, S., Prieto Lewis, N. & Rubenstein, L. (2001). Fall risk assessment measures: An analyt ic review. Journals of Gerontology Series A: Biological Sciences and Medical Sciences 56(12), 761 766. Ratitch, B. (2011). Implementation of pattern mixture models using standard SAS /STAT procedures Paper SP04. Raudenbush, S. (2001). C omparing persona trajectories and drawing causal inferences from longitudinal data. Annual Review of Psychology, 52, 501 525. Reason, J. (1995). Understanding adverse events: Human factors. In: Vincent, C, editor. Clinical Risk Management. London: Br iti sh Med ical J ournal Publishing Group, 31 54. Richards, L. & Morse, J. M. (2006). methods Thousand Oaks, CA: Sage Publications, Inc. Rivers, P., Tsai K. & Munchus, G. (2005). The financial impacts of th e n ursing s hortage. Journal of Health Care Finance 31(3) 52 64. rule. N ew Engl and J ournal of Med icine 357(16), 1573 1575. Rubenstein, L., & Josephson, K. (2002). The epid emiology of falls and syncope. Clin ical Geriatr ic Med icine 18, 141 158 Rubenstein, L. (2005). Evidence Based Fall Prevention. Paper presented at the Evidence Based Strategies for Patient Falls and Wandering, Clearwater, FL.

PAGE 224

224 Rubin, D. (1987). Multip le imputation for nonresponse in s urveys New York NY : John Wiley & Sons. SAS Institute (2004). SAS System for Microsoft Windows (version 9.1) [computer software]. Cary, NC: SAS Institute, Inc. Schmiege, S., Meek, P., Bryan, A. & Petersen, H. (201 2). Latent variable mixture modeling: A flexible statistical approach for identifying and classifying heterogeneity. Nursing Research, 61(3), 204 212. Schwendimann, R. (1998). Frequency and circumstances of falls in acute care hospitals: A pilot study. Pflege 11 335 341 Schwendimann, R., Buhler, H., De Geest, S. & Milisen, K. (2006). Falls and consequent injuries in hospitalized patients: Effects of an interdisciplinary falls prevention program. BMC Health Services Research 6(69), 1 7. Schwendima nn, R., Buhler, H., De Geest, S. & Milisen, K. (2008). Characteristics of hospital inpatient falls across clinical departments. Gerontology, 54(6), 342 348. Scott, J., Sochalski, J. & Aiken, L. (1999). Review of Magnet hospital research: Findings and implications for professional nursing practice J ournal of Nurs ing Adm inistration 29(1), 9 19. Seago, J., Spetz, A., & Mitchell, S. (2004). Nurse staffing and hospital ownership in California. Journal of Nursing Administration, 34 831 852. Sehested, P ., & Severin Nielson, T. (1977). Falls by hospitalized elderly patients: Causes, prevention. Geriatrics, 32 101 109. Shekelle, P., Rubenstein, L., Maglione, M., Chang, J., Mojica, W., Morton, S., Suttorp, M., Roth, E., Rhodes, S., Wu, S., & Newberry, S (2003). Falls prevention interventions in the Medicare population. Baltimore, MD: U.S. Department of Health and Human Services, Health Care Financing Administration. Shi, L. & Singh, D. (2008). Delivering health care in America: A systems a pproach (4 th ed ). Sudbury, Massachusetts: Jones and Bartlett. Shorr, R., Guillen, M., Rosenblatt, L, Walker, K., Caudie, C & Kr itchevsky, S (2002). Restraint use, restraint orders and the risk of falls in hospitalized patients. J ournal of the Am erican Geriatr ic s Soc iety 50, 526 529. Shuldham, C. (2004). Commentary. Nursing skill mix and staffing. Journal of Nursing Management, 12, 385 387.

