<%BANNER%>

Patient-Reported Outcomes in Hematopoietic Stem Cell Transplant Survivors

Permanent Link: http://ufdc.ufl.edu/UFE0045286/00001

Material Information

Title: Patient-Reported Outcomes in Hematopoietic Stem Cell Transplant Survivors
Physical Description: 1 online resource (205 p.)
Language: english
Creator: Kenzik, Kelly M
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2013

Subjects

Subjects / Keywords: cancer -- hrqol -- outcomes -- patient -- reported -- survivor -- symptom
Epidemiology -- Dissertations, Academic -- UF
Genre: Epidemiology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Advancements in hematopoietic stem cell transplantation (HSCT) result in a growing number of long-term survivors.  Recent research has shifted attention from HSCT side-effects to patient-reported outcomes (PROs) including symptoms and health-related quality of life (HRQOL).  The three aims were to 1) refine the HSCT-specific physical symptom scale for long-term survivors based on the Functional Assessment of Cancer Therapy–Bone Marrow Transplant (FACT-BMT), 2) develop and test a conceptual framework of factors influencing PROs, and 3) examine population heterogeneity based on PTSS and PTG in HSCT survivors and the impact on HRQOL. This study is a secondary data analysis of PROs measures collected from 662 subjects in the RO1 parent study, “Quality of Life and Relationships in BMT Survivors”; PI: Dr. John Wingard.  First, the FACT-BMT scale was refined using item response theory methods and validated against known markers of health status.  Second, structural equation modeling was used to test relationships between physical and psychological symptoms, psychosocial factors, and HRQOL outcomes.  Third, mixture modeling was used to identify classes of survivors based on items measuring PTSS and PTG.  Demographic and clinical factors were evaluated as predictors of class membership using multinomial logistic regression and HRQOL outcomes were compared across classes. The revised unidimensional FACT-BMT resulted in a 13-item scale with strong known-groups validity with physical HRQOL.  In the SEM, physical symptoms, as measured by the revised scale in Aim 1, were more strongly associated with physical HRQOL compared to mental HRQOL.  Psychosocial factors played a larger role to influence the relationship of physical symptoms with physical HRQOL than with mental HRQOL.  A 2-factor 4-class mixture model partially relaxing the conditional independence assumption generated the following classes: low PTSS/low PTG (16.3%), low PTSS/high PTG (35.5%), high PTSS/low PTG (26.6%), and moderate PTSS/moderate PTG (21.6%).  Class membership was more strongly associated with mental HRQOL (low PTSS/high PTG vs. high PTSS/low PTG; p<0.01) compared to physical HRQOL. Overall, physical and psychological symptoms were strongly associated with HRQOL.  Conceptual models assisted in mapping out relationships between symptoms, psychosocial factors and HRQOL.  Finally, mixture modeling provided insight into complex psychological experiences of long-term survivors of HSCT.
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 Kelly M Kenzik.
Thesis: Thesis (Ph.D.)--University of Florida, 2013.
Local: Adviser: Huang, I-Chan.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2013
System ID: UFE0045286:00001

Permanent Link: http://ufdc.ufl.edu/UFE0045286/00001

Material Information

Title: Patient-Reported Outcomes in Hematopoietic Stem Cell Transplant Survivors
Physical Description: 1 online resource (205 p.)
Language: english
Creator: Kenzik, Kelly M
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2013

Subjects

Subjects / Keywords: cancer -- hrqol -- outcomes -- patient -- reported -- survivor -- symptom
Epidemiology -- Dissertations, Academic -- UF
Genre: Epidemiology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Advancements in hematopoietic stem cell transplantation (HSCT) result in a growing number of long-term survivors.  Recent research has shifted attention from HSCT side-effects to patient-reported outcomes (PROs) including symptoms and health-related quality of life (HRQOL).  The three aims were to 1) refine the HSCT-specific physical symptom scale for long-term survivors based on the Functional Assessment of Cancer Therapy–Bone Marrow Transplant (FACT-BMT), 2) develop and test a conceptual framework of factors influencing PROs, and 3) examine population heterogeneity based on PTSS and PTG in HSCT survivors and the impact on HRQOL. This study is a secondary data analysis of PROs measures collected from 662 subjects in the RO1 parent study, “Quality of Life and Relationships in BMT Survivors”; PI: Dr. John Wingard.  First, the FACT-BMT scale was refined using item response theory methods and validated against known markers of health status.  Second, structural equation modeling was used to test relationships between physical and psychological symptoms, psychosocial factors, and HRQOL outcomes.  Third, mixture modeling was used to identify classes of survivors based on items measuring PTSS and PTG.  Demographic and clinical factors were evaluated as predictors of class membership using multinomial logistic regression and HRQOL outcomes were compared across classes. The revised unidimensional FACT-BMT resulted in a 13-item scale with strong known-groups validity with physical HRQOL.  In the SEM, physical symptoms, as measured by the revised scale in Aim 1, were more strongly associated with physical HRQOL compared to mental HRQOL.  Psychosocial factors played a larger role to influence the relationship of physical symptoms with physical HRQOL than with mental HRQOL.  A 2-factor 4-class mixture model partially relaxing the conditional independence assumption generated the following classes: low PTSS/low PTG (16.3%), low PTSS/high PTG (35.5%), high PTSS/low PTG (26.6%), and moderate PTSS/moderate PTG (21.6%).  Class membership was more strongly associated with mental HRQOL (low PTSS/high PTG vs. high PTSS/low PTG; p<0.01) compared to physical HRQOL. Overall, physical and psychological symptoms were strongly associated with HRQOL.  Conceptual models assisted in mapping out relationships between symptoms, psychosocial factors and HRQOL.  Finally, mixture modeling provided insight into complex psychological experiences of long-term survivors of HSCT.
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 Kelly M Kenzik.
Thesis: Thesis (Ph.D.)--University of Florida, 2013.
Local: Adviser: Huang, I-Chan.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2013
System ID: UFE0045286:00001


This item has the following downloads:


Full Text

PAGE 1

1 PATIENT REPORTED OUTCOMES IN HEMATOPO IE TIC STEM CELL TRANSP LANT SURVIVORS By KELLY KENZIK A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2013

PAGE 2

2 2013 Kelly Kenzik

PAGE 3

3 To my family: Mom & Dad Eric, Luke Jenna, and my Aunt Karen & Uncle John

PAGE 4

4 ACKNOWLEDGMENTS I would especially like to thank my men tor, Dr. I Chan Huang for his guidance and leadership throughout my doctoral program. I would also like to thank my committee member Dr. Elizabeth Shenkman, for her incredible support during my fellowship with the Institute for Child Health Policy. I wou ld also like to acknowledge the contributions and feedback from my committee members Dr. Mildred Maldonado Molina, Dr. Robert Cook, and Dr. John Wingard I am incredibly grateful to Dr. Wingard for allowing me to use his rich dataset to complete this dis sertation.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMEN TS .................................................................................................. 4 LIST OF TABLES ............................................................................................................ 8 LIST OF FIGURES ........................................................................................................ 10 LIST OF ABBREVIATIONS ........................................................................................... 11 ABSTRACT ................................................................................................................... 12 CHAPTER 1 INTRODUCTION TO HEMATOPOIETIC STEM CELL TRANSPLANT SURVIVORSHIP ..................................................................................................... 14 Epidemiology of Health Conditions in HSCT Survivors ........................................... 14 PROs in Long-Term HSCT Survivor Research ....................................................... 16 Sp ecific Aims of Dissertation .................................................................................. 23 Specific Aim 1: Refine the HSCT-Specific Physical Symptom Scale ................ 23 Specific Aim 2: Develop and Test a Conceptual Framework of Factors Influencing PROs of HSCT Survivors ............................................................ 24 Specific Aim 3: Examine the Population Heterogeneity Based on PTSS and PTG in HSCT Survivors and the Impact on HRQOL ..................................... 26 Study Design .......................................................................................................... 29 Participant Recruitment .................................................................................... 29 Study Measures ............................................................................................... 31 Advantages of Study Population ...................................................................... 31 REFINING SYMPTOM MEASUREMENT TOOL FOR LONG-TERM SURVIVORS OF HEMATOPOIETIC STEM CELL TRANSPLANT ......................... 40 Introduction ............................................................................................................. 40 Methods .................................................................................................................. 43 Participants and Data Collecti on ...................................................................... 43 Study Measures ............................................................................................... 44 Analytic Strategy .............................................................................................. 46 Instrument refineme nt ...................................................................................... 46 Results .................................................................................................................... 51 Study Sample Characteristics .......................................................................... 51 Instrument Refinement ..................................................................................... 52 Known-Groups Validity Results ........................................................................ 54 Discussion .............................................................................................................. 54 Limitations ............................................................................................................... 58 Conclusion .............................................................................................................. 59

PAGE 6

6 3 EXAMINING THE RELATIONSHIP BETWEEN SYMPTOMS, PSYCHOSOCIAL FACTORS AND HEALTH RELATED QUALITY OF LIFE IN HEMATOPOIE TIC STEM CELL TRANSPLANT SURVIVORS ................................ ............................. 68 Introduction ................................ ................................ ................................ ............. 68 Methods ................................ ................................ ................................ .................. 71 Participants and Data Collection ................................ ................................ ...... 71 Study Measures ................................ ................................ ............................... 72 Hypothesis ................................ ................................ ................................ ........ 76 Analytic Strategy ................................ ................................ ............................. 77 Results ................................ ................................ ................................ .................... 81 Study Sample Characteristics ................................ ................................ .......... 81 Bivariate Correlations ................................ ................................ ....................... 82 Pathways and Framework Results ................................ ................................ ... 82 Discussion ................................ ................................ ................................ .............. 86 Limitations ................................ ................................ ................................ ............... 94 Conclusion ................................ ................................ ................................ .............. 95 4 HETEROGENEITY IN POSTTRAUMATIC STRESS AND POSTTRAUMATIC GROWTH AMONG HS CT SURVIVORS ................................ .............................. 108 Introduction ................................ ................................ ................................ ........... 108 Relationship of PTSS and PTG ................................ ................................ ...... 109 Measurement Difficulties ................................ ................................ ................ 110 Application of Factor Mixture Model ................................ ............................... 111 Methods ................................ ................................ ................................ ................ 113 Participants and Data Collection ................................ ................................ .... 113 Study Measures ................................ ................................ ............................. 113 Hypothesis ................................ ................................ ................................ ...... 115 Analytic Strategy ................................ ................................ ............................ 115 Model Building ................................ ................................ ................................ 116 Predictors and Distal Outcomes Associated with Class Members hip ............. 118 Results ................................ ................................ ................................ .................. 121 Study Sample Characteristics ................................ ................................ ........ 121 Model Build ing ................................ ................................ ................................ 121 Predictors and Distal Outcomes Associated with Class Membership ............. 124 Discussion ................................ ................................ ................................ ............ 125 Limitations ................................ ................................ ................................ ............. 131 Conclusion ................................ ................................ ................................ ............ 131 5 IMPROVING PATIENT REPORTED OUTCOMES RESEARCH IN LONG TERM SURVIVORS OF HEMATOPOIETIC STEM CELL TRANSPLANT: DISCUSSION OF RESULTS ................................ ................................ ................ 144 Review of Findings ................................ ................................ ............................... 144 Chapter 2: Refine the HSCT Specific Phys ical Symptom Scale ..................... 145

PAGE 7

7 Chapter 3: Develop and Test a Conceptual Framework of Factors Influencing PROs of HSCT Survivors ................................ .......................... 148 Chap ter 4: Examine the Population Heterogeneity Based on PTSS and PTG in HSCT Survivors and the Impact on HRQOL ................................ ... 150 Evaluating Disease and Health Outcomes ................................ ............................ 154 Limitations ................................ ................................ ................................ ............. 155 Conclusion ................................ ................................ ................................ ............ 156 APPENDIX A CHAPTER 2 DETAILED METHODS FOR DIF AND ITEM SELECTION .............. 158 Differential Item Functioning ................................ ................................ ................. 158 Item evaluation ................................ ................................ ................................ ..... 159 B CHAPTER 2 DETAILED RESULTS FOR ITEM SELECTION .............................. 161 C CHAPTER 3 INSTRUMENT MEASUREMENT PROPERTIES ............................ 165 D CHAPTER 4 DETAILED METHODS ................................ ................................ .... 168 Factor Mixture Models ................................ ................................ .......................... 168 Fitting Factor Model and Latent Class Model ................................ ................. 170 Fitting the Factor Mixture Model ................................ ................................ ..... 171 Adding Predictors of Class Membership to Final Model ................................ 172 Evaluating Class Membership and Distal Outcomes ................................ ...... 175 Mplus Sample Code for 2 Class 2 Factor Mixture Model Variations ..................... 176 E CHAPTER 4 DETAILED RESULTS TABLES ................................ ....................... 183 LIST OF REFERENCES ................................ ................................ ............................. 185 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 205

PAGE 8

8 LIST OF TABLES Table page 1 1 Comparison of statistical models for exam ining heterogeneous populations ..... 32 1 2 Study characteristics ................................ ................................ .......................... 34 1 3 Clinical variables and patient reported outcomes (PROs) instruments ............... 36 2 1 Study characteristics ................................ ................................ .......................... 60 2 2 Criteria for item removal based on content, ceiling effects, local dependency, IRT properties, and differential item functioning: Stage 1 (25 items) .................. 62 2 3 Criteria for ite m removal based on content, ceiling effects, local dependency, IRT properties, and differential item functioning: Stage 2 (20 items) .................. 63 2 4 Criteria for item removal based on content, ceiling effects, local dependency, IRT properties, and differential item functioning: Stage 3 (16 items) .................. 64 2 5 Stage 4 final item set measurement properties (13 items) ................................ 65 2 6 Unadjusted and adjusted mean latent HSCT symptom scores by known groups ................................ ................................ ................................ ................ 66 2 7 Difference in latent mean scores and effect sizes for known groups .................. 67 3 1 Study Characteristics ................................ ................................ .......................... 97 3 2 Bivariate standardized correlation matrix of latent factors ................................ .. 99 3 3 Direct, indirect and total effects of pathways from physical symptoms through depression and PTSS to HRQOL ................................ ................................ ..... 102 3 4 Direct, indirect and total effects of concep tual framework for physical symptoms, optimism, coping, depression & HRQOL ................................ ........ 106 4 1 Study characteristics ................................ ................................ ........................ 132 4 2 Latent class an d factor mixture model accuracy and fit parameters ................. 135 4 3 FM 3 four class model item threshold values ................................ ................... 136 4 4 Description of fo ur classes in FM 3 model ................................ ........................ 139 4 5 FM 3 factor means and factor loadings for posttraumatic stress and posttraumatic growth by class membership ................................ ...................... 140

PAGE 9

9 4 6 Results from multinomial logistic regression evaluation effects of covariates on class membership ................................ ................................ ........................ 142 4 7 Results from the Wald Chi 2 ) tests of mean equality of t he auxiliary analyses of outcomes ................................ ................................ ....................... 143 C 1 Measurement properties of instruments ................................ ........................... 165 D 1 Parameter specifications for mixture mod el variations ................................ ..... 180 E 1 Measurement properties of Impact of Events Scale (IES) and Posttraumatic Growth Inventory (PTGI) ................................ ................................ .................. 183 E 2 T wo and three class factor mixture model comparison ................................ ..... 184

PAGE 10

10 LIST OF FIGURES Figure page 1 1 Flow chart summarizing recruitment of HSCT survivor group ............................ 39 3 1 Hypothesized conceptual models for pathways and framew ork of factors influencing HRQOL ................................ ................................ ............................. 96 3 2 Symptoms and physical HRQOL p athways ................................ ...................... 100 3 3 Symptoms and mental HRQOL pathways ................................ ........................ 101 3 4 Symptoms, psychosocial factors and physical HRQOL framework .................. 104 3 5 Symptoms, psychosocial factors and mental HRQOL framework .................... 105 4 1 General latent variable mixture model ................................ .............................. 134 B 1 Test information function for final 13 item physical symptom scale .................. 164 D 1 Graphic representation of allowing variation within classes ............................. 179 D 2 Common factor model, latent class model, and mixture model variations ........ 182

PAGE 11

11 LIST OF ABBREVIATIONS C GVHD Chronic graft versus host disease FMM Factor mixture mod el HRQOL Health related quality of life HSCT Hematopoietic stem cell transplantation IRT Item response theory LCA Latent class analysis LPA Latent profile analysis MCS Mental component summary PCS Physical component summary PROs Patient reported outcome s PTG Posttraumatic growth PTSS Posttraumatic stress symptoms SEM Structural equation modeling

PAGE 12

12 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the De gree of Doctor of Philosophy PATIENT REPORTED OUTCOMES IN HEMATOPOIE TIC STEM CELL TRANSP LANT SURVIVORS By Kelly Kenzik May 2013 Chair: I Chan Huang Major: Epidemiology A dvancements in hematopoietic stem cell transplantation (HSCT) result in a growing number of long term survivors. R ecent research has shifted attention from HSCT side effects to patient reported outcomes (PROs) including symptoms and health related quality of lif e (HRQOL) The three aims were to 1) refine the HSCT specific physical sym ptom scale for long term survivors based on the Functional Assessment of Cancer Therapy Bone Marrow Transplant (FACT BMT) 2) develop and test a conceptual framework of factors influencing PROs and 3) examine population heterogeneity based on PTSS and PTG in HSCT survivors and the impact on HRQOL This study is a secondary data analysis of PROs measures collected from 662 subjects in the RO1 parent study, Quality of Life and Relationships in BMT Survivors ; PI: Dr. John Wingard First, the FACT BMT scale was refined using item response theory methods and validated against known markers of health status. Second, structural equation modeling was used to test relationships between physical and psychological symptoms, psychosocial factors and HRQOL outcomes Third, mixture mode ling was used to identif y classes of survivors based on items measuring PTSS and PTG. Demographic and clinical factors were evaluated as predictors of class

PAGE 13

13 membership using multinomial logistic regression and HRQOL outcome s were com pared across classes The revised unidimensional FACT BMT resulted in a 13 item scale with strong known groups validity with physical HRQOL. In the SEM, physical symptoms, as measured by the revised scale in Aim 1 w ere more strongly associated with phys ical HRQOL compared to mental HRQOL Psychosocial factors played a larger role to influence the relationship of physical symptoms with physical HRQOL than with mental HRQOL A 2 fact or 4 class mixture model partially relaxing the conditional independence assumption generated the following classes: low PTSS/low PTG (16.3%), low PTSS/high PTG (3 5.5 %) high PTSS/low PTG (26.6%) and moderate PTSS/moderate PTG (21.6%) Class membership was more strongly associated with mental HRQO L ( low PTSS/high PTG vs. hig h PTSS/low PTG ; p<0.01) compared to physical HRQO L Overall, physical and psychological symptom s were strongly associated with HRQOL. Conceptual models assist ed in mapping out relationships between symptoms, psychosocial factors and HRQOL. Finally, mixt ure modeling provide d insight into complex psychological experiences of long term survivors of HS CT.

PAGE 14

14 CHAPTER 1 INTRODUCTION TO HEMATOPOEITIC STEM CELL TRANSPLANT SURVIVORSHIP Epidemiology of Health Conditions in HSCT Survivor s Hematopoie tic stem cell tra nsplantation (HSCT) is a medical procedure conducted to replace the stem cells from bone marrow, peripheral blood, or umbilical cord blood in patients with damaged bone marrow or immune systems. HSCT is typically used for treating patients with severe mal ignant diseases (~95%) and non malignant hematological diseases (~5%) [1] The most recent estimates indicate the majority of HSCT recipients are diagnosed with lymphoma (54.5%), followed by leukemia (33.8%), and so lid tumors (5.8%) [1] The treatment is delivered to more than 45,000 patients annually worldwide [2] In the United States, about 50,000 HCST recipients survive for more th an 5 years [3] and more than 85% of the recipients survive for more than 10 years [4] However, the overall life expectancy of HCST recipients remains lower than the genera l population [5] With the increasing long term survival rates, attention has shifted to long term side effects or late effects due to the diseases and treatment [6] Two types of HSCT are utilized with potential differences in long term late effects: allogeneic and autologous HSCT. Allogeneic HSCT uses donor stem cells and is associated with graft versus host disease (GVHD) [7] Gr aft versus host disease, classified as acute or chronic (cGVHD), is the most frequent complication of allogeneic transplantatio n [7] attacks the pat sues. While it is typically seen on the skin or mouth, cGVHD can manifest in any organ system. cGVHD occurs in 40% to 70% of the allogeneic HCST recipients [8, 9] In contra st, autologous transplantation uses the

PAGE 15

15 may be lost leaving the patient at risk of infection or bleeding, the risk of cGVHD is very minimal. Allogeneic and autologous HS CT recipients are at risk of impaired health outcomes immediately following transplant and in the subsequent years. The cumulative incidence of chronic health conditions at 10 years is 60% and the incidence of a severe or life threatening condition is 35% [10] The most common comorbid conditions for HSCT survivors include cardiac and cardiovascular complications, pulmonary complications, endocrine issues (i.e. thyroid conditions, metabolic syndrome), fertility, muscul oskeletal complications, chronic kidney problems, visual problems, and subsequent malignant neoplasms [11] Chronic conditions or other subsequent long reported outcomes ( PROs), which is a rubric term comprised of the concepts of symptoms, functional status and health related quality of life (HRQOL ) effects most proximal to the disease or treatment process [12] Functional status refers to day ability to conduct normal activities (i.e. role functioning in work or family). status and subjective well being. The general consensus for domains of HRQOL includes physical, psychological, spiritual and social well being [12 14] While the distribution and risk factors for chronic conditions post HSCT have report s of symptoms and HRQOL have resulted in heterogeneous findings. Some studies suggest that long term HSCT survivors recover to a state comparable to the general population in some domai ns of HRQOL [15 17]

PAGE 16

16 while others show continued impairment physically or psychologically over time [17 22] It is unclear which factors contribute to some survivors reporting an almost complete restoration of HRQOL while others continue to have impaired health outcomes L imited PROs measurement tools and a lack of a clear conceptual framework to guide studying important factors associated with PROs may contribute to the inconsistent findings. PROs in Long Term HSCT Survivor R esearch PROs are information obtained directly f rom the patient regarding their health status or well being [23] The term encompasses a variety of health outcomes, including somatic symptoms such as pain fatigue, vomiting insomnia, and physical psychological, and social functional relating to disease and treatment [12] In medicine, PROs measures can help evaluate whether treatments are doing more harm than good. Similarly, PROs can be delivered as end points to clinical trials or measured over time to assess for changes in health status. In public health, PROs are a feasible method of measures are incorporated into population surveys and epidemiologic studies to compare population outcomes [24] consequences of cancer an d its treatment These consequences may include, but are not limited to, problems with pain, fatigue, and role functioning [25] Measuring HRQOL, symptoms, and functional status and identifying fa ctors associated with symptoms and HRQOL is now a priority. Wingard and colleagues evaluated HRQOL and the factors ( i.e. time since HSCT comorbid conditions, optimism anxiety and other PROs) associated with HRQOL outcomes in two studies through bivariat e and multivariate analysis [26] The findings suggest that demographic, clinical, and social factors

PAGE 17

17 significantly influence d health outcomes such as HRQOL, and that further study is warranted to better under stand the complex relationships and mechanisms among these factors It is evident that PROs play an important role in evaluating health status of long term HSCT survivors, but several research gaps exist in the use of PROs in this population. First, the re are no validated instrument s for measuring unique a set of physical symptoms for long term HSCT survivors Second, no evidence based model is available to describe the complex relationships among physical and psychological symptoms, intrapersonal facto rs (e.g., personality), social influences, coping strategies PROs (symptoms and HRQOL), and the joint effects of these factors on HRQO L. Third, limited studies focu s on positive psychological effects in long term HSCT survivors. P revious research sugges ts that experience from cancer and treatment may result in negative psychological reactions such as posttraumatic stress [27] However, additional early evidence also points that posttraumatic growth, benefit finding, or stress related growth may lead to a perceived gain from cancer and HSCT [28] It is also suggested that both positive and negative responses can co occur and are not mutually exclusive [29, 30] While both the negative and positive effects of HSCT have been found to impact HRQOL, few studies have attempted to use sophisticated methods to untangle the complex relationship of these psychological responses and their effects on HRQOL [31, 32] These gaps provide the foundation for the overall goal of this dissertation to attempt to address the meas urement and evaluation issues related to PROs in long term HSCT survivors.

PAGE 18

18 S ymptom measurement tool for long term HSCT s urvivors: Utilizing high quality measurement tools is paramount to the design of an epidemiologic study. Specifically, w hen designin g a study to evaluate health outcomes in a population it is important to consider that clinical measures and PROs capture different aspects of a [12] PROs such as physical symptoms (fatigue and pa in) and HRQOL judgment, and developing appropriate tools to accurately and reliably measure PROs is essential in clinical and public health research A review of the literature revealed the availability of symptom m easurement tools specific to HSCT survivors is limited, particularly for long term survivors (5+ years) [16, 33, 34] Type and severity of symptoms in recent HSCT sur vivors may not appear the same as the long term survivors (e.g. anxiety about treatment effectiveness vs fertility concerns or cognitive issues). Because HSCT is used in severe disease states and with the treatment having a potentially greater impact on physical functioning, HSCT survivors are likely to have different symptoms persisting over time compared to cancer survivors receiving standard therapy [35] The most renowned HSCT symptom scale, the Functional Ass essment of Cancer Therapy Bone Marrow Transplant scale ( FACT BMT ) was developed using classical test theory (CTT) methods and then validated using a sample of HSCT recipients who were at most 1 year post treatment [36] The sample contained a small proportion of allogeneic patients (20%) who may have different item response pattern s in terms of symptom type or severity compared to autologous patients [37] In addition, the use of CTT encounters some limitations in instrument development and validation ( Chapter 2 ). The item set was designed as an extension to the Functional Assessment of Cancer

PAGE 19

19 Therapy General scale, without evaluation of its properties as a stand alone instrument. Analyzing measurement properties of the individual items using advanced methodology such as item response theory (IRT) to better capture symptom severity is warranted (Chapter 2) The use of the extant FACT BMT scale may limit our ability to accurately measure symptoms as a result of treatment in long term survivors. An HSCT specific instrument for long term survivors is preferred for measuring the level of symptoms experienced that are most proximally attributable to the HSCT treatment. A mo re HSCT sensitive tool to identify symptoms related to HSCT may help address the overreliance on the use of generic or non disease specific tools to measure symptom burden for long term HSCT survivors [38] Refi ning existing instrument s such as the FACT BMT is an efficient method to improve symptom measurement systems for HSCT [39] Complex relationships among demographic and clinical characteristics, symptoms and social resources on HRQOL in HSCT survivor: To our best knowledge, n o conceptual model is available to guide investigations on the relationships between demographic and clinical characteristics, symptoms, social resources, and HRQOL for long term survivors of H SCT. Previous research has produced inconsistent findings with respect to the relationships between symptom reports and HRQOL in HSCT survivors [40 43] Some long term HSCT survivors report ed comparable HRQOL in contrast to the general population or their pre treatment levels despite the presence of physical symptoms [15 17, 22] while other survivors continue to suffer from impaired HRQOL [18 21, 26] The inconsistencies may be a result of true differences in survivors' HRQOL the sel ection of different factors (e.g. psychological symptoms,

PAGE 20

20 coping strategies) and the hypothes ized role of these factors (i.e. direction of influence) in influencing HR QOL across different stud ies Unfortunately, research on the interrelationships between the factors themselves and their influence on HRQOL is sparse [42] Demographic characteristics such as age, sex, soc ioeconomic status, education level, and marital status have been associated with HRQOL [26, 43, 44] Clinical factors deemed important in their association with PR Os ; these factors include disease diagnosis, treatment, severity of HS CT experience (autologous vs. allogeneic and presence of cGVHD), intensity of treatment before HSCT (e.g. chemotherapy intensity) and the time since HSCT [45] In addition, the presence of comorbid conditions and current symptom status are important clinical factors contributing to poor HRQOL [46] Research also shows that psychosocial factors such as social support, social constraints, and coping are associated with HRQOL in HSCT survivor in multivariate analysis [26] and may act as mediating or moderating factors bet ween physical symptoms and/ or psychological symptoms and HRQOL [47] Given the importance of identifying risk factors for improving HRQOL the development of comprehensive conceptual model can better lead h ypothesis testing for the relationship of demographic, clinical, symptomatic, and psychosocial factors with HRQOL Haase & Braden described how the la ck of theoretical framework can lead to failure in a study [48] First, the relationship between domains cannot be determined and interpretability of relationship patterns is misleading. Second, there is no basis for determining moderation or mediation among factors. A review by Mosher and colleagues [42] suggests the key factors to consider in HRQOL

PAGE 21

21 research include clinical variables related to severity of treatment experience, physical and psychological symptoms, and social and coping resources of the HSCT survivors. Conclusions of the review by Mosher indicated that the domains of HRQOL were often assessed without understanding their interrelationships and relating to clinical information and symptom variables such as distress or anxiety [42] Importantly, no framework is available to guide a structured analytic procedure and compare the results across different studies. Evaluating the mechanism through which these factors influence HRQOL may provide insight to why pat ients with similar clinical reports may have variable outcomes. Positive and negative psychological experiences resulting from HSCT: According to the American Psychiatric Association [49] the experience of cancer psychological symptoms and HRQOL is of research and clinical interest. However, it is unclear if or how this traumatic event plays a role in the HRQOL o f long term survivors through psychological effect s seen in traditionally defined traumatic events su ch as natural disasters or assau lts [28, 50] I t is possible that a cancer related ev ent has similar impacts on psychological health and different aspects of HRQOL compared to other traumatic events such as natural disasters, accidents, or violent attacks. Individuals suffer ing from traumatic events (e.g., natural disasters, accidents, at tacks) often reported both negative psychological state (e.g., posttraumatic stress symptoms (PTSS) ) and positive psychological state (e.g., posttraumatic growth (PTG) ) [5 1 53] It is theorized that ma n y individuals with HSCT experience may also report experiencing PTSS and PTG [50, 54,

PAGE 22

22 55] PTSS are a set of 17 behavioral and psyc hological symptoms that make up the criteria for establishing a diagnosis of posttraumatic stress disorder (PTSD) [49] PTSS are characterized by three clusters including avoidance, intrusive t houghts, and hyperarousal that develop in response to the trauma itself. The symptoms result ing from the struggle to deal with the shock of a traumatic event may or views about oneself and the world [56] PTG was defined by Calhoun and [57] PTG occurs when an individual reinterpr ets a traumatic event resulting in positive personal change [58] Evidence suggests that PTSS and PTG do not necessarily occur in isolation from one another [29, 54] Previous examination of the relationship between PTSS and PTG within the cancer and HSCT survivor populations has resulted in mixed findings. Several cross sectional studies using standard measures found no association between PTSS and P TG [55, 59] However, some studies found those with higher level of distress reported greater growth [28] whereas others found those with lower level of distress reported higher levels of positive growth [60] Given the possible co existence or co occurr ence of PTSS and PTG in a survivor, questions of how to classify individuals based on different levels of PTSS and PTG, identify factors related to the classification, and link the classification results to HRQOL outcomes become critical. In assessing these concepts, several classes of individuals are likely to emerge which is associated with different combinatio ns of item responses to PTSS and PTG measures One approach that has yet to be uti lized to explore this issue is latent variable mixture modeling or factor mixture modeling (FMM)

PAGE 23

23 [61] (Chapter 4) T his analytic method allows for examining heterogeneous populations of HSCT survivor population who experienc e both PTSS and PTG at varying levels [61, 62] Specific Aims of Disse rtation This dissertation propose s three aims to address the gaps in the current literature. The first aim was to refine the HSCT specific physical symptom scale for long term survivors ; the second aim was to develop and test a conceptual framework of fact ors influencing PROs of HSCT survivors ; the third aim was to examine the population heterogeneity based on PTSS and PTG in HSCT survivors and the impact on HRQOL. Specific Aim 1: Refine the HSCT S pecific P hysical S ymptom S cale The first aim was to develop a HSCT specific physical symptom scale for long term survivors This possesses particularly important clinical implications This is because c ondition health status that is proximally re lated to the disease or treatment of interest compared to generic PRO measures that evaluate the overall status of the individual. Refining an instrument that is brief in length, covers appropriate symptoms related to long term HSCT survivors and demonst rates acceptable psychometric properties will help detect the variation of symptom burden and increases the utility of clinical practice [63] The first aim had two main objectives. The first objective was to u se item response theory (IRT) to refine the Functional Assessment of Cancer Therapy Bone Marrow Transplant (FACT BMT) scale to better measure physical symptom s It was hypothesized that IRT can help select appropriate items to measur e physical symptoms specific to long term survivors and the modified scale will demonstrate acceptable

PAGE 24

24 psychometric properties The second objective was to v alidate the newly refined sc ale using health status markers. These include the MOS SF 36 Physical Component Summary s core (PCS) and Mental Component Summary score (MCS) [64] Karnofsky Performance Scale (KPS) [65] presence of comorbid conditions, and severity of treatment experienc e. It was hypothesized that physical symptom s measured by this modified FACT BMT scale will be most strongly associated with the MOS SF 36 PCS scale and severity of treatment experience compared to other health status markers A ma in contribution of this a im is t he modified FACT BMT scale will be appropriate for measuring physical symptoms for a long term survivor population (5+ years) as opposed to original development in the immediate survivor population (up to 1 year) [36] An additional advantage is the application of IRT methodology to refine the scale will be sample and scale independent using items with the best measurement properties [66, 67] IRT approach focuses on the item level analysis. Information obtained from a questionnaire will be use ful if the contents of items are appropriate and valid for the population of interest. Specific Aim 2: D evelop and T est a Conceptual Framework of Factors Inf luencing PROs of HSCT S urvivors The second aim was to develop a conceptual framework to test the relationship between physical symptoms, psychological symptoms, psychosocial variables (e.g., optimism, social constraint, and coping), and HRQOL through struc tural equation modeling (SEM) approach This study collected comprehensive information from long term HSCT survivors with respect to demographic, clinical, symptom, and psychosocial factors in addition to HRQOL outcomes, which allow for pursuing this spec ific aim. Generating an evidence based framework for assessing important factors contributing

PAGE 25

25 to HRQOL of long term HSCT survivors will provide constructive information for guiding future research and identify risk factors of poor HRQOL for further target interventions This aim had two major objectives. The first objective was to evaluate whether psychological symptoms including depressive symptoms and PTSS mediate the effect of physical symptoms on HRQOL in long term survivors of HSCT through path ana lysis. We hypothesized that physical symptoms had a direct effect on both physical and mental HRQOL as well as an indirect effect on HRQOL through psychological symptoms. We also hypothesize that the psychological symptoms play a more significant role to mediate the relationship of physical symptoms with mental HRQOL than with physical HRQOL. The secondary objective was to test the overall conceptual model of factors in contributing to HRQOL, which include psychosocial factors, such as optimism social c onstraints and coping in addition to physical and psychological symptoms. Specifically, this objective was to identify specific pathways through which physical symptoms, depressive symptoms and psychosocial variables influence physical and mental HRQOL. We hypothesize that the psychosocial variables ( i.e., optimism, social constraints and coping ) will explain the pathway from physical symptoms through depressive symptoms to physical HRQOL as identified in objective 1 but not mental HRQOL. Specificall y, we hypothesize that the pathways of physical symptom s coping physical HRQOL and optimism physical symptom s physical HRQOL will explain the physical symptom s depression physical HRQOL relationship. However, f or mental HRQOL, we hypothesize that the same pathways ( physical symptom s coping mental

PAGE 26

26 HRQOL and optimism physical symptom s mental HRQOL ) will not play a significant role and that depressive symptoms will remain an important mediator. Establishing and testing PROs conceptual framework s will be provi de a critical foundation for guiding a variety of PROs research A framework helps describe the interrelationships and pathways among factors toward PROs outcomes identify specific antecedent factors and mediator variables of poor PROs outcomes and desi gn interventions targ eting risk factors that influence PROs. Lastly, an evidence based framework can highlight potential gaps or inconsistencies in research to drive future hypotheses. Within epidemiology, frameworks can guide to mapping relationships a mong variables and are useful to better describe and study complex interrelationships. Combining a theoretical foundation of why certain relationships exist with how the relationships exist ( depicted by the model or framework) is essential to understanding disease processes and health outcomes. Specific Aim 3: E xamine the Population Heterogeneity B ased on PTSS and PTG in HSCT Survivors and the I mpact on HRQOL The third aim was to describe complex relations between PTSS and PTG by categorizing HSCT survivo rs into different classes using a factor mixture model (FMM) We hypothesize that PTSS and PTG may co exist in the form of different combinations among long term HSCT survivors. Individuals may possess one of the unique combin ations of p sychological stat es : high PTSS /high PTG, low PTSS /low PTG, high PTSS / low PTG, and low PTSS / high PTG. Traditional analytic strategies simply add PTSS and PTG as the independent variables in the regression model and assume the effect of PTSS and PTG is subtractive. In addi tion, the tradition al approach relies on scale level rather than item level information in the analysis, ignoring the measurement

PAGE 27

27 errors derived from scales. In this study, we use an innovative strategy to a nalyz e PTSS and PTG simultaneously and create a class or profile to help characterize psychological symptoms of individual HSCT survivor s I t is preferred to use a method that allows for evaluating both PTSS and PTG without diluting the outcome (i.e., adding positive to negative scores). As a result, t hese unique combinations of positive and negative psychological symptoms should have differ ent effects on HRQOL. PTSS and PTG may coexist because t he negative consequences of stress are important to proceeding through the cognitive processing that may le ad to a positive impact [68] Park & Helgeson recognized this methodological issue in their study on PTSS and PTG, and suggest that it is not necessary to assume the two concepts are on the opposite ends of a conti nuum and advanced statistical methods should be used to tease out the true relationship [31] FMM is a form of generalized latent variable modeling that can be used to examin e the complex constructs of psychological st ates by allowing the underlying structure to be both categorical and continuous (dimensional) [61, 69] It is a hybrid approach of the common factor model and latent class model. The FMM method ology is described in detail in Chapter 4 and Appendix D. Table 1 1 provides a comparison across different type of latent methods for psychological and PROs research Briefly, FMM combines different aspects of the common factor model and l atent class model to evaluate the class membership among individuals and account for measurement error of item s measuring psychological sates [69] (Table 1 1). In this study, FMM will be used to differentiate sub grou p s of survivors into different classes based on their

PAGE 28

28 responses to both the PTSS and PTG scales while also accounting for the multidimensionality of the scales A FMM approach will be used to identify heterogeneous classes of survivors based on items of P TSS and PTG scales As described in the previous paragraph, we anticipated four unique classes will be generated : high PTSS/ high PTG, low PTSS/ low PTG, high PTSS/low PTG, and low PTSS/high PTG. We also hypoth esized that individuals with younger current age, lower education, and greater s everity of treatment experience are likely to relate to the class with symptoms of higher PTSS / higher PTG. Individuals with o lder age, lower severity of treatment experience, and no comorbid conditions are likely to rel ate to the class with symptoms of lower PTSS / lower PTG. In addition, we hypothesized that the classes comprised of individuals with low PTSS / high PTG are associated with higher physical HRQOL and mental HRQOL scores compared to c lasses comprised of survi v ors with high PTSS / low PTG. Compared to traditional latent class approach, the use of FMM approach is flexible to identify heterogeneous classes of PTSS and PTG and also allow for the individual classes to have within class heterogeneity. These characteri stics are particularly useful to PRO s research because individuals may potentially endorse both positive and negative outcomes in any one of multiple combinations (i.e. high PTSS/high PTG, low PTSS/low PTG, high PTSS/low PTG, and low PTSS/high PTG ). In th e broader context of epidemiology, individual diversity can greatly influence disease or health outcomes. However, t raditional statistical approaches do not take into account the issue of heterogeneity and may result in limited inferences for the results

PAGE 29

29 Study Design Participant Recruitment This study is a secondary data analysis of a large multi site study (40 centers) comprised of long term cancer/HSCT survivors ( the parent study: NIH RO1 Quality of Life and Relationships in BMT Survivors ; PI: Dr. John Wingard). Potential participants were HSCT recipients at participating transplantation centers with records identified from International Bone Marrow Transplant Registry/Autologous Blood and Marrow Transplant Registry (IBMTR/ABMTR). The participati ng centers register transplants performed at their center and submitted demographic and clinical information on eligible patients to the IBMTR/ABMTR Statis tical Center. The preliminary eligibility criteria for study participation included the age at 18 ye ars or older, a single allogeneic or autologous HSCT, at least 12 months post HSCT, specific diagnosis of cancer ( e.g., chronic myelogous leukemia, acute leukemia, lymphoma, or breast cancer), continuous remission since HSCT, and able to read and understan d English. Initially, 2,447 eligible participants were identified between 2000 and 2002. Patients at each center who met the first three criteria were then stratified by diagnosis, transplant type, years since HSCT (<5 vs >= 5 years), and intensity of pre transplantation therapy at each center (less intense vs more intense). Less intense included survivors who received transplant for chronic phase chronic myelogenous leukemia (CML) within one year of diagnosis, those who received transplant for acute leukemia or lymphoma in their first year of remission, and survivors who received transplant for adjuvant treatment of Stage II or III breast cancer. More intense pre transplant therapy included those who were transplanted for chronic phase CML over 1 yea r from diagnosis, survivors transplanted for accelerated or blast phase CML,

PAGE 30

30 survivors transplanted for acute leukemia or lymphoma beyond the first remission, and survivors transplanted for metastatic breast cancer. After stratification, survivors were ra ndomly selected to be contacted for eligibility confirmat ion. The flow chart in Figure 1 1 demonstrates the process of subject recruitment after the random selection of 1,946 potentially eligible survivors from the 2,447 identified. Of the 1, 946 survivor s identified, 295 were ineligible ( n= 133 : death and n= 134 : disease relapse ). For the remaining 1,399, contact was attempted and was successful for 960 survivors. Among the 960 survivors 704 provided written consent ( n= 118 : declined participation and n= 1 38 : verbal but not written consent). Once informed consent was received via postal mail (73.3% of eligible survivors that were contacted), a telephone interview was scheduled by Center on Outcomes, Research and Education (CORE) at Evanston Northwestern He althcare Center (Evanston, IL). Forty two survivors were withdrawn from the study after consent. Reasons for withdrawal included voluntary withdrawal from the study (n=16), ineligibility (n=12), and loss to follow up ( n =10). Participants were informed t hat they would receive a mailed packet of questionnaires with half of the measures to be completed prior to the phone interview. After the mailed packets were completed and mailed back to the research team a phone interview was conducted to ask the remain ing half of the questionnaires and to obtain basic demographic information After phone interview and mailed questionnaire were completed, records were abstracted from IBMTR/ABMTR to obtain date and type of initial diagnosis, date and type of HSCT, and th e nature of donor relationship (allogeneic). Table 1 2 shows the characteristics of study sample used in this study.

PAGE 31

31 Study M easures Several questionnaires were used in this study The description of the questi onnaires is provided in Table 1 3. The items in each PRO s questionnaire contain s state at the time of survey, past 7 days or past one month in addition to a prompt to refer the response to having had cancer or cancer treatment. Advantages of Study P opula tion D ata collected from the RO1 parent study provide a rich data source with extensive demographic, clinical, and PRO information on a large sample of long term HSCT survivors. Th e parent study provides the most comprehensive set of information to be a ble to accomplish the three aims proposed in this study. T he large sample size used in this study provides sufficient statistical power for conduct ing sophisticated analyses to accomplish three aims: IRT for Aim 1 structural equation modeling and path an alysis for Aim 2 and FMM for Aim 3 Finally, several previous studies based on this data source also build the foundation for pursuing the speci fic aims proposed in this study [26, 33, 70 72]

PAGE 32

32 T able 1 1. Comparison of statistical models for examining heterogeneous populations [62] Statistical Model Utiliza tion Assumptions Advantages Disadvantages General linear models (e.g., discriminant analysis, logistic regression, and MANOVA) Source of heterogeneity is observed or explicit (e.g., male and female) Focus typically on relationship among variables of inte rests Relate the observed group membership, other independent variable, and outcomes Linearity Independence of observations Normality of residuals Independent variables are measured without error Easily implemented Simple interpretation Equal weight for all items on the same factor Treat population as homogenous Assume unidimensional factor (or outcome of interest) Common factor model (e.g., exploratory factor analysis, multi group confirmatory factor analysis) Source of heterogeneity is observe d or explicit (e.g., male and female) Multidimensional constructs Identify latent factors comprising of item measuring the factors Latent factor describes all correlations among items measuring that factor Local independence (once latent factor is assume d, the correlations are all explained by the latent factor) Allow for continuous latent factor representing the underlying latent trait Allow for multidimensionality Different weights for different items Treat population as homogenous Latent class model (i.e., latent class analysis, LCA) Source of heterogeneity is not directly observed or explicit Focus on relationship/classification among individuals Categorical items or variables Creating classes of individuals that homogenous within class and heterogeneous across classes Useful when classification is primary focus Local independence of items within a class (within class homogeneity) Non linearity of the data Non normal distribution of data Latent class membership can be determined Other advant ages remain the same as common factor model Not allow for within class heterogeneity Categorical observed items only

PAGE 33

33 Table 1 1. C ontinued Statistical Model Utilization Assumptions Advantages Disadvantages Latent profile model (i.e., la tent profile analysis, LPA) Source of heterogeneity is not directly observed or explicit Focus on relationship/classification among individuals Continuous items or variables Creating classes of individuals that homogenous within class and heterogeneous ac ross classes Useful when classification is primary focus Local independence of items within a profile (within profile homogeneity) Conditional normal distribution Continuous observed variable that may provide more information compared to categorical obser ved variable Other advantages remain the same as latent class model Not allow for within profile heterogeneity Factor mixture model (i.e., factor mixture analysis) Source of heterogeneity is not directly observed or explicit Focus on relationship /classification among individuals Allow categorical, ordinal or continuous items to determine class membership Population heterogeneity anticipated within classes Theoretical or practical foundation required for relaxing local independence assumption Dis tributional assumptions based on observed item or variable type (categorical, ordinal, continuous) Incorporates the advantages of common factor model and LCA Items may be more representative of varying levels of latent trait (within class heterogeneity) Factor means variances can vary across classes (within class heterogeneity) Allows for inclusion of continuous, ordinal and categorical observed items Complex modeling process which involves problems of model identification and result convergence

PAGE 34

34 Ta ble 1 2 Study characteristics N Distribution, % Demographics Age, years 662 Mean (SD) 42.1 (11) Median (Range) 42.4 (18 71) <35 182 28 35 39 95 14 40 44 106 16 45 49 120 18 >50 159 24 Sex Male 251 38 Female 411 62 Race White 6 03 92 Other 56 8 Education 658 High school or below 194 30 Some college or technical education 209 32 College degree 122 18 > College degree 133 20 Occupational status Working or student 484 73 Not working 100 15 Retired 75 11 Marital status 659 Married/living with partner/ committed 483 73 Other 176 27 Annual family income < $20,000 70 11 $20,000 $40,000 141 22 $40,000 $60,000 156 24 $60,000 $80,000 100 15 >$80,000 181 28 Insurance status 597 Public 193 32 Private 387 65 No insurance 17 3 Clinical variables Time since HSCT, years Mean (SD) 662 7.0 (3.1) Median (Range) 6.6 Type of Transplant Allogeneic 272 41.1 Autologous 390 58.9

PAGE 35

35 Table 1 2. Continued N Distribution, % Donor relationship HLA identical sibling 187 70 Alternative related donor 11 40 Unrelated donor 33 12 Alternative related donor 11 40 Unrelated donor 33 12 Other or missing 36 13 Malignant disease at initial diagnosis Acute leukemia (acute myelogous leukemia or acu te lymphoblastic leukemia) 243 37 Chronic myeloid leukemia 131 20 Breast cancer 156 24 132 20 Severity of treatment Low autologous no GVHD 390 60 Moderate allogeneic no GVHD 168 26 High allogeneic GVHD 88 14 Comorbid conditions Presence of comorbid conditions at survey 104 16 Intensity of treatment before HSCT* Less intense 441 66.6 More intense 221 33.4 Karnofsky Performance Score at last follow up Mean (SD) 597 90.2(10.1) Median (Range) 10 10 0 Less intense treatment category includes patients who underwent treatment with the first year of being diagnosed with CML, patients with acute leukemia or lymphoma in remission, or treatment of stage II or stage III breast cancer. Higher intensity cat egory included patients who had undergone BMT for chronic CML more than a year after diagnosis, accelerated CML, acute leukemia or lymphoma b.

PAGE 36

36 Table 1 3 Clinical variables and patient reported outcome s (PRO s ) instruments Variable Item description Respo nse options /Domains (items) Items Scoring Direction Pre transplantation treatment intensity Based on status and duration of disease before transplantation Less intense= patients who underwent transpla nta tion for chronic phase CML within 1 year, acut e leukemia or lymphoma in fi rst complete remission, or adjuv ant treatment of high risk stage II or III breast cancer More intense= transplantation for chronic phase CML >1yr from diagnosis, accelerated or blast phase CML, acute leukemia or lymphoma beyond first remission, or metastatic breast cancer Time since diagnosis Years since diagnosis date Severity of treatment Type of HSCT treatment and presence of graft versus host disease Low= autologous (no GVHD) Moderate=allogeneic (no GVHD) High = allogeneic (GVHD) C omorbid conditions Comorbid conditions present at survey No/Yes Functional status outcome Instrument Domains (items) Physical functioning MOS SF 36* Physical Component Summary (PCS) Physical functioning (10) Role limitations due to physical problems (4) Bodily pain (2) General h ealth perceptions (5) 21 Higher scores indicate better functioning Mental functioning MOS SF 36* Mental component summary (MCS) Energy/ v itality (4) Social functioning(2) R ole limita tions due to emotional health problems (3) Mental health (5) 14 Higher scores indicate better functioning

PAGE 37

37 Table 1 3. C ontinued Variable Item description Response options /Domains (items) Items Scoring Direction Post traumatic growth Post Traum atic Growth Inventory (PTGI) Relating to others (7) New possibilities (5) Personal strength (4) Spiritual change (2) Appreciation of life (3) 21 Higher scores indicate greater growth Psychological Symptoms Depressive symptoms Center for Epidemi ologic Studies Depression scale (CES D) Depressed affect Positive affect Somatic Interpersonal 10 Higher score indicate worse symptoms Posttraumatic stress symptoms Impact of Events Scale (IES) Intrusion (7) Avoidance (8) 15 Higher score indicate worse symptoms Physical Symptoms Treatment specific symptoms Functional Assessment of Cancer Therapy Bone marrow transplant scale (FACT BMT) N/A 25 Higher scores indicate better health (fewer symptom experience) Social resources Social constraints Social Constraints Scale (SCS) N/A 16 Higher scores indicate greater constraints

PAGE 38

38 Table 1 3. C ontinued Variable Item description Response options /Domains (items) Items Scoring Direction Coping Brief COPE Active copi ng (2) Planning (2) Positive reframing (2) Acceptance (2) Humor (2) Religion (2) Using emotional support (2) Using instrumental support (2) Self distraction (2) Denial (2) Venting (2) Substance use (2) Behavioral disengagement (2) Self blame (2) 28 Higher scores indicate greater coping Personality characteristics Optimistic personality Life Orientation Test (LOT 12) N/A 12 Higher scores indicate greater optimism

PAGE 39

39 Figure 1 1 Flow chart summarizing recruitment of H SCT survivor group (IBMTR/ABTR: International Bone Marrow Transplant Registry/Autologous Blood and Marrow Transplant Registry )

PAGE 40

40 CHAPTER 2 REFINING SYMPTOM MEASUREMENT TOOL FOR LONG TERM SURVIVORS OF HEMATOPOIETIC S TEM CELL TRANSPLANT Introduction Hematopo ie tic stem cell transplantation (HSCT) is a medical procedure that replac es stem cells from bone marrow, peripheral blood, or umbilical cord blood in patients with damaged bone marrow or immune systems due to malignant disease such as cancer [3] Approximately 85% of HSCT recipients are now living 10 years or longer, and many of them are suffering from late effects such as chronic conditions [4] Evidence suggests that the occurrence of chronic condition s or complication s was 60% in survivors 10 years from diagnosis and the occurrence of life threatening condition s was 35% [10] In addition, HSCT relevant symptoms can persist in many su rvivors after the primary disease and treatment have been resolved [73, 74] The longer survival time together with the persistence of symptoms and long term chronic or life threatening condit related quality of life (HRQOL) [10] Symptoms rep orted by HSCT survivors include changes to bodily appearance, fatigue pain, loss of strength, an d appetite loss [42, 75] Measuring treatment related symptoms and HRQOL through patient reported outcome s ( PRO s ) measure is important to obtaining the complete picture of a survivor [76] The use of diagnoses (i.e. chronic obstructive pulmonary disease and cardiomyopathy) or clinical indicators (i.e. high cholesterol and imbalanced hormone levels) [6, 10] is not adequate to determine the impact of the long term effects of HSCT treatment. Evidence suggests self reported symptom s might be more sensitive to relevant health status compar ed to clinical measures [76] Compared to psychological

PAGE 41

41 symptoms, physical symptoms are more proximally related to the cancer and HSCT treatment and are a direct result of the intensity of disease an d physical invasiveness of the treatment [77] Physical symptoms have been identified as pre cursors to other long term effects such as psychological symptoms and health related quality of life (HRQOL) [78] line assessment. The availability of instrument s to measure HSCT specific symptom is limited [16, 33, 34] Previous studies assessing HSCT survivor symptoms have utilized symp tom specific scales (e.g. McGill Pain Questionnaire (MPQ) for pain ) [79] disease specific scales (e.g Functional Assessment of Cancer Therapy General) [80] or other scales symptoms of HSCT survivor [81] A limitation o f these scales is that each scale was designed to measure a specific symptom (e.g., pain) or for cancer patients currently receiving therapy and may not contain items measuring key symptoms applicable to long term HSCT survivors. Developing a new tool or r efining existing tools to measure comprehensive physical symptoms in long term HSCT survivors will be beneficial for long term survivorship research and clinical practice. Given the fact that developing a new scale involves painstaking effort r efining an existing tool, Functional Assessment of Cancer Therapy Bone Marrow Transplant Scale (FACT BMT) [37] to measure physical symptoms for HCST survivors is a practical way to create a survivor specific tool In con trast to the FACT BMT, the previous scales were not developed with a sample of HSCT recipients and the goal of measuring their unique symptoms This limitation of the previous scales makes the FACT BMT which was created with HSCT recipients, an efficient instrument

PAGE 42

42 to refine for a long term HSCT survivor symptom specific scale. However, t he original FACT BMT scale has several limitations for a long term survivor population. First, the subjects used for developing FACT BMT consisted of those who finished HSCT within 1 year. S urvivors in the first 12 months typically experience physical symptoms directly related to treatment and are different from long term survivors who have late symptoms or complications [6] Second, conventional measurement methodology (i.e., classical test theory (CTT)) was used to develop the original FACT BMT scale. The CTT approach to construct and evaluate PRO s is limited by the characteristics of scale dependence (observed score s are hypothesi zed to represent the true score s and tied to a particular set of items) and sample dependency ( measurement properties such as reliability and validity depend upon the sample under investigation ) [82] In additio n, CTT methods for scale development do not allow for a distinction between the ability of the scale to measure the latent trait (i.e. depr ession, physical functioning) and the underlying severity of the subjects [83] Third, the original FACT BMT scale has the potential to be shortened in length. Developing a physical symptom scale which is short in content and possesses good measurement properties may increase the likelihood of use in clinical practice and decre ase the response burden. A previous analysis by Huang and colleagues used a CTT approach to create a shortened subscale of the FACT BMT items [84] The items selected were subject to the limitations of CTT and cited a modern test theory approach, such as item response theory (IRT), as the next step to refining the scale.

PAGE 43

43 IRT is comprised of a variety of modeling methods for analyzing specific items that are designed to capture a particular latent trait (i.e. physic al symptom s ) [85, 86] IRT allows for creation of a latent trait scale measured by a set of items that are calibrated on the same metric as the latent trait of survivors IRT provides the adv antages of sample and scale independence to address the previously stated problems of CTT In addition, IRT gives precise information for selecting best performing items to measure the underlying burden of physical symptom s for HSCT survivors Burket t and colleagues identified several future directions for symptom research in cancer survivors including the recommendation to increase use of assessment tools that are validated, simple, and for multiple symptoms [ 74] A refined item pool may result in stand alone scale that can reduce the response burden for patients. Given the importance of physical symptom burden on HSCT survivors and the needs of developing a scale to monitoring progress of physical symptoms of this population, t he first objective of this study was to use IRT methodology to refine the FACT BMT scale for measuring physical symptom s for long term survivors. The second objective was to validate the newly refined scale using PRO meas ures and cli nical information. Methods Participants and Data C ollection This study is a secondary data analysis of the multi site study (40 centers) described previously in Chapter 1, Study Design Briefly, p articipants were HSCT recipients at participating transplan tation centers with records identified from International Bone Marrow Transplant Registry/Autologous Blood and Marrow

PAGE 44

44 Transplant Registry (IBMTR/ABMTR ). Inclusion criteria for study participation are: age at 18 years or older, a single allogeneic or autol ogous HSCT, at least 12 months post HSCT, specific diagnosis of cancer (chronic myelogous leukemia, acute leukemia, lymphoma, or breast cancer), continuous remission since HSCT, and able to read and understand English Eligible s urvivors were randomly sel ected to be contacted for eligibility confirmation. The flow chart i n Chapter 1 Figure 1 1 demonstrates the process of subject recruitment Data were collected from mail surveys and phone interviews. Of the 704 survivors that provided written consent, 6 62 completed the study ; 658 who complet ed the FACT BMT with no missing data were available for instrument development. Clinical information was extracted from medical records and the IBMTR/ABMTR. Study Measures Demographic information included age at sur ve y, gender, race, education, occupational status, family income and marital status. Clinical information used included presence of comorbid conditions, severity of treatment experience and Seve rity of treatment experience was defined as low severity for autologous and no cGVHD ; moderate severity for allogeneic and no cGVHD ; high severity for allogeneic and cGVHD Clinical information was obtained through the registry and medical record informat ion. PRO s measures used include physical symptoms measured by the modified FACT BMT and HRQOL measured by the Medical Outcomes Study 36 item short form health survey instrument ( MOS SF 36 ) [64] The modified F ACT BMT contains 25 items instructing participants to respond with how much they experience the physical symptom s The items are based on a

PAGE 45

45 L ikert type , ery much scores indicate better health or less symptoms and negatively worded items were reverse scored. The item set was previously modified by expert panel prior to survey administration to eliminate items not relevant to long term survivors a nd add new items determined to be important to survivors by the panel [84] Of the 23 items in the original FACT BMT [37] 6 items appropriate for current those who were cur rently receiving HSCT or had just completed the treatment will be excluded These 6 items include of treatment are worse than I imagined the bone marrow t ransplant by expert panel to better address issues previously reported by long term survivors. These 8 items This resulted in the 25 item modified FACT BMT scale used in the present study The MOS SF 36 is a generic HRQOL instrument comprised of 36 items designed to measure 8 domains (physical functioning, role limitations due to physical health problems role limitations due to emotional health proble ms social functioning, bodily pain, mental health, energy vitality, and general health perceptions ) In addition, two component scales ( Physical Component Summary (PCS) and Mental Component Summary (MCS) ) were created to summarize physical and mental asp ects of health. In

PAGE 46

46 this study, the PCS and MCS were treated as the main outcome variables. The scores of the PCS and MCS scale are normalized with a mean of 50 and a SD of 10 [64] The KPS scale is the clini per formance status based on a 0 100 scale [65] The performance status is determined ve work, Analytic Strategy Two phases of analysis were used : instrument refinement and instrument validation Instrument refinem ent was an iterative process using quantitative ( IRT methodology ) and qualitative information to select and remove items for the symptom scale. The instrument validation was conducted based on known groups validity which is determined by the extent to whi ch scale scores can distinguish between clinical known groups that are related to physical symptoms. Instrument R efinement A four st ep approach w as conducted to accommodate the iterative process for item selection The se st eps include : st ep 1 for exam ining the content of individual item s ; step 2 for inspecting the distribution of item response ; step 3 for testing IRT assumptions ( i.e., unidimensionality and local independence) ; step 4 for evaluating item level measurement properties based on IRT method s ( i.e., item discrimination item difficulty item characteristic curve, and differential item functioning ) I nstrument refinement is a reiterated p r ocess comprised of different stages of analyses. St eps 1 and 2 were implemented in Stage 1, and Steps 3 and 4 were implemented in multiple Stages until the results were satisfied (i.e., meeting certain level s of evaluation criteria)

PAGE 47

47 The following section describes the criteria used to evaluate the quality of items across different steps : St ep 1 : It em content was examined for its ability to assess physical symptoms and the content similarity to other items in the scale. We aimed to retain no more than 25 items measuring a unidimensional concept of physical symptoms for long term HSCT survivors. If an item had questionable content (e.g. not physical symptom) it was considered for removal at this step St ep 2: I tem response distributions were inspected for ceiling and floor effects. Ceiling effect refers to when the scores are at th e maximum value for the domain. Floor effect refers to when the scores are at the minimum possible value for the domain If 80% or more of respondents respond to the highest category or lowest category (highest category indicate no symptom, lowest indicate greatest symp tom) then the item was regarded as having ceiling or floor effects, respectively. In this study, we collapsed the item categories from 5 categories to 3 categories to address the skewed item distributions St ep 3: IRT assumptions (unidimensionality and local dependency) were tested b ased on a confirmatory factor analysis. I f the criteria of C omparative Fit Index ( CFI ) >0.95, Tucker Lewis Index (TLI) >0.95, and Root Mean Square Error of Approximation ( RMSEA ) <0.06, the assumption of unidimensionality is held [87 89] If the residual correlation among any pairs of items <0.2, the assumption of local independency is held [87] If items demonstrated residual correlations with each other, the item with the greater number of criteria violations was removed.

PAGE 48

48 St ep 4: IRT measurement properties (i.e., item discrimination item difficulty item characteristic curve, and differential item functioning ) were evaluated graded response model (GRM) which is a two parameter IRT model [90] The two parameters under estimation were the item discrimination and item difficulty. The item d iscrimination (slope a parameter ) in GRM describes how strongly an item is related to the underlying latent trait of physical symptoms or the degree to which the item discriminates between individuals along the latent trait. It is the slope where subjects endorse an item with a probability of 50% [12, 66] Higher discrimination values are better, with values greater than 1.0 being ideal, 2.0 indicating perfect discrimination Item difficulty (l ocation b parameter ) in GRM describes how easy or difficult it is for a subject to endorse an item for a concept the item intends to measure [12, 66] It is the location of items on the latent c ontinuum of physical symptoms where subjects endorse items with a probability of 50% with higher scores indicating better health. The GRM estimates the difficulty and discrimination parameters for each item and the underlying trait of physical symptom status (theta; ) for each subject. Given the five response categories for each item, the use of GRM estimates one slope parameter and four threshold parameters. The underlying latent trait ( 3 to +3 wit h higher scores indicating better health ( lower level of physical symptoms) In this study, a discrimination value less than 0.50 and/or a difficulty value greater than 2 standard deviations from the mean of 0 were the criteria applied to remove items [91] Item parameters are estimated using marginal maximum likelihood estimation v ia the expectation maximization (EM) algorithm. When an item had a discrimination

PAGE 49

49 parameter below the desired value and /or a difficulty parameter above the desired value [91] it was retained ( if this was the only violation identified for the item ) The item characteristic curve (ICC) describes the probability of endorsing an item conditioning (or regr essing) on the level of the underlying physical symptom trait [83] Easier items will have higher probabilities of being endorsed at the lower end of the latent trait (moving from left to right on trait scale, h igher levels on the right end). H arder items will have greater probability of being endorsed at the upper end of the latent trait [91] Differential item functioning (DIF) is an additional criterion to assist in inst rument refinement DIF analysis is a method that can be used to identify the items we intended to measure ( i.e., physical symptoms) are responded differently by sub groups (e.g. gender or treatment type) after controlling for the underlying latent trait o f physical symptom s [92, 93] In this study, we test ed for t he presence of DIF by gender or treatment type that can threaten the validity of an instrument The chi square tests of item locatio n and item slope contrasts were evaluated for significant DIF in each item C hi square values >10 were considered a minimal clinically important difference [94] Technical details of DIF methodology are available in Appendix A. The item information function (IIF) and test information function (TIF) were estimated at each stage to describe the reliability and measurement error at the item and scale levels, respectively [66, 95] IIF was estimate s to demonstrate the precision of measurement across different levels of the underlying latent trait. The more information the item provides, the more precisely the estimates will fall around the true ability. TIF was estimated to determine to what extent the unid i mensional scale can

PAGE 50

50 reliably measure different levels of underlying trait [66, 95] TIF is a summation of IIF for a given level of the un derlying trait. The value of TIF of 10 or higher (equivalent to was considered as a cutoff to determine measurement precision of the scales [91] Instrument V alidation Validation anal ysis for the revised scale of physical symptoms was conducted based on known groups validity which evaluate s the ability of the scale to discriminate between survivors with known levels of health status determined by other markers. H ealth status markers i nclude d the MOS SF 36 PCS and MCS, the KPS [65] presence of comorbid conditions, and severity of treatment experience (low, moderate, and high). For the SF 36, three modified cut offs f rom the standard PCS and MCS cut offs (<40 poor, 41 60 norm, >60 above norm) cut offs were used to accodomate the distribution of the scores in the sample ( PCS: <40 poor HRQOL, 41 50 n orm HRQOL and >50 above norm HRQOL; MCS: <45 for poor HRQOL 45 55 for norm HRQOL >5 5 for above norm HRQOL ) were used to facilitate analysis [96] For t he KPS because no standard cut offs are available we dichotomized the KPS scores by high and low values based on the median score (0 for performance s t arus and 1 00 for better performance status ). To avoid overestimating known groups validity of the instrument the sample was randomly split into two equal parts. The SAS procedure surveyselect with th e simple rando m sampling option was used to split the sample ( training and validation sample each at N=331). To evaluate known groups validity, linear regression was conducted to estimate the difference in latent physical symptom scores between survivors with high and low health status as measured by the other markers. In the regression analyses,

PAGE 51

51 the latent score s were used as dependent variable and health status marker s (known groups) as the independent variable. For h ealth status markers d ummy variables were create d the markers with more than two groups ( i .e. PCS, MCS, and severity of treatment experience). Two analytic strategies were performed: one without covariat e adjustment and another with covariate adjustment. The selection of c ovariates was based on the va riables that may confound the relationship between physical symptoms and health status markers including age at survey gender, race, marital status, education, and family income This procedure was conducted in both the training and validation samples. Ef difference in scores of the symptom scale for specific health markers Each effect size was calculated based on the difference in the mean physical symptom scores between differ ent categories of health markers divided by the pooled standard deviations of the mean scores Effect sizes <0.02 are classified as negligible, 0.2 0.49 as small, 0.5 0.79 as moderate and >0.8 as large [97] Effect s izes of 0.5 or greater are considered clinically important difference [98] Data management and known groups validity methods were conducted in SAS 9.2 [99] The CF A was c onducted using Mplus Version 7 [100] and the IRT procedures used PARSCALE Version 4 [101] Results Study Sample C haracteristics The mean age of participants at the time of the survey was 42 years old (SD=11) (Table 2 1) The majority of the sample was White (92%) and over half was female (62%). One third of the sample (32%) reported some college or technical education

PAGE 52

52 and another 30% reported a high school education or le ss. Approximately 70% of the participants reported annual family income s of $40,000 or more. Over half of the sample (60%) reported autologous HSCT with no chronic graft versus host disease (cGVHD) and 14% reported allogeneic transplant with cGVHD. Amon g the participants, 16% reported the presence of comorbid conditions at the time of the survey. Instrument R efinement Stage 1: Table 2 2 shows the results derived from this stage which includes 25 items for analyses T he dimensionality assessment using C FA for the initial 25 items was not fully satisfied (CFI=0.93 RMSEA=0.08). I tem response distributions were skewed left with some ceiling effects (>80% of responses in the highest category). In this ro und, fi ve items were removed: eping my job able to get around by myself difficulty Content issues were raised for BMT1 due to its lack of assessing physical symptom. BMT5 was flagged for ceili ng effects (CE=84.5%). BMT10 had local dependency ( residual correlation >0.20 ) able to concentrate BMT10 and BR1 items were conceptually similar but BMT10 demonstrated worse discrimination and difficulty parameters compared to BR1 P7 was identified to with ceiling effects (CE=87.5%) as well as local dependency with SCL3 and poor difficulty ( b = 3.25 ). Table 2 2 show items flagged for removal at each stage. At the end of this stage 20 out of 25 remained for analys e s at Stage 2. Stage 2 : Table 2 3 shows the results based on this stage. The dimensionality assessment based on CFA of the 20 items in this stage showed improved fit after the removal the first five i tems (CFI=0.96 RMSEA=0.06) (Table 2 2) However, the four

PAGE 53

53 items further removed included ability to have children Poor discrimination was noted for BL4 ( a =0.34 ) and BMT7 ( a= 0.12 ). BMT7, BMT14, and SCL2 had difficulty parameters outside of the ideal range ( b = 10.69 b = 2.63 b= 2.03 respectively). In addition, BL4 was identified with large magnitude of DIF by gender. At the end of this stage 16 ou t of 20 items remained for analyses at Stage 3 Stage 3 : Table 2 4 shows the results based on this stage The dimensionality assessment based on CFA of 16 items w as satisfied and consistent with that of the 20 item model (CFI=0.96 RMSEA=0.06). Three ite m s were further removed: BMT11 and SCL4 discrimination value of 0.50 and was also identified with DIF by treatment type. BMT15 was identified with significant DIF by treatment type and threshold values indicated it covered a limited level of the latent trait. The discrimination and difficulty of SCL4 fell below the acceptable criteria ( a =0.45 b =2.03). At the end of this stage 13 out of 16 items w as remained for analyses at Stage 4 Stage 4 : Table 2 5 Fit indices are improved from the initial set of 25 items (CFI=0.96, RMSEA=0.06). Item d ifficulties ranged between 1.15 and 2.19 SCL3 had lower discriminati on at this stage but had good item properties in each of the previous stages. It was reta ined due to clinical relevance, item properties are only marginally outside of the desired range, and the item fit p value did not indicate removal (p=0.97 ). SCL6 showed mild DIF by gender, but the

PAGE 54

54 magnit ude was low ( 2 <10). Figure B 1 in Appendix B displays the TIF for the final 13 item scale. Detailed results for each stage are also discussed in Appendix B Known Groups Validity R esults D emographic and clin ical characteristics between the training and validation samples were similar and no t significant ly differen t (Table 2 1 ). There were no significant differences between the mean ( M) latent symptoms scores of the training and validation samples (Table 2 5) T hose with higher levels of physical symptom s (lower latent scores indicate more symptoms) reported MCS scores in the lowest category (<45) Those reporting the least amount of symptoms were in the > 5 0 PCS category Table 2 6 displays the difference between the latent mean scores among the different categories of known groups as well as the respective effect sizes (ES) for the unadjusted and adjusted analyses in both sample s The largest ES were observed for the PCS known group s with high and low ph ysical health (adjusted ES =1.21 for the training sam ple and adjusted ES = 1.19 for the validation sample). The KPS demonstrated the lowest ES across both groups ( adjusted ES=0.02 for the training sample and ES=0.06 for the validation sample ). Severity of t reatment experience and comorbid conditions had moderate to negligible ES across both samples Discussion In this study, we used qualitative approach combined with IRT methodology to refine the FACT BMT to facilitate measur ing physical symptom s for long t erm HSCT survivors. We retained 13 (out of 25) items for a refined FACT BMT and validated this scale against health status known groups including clinical and HRQOL measures. The refined scale demonstrated good measurement properties and strong known gro ups

PAGE 55

55 validity related to physical and mental HRQOL measured by PCS and MCS scales K nown groups validity related to clinical markers were negligible or moderate Interpretation of final items : The majority of items in the refined FACT BMT had difficulty parameters clustered towards the lower end (or less severe) latent trait of physical symptoms with most having restricted threshold values (majority of thresholds between 2 .00 to 1 .00 ) The restricted threshold range is consistent with other PROs studies using IRT methods [102 105] meaning extant items (or item banks) measuring specific symptoms and functional status may only capture a limited range of the severity typically less severe /less challenge symptoms and functional status values of difficult y parameters thus capturing the highest level of the underlying t rait of physical symptom s Given higher scores in the physical symptom scale represent less symptom burden, the higher positive values of difficulty parameters mean these two items capture more challenge level of underlying physical symptom burden compare d to the other items Because item parameters based on IRT methodology are calibrated on the distribution of population, i t also implies that relative few long term HSCT survivors are able to achieve satisfaction with appearance of body and not getting tired easily compared to endorsing other items in our physical symptom scale In contrast, i tems with the lower values of difficulty parameters (e.g., headaches ) capture less challenge level of underlying physical symptom burden compared to the other items. Two items ( value of discrimination parameter compared to other items, implying these two items are

PAGE 56

56 able to discriminate bet ween participants separated by small differences in the underlying physical symptom s Items demonstrating the lowest discrimination ( SCL3 BMT12 not only contain the symptom content that is relevant to long term HSC T survivors, but also relevant to a decline in physical health of individuals in the process of aging Survivors may not perceive it as a HSCT treatment IRT findings compared with CTT findings : Nine items were co nsistent ly retained based on this IRT analysis (13 items total) and the previous CTT analysis (12 items total ) conducted by Huang and colleagues [84] The common items shared by these two studies included BMT6, BMT12 GF7, SCL1, BR1, SCL2, C6, B1, and SCL6. Items retained from this study that were not in the CTT analysis are C7, BMT16, BMT13, and SCL3. The CTT analysis included it was excluded from this study because of residual corre lation with BR1 able to and the measurement non invariance related to gender as identified by DIF test Effect sizes for detecting differences in clinical markers were lower from this IRT study using the latent mean scores compared to the o bserved mean scores from the CTT analysis For example, t he effect sizes for the KPS in this study were negligible the effect size for the difference in observed scores be tween different categories of comorbid condition status and the severity of treatment experience ranged from negligible to low in this IRT study ; whereas the magnitude was moderate in the CTT study.

PAGE 57

57 The differences between our study and the CTT study ma y be in part due to the measurement error s introduced by confounding of the summed observed scores with the sample charact eristics (sample dependence) based on CTT analysis The parameters for symptom items derived from IRT are independent of our sample a nd may be a better estimate of long term survivors compared to the use of summed obser ve d score s If the IRT scores are more accurate, higher ES in the CTT study may overestimate the ability of the items to distinguish between groups. S ymptom measures for H SCT survivors and clinical i mplication s : It was hypothesized that the re fined FACT BMT scale would moderately differentiate between the presence of comorbid conditions and severity of treatment experience as clinical markers. Our find ings meet this hypothesis given the fact that symptoms are more proximal to the disease process and are a summative indicator of the severity of disease [106] In addition s ymptoms are distinct from the concept of HRQOL that captures the impact of diseases and related symptoms [14] Unfortunately many instruments designed to measure PROs did not distinguish the concepts of symptoms from HRQOL. Although clinical practice oftentimes relies on clinical evaluation, such as presence of diseases, sign, laboratory data, and KPS rated by clinicians this strat egy tends to underestimate and is unable to capture the comprehensive picture of health status that includes burden of symptoms functional status, and HRQOL [15 17] The ref ined d FACT BMT provides a useful tool to assist clinicians to measure the burden of physi cal symptoms for long term HSCT survivors. Of note, the relationship of physical symptoms with HRQOL is not straightforward in H SCT survivors, and

PAGE 58

58 ps ychological sympt oms and psychosocial variabl es will play a role to influence the relationship. This topic is important and will be investigated in Chapter 3. The findings from this study suggest that in a long term survivor population, the refined FACT BMT scale i s more strongly related to HRQOL and is better able distinguish between survivors HRQOL status compared to their clinical characteristics such as severity of treatment experience or current comorbidities. This is consistent with previous literature suggest ing discordance between patient and physician reports of health status [107, 108] supporting evidence that symptoms measure a different aspect the items in this symptom instrument measured a restricted threshold range or restricted range of the severity of the latent trait the items were trying to measure. Qualitative approaches such as focus groups or interviews with survivors may help to dev elop new items for this population. Additionally, longitudinal studies will provide evidence for the reliability over time and sensitivity to change. Limitations Several limitations are apparent in this study. First, this study is cross sectional and no causal relationships can be inferred. Additionally, we are unable to establish the only one time point Second, the study may be subject to selection bias because stu dy participation may indicate that the survivor is functioning well compared to those who decline to participate due to poor health [109] This selection bias of a healthier sample indicates that we may underestimate the reporting of symptoms or overestimate the overall HRQOL status. Third, th e study population is largely homogenous (91.3%

PAGE 59

59 White). Therefore, r esults may not be generalized to populations comprised of different clinical and demographic characteristics as well as psychosocial and symptom burden Conclusion Modern test theory approaches provide unique advantages to help refine the FACT BMT scale for measuring physical symptoms The revised FACT BMT scale demonstrates good measurement properties and kno wn groups validity. Instrument development, evaluation, and revision are ongoing processes. While the items demonstrated adequate measurement properties and the ability to distinguish between other clinical markers of health status, the scale could be im proved through future studies. An instrument should have items geared to a variety of severity levels of the trait they are trying to measure. Including qualitative approaches as well as a longitudinal design to test the reliability over time and sensiti vity to change are necessary steps to improving the measurement properties for symptom scales

PAGE 60

60 Table 2 1. Study characteristics Overall Training Sample Validation sample N % N % N % Demographics Age, years 662 331 331 Mean (SD) 42 .1 (11) 48.9 (10) 49.4(10) Median (Range) 42.4(18 71) 49.6() 49.3() <35 182 28 29 8.8 37 11.2 35 39 95 14 40 12.1 31 9.4 40 44 106 16 45 13.6 45 13.6 45 49 120 18 56 15.9 65 19.6 >50 159 24 131 48.6 153 46.2 Sex Male 251 38 114 34.4 137 41.4 Female 411 62 217 65.6 194 58.6 Race White 603 92 304 92.1 299 90.9 Other 56 8 26 7.9 Education 658 High school or below 194 30 95 28.8 99 30.1 Some college or technical education 209 32 100 30.3 109 33.2 College degree 122 18 65 19.7 57 17.4 > College degree 133 20 70 21.2 63 19.2 Occupation Working 484 73 240 72.7 244 74.2 Not working 100 15 55 16.7 45 13.7 Retired 75 11 35 10.6 40 12.2 73 Marital status 659 330 Married/living with partner/ committed 483 7 3 237 71.8 246 74.8 Other 176 27 93 28.2 83 25.2 Annual family income < $20,000 70 11 37 11.5 33 10.1 $20,000 $40,000 141 22 62 19.3 79 24.2 $40,000 $60,000 156 24 77 23.9 79 24.2 $60,000 $80,000 100 15 52 16.2 48 14.7 >$80,000 181 28 94 29.2 87 26.7 Clinical variables Severity of treatment Low autologous no GVHD 390 60 197 60.8 193 59.9 Moderate allogeneic no GVHD 168 26 83 25.6 85 26.4 High allogeneic GVHD 88 14 44 13.6 44 13.7 Comorbid conditions Presence of comorbid conditions at survey 104 16 52 15.8 52 15.8 Karnofsky Performance Score at last follow up Mean (SD) 597 90.2(10.1) 90.2 (10.5) 90.1(9.8) Median (Range) 10 100 90( 10 100 ) 90( 10 100 )

PAGE 61

61 Table 2 1. C ontinued Overall Tr aining Sample V alidation sample N % N % N % PROs FACT BMT ^ score(observed) Mean (SD) 658 76 (13.9) 76.3 (14.6) 77.1(13.2) Median (Range) 16 100 80(16 90) 80(36 100) MOS SF 36 Physical component s ummary(PCS) Mean (SD ) 44.5 (11.6) 44.3 (11.6) 44.6(11.5) Median (Range) 6.4 64.5 47.4( 14.7 64.5 ) 47.9( 6.6 63.6 ) MOS SF 36 Mental component summary (MCS) Mean (SD) 50.6 (10.4) 50.6 (10.5) 50.7(10.3) Median (Range) 10.1 70.2 54.3( 10.1 70.2 ) 53.7( 18.8 69. 9 ) ^ FACT BMT scale is the 25 item modified version and was standardized to a T score with a mean of 50 and standard deviation of 10 on a 0 100 scale. MOS SF 36: Medical Outcomes Study 36 item short form health survey

PAGE 62

62 Table 2 2 Criteria for item removal based on content, ceiling effects, local dependency, IRT properties, and differential item functioning: Stage 1 (25 items) Item Variable content Content* CE* LD^ a b ICC DIF BMT1 Concern about keeping my job X X X BMT2 Feel distant from other p eople C6 I have a good appetite C7 Like the appearance of my body BMT5 Be able to get around by myself X X X BMT6 Get tired easily BL4 Be interested in sex BMT7 Concern about my ability to have children BMT10 Can remember things X X X X X BR1 Be able to concentrate BMT11 Have frequent colds/infections BMT12 Eyesight is blurry BMT13 Be bothered by a change in the way food tastes BMT14 Have tremors B1 Have been s hort of breath BMT15 Be bothered by skin problems BMT16 Have trouble with my bowels SCL6 Have headaches SCL1 Have dizzy spells SCL2 Have stiff joints P7 Have urinating difficulty X X X SCL3 Have hearing loss SCL4 Have sleep problems SCL5 Have mouth sores X X GF7 Be content with quality of my life right now Overall model fit indices CFI RMSEA 0.93 0.08 *Content and ceiling effect (CE) are descriptive criteria used in the f irst stage of item selection. Ceil ing effect: over 80% of the respondents endorse the category that indicates the greatest health status. ^LD (Local dependency): residual co rrelation between this item and a nother (cut off at r= 0.20); a (slope): how strongl y the item is related to the domain of physical symptoms (higher values are better)b (location): severity level the item intends to capture (metric: 3 to +3); ICC (Item characteristic curve) :describes probability of endorsing an item conditioning on the level of latent trait; DIF ( Differential item functioning ) : extent to which an item may be answered differently by a subgroup (e.g. gender, treatment type) given the same level of latent trait (or severity) of physical symptoms.

PAGE 63

63 Table 2 3 Criteria for item removal based on content, ceiling effects, local dependency, IRT properties, and differential item fun ctioning: Stage 2 (20 items) Item Variable content Content* CE* LD^ a b ICC DIF BMT1 Concern about keeping my job BMT2 Feel distant fr om other people C6 I have a good appetite C7 Like the appearance of my body BMT5 Be able to get around by myself BMT6 Get tired easily BL4 Be interested in sex X X BMT7 Concern about my ability to have childre n X X X BMT10 Can remember things BR1 Be able to concentrate BMT11 Have frequent colds/infections BMT12 Eyesight is blurry BMT13 Be bothered by a change in the way food tastes BMT14 Have tremors X X X B1 H ave been short of breath BMT15 Be bothered by skin problems BMT16 Have trouble with my bowels SCL6 Have headaches SCL1 Have dizzy spells SCL2 Have stiff joints X X X P7 Have urinating difficulty SCL3 Hav e hearing loss SCL4 Have sleep problems SCL5 Have mouth sores GF7 Be content with quality of my life right now Overall model fit indices CFI RMSEA 0.96 0.06 *Content and ceiling effect (CE) are descriptive criteria used in the first stage of item selection. Ceil ing effect: over 80% of the respondents endorse the category that indicates the greatest health status. ^LD (Local dependency): residual co rrelation between this item and a nother (cut off at r= 0.20); a (slope): how strongly the item is related to the domain of physical symptoms (higher values are better)b (location): severity level the item intends to capture (metric: 3 to +3); ICC (Item characteristic curve) :describes probability of endorsing an item conditionin g on the level of latent trait; DIF ( Differential item functioning ) : extent to which an item may be answered differently by a subgroup (e.g. gender, treatment type) given the same level of latent trait (or severity) of physical symptoms.

PAGE 64

64 Table 2 4 Crite ria for item removal based on content, ceiling effects, local dependency, IRT properties, and differential item functioning: S tage 3 (16 items ) Item Variable content Content* CE* LD^ a b ICC DIF BMT1 Concern about keeping my job BMT2 Feel dis tant from other people C6 I have a good appetite C7 Like the appearance of my body BMT5 Be able to get around by myself BMT6 Get tired easily BL4 Be interested in sex BMT7 Concern about my ability to have ch ildren BMT10 Can remember things BR1 Be able to concentrate BMT11 Have frequent colds/infections X X BMT12 Eyesight is blurry BMT13 Be bothered by a change in the way food tastes BMT14 Have tremors B1 Have been short of breath BMT15 Be bothered by skin problems X X BMT16 Have trouble with my bowels SCL6 Have headaches SCL1 Have dizzy spells SCL2 Have stiff joints P7 Have urinating difficulty SCL3 Hav e hearing loss SCL4 Have sleep problems X X SCL5 Have mouth sores GF7 Be content with quality of my life right now Overall model fit indices CFI RMSEA 0.96 0.06 *Content and ceiling effect (CE) are descriptive criteria use d in the first stage of item selection. Ceil ing effect: over 80% of the respondents endorse the category that indicates the greatest health status. ^LD ( Local dependency ) : residual co rrelation between this item and a nother (cut off at r= 0.20); a (slope): h ow strongly the item is related to the domain of physical symptoms (higher values are better)b (location): severity level the item intends to capture (metric: 3 to +3); ICC (Item characteristic curve) :describes probability of endorsing an item condition ing on the level of latent trait ; DIF ( Differential item functioning ) : extent to which an item may be answered differently by a subgroup (e.g. gender, treatment type) given the same level of latent trait (or severity) of physical symptoms.

PAGE 65

65 T able 2 5 S tage 4 f inal item set measurement properties (13 items) Items Contents Slope ( a )* Threshold Location ( b ) ** Threshold 2 Item fit p C7 Like the appearance of my body 0.60 0.73 1.15 3.02 0.02 BMT6 Get tired easily 1.16 0.6 0 0.34 1.29 0.38 BMT1 2 Eyesight is blurry 0.6 8 1.65 0.60 0.44 0.16 GF7 Be content with quality of my life right now 0.73 1.90 0.62 0.66 0.80 SCL1 Have dizzy spells 0.90 1.67 0.90 0.14 0.14 BMT16 Have trouble with my bowels 0.77 1.91 0.97 0.03 0.65 BR1 Be able to concentrate 0.74 2.43 0.99 0.46 0.1 7 BMT2 Feel distant from other people 0.73 2.33 1.17 0.01 0.25 BMT13 Be bothered by a change in the way food tastes 0.86 1.86 1.18 0.4 5 0.36 C6 I have a good appetite 0.68 2.62 1.40 0.17 0.18 B1 Have been short of breath 0.61 2.53 1.46 0.40 0.07 SCL6# Have headaches 1.12 2.34 1.57 0.81 0.80 SCL3 Have hearing loss 0.41 3.58 2.19 0.81 0.97 Model fit statistics CFI RMSEA 2 (df) 0.96 0.06 347 (104) *Slope (a): how strongly the item is related to the domain of physical symptoms (higher values are better) : severity of an item response or location fo r that response category (i.e. three response categories mak es two thresholds). **Location (b): severity level the item intends to capture (metric: 3 to +3); values indicate worse fit. # Di fferential item functioning (DIF): extent to which an item may be r answered differently by a subgroup given the same level of latent trait (or severity) of physical symptoms. DIF by gender was found for SCL6 but the location contrast was mild (0.61) H olding the latent trait constant, females endorsed headaches slightly more than males No DIF by treatment type in final item set.

PAGE 66

66 Table 2 6 Unadjusted and adjusted mean latent HSCT symptom scores by known groups Training Sample 1 N=331 Validation Sample 2 N=331 N Unadjusted Mean (SD) Adjusted Mean (SE) N Unadjusted Mean (SD) Adjusted Mean (SE) PROs MO S SF 36 PCS Poor <=40 141 0.62 (0.91 ) 0.52 (0.28 ) 139 0.53 (0.75 ) 0.32 (0 .23 ) Norm 41 50 125 0.27 (0.69 ) 0.31 (0.28 ) 125 0.35 (0.68 ) 0.49 (0.25 ) Above norm >50 63 0.61 (0.69 ) 0.69 (0.29 ) 65 0.63 (0.76 ) 0.78 (0.25 ) MOS SF 36 MCS Poor <=45 84 0.88 (0.85 ) 0.80 (0.27 ) 77 0.76 (0.71 ) 0.61 (0.2 4 ) Norm 45 55 104 0.10 (0.76 ) 0.16 (0.27 ) 106 0.04 (0.70 ) 0.09 (0.25 ) Above norm >55 141 0.50 (0.70 ) 0.49 (0.27 ) 146 0.51 (0.75 ) 0.55 (0.24 ) Clinical indicator Karn ofsky Performance Score^ Low <=90 109 0.09 (0.98 ) 0.27 (0.35 ) 98 0.05 (0.86 ) 0.06 (0.31 ) High >90 196 0.0 1 (0.90 ) 0.24 ( 0.37 ) 194 0.08 (0.90 ) 0.14 (0.30 ) Comorbid conditions at time of survey Yes 52 0.64 (0.94 ) 0.55 ( 0.32 ) 52 0.41 ( 0.90 ) 0.27 (0.29 ) No 276 0.07 (0.89 ) 0.07 (0.31 ) 276 0.12 (0. 90) 0.11 (0.27 ) Severity of HSCT treatment experience Low (autologous, no cGVHD) 197 0.03 (0.87 ) 0.10 (0.32 ) 193 0.11 (0.83 ) 0.11 (0.28 ) Moderate (allogeneic, no cGVHD) 83 0.15 (0.93 ) 0.0 2 (0.32 ) 85 0.02 (0.89 ) 0.15 (0.29 ) High (allogeneic, cGVHD) 44 0.37 (1.16 ) 0.50 (0.34 ) 44 0.25 (0.96 ) 0.15 (0.29 ) Adjusted for current age, gender, race, marital status, highest education attained, and family income. ^PCS and MCS categories are based on revised cutoffs to the standard ized cut off values of <40, 41 50, 5 0+, to accommodate the distribution in our samples ; KPS is based on median split (median=90) as there is no standardized cutoff

PAGE 67

67 Table 2 7 Difference in late nt mean scores a nd effect sizes for known groups in random sample splits Training Sample n=331 Validation Sample 2 n=331 Difference 1 ( ES ) t statistic (p value) Difference 2 ( ES ) t statistic (p value) Difference 1 ( ES ) t statistic Difference 2 ( ES ) t statistic PROs MOS SF 36 PCS^ Above norm vs p oor(ref) 1.220 (1.29) 10.16 (<0.0001) 1.200 (1.21) 9.51 (<.0001) 1.162 (1.33) 10.65 (<.0001) 1.10 (1.19) 9.50 (<.0001) Norm vs p oor(ref) 0.885 (1.16) 9.09 (<0.0001) 0.820 (1.06) 8.28 (<.0001) 0.886 (0.89) 9.89 (<.0001) 0.810 (0.76) 8.37 (<.0001) Above norm vs n orm(ref) 0.335 (0.25) 2.74 (0.0065) 0.380 (0.28) 3.03 (0.002) 0.276 (0.31) 2.49 (0.0132) 0.290 (0.33) 2.60 (0.010) MOS SF 36 MCS^ Above norm vs p oor(ref) 1 .378 (1.12) 13.09 (<.0001) 1.293 (1.05) 12.22 (<.0001) 1.267 (1.05) 12.39 (<0.001) 1.17 (0.95) 11.17 (<.0001) Norm vs p oor(ref) 0.783 (0.70) 6.99 (<.0001) 0.645 (0.58) 5.76 (<.0001) 0.721 (0.66) 6.64 (<.0001) 0.706 (0.64) 6.42 (<.0001) Above norm vs N orm(ref) 0.596 (0.52) 6.05 (<.0001) 0.648 (0.57) 6.65 (<0.001) 0.546 (0.50) 5.92 (<.0001) 0.460 (0.41) 4.89 (<.0001) Clinical indicators Karnofsky Performance Score^ High vs. l ow (ref) 0.076 (0.05) 0.67 (0.503) 0.028 (0.02) 0.25 (0.802) 0.130 ( 0.09) 1.18 (0.240) 0.086 (0.06) 0.80 (0.424) Comorbid conditions at time of survey No condition vs p resence of conditions (ref) 0.711 (0.33) 5.22 (<.0001) 0.620 (0.29) 4.47 (<0.0001) 0.533 (0.26) 4.10 (<.0001) 0.379 (0.19) 2.90 (0.004) Severity of HSCT treatment Low severity vs m oderate(ref) 0.177 (0.16) 1.45 (0.149) 0.081 (0.07) 0.65 (0.519) 0.090 (0.09) 0.80 (0.425) 0.031 (0.03) 0.27 (0.785) Low severity vs High s everity (ref) 0.339 (0.33) 2.19 (0.029) 0.401 (0.38) 2 .52 (0.012) 0.351 (0.37) 2.41 (0.0167) 0.263 (0.28) 1.82 (0.069) Modera te vs high s everity(ref) 0.517 (0.46) 2.97 (0.003) 0.482 (0.42) 2.77 (0.006) 0.261 (0.25) 1.61 (.1089) 0.294 (0.28) 1.84 (0.066) Effect size <0.2 as negligible, 0.2 0.49 as small, 0.5 0.79 as moderate, and >0.8 as large; ( 0.5 or greater is considered as a clinically important difference ). Adjusted for current age, gender, race, marital status, and highest education attained, and income. ^PCS and MCS categories are based on revise d cutoffs to the st andardized cut off values of < 4 0 41 50 50 +, to accommodate distribution; KPS is based on median split (median=90) as there is no standardized cutoff

PAGE 68

68 CHAPTER 3 EXAMINING THE RELATIONSHIP BETWEEN SYMPTOMS, PSYCHOSOCIAL FACTORS AND HEAL TH RELAT ED QUALITY OF LIFE IN HEMATOPOIE TIC STEM CELL TRANSPLANT SURVIVORS Introduction The increased use and success of hematopoietic stem cell transplantation (HSCT) has shifted the research focus from how long cancer survivors are living to how well the y are living [110] All long term HSCT survivors are at risk for increased mortality [4, 5] chronic conditions or other subsequent long term effec t s that may reported outcomes (PROs) such as symptoms, functional status and/or health rel ated quality of life (HRQOL) [46, 111, 112 ] While the majority of survivors report a health status similar to pre treatment or comparable to the general population, a portion of survivors have persistent adverse effects (or late effect) as a results of cancer and treatment [113] The impact of HSCT leaves survivors at risk for physical (i.e. pain, sleep problems) and psychological symptoms that contribute to decreased mental and physical aspects of HRQOL [26, 38, 112] Estimates of emotional distress, such as depression and posttraumatic stress symptoms (PTSS) are varied in the literature. A review by Mosher and colleagues report ed prevalence estimates of diag nosed posttraumatic stress disorder (PTSD) in HSCT survivors range from 5 to 19% [42] Estimates of any PTSS in HSCT survivors are as high as 50% [114, 115] Estimates of depression in HSCT are between 25% and 50% [21, 44] and are higher than the general cancer survivors (10% to 25%) [116] Man y studies have focused on the impact of treatment factors and other clinical or demographic characteristics on HRQOL [117 119] However, these factors are beyond the point of intervention to improve HRQOL [40, 120] Physical and

PAGE 69

69 psychological symptoms as well as other psychosocial factors, such as coping or social constraints, may be viable points for targeting interventions [40, 121] The relationship among physical symptoms, psychological symptoms, and HRQOL is complex and may contribute to the heterogeneous findings on health outc omes among survivors. Survivors with similar physical symptom reports can have differing HRQOL outcomes [42] Unfortunately, m ost studies lack a conceptual framework to help test the relationship between d ifferent factors and HRQOL in HSCT survivors [42, 122] The heterogeneous findings may be a result of true differences in survivor outcomes between studies, the selection of different f a ctors (i.e. psychological symptoms, coping strategies) and /or the hypothesized role of these factors (i.e. direction of influence) in influencing HRQOL vary from study to study making comparisons difficult and potentially leading to the lack of conclusive evidence. R esearch on the interrelationships between the factors themselves is sparse but is needed to begin mapping out the long term effects of HSCT [42] Using a structural equation modeling (SEM) appro ach to test pathways enables examination of the direct and indirect effects (or pathways) among variables within the framework to facilitate evaluation and interpretation of PROs in HSCT survivors [123] Direct effects are the direct relationships between the variables. Indirect effects are the effects of one variable on another through a mediating variable. A total effect is the sum of the direct and indirect effects. Another advantage of the SEM approach is the abil ity to incorporate unobserved or latent variables.

PAGE 70

70 Physical symptoms psychological symptom HRQOL: A robust conceptual framework is required to carefully model the complex relationships between variables because multiple factors may play a role in influenc ing HRQOL, (Figure 3 1). Specifically, there are several possible pathways from physical symptoms through psychological symptoms to physical and mental HRQOL (Figure 3 1A). Identifying and quantifying these pathways may provide clarification on the role that physical symptoms and psychological symptoms play in influencing physical and mental HRQOL. Determining whether psychological symptoms, such as depressive symptoms and PTSS mediate the effect from physical symptoms to mental and physical is HRQOL is important when considering which patient factors are significant points of intervention to improve survivor HRQOL. Framework including psychosocial factors : While physical and psychological symptoms are hypothesized to have the most direct impact on HRQOL [26] other factors that might influenc e HRQOL include a survivor social constraints and approach to coping [124 126] Figure 3 1B display s the hypothesized conceptual pathway from physical symptoms to physical and mental HRQOL and includes the relationships of depressive symptoms and psychosocial factors These psychosocial factors were associated w ith PROs in HSCT survivor in previous studies [26] Literature also suggests these factors may act as mediating or moderating factors on the relationship between symptoms and HRQOL [47] Specifically, c oping was identified as a mediator to buffer the negative effect of physical symptoms or disease status on HRQOL. Individuals with h igher dispositional optimism have not only been associated with better physical symptom repo rts [127] but also

PAGE 71

71 related to better physical and mental HRQOL compared to those with lower dispositional optimism [128] One study on lung and liver cancer patients repor ted that great optimism and sense of coherence predicted better mental health status 12 months post transplant [129] The unique aspect of this study was to incorporate include multiple physical, psychological, and psychosocial factors in conceptual models to test the hypothesized relationships among important factors on HRQOL while controlling for demographic and clinical characteristics. The primary objective was to evaluate whether the presence of psychological symptoms (depressive symptoms and PTSS) mediates the effect of physical symptoms on HRQOL outcomes in long term survivors of HSCT through path analysis. The secondary objective was to test the overall conceptual model of factors in contributing to HRQOL, which include psychosocial factors such as personality, social resources, and coping methods in addition to physical and psychological symptoms. Specifically, th is objective is to identify pathways through which physical symptoms, depressive symptoms and psychosocial variables influence physical and mental HRQOL. Methods Participants and Data Collection This study used secondary data derived from a large multi site study (40 centers) comprised of long term cancer/HSCT survivors. A detailed description of data collection has been previously described (Chapter 1, Study Design ). Potential participants were HSCT recipients at participating transplantation centers with records identified from International Bone Marrow Transplant Registry/Autologous Blood and Marrow Transplant Registry (IBMTR/ABMTR). Eligibility criteria included the age at 18 years or

PAGE 72

72 older, a single allogeneic or autologous HSCT, at least 12 months post HSCT, specific diagnosis of cancer (chronic myelogous leukemia, acute leukemia, lymphoma, or breast cancer), continuous remission since HSCT, and able to read and understand English. This study randomly selected 1,946 out of 2,447 eligible survivors for recruitment The flow chart in Chapter 1, Figure 1 1 displays the recruitment process. Que stionnaires were completed through mailing and phone interview ; medical records were abstracted from IBMTR/ABMTR to determine the type of initial diagnosis, type of HSCT, and the nature of donor relationship (allogeneic). Study Measures HRQOL outcomes : HRQOL outcome measures were two component scales ( Physical Component Summary (PCS) and Mental Component Summary (MCS) ) of the Medical Outcomes Study 36 item s hort f orm health survey (MOS SF 36) [64] The MOS SF 36 is a generic HRQOL measurement with 36 items designed to measure 8 domains (physical functioning, role limitations due to physical problems role limitations due to emotional problems social functioning, bodily pain, mental health, energy vitality, and general health perceptions. The PCS and MCS are component measures of the physical and mental aspects of HRQOL. In this study, PCS was composed of items measuring physical functioning, role limitations due to physical health problems, and bodily pain MCS was comprised of items measuring role limitations due to emotional health problems social functioning, energy/vitality, and mental health PCS and MCS were treated as the main outcome variables for the SEM. The scores of the PCS and MCS scale are normalized with a mean of 50 and a SD of 10 [64] The PCS and MCS are scored so higher scores indicate better health. Measurement properties are available for all instruments in Appendix C Table C 1.

PAGE 73

73 P hysica l symptoms : Physical symptoms related to HSCT were measured using the modified version of the Functional Assessment of Cancer Therapy Bone Marrow Transplant (FACT BMT) [37] This modified scale is comprised of 13 item s that capture physical symptoms in long term HSCT survivors. Chapter 2 provides the process of scale development and validation. Higher scores indicate worse symptoms This scale comprised of 13 items to measure a unidimensional latent concept of ph ysical symptoms. Psychological symptoms : The Impact of Events Scale ( IES ) was used to measure PTSS [130] The IES [130] is comprised of 15 items measuring two concepts/domain s: intrusion (7 items) and avoidance (8 items) The IES used in this study was developed prior to the creation of the post traumatic stress disorder (PTSD) diagnosis in the Diagnostic and Statistical Manual of Mental Disorders IV (DSM IV ) [131] The IES used in this study has been widely used for many types of traumatic events (e.g., disasters, assaults ) [130] The scale can discriminate between individuals with mild to severe stress responses and retains good reliability over time despite the lack of hyperarousal domain [132] Domain scores were calcul ated by summing the item responses to create a continuous score. Higher scores indicate worse PTSS. Depressi ve symptoms were assessed using the Center for Epidemiologic Studies Depression (CES D) [133] T he CES D scale has 20 items and is comprised of four domains: depressed affect (7 items), somatic complaints (7 items), interpersonal difficulties (2 items), and low positive affect (4 items). Respondents are asked to indicate how often the depressive sym ptom is experienced in the past week based on a

PAGE 74

74 4 Domain scores were calculated by summing the item responses within each domain. Higher scores indicate worse depressive symptoms. Psychosocial variables : The Life Orientation Test (LOT) is a unidimensional scale (12 items) measur ing pessimism [127] Respondents are asked to indicate their agreement with the statement for 12 items based on a L ikert type scale of 5 response categories ranging The LOT is scored so higher scores indicate greater optimism. This scale comprised of 12 items to measure a unidimensional latent concept of optimism The Social Constraints Scale (SCS) is a 16 item unidimensional scale measuring the degree to which individuals feel their social relationships are strained and the degree to which they feel constrained in discussing their trauma related thoughts (HSCT related thou ghts) at the time of the survey [134] The SCS is scored so higher scores indicate more social constraint. This scale comprised of 16 items to measure a unidimensional latent concept of social constraints The Brief COPE is a well validated 28 item scale designed to measure coping strategies in stressful situations [135] The full form of the COPE is c omprised of 60 items [136] In this study, a short form of the COPE was used that measures 14 domains ( 2 items each) of coping strategy. The 14 domains of the Brief COPE include 1 ) active coping, 2) planning, 3) us ing instrumental support, 4) using emotional support, 5) venting, 6) behavioral disengagement, 7) self distraction, 8) self blame, 9) positive reframing, 10) humor, 11) denial, 12) acceptance, 13) religion, and 14) substance use

PAGE 75

75 Items are scored on 4 poi Acknowledging the limitations of the factor analysis for this scale (high residual correlations among the sub domains comprised of two items) the developer recomm ended researchers use the items of the BRIEF COPE in a flexible manner and select specific sub scales or combi ne sub scales [135, 137, 138] For this study, we used an exploratory factor analysis (EFA) as well as findings from previous studies that have combined sub domains of the BRIEF COPE [137 139] to select the domains for this study The main domains found from our EFA were problem solving/ approach coping (7 items) and avoidant/maladaptive coping (7 items) and demonstrated adequate fit statistics Comparative Fit Index ( CFI ) =0.92; Root Mean Square Error of Approximation ( RM SEA ) =0.10) These two domains and items within each are consistent with previous studies two domain breakdown of coping strategies [137 139] The two domains were ca lculated by summing the item responses in each domain and were scored so higher scores indicate better coping. Covariates: Several factors that significantly contribute to physical and psychological symptoms as well as PCS and MCS were included in the sta tistical analyses. Important demographic characteristics to be considered include current age, gender, occupational status, education, and marital status [27, 42, 43, 109] Survivors with f emale gender, lower income, lower education level, and being single or non married are predictors of reporting poor physical and mental functioning status than those with male gender, higher income, higher educational level, and being married [30, 42, 140] Survivors with these characteristics ( i.e., female gender, lower income and education) tend to report co occu rrence of greater distress and greater positive

PAGE 76

76 psychological impacts from the treatment [30, 42, 140] In addition, older survivors reported great er impairment i n physical health than younger survivors [43] while younger surv ivors reported significant PTSS than older survivors [28] C linical variables include time since HSCT ( unit: years), severity of treatment experience and comorbid conditions [27, 42, 43, 109] Severity of treatment experience was defi ned as low severity for autologous and no cGVHD ; moderate severity for allogeneic and no cGVHD ; high severity for allogeneic and cGVHD The presence of comorbid conditions at the time of the survey was dichotomized to yes/no. A longer t ime since HSCT has been identified as a significant predictor of better PROs such as improved physical functioning but its relationship with PROs is not always linear. Some studies show PROs of survivors were improve d continuously over time since the treatment [18] and then reach ed a plateau after 2 years [40, 45] Other studies show PROs of survivors were improved immediately (typically physical functioning) in t he first year, and then declined subsequent ly [41] Higher severity of HSCT treatment experience as reflected by type of treatment has been associated with poor physical functio nal status [26] Some studies demonstrated null relationships between severity and HRQOL with the authors noting that additional factors (psychological symptoms and psychosocial factors) not measured in the study may likely mediate the outcome [42] Hypothesis For the primary objective, we hypothesized that physical symptoms had a direct effect on both physical and mental HRQOL as well as an indirect effect through psychological symptoms (depressive sy mptoms and PTSS) We also hypothesized that the psychological symptoms play a stronger mediating role for mental HRQOL than

PAGE 77

77 physical HRQOL. Our conceptual pathway for the primary objective is presented in Figure 3 1A. For the secondary objective, we ass ume that the psychosocial variables (optimism, coping, social constraints) will explain the variation in the pathway from physical symptoms through depressive symptoms to physical HRQOL (Figure 3 1B). Specifically, we hypothesize that the pathway from phys ical symptom coping physical HRQOL will explain the physical symptom depressive symptoms physical HRQOL relationship. For mental HRQOL, we hypothesize that the same pathways ( physical symptom coping mental HRQOL ) will not play as significant a role and de pressive symptoms will remain an important mediator. Our conceptual framework fo r the secondary objective is presented in Figure 3 1 B. Analytic Strategy Symptom HRQOL pathways: The main pathways (physical symptom HRQOL, physical symptom PTSS depressive symptom HRQOL, physical symptom depressive symptom HRQOL, and physical symptom PTSS HRQOL) were specified in a diagram followed by identification and evaluation of the measurement and structural portion of the SEM [123, 141] The measurement portion of the SEM is comprised of the relationship between measured indicators (i.e., items of the scales) and their respective latent constructs /latent variables the scales intend to measure The structu ral portion of the SEM is comprised of the relationships between different latent variables, which also serve as a foundation for testing the pathways between different latent variables. As displayed in Figure 3 1, l atent variables are represented in circl es (symptoms and HRQOL) and observed variables are displayed in boxes. The covariates in this study were observed variables hypothesized to be precursors to both health outcomes and mediators including demographic characteristics and clinical variables.

PAGE 78

78 In this study, the observed and latent domain scores of different scales were used to build SEM given a great number and the complexity of measurement scales. For multidimensional scales, including CES D, PTSS, BRIEF COPE, PCS and MCS, the observed domai n scores were treated as measured indicators for the respective latent factor and included in SEM. In contrast, for unidimensional factors, including the modified FACT BMT scale, LOT, and SCS, individual items w ere treated as measured indicators and inclu ded in SEM. To assess the model improvement for the measurement portion of the model, we used the modification indices (MIs), which are the predicted decrease in the model chi square 2 ) value after allowing for the relationship to be freely estimated between two variables Using the largest MI first, items (or domains) measuring the same latent construct were allowed to be freely estimated; for example, items within the physical sy mptom instrument were allowed to be freely correlated with each other, but not with items in the depressive symptom scale. For the path analysis section of this study, the default parameterization of Mplus was used, with the first indicator variable fixed to 1 and the variance of the latent variable freely estimated. Overall fit indices for the measurement model included the Comparative Fit Index (CFI), Root Mean Square Error of Approximation (RMSEA), and an examination of the residual correlations betwee n items or domains measuring the same latent construct [87, 142] acceptable [87, 142] [143] Residual correlations among items or domains measuring the same latent construct <0.20 were considered acceptable to meet the as sumption of local independence [87]

PAGE 79

79 To test the pathways through which depressive symptoms and PTSS influence the relationship between physical symptoms and physical and mental HRQOL, t he full pathw ays in the concep tual framework (or baseline models) were evaluated for physical HRQOL (Figure 3 2A ) and mental HRQOL (Figure 3 3A ) separately based on t he structural portion of the SEM The 2 statistic, CF I RMSEA, and residual correlations. The first pathway to be tested is the relationship between physical symptoms, PTSS, depressi ve symptoms and physical and mental HRQOL (Figure 3 2A and Figure 3 3A ). T he fit indices, direct, indirect, and total effect effects were obtained for the desired pathways in the first model. The second path model specifically tested the mediating effect of depressi ve symptoms on the relationship between physical symptom s and p hysical HRQOL (Figure 3 2B) as well as the relationship between physical symptom s and mental HRQOL (Figure 3 3B). The third path model specifically tested the mediating effect of PTSS on the relationship between physical symptom s and physical HRQOL (Figur e 3 2C) as well as the relationship between physical symptom s and mental HRQOL (Figure 3 3C). To conduct the second model which is nested within the first model, all relationships with depressive symptoms were constrained to zero. The same procedure was done for the third model (i.e., PTSS nested model). The two nested models were individually tested against the full path model usi ng the Mplus DIFFTEST procedure for fit comparisons. The DIFFTEST procedure i 2 difference test that examine if the nested, or more restrictive model, has worse fitting function and fewer parameters than the full model. Significant p values indicate that the nested model is worse fitting compared to the full model.

PAGE 80

80 Direct, ind irect, and total effects were assessed using the MPlus Model INDIRECT command. Unstandardized (b) and standardized ( b ) path estimates were estimated for each latent variable. The estimates were standardized using the variances of the continuous latent va riables as well as the covariates and outcome variables. All analyses adjusted for covariates ( marital status, education, occupational status, years since HSCT and severity of treatment experience ) Symptom psychoso cial HRQOL framework : Important psychosocial variables including optimism, social constraints, and coping were included to better understand their influences on the relationships among depressive symptoms, physical symptoms, and HRQOL outcomes. Depressive symptoms were selected as the measure of psychological health given the significant findings on the relationship between physical symptoms and HRQOL. In addition, literature supports depressive symptoms were associated with the psychosocial variables such as coping and social constraints [134, 144] To test the pathways through which psychosocial variables influence the relationships of physical and depressive symptoms with physical and mental HRQOL, t he full pathw ays in the conceptual framework (or baseline models) were evaluated for physical HRQOL (Figure 3 4A) and mental HRQOL (Figure 3 5A ) separately based on t he structural portion of the SEM. The measurement portion of the model wa s evaluated using the same criteria for CFI and RMSEA addressed in the previous section. To assess the model improvement, the MIs were used to iteratively add item or domain correlations within the same latent variable [145]

PAGE 81

81 The full baseline model s with physical HRQOL (Figure 3 4A) and mental HRQOL (Figure 3 5A) as outcome variables were tested first. Iteratively, the r elationships among symptoms and psychosocial variables were constrained to zero in the sub sequent models if the relationships were not statistically significant in the previous models. The DIFFTEST procedure was used to assist in determining if the constrained model was significantly worse from the previous model s with a chi square test [145] The model s were iteratively modified based on the literature and the improvement in the overall fit statistics to establish the most parsimonious and theoretically reasonable model. Unstandardized and standar dized coefficients were estimated for each latent variable in both physical and mental HRQOL models The model INDIRECT command was used to obtain direct, indirect, and total effect information for the pathways from physical symptom and coping. All model s were adjusted for covariates that included the years since HSCT and severity of treatment experience. Mplus version 7 [100] was used with the mean and variance adjusted weight least squares (WLSMV) robust estimator [145] Results S tudy Sample C haracteristics The mean age of participants at the time of the survey was 42 years old (SD=11), the majority of the sample was White (92%) and over half was female (62%) (Table 3 1). One third of the sample (32%) reported some college or technical education and another 30% reported a high school education or less. Approximately 73% of the survivors reported a working or student occupational status. The mean number of years

PAGE 82

82 since HSCT was 7 years with a range of 2 to 22 years. The majority reported low severity of treatment experience (60%) and a less intense previous treatment (66%). Among the participants, 16% reported the presence of comorbid conditions at the time of the survey. Due to missing data on covariates, 45 (6.8%) cases were excluded from the analysis, leaving a sample of 617 observations for the path analysis and framew ork testing. Bivariate Correlations The correlations in Table 3 2 display the bivariate associations among latent variables of interests. Physical symptoms were strongly correlated with depressive symptoms (r=0.85), mental HRQOL (r= 0.88), and physical HRQOL (r= 0.80). Depressive symptoms were strongly correlated with mental HRQOL (r= 0.97), coping (r= 0.74), optimism (r= 0.69), and physical HRQOL (r= 0.66). In addition, optimism was strongly correlated with mental HRQOL (r=0.63). The magnitudes of th e remaining pair wise correlations were moderate (0.4 < r < 0.6) or small (r<0.4). Pathways and Framework Results Symptom p hysical HRQOL p athways : Figure 3 2A displays associations between physical symptoms, depressive symptoms PTSS, and physical HRQ OL based on the full path model The unstandardized and standardized results of the full path model indicate that physical symptoms was significantly associated with physical HRQOL ( b = 0.92 ). In addition, p hysical symptoms were significantly associated with depressive symptoms ( b =0.84 ) and PTSS ( b = 0.38 ) ; PTSS was significantly associated with depressive symptoms ( b = 0.19) The indirect effects of physical symptoms on physical HRQOL through PTSS were mildly significant ( b = 0.04); however, t he indirect eff ects of physical symptoms on physical HRQOL through depressive symptoms were

PAGE 83

83 not significant ( b = 0.17) (Table 3 3) The total indirect effects of physical symptoms on physical HRQOL through psychological symptoms were not significant but accounted for app roximately 19 % of the total effects on physical HRQOL ( b =0.15) (Table 3 3). The two nested models (Figures 3 2B C ) tested only the path way of physical symptoms to physical HRQOL through depressi ve symptoms as well as the pathway through PTSS respectively. Results showed that physical symptoms were significantly related to two psychological symptoms (i.e., depressive symptoms and PTSS); however, the relationship of psychological symptoms with physical HRQOL were non significant ( b =0.1 2 and b = 0.05 ) The t otal indirect effects for the depressi ve symptoms nested model ( b = 0.1 1) and the PTSS nested model ( b =0.02) did not significantly impact the total effect of physical symptoms on physical HRQOL. The DIFFTEST procedure comparing the performance of the nest ed pathways versus the full pathway showed a significant 2 test ( 2 (3 ) =142.68 p<0.001 and 2 (3) =294, p<0.001 ) indicating both nested models had a worse model fit compared to that of a full model Symptom m ental HRQOL p athways : Figure 3 3A shows the associations between physical symptoms, psychological sym ptoms and mental HRQOL. P hysical symptoms was significantly associated with m ental HRQOL ( b = 0.2 5) with depression symptoms ( b = 0.79 ), and with PTSS ( b = 0.37). The specific pathway from physical symptoms through depression had the strongest relationship with mental HRQOL ( b = 0.59) compared to the other two pathways ( b =0.01 for the pathway through PTSS alone and b = 0.05 for the pathway through PT SS and depressive symptoms) (Table 3 3). However, t he direct effect of physical symptoms o n the mental HRQOL in full path

PAGE 84

84 model did not have as large of a magnitude compared that on physical HRQOL ( b = 0.25 vs b = 0.92 ). I n the nested models t he effects of physical symptoms through depressi ve symptoms and PTSS o n menta l HRQOL were all significant (Figure 3 3B and 3 3C ). The magnitude of the direct effect of physical symptoms in the depressi ve symptoms only model was similar to the full model direct effect ( b = 0.25 vs b = 0.26 ). The indirect effect of physical symptoms through depressi ve symptoms alone on mental HR QOL was stronger compared to indirect effect through PTSS alone ( b = 0.61 vs b = 0.04). The DIFFTEST procedure for each nested model indicated both nested models had a worse model fit compared to that of a full model ( 2 (2 ) = 33.24 p<0.001 and 2 (2 ) = 294 p<0.001). Symptoms p sychosocial p hysical HRQOL pathways : The full path model for the relationship between physical symptoms and physical HRQOL through psychosocial variables (Figure 3 4A) had adequate fit statistics (CFI=0.93, RMSEA=0.04) despite several i nsignificant pathways. The direct effect of physical symptoms on physical HRQOL was significant ( b = 1.01) Indirect pathways from physical symptoms through coping on physical HRQOL were insignificant, suggesting negligible indirect effects. Although dep ressive symptoms and coping were statistically insignificant with physical HRQOL, their direct relationships with physical HRQOL were retained in the next sets of analyses because these two variables were significantly associated with other variables such as optimism, social constraints, and physical symptoms. Based on the other insignificant relationships from the full model, the first pathway to be removed from the analysis (i.e., coefficients constrained to zero ) was from physical symptoms to coping.

PAGE 85

85 T he next two pathways to be removed one at a time were the path from social constraints to coping followed by the path from optimism to depressi ve symptoms The fit statistics were similar for each modification (CFI=0.92 and 0.93 and RMSEA=0.04 for both models ) and the DIFFTEST results indicated that the final reduced model was more parsimonious compared to the full model with a p value >0.05 2 ( 3 ) = 2.59 p =0.46) (Table 3 4). While t he reduced model retained the same in significant indirect effect of physical symptoms on physical HRQOL and the effect of coping through depressi ve symptoms on physical HRQOL the direct effect of depressive sympt oms and coping was marginally significant. Symptoms psychosocial mental HRQOL pathways : The full path model for the relationship between physical symptoms and mental HRQOL through psychosocial variables (Figure 3 5A) had adequate fit statistics (CFI=0.9 2 RMSEA=0.04) despite several insignificant pathways. Insignificant relationships included the pathway of optimism through social constraints on mental HRQOL the pathway of optimism through depressi ve symptoms on mental HRQOL, the pathway of physical sy mpt oms through coping on mental HRQOL, and the pathway of optimism through social constraints and coping on mental HRQOL The direct effect of physical symptoms on mental HRQOL was small ( b = 0.21) while the direct effect of depressive symptoms on mental HRQO L was large ( b = 0.88). The indirect effects from physical symptoms through depression were moderate ( b = 0.58). Coping had a small significant direct effect ( b = 0.15) on mental HRQOL and indirect effect through depressi ve symptoms ( b= 0.18) on mental HR QOL. The DIFFTEST results indicated that the final reduced model was more parsimonious than the full model with a p value >0.05 2 ( 5 ) = 4.98

PAGE 86

86 p =0.42) (Table 3 4). The indirect and direct effects through physical symptoms, coping and depression remained o f the same magnitude and significance compared to the baseline model. Discussion This study examine d the relationships among physical symptoms, psychological symptoms, psychosocial factors and HRQOL in long term HSCT survivors using SEM technique while con trolling for important demographic and clinical characteristics. The primary objective was to evaluate whether the presence of psychological symptoms mediates the relationship between physical symptoms and HRQOL outcomes We found that compared to psycho logical symptoms (depressive symptoms and PTSS), phy sical symptoms had the stronger direct effect on physical HRQOL In contrast, compared to phy sical symptoms, psychological symptoms had the stronger direct effect on mental HRQOL In both HRQOL outcome models, the full path model had a better fit compared to the nested models. Th e secondary objective was to test the overall conceptual model on how different factors contribut e to HRQOL specifically by identifying and quantifying the pathways through whi ch physical symptoms, depressive symptoms, and psychosocial variable s (optimism, social constraints, and coping) influence physical and mental HRQOL. We found that after taking psycho social factors into account, the direct effect of physical symptoms on p hysical HRQOL remained statistically significant. There was no significant direct or indirect effect through the psychological symptom variable The non significant relationship was in part explained by one pathway from physical symptoms through coping t o physical HRQOL and another pathway from optimism through physical symptoms to physical HRQOL. In contrast, after taking psycho socia l factors into account, the direct effect of both

PAGE 87

87 depressive symptoms and physical symptoms on mental HRQOL remained signi ficant. In addition, the significant pathway from physical symptoms through depressive symptoms to mental HRQOL as seen in objective 1 remained significant despite the fact that the pathway of optimism physical symptoms mental HRQOL was significant and de tracted some effects from the relationship of physical symptoms with mental HRQOL. Relationship between s ymptom s and HRQOL : The findings that physical symptoms had the strongest direct association with physical HRQOL compared to psychological symptoms and that psychological symptoms had the strongest direct association with mental HRQOL are consistent hypothesis The findings are also consistent state has the most immediate imp act on physical HRQOL compared to psychological symptoms [77, 146] and that psychological symptoms have the strongest direct impact on mental HRQOL [118, 147] Direct effects of psychological symptoms were non significant with physical HRQOL but the i ndirect effects of psychological symptoms ( PTSS) were significant in physical HRQOL. In contrast, psychological symptoms were both signifi cantly directly associated with mental HRQOL and also significantly mediated the effects of physical symptoms on mental HRQOL From an analytical perspective, the results suggest that when physical symptoms are examined with mental HRQOL without a psychol ogical symptom measure, in a long term HSCT survivor sample, a significant amount of the effect of physical symptoms on mental HRQOL that occurs through psychological symptoms may be missed. From a clinical perspective, addressing physical symptoms remain s the most direct point of intervention to improve

PAGE 88

88 physical HRQOL in this long term survivor population while psychological symptoms, particularly depressive symptoms, are a point of intervention for mental HRQOL. In our conceptual framework, PTSS was a ssumed to influence depressive symptoms (arrow from PTSS to depression in Figure 3 1A C) based on the paradigm that continued chronic stress contributes to major depressive disorder [148] Depressive symptoms and PT SS had mediating effects on the relationship between physical symptoms and physical HRQOL but no significant direct effects on physical HRQOL Similar findings were reported in a previous study examining HRQOL of cervical cancer survivors using SEM [149] Psychological distress was not significantly associated with physical HRQOL in this study (measured by the PCS), but was significantly negatively correlated with physical HRQOL in bivariate analysis of the sa me study For the mental HRQOL path models, the significant relationships among physical symptoms, depressive symptoms, and PTSS are supported by previous studies [147, 150] We foun d physical symptoms was not only directly associated with mental HRQOL, but also indirectly through depressive symptoms and PTSS. The full path model suggests that depressive symptoms have the overriding impact on mental HRQOL compared to PTSS. This is e specially the case in comparing across the two nested models where depressive symptoms act more strongly as a mediator than PTSS, although both are significant. The finding of the mental HRQOL pathway suggests that depressive symptoms and PTSS, alongside physical symptoms, may be viable targets for clinical interventions to improve mental HRQOL of HSCT survivors. Previous literature suggests that

PAGE 89

89 physical symptom related depression or distress decreases survival and contributes to higher suicide rates in long term HSCT survivors, although more evidence is needed [151, 152] Cooke and colleagues reviewed the psychological consequences of HSCT a nd the intervention research available on impro vin g psychological symptoms in survivor population s [153] Some of the suggested interventions to target psychological symptoms included antidepressants, psycho educational interventions, support groups, and complementa ry and alternative medicine (CAM). Of note, the CAM therapy focuses on encouraging a positive re interpretation of the HSCT treatme nt and re defining the Future studies should consider positive psychological imp acts of HSCT survivorship, such as posttraumatic growth [154] Often, these reports of growth occur in conjunction with reports of distress [155] The interplay of negative and positive psychological consequences of receiving HSCT remains unclear and is the main focus of Chapter 4. Findings from the first objective shed light on previous literature in several ways. First, the majority of previous studies on HSCT survivors only examined the effects of a subset of the factors included in this study. There are a substantial number of studies modeling physical symptoms and demographic and clinical characteristics as predictors of psychological distress [156] or HRQOL [21] while others examined the influence of depressive symptoms or PTSS on HRQOL [157] We incorporated multiple factors and controlled for the influence of important demographic and clinical characteristics, providing additional evidence that having fewer physical symptoms is the strongest protective factor for physical HRQOL. Second, this objective also provides evidence that psychological symp toms, especially depressive symptoms, play a stronger role

PAGE 90

90 directly on mental HRQOL and as a mediator between physical symptoms and mental HRQOL compared to physical HRQOL. The roles of p sychosocial factors on the relationship between symptoms and HRQOL : The second objective of this study was to examine the influence of psychosocial factors on the relationship of physical and depressive symptoms with HRQOL in long term HSCT survivors Our findings with respect to the role of psychosocial variables on phy sical and mental HRQOL warrant further discussion. First, coping was hypothesized to play a larger role, but was not significantly associated with physical symptoms or physical HRQOL. Second, the pathways from physical symptoms through the psychosocial v ariables to physical HRQOL may explain the variation in the pathway from the physical symptoms through psychological symptoms that w as found in objective 1. Including the same psychosocial variables did not have the same effect on mental HRQOL; the mediat ing effect of depressi ve symptoms on the relationship between physical symptoms and mental HRQOL remained intact Our findings that coping was not significantly directly associated with physical symptoms are consistent with the findings of an SEM study by Nuamah and colleagues [157] but in contrast to previous studies findings that physical symptom s influences a [38, 158] Additionally, the secondary objective demonstrated that coping was not directly associated with physical HRQOL. While contrary to our hypothesis, this finding is in line with one previous cancer study showing that coping was non significantly associate d with physical HRQOL after adjusting for social support measure and physical symptoms in SEM analyses [159] A nother study also reported that both positive and negative coping strategies were associated with men tal HRQOL,

PAGE 91

91 but only negative coping was mildly associated with physical HRQOL [140] This significance of better coping approaches with better mental HRQOL is supported by other studies as well [160, 161] Of note the coping variable included in our analysis was a combined latent factor of positive and negative coping strategies, potentially losing some variability if modeled as two separate concept s The relationships of the psychosocial factors of optimism, social constraints, and coping with physical symptoms and physical and mental HRQOL were complex. As previously discussed, coping did not play as strong of a role with physical HRQOL as antici pated However, the direct effects of optimism and social constraints and mediating effects of the psychosocial variables did play a role on physical HRQOL. The significant direct effects of optimism and social constraints and the significant pathways fr om optimism to physical HRQOL are likely the mechanism that explain the physical symptom psychological symptom physical HRQOL relationship that were present in objective 1 although PTSS was not directly tested in the full model Depressive symptoms remain ed a significant mediator between physical symptoms and mental HRQOL despite the presence of psychosocial variables. Coping was the only psychosocial variable significantly associated with mental HRQOL, and the magnitude of the relationship was small. Th e small effect of this relationship may explain why the magnitude of the effect of both physical symptoms and depressive symptoms on mental remained almost unchanged in the reduced model. Social constraints w ere associated with physical HRQOL both direct ly and as a mediator between optimism and physical HRQOL. Similar to a study by Nuamah and colleagues the magnitude of the direct effect of social constraints on HRQOL w as not

PAGE 92

92 as strong as the direct effect of physical symptoms on physical HRQOL [157] Social constraints also indirectly influenced physical HRQOL through two pathways: optimism social constraints physical HRQOL and optimism social constraints depressive symptoms physical HRQOL. This significant rel ationship supports a previous finding that a positive social environment or fewer social constraints is associated with greater physical activity participation and better physical functioning [162] In our study, soci al constraints w ere not directly associated with mental HRQOL but w ere significant in the pathway optimism social constraints depressive symptoms mental HRQOL. This is in line with a path analysis by Mosher and colleagues in a sample of HSCT survivors sug gesting that social constraints w ere moderately associated with depression (modeled as the outcome) [121] The direction and the significan ce in the association between optimism and physical HRQOL are inte resting We expected the magnitude of correlation to be moderate between optimism and physical HRQOL and stronger between optimism and mental HRQOL. In the bivariate analysis, higher optimism was significantly associated w i th better physical and mental H RQOL; however, the direction for the relationship of optimism with physical HRQOL variable was inversed after adjusting for other symptom variables and covariates, and a negative significant relationship was found with physical HRQOL. This finding is in c ontrast to that of some previous studies [128, 163] One of the previous studies conducted by Hochausen and colleagues [128] examined the rel ationships of pre HSCT optimism, social support, self efficacy and physical and mental well being one year after HSCT and found pre HSCT optimism was moderately positively associated with well being one year post HSCT ( b= 0 2 6 ). However, in the

PAGE 93

93 Hochhausen study, optimism was evaluated independently with the physical symptoms (optimism as independent variable and physical symptoms as dependent variable), depression (optimism as independent variable and depression as dependent variable), and well being measur es (optimism as independent variable and well being measures as dependent variable). The inclusion of all three of these variables simultaneously is the likely culprit of the unexpected findings in our study. To make our study comparable to the Hochhause changes in the relationship between optimism and physical HRQOL when physical HRQOL was regressed on optimism independently from physical symptoms and depression based on our HSCT data. We study indicating that optimism was positively associated with both physical and mental HRQOL and negatively associated with physical and depressive symptoms when conducted as separate analyses. At this time the true answers are still unknown and future studies are needed to understand the complex mechanism among optimism, psychosocia l variables, and HRQOL outcomes. The second objective builds on the findings from the first objective and provides additional evidence on what role psychosocial factors are important and the role they play in influencing HRQOL in this long term survivor population. There are several approaches to improving HRQOL among those at risk of experiencing physical symptoms. Interventio ns for physical HRQOL with the greatest reported impact include exercise interventions and behavioral interventions [164] As was previously discussed, psychosocial interventions (cognitive therapy and support grou p) are typically designed with the goal of addressing psychological symptoms [153] and these interventions have

PAGE 94

94 not had as strong of an effect on physical HRQOL compared to mental HRQOL [165] Our study found that psychosocial variables did mediate the relationship between physical symptoms and physical HRQOL and explained the relationship between physical symptoms through depression to physical HRQOL. Interventions targeting p sychosocial issues (i.e. cognitive behavioral therapy and group therapy) may be a viable route to pursue for survivors with poor physical HRQOL in our population. J ournal of Clinical Oncology report on the progress of cancer survivorship care and research highlights the psychosocial interventions as an integral component of care and transplantation survivors as those in particular need of this care [166] Our study supports previous literature suggestion that i ntervening on psychological symptoms to is still the most direct approach to improv e mental HRQOL [166, 167] Limitations The findings of this study should be interpreted with the acknowled gement of several limitations. First, the study sam ple is largely homogenous (92% W hite) limiting generalizability to other racial or ethnic groups. Second, the cross sectional nature of the data does not allow for estimating the causal relationship bet ween variables of interests. Specifically, for developing conceptual models and conducting the path analysis, the findings can be interpreted as significant associations but not as a causal pathway among variables Future studies may test the validity of the framework with longitudinal data. Third, t he hypothesized psychosocial pathways are based previous literature and there are several other plausible pathways with differences in directio nality of the relationships. For example, a previous study posit ed that coping [126] Finally, while this study includes measures not previously combined in an SEM analysis, there are still other

PAGE 95

95 possible contributing factors t hat were not addressed. These include cognitive functioning, self efficacy, self esteem and health behaviors [121, 128, 168] Co nclusion This study demonstrate s that having a strong conceptual model is important when evaluating the complex factors influencing HRQOL in survivorship research. The findings highlight that the relationship between physical symptoms and HRQOL is not straightforward because psychologi cal symptoms (e.g., depressive symptoms and PTSS) and psychosocial variables (e.g., optimism, social constraints, and coping) will play a critical role. Specifically, in the first objective phy sical symptoms had the stronger direct effect on physical HRQ OL compared to psychological symptoms but psychological symptoms significantly mediated the relationship between physical symptoms and physical HRQOL. However, in the second objective, when psychosocial variables were included in the framework for the se cond objective, psychosocial factors played a more significant role than psychological symptoms on the relationship of physical symptoms with physical HRQOL In contrast, psychological symptoms had the stronger direct effect compared to physical symptoms on mental HRQOL. P sychosocial variables did not play a significant role, with the exception of coping, on mental HRQOL and the indirect pathways did not change the magnitude of the effect of depressive symptoms on mental HRQOL. Our findings suggest physi cal symptoms and psychosocial factors are important points of intervention to address physical HRQOL and that to improve mental HRQOL, psychological symptoms, particularly depressive symptoms, should be targeted.

PAGE 96

96 A B Figure 3 1. Hypothesized con ceptual models for pathways and framework of factors influencing HRQOL. A) Hypothesized pathway from physical symptoms through psychological symptom to HRQOL. B) Overall conceptual model for factors influencing HRQOL. *Model B is also adjusted for covari ates presented in A. ( Circles= latent variable )

PAGE 97

97 Table 3 1. Study Characteristics N % Demographics Age, years 662 Mean (SD) 42.1 (11) Median (Range) 42.4(18 71) <35 182 28 35 39 95 14 40 44 106 16 45 49 120 18 >50 159 24 Sex Male 251 38 Female 411 62 Race White 603 92 Other 56 8 Education 658 High school or below 194 30 Some college or technical education 209 32 College degree 122 18 > College degree 133 20 Marital status 659 Married/living with partner/ committed 483 73 Other 176 27 Occupational status Working or student 484 73 Not working 100 15 Retired 75 11 Annual family income < $20,000 70 11 $20,000 $40,000 141 22 $40,000 $60,000 156 24 $60,000 $80,000 100 15 >$80,000 181 28 Clinical variabl es Years since diagnosis Mean (SD) 662 7.0 (3.1) Median (Range) 6.6(1.8 22) Severity of treatment Low autologous no GVHD 390 60 Moderate allogeneic no GVHD 168 26 High allogeneic GVHD 88 14 Comorbid conditions Presence of comorbid conditions at survey 104 16 Intensity of previous treatment* Less intense 441 66.6 More intense 221 33.4

PAGE 98

98 Table 3 1. Continued N % PROs Modified FACT BMT score(observed) Mean (SD) 658 37.4 (4.7) Median (Range) 21(13 39) MOS SF 36 Physical Component Summary(PCS) Mean (SD) 658 44.5 (11.6) Median (Range) 48(6.4 64.5) MOS SF 36 Mental component summary (MCS) Mean (SD) 658 50.6 (10.4) Median (Range) 54(10.1 70.2) *Intensity of previous treatment is based on status and durat ion of disease before translation. Less intense=patients who underwent transpla nta tion for chronic phase CML within 1 year, acute leukemia or lymphoma in fi rst complete remission, or adjuv ant treatment of high risk stage II or III breast cancer More intens e= transplantation for chronic phase CML >1yr from diagnosis, accelerated or blast phase CML, acute leukemia or lymphoma beyond first remission, or metastatic breast cancer FACT BMT using 13 items selected from previous study (Chapter 2) l Outcomes Study 36 item short form health survey

PAGE 99

99 Table 3 2 Bivariate standardized correlation matrix of latent factors 1. 2. 3. 4. 5. 6. 7. 1. Physical HRQOL 1.0 0 2. Mental HRQOL 0.87*** 1.00 3. Depression symptoms 0.66*** 0.97 *** 1 .0 0 4. PTSS 0.36*** 0.4 6*** 0. 49*** 1.00 5. P hysical symptom s 0.80*** 0.88*** 0.85*** 0.39*** 1.00 6. Coping 0.38** 0.5 8*** 0.74** 0.46** 0.55*** 1.00 7. Social constraints 0.42*** 0.49 *** 0.53*** 0.48*** 0.51*** 0.40** 1.00 8. Optimism 0.36*** 0.6 3*** 0.69*** 0.35*** 0.57*** 0.60** 0.37*** Latent variable is positively scored so higher scores mean better health ; negatively scored with higher scores meaning worse health. *p<0.05, **p<0.01, ***p<0.001

PAGE 100

100 A B C Figure 3 2. Symptoms and physical HRQOL pathways. A) Full mode l B) Nested model 1: Depression symptoms only and C) Nested model 2: PTSS only. C ircles= latent variable; solid lines= significant associations; dashed line s= non significant associatio ns. *p<0.05, **p<0.01, ***p<0.001 0.84*** 0.20 0.92*** 0.38*** 0.10* 0.87*** 0.92*** 0.12 0.75*** 0.05 0.38*** 0.19***

PAGE 101

101 A B C Figure 3 3. Symptoms and mental HRQOL pathways A) Full model B) Nested model1: Depression symptoms only and C) Nested model 2: PTSS only. C ircles= latent variable; solid lines= significant asso ciations; dashed line s= non significant associations. *p<0.05, **p<0.01, ***p<0.001 0.25*** 0.18*** 0.03 0.37*** 0.79*** 0.74*** 0.26*** 0.86* ** 0.71*** 0.83*** 0.37*** 0.11***

PAGE 102

102 Table 3 3. Direct, indirect and total effects of pathways from physical symptoms through depression and PTSS to HRQOL Physical HRQOL Mental HRQOL Unstandardized b S TDYX b Unstandardized b STDYX b Full Model^: P hysical symptoms PTSS, depression symptoms & HRQOL Direct Effects P hysical symptoms HRQOL 21.81*** 0.92 5.88*** 0.25 Specific Indirect Effects P hysical symptoms depression sym ptoms HRQOL 3.95 0.17 13.97*** 0.59 P hysical symptoms PTSS HRQOL 0.84* 0.04 0.24 0.01 P hysical symptoms PTSS depression symptoms HRQOL 0.34 0.02 1.18*** 0.05 Total Indirect Effects 3.45 0.15 14.91*** 0.63 Total Eff ects 18.36*** 0.77 20.79*** 0.88 Model Fit CFI/TLI RMSEA CFI/TLI RMSEA 0.92/0.91 0.04 0.92/0.90 0.04 Nested Model 1 ^ : P hysical symptoms depression symptoms & HRQOL Direct Effects P hysical symptoms HRQOL 36.93*** 0.87 6.29 *** 0.26 Indirect Effects P hysical symptoms depression symptoms HRQOL 4.68 0.11 14.61*** 0.61 Total Effects 32.25*** 0.76 20.89*** 0.87 Model Fit CFI/TLI RMSEA CFI/TLI RMSEA 0.80/0.76 0.06 0.78/0.74 0.07 DIFFTEST X 2 (df) X 2 (d f) 130.30(3) P<0.001 142.68(3) P<0.001

PAGE 103

103 Table 3 3. Continued Physical HRQOL Mental HRQOL Unstandardized b STDYX b Unstandardized b STDYX b Nested Model 2 ^ : P hysical symptoms PTSS, & HRQOL Direct Effects Physical symptoms HRQOL 38.98*** 0.75 22.74*** 0.83 Indirect Effects Physical symptoms PTSS HRQOL 1.02 0.02 1.09*** 0.04 Total Effects 40.00*** 0.77 23.83*** 0.87 Model Fit CFI/TLI RMSEA CFI/TLI RMSEA 0.44/0.33 0.11 0.35/0.24 0.11 D IFFTEST X 2 (df) X 2 (df) 294(3) P<0.001 529.47(3) P<0.001 Latent variable is negatively scored with higher scores meaning worse symptoms ; positively scored so higher scores mean better ; ^Adjusted for age at survey, gender, race, marital status, educ ation, occupational status, severity of treatment experience, years since HSCT and intensity of previous treatment ; *p<0.05, **p<0.01, ***p<0.001

PAGE 104

104 A B Figure 3 4. Symptoms, psychosocial factors and physical HRQOL framework A) Full model and B) Reduced model All latent variables were adjusted for age at survey, severity of treatment experience, race, gender, marital status, education, occupational status, and years since HSCT *p<0.05, **p<0.01, ***p<0.001 1.01*** 0.49*** 0.19*** 0.05 0.06 0.07 0.03 0.53** 0.60*** 0.70*** 0.49*** 0.13 0.18*** 0.71*** 0.67*** 0.27 1.18 *** 0.49*** 0.26*** 0.20*** 0.55*** 0.30*** 0.61*** 0.40 0.26***

PAGE 105

105 A B Figure 3 5. Symptoms, psychosocial factors and mental HRQOL framework A) Full model and B) Reduced model All latent variables were adjusted for age at survey, severity of treatment experience, race, gender, marital status, education, occu pational status, and years since HSCT *p<0.05, **p<0.01, ***p<0.001 0.66*** 0.88*** 0.21* 0.67*** 0.08 0.15** 0.05 0.12 0.20*** 0.21** 0.01 0.46*** 0.05 0.43** 0.68*** 0.67*** 0.85*** 0.21*** 0.20** 0.24*** 0.15** 0.60*** 0.47***

PAGE 106

106 Table 3 4 Direct, indirect and total effects of conceptual f ramework for optimism coping depression & HRQOL Physical HRQOL Mental HRQOL Unstandardized b STD YX b Unstandardized b STDYX b Full Model^: Physical symptoms optimism, social resources, depression & HRQOL Path from physical symptoms : Direct Effects Physical symptoms HRQOL 10.69*** 1.01 2.12* 0.21 Specific Indirect Effects P hysical symptoms depression symptoms HRQOL 0.45 0.04 5.96*** 0.58 Physical symptoms coping HRQOL 0.02 0.002 0.19 0.02 P hysical symptoms coping depression symptoms HRQOL 0.01 0.00 1 0.23 0.02 Total Indirec t Effects 0.44 0.04 6.01*** 0.58 Total Effects 10.24*** 0.97 8.14*** 0.78 Path from coping: Direct Effects Coping HRQOL 0.86 0.06 2.01** 0.15 Indirect Effects Coping depression symptoms HRQOL 0.19 0.01 2.44** 0 .18 Total Effects 0.67 0.05 0.43 0.03 Model Fit CFI/TLI RMSEA CFI/TLI RMSEA 0.93/0.92 0.04 0.92/0.91 0.04 Reduced Model ^ :P hysical symptom, depression symptoms physical HRQOL ` Path from physical symptoms : Direct Effects P hysical symptom HRQOL 12.30*** 1.12 2.07** 0.20 Indirect Effects P hysical symptoms depression symptoms HRQOL 1.90 0.18 5.89*** 0.57 Total Effects 10.39*** 0.94 7.96*** 0.77 Path from coping: Direct Effects Coping HRQOL 0.9 7 0.15 1.97** 0.15 Indirect Effects Coping d epression symptoms HRQOL 4.46 0.07 2.65*** 0.21 Total Effects 2.01* 0.08 0.68 0.05

PAGE 107

107 Table 3 4 Continued Physical HRQOL Mental HRQOL Unstandardized b STDYX b Unstandardized b STDYX b Model Fit CFI/TLI RMSEA CFI/TLI RMSEA 0.93 /0.92 0.04 0.92/0.91 0.04 DIFFTEST X 2 (df) p value X 2 (df) p value 2.59(3) 0.46 4.98(5) 0.42 Latent variable is negatively scored with higher scores meaning worse symptoms ; positively scored so higher s cores mean better ; ^Adjusted for age at survey, gender, race, marital status, education, occupational status, severity of treatment experience, years since HSCT and intensity of previous treatment ; *p<0.05, **p<0.01, ***p<0.001

PAGE 108

108 CHAPTER 4 HETEROGENEIT Y IN POSTTRAUMATIC STRESS AND POSTTRAUMATIC GROWTH AMONG HSCT SURVIVORS Introduction As a life threatening illness requiring intensive treatment cancer was first included as a traumatic event in the Diagnostic and Statistical Manual of Mental Disorders, I V (DSM IV) [131] Literature also suggests that t he extreme physical invasiveness of hematopoietic stem cell transplant (HSCT) can add additional stressors to the initial cancer diagnosis [169, 170] The cancer and HSCT experience is more complex compared to discreet traumatic events (e.g., natural disasters, assaults) because the set of stressors consists of a multitude of fact ors related to the diagnosis, treatment, long term complications, financial issues, and family implications [55, 112] Additionally, a perception of internal or personal control exists relat ed to the survivor being responsible for follow up care and engaging in preventive behaviors (e.g. screening tests and h ealthy behaviors) that adds additional stress in contrast to someone who experienced discreet event s caused by external factors [55] Earlier research indicates cancer s urvivors report ed both negative psychological health responses such as posttraumatic stress symptoms (PTSS) and positive responses such as posttraumatic growth ( PTG ) [50, 54, 55] PTSS are a collection anxiety related symptoms occurring after a terrifying or intensely threatening event [4 9] The prevalence of any PTSS in HSCT survivors is varied due to different cancer diagnoses and PTSS measurement methods, but estimates are as high as 50% [114, 115] A review study suggests that 5% to 19% of HSCT survivors have symptoms

PAGE 109

109 consistent with posttraumatic stress disorder, but that the occurrence of any PTSS is higher [42] The term PTG describe s the perceived positive change in an i ndividual over and above the previous level of psychological adaption a s a result of a traumatic event [171] S imilar terms include perceived positive impact, stress related or adversarial growth, and benefit find ing [172 174] A review by Stanton and colleagues (2006) estimate perceived psychological growth was between 60 % and 95% of cancer survivors [175] and another review suggests that prevalence was between 83% and 98% in breast cancer survivors [176] The core hypothesis is that traumatic events can views about themselves and the world, leading to cognitive processing that rebuild s the beliefs and in turn, psychological growth The model proposed by Tedeschi and Calhoun suggests that to achieve PTG, a traumatic event causing emotional distress is exp erienced, followed by rumination or reappraisal of events, and then behaviors are initiated to reduce distress [171] Previous studies have hypothesized that PTG can contribute to a positive reinterpretation of th e cancer experience or a ct as a positive buffer to PTSS [177] Relationship of PTSS and PTG The relationship of PTSS with PTG and the combined effects on health outcomes such as health related quality of life ( HRQOL) remain unclear In Tedeschi and PTG is not exclusive of negative psychological outcomes, such as stress or anxiety. Th ey hypothesized because PTG occurs in the context of traumatic and stressful events, it can co occur with P TS S [171] The theory states it is not necessary that everyone have completely positive or negative psychological response [58] and this concept is supported in sub sequent research [31] Some studies reported

PAGE 110

110 positive associations (greater PTSS, greater PTG), whereas other studies found no association between PTSS and PTG, and still others find an inverse relationship (lower PTS S greater PTG) [31] Measurement D ifficulties Measuring the concepts of PTSS and PTG together has proved difficult in previous studies. These two concepts may share common variance in psychological aspects which is f urther associated with health outcomes such as physical and mental health related quality of life (HRQOL) A previous study on PTSS and PTG by Park & Helgeson identified this methodological issue and suggest that it is not necessary to assume the two conc epts are on the opposite ends of a continuum and advanced statistical methods should be used to tease out the true relationship [31] PTSS and PTG have similar initial pathways from the traumatic event (both theorize that distress and cognitive processing occur ), making it especially important to carefully analyze these concepts together [56, 178] T raditional general linear model approaches typically ai m to predict outcomes and establish relationships between predictive and outcome variables (variable centered) [179] The se approaches maintain the assumption that all individuals interpret the items in a consistent way [180] Alternatively, person centered analytic approaches focus on the relationship among individuals and determining which individuals are similar to each other (aggregating into groups, clusters, or class es) and which individuals are different based on measured characteristics [179] It is useful to apply a person centered method to distinguish the heterogeneity among survivor s as a result of experiencing both PTSS an d PTG and subsequently identify the factors associated with the heterogeneous groups of survivors This study hypothesizes that there is a unique latent (unobservable)

PAGE 111

111 construct comprised of both PTSS and PTG measures. The latent construct is associated with the formulation of heterogene ous survivors who experience various types/levels of PTSS and PTG (Appendix D Figure D 1) Application of Factor Mixture M odel The use of traditional regression based statistical approaches (i.e. discriminant analysis, logistic regression, multivariate analysis of variance (MANOVA)) for evaluating population heterogeneity in PROs often ignore some characteristics of PRO s that may violate standard assumptions of these techniques [62] Chapter 1, Table 1 1 provides a co mparison