PAGE 225

225 Shuldham, C., Parkin, C., Firouzi, A., Roughton, M. & Lau Walker, M. (2009). The relationship between nurse staffing and patient outcomes: A case study. International Journal of Nursing Studies 46, 986 992. Simon, M., Yankovskyy, Y., Klaus, S., Gajewski, B., & Dunton, N. (2010). Midnight census revisited: Reliability of patient day measurements in US hospital units. Int ernational J ournal of Nurs ing Stud ies 48 56 61. Simon, M., Yankovskyy, Y., & Dunton, N. (2010). Solving the mystery of patient days and the midnight census Nurs ing Manage ment 41(2), 12 14 Sovie, M. & Jawad, A. (2001). Hospital restructuring and its impact on outcomes. J ournal of Nursing Administration 31(12), 588 600. Stevens, J., Corso, P ., Finkelstein, E., & Miller, T. (2006). The costs of fatal and non fatal falls among older adults. Inj ury Prev ention 12(5), 290 295. Thurston, G. (1957) Fatal hospital falls. B ritish M edical J ournal 16(51), 396 397. Tideiksaar, R. (1993). Falls in older person: Prevention and management in hospitals and nursing homes. Boulder CO : Tactilitics. Tideiksaar, R., Feiner, C., & Maby, J. (1993). Falls prev ention: The efficacy of a bed alarm system in an acute care setting. M oun t Sinai J ournal of Med icine 60(6), 522 527. Tutaurini, J., de Haan, R., & Limburg, M. (1993). Patient outcomes: Are they linked to registered nurse absenteeism, separation, or wor k load? Journal of Nursing Administration, 2(45), 48 55. Twisk, J. & de Vente, W. (2002). Attrition in longitudinal studies: How to deal with missing data. Journal of Clinical Epidemiology, 55(4), 329 337. Umberson, D., Williams, K., Powers, D. Liu, H. & Needham, B. (2006). You make me sick: Marital quality and health over the life course. Journal of Health and Social Behavior, 47(1), 1 16. Unruh, L. (2002). Trends in adverse events in hospitalized patients J ournal of Healthc are Qual ity 24 4 10. Unruh, L. (2003). Licensed nurse staffing and adverse events in hospitals. Medical Care 41(1), 142 152. Unruh, L (2008). Nurse staffing and patient, nurse, and financial outcomes. A merican J ournal of N ursing 108(1), 62 70.

PAGE 226

226 Unruh, L., & Wan, T. (2 004). A systems framework for evaluating nursing care quality in nursing homes. Journal of Medical Systems 28(2), 197 214. Unruh, L., & Zhang, N. (2012). Nurse staffing and patient safety in hospitals. Nursing Research, 61(1), 3 12. U.S. Department o f Health and Human Services. (2000). Healthy people 2010 with understanding and improving health and objectives for improving health ( 2 nd ed ) Washington D C : U.S. Government Printing Office. Van den Heede, K., Clarke, S., Sermeus, W., Vleugels, A., & Aiken, L. (2007). outcome literature. Journal of Nursing Scholarship 39 290 297. Vassall o, M., Vignaraja, R., Sharma, J., Briggs, A. & Allen, S. (2004). Predictors for fa lls among hospital inpatients with impaired mobility. J ournal of the R oyal Soc iety of Med icine 97(6), 266 269. Vellas, B., Wayne, W., Romero, L., Baumgartner, R., & Garry, P. (1997). Fear of falling and restriction of mobility in elderly fallers Age an d Ageing, 26 189 193. Vlahov, D., Myers, A. & Al Ibrahim, M (1990). Epidemiology of falls among patients in a rehabilitation hospital. Arch ives of Phys ical Med icine and Rehabil itation 71 8 12. Wa ng, L., Zhang, Z., McArdle, J., & Salthouse, T. (2008 ). Investigating ceiling effects in longitudinal data analysis. Multivariate Behavioral Research, 43, 476 496. West, E., Griffi th, W. & Iphofen, R. (2007). A historical perspective on the n ursing s hortage. MedSurg Nursing 16(2), 124 129. Wooldridge, J. (2006). Introductory econometrics: A modern a pproach ( 3 rd ed). Thom p son South Western. Whitman, G., Kim, Y., Davidson, L., Wolf, G. & Wang, S. (2002). The impact of staffing on patient outcomes across specialty units. J ournal of Nursing Administration 32(12), 633 639 Wishall, M. (2008). Nurse retention: One solution for resolving the nursing s hortage. The Kan as Nurse 83(5), 3 5. Wong, C., Recktenwald, A., Jones, M., Waterman, B., Bollini, M. & Dunagan, C. (2011). The cost of se rious fall related injuries at three Midwestern hospitals The Joint Commission Journal on Quality and Patient Safety 37(2), 81 87.

PAGE 227

227 Wunderlich, G. Sloan, F. & Davis, K. (1996). Nursing staff in hospitals and nursing homes: Is it adequate? Washington, D.C.: National Academy Press. Zuckerman, J. (1996). Hip fracture. N ew Engl and J ournal of Med icine 334(23), 1519 1525.

PAGE 228

228 BIOGRAPHICAL SKETCH Damian Everhart was born in Medina, Ohio to Charles Everhart and Deborah the health care system initially led him to become an emergency medical technician (EMT), working for the Alachua County Fire Rescue department during his undergraduate studies. After completion of a Bachelor of Science in Nursing in August of 2005 Damian continued his work as an emergency room nurse treating patients at a level II trauma center in Lakeland, FL. Following this work, Damian obtained a Master of Science in Management at the University of Florida in 2006 and then pursued a PhD in Health Serv ices Research at the University of Florida. His training during the PhD program focused on improving the financing, delivery, access and quality of health care in the United States including conducting studies on patient safety and quality of care in hosp ital organizations, which he will continue after his graduation in December of 2012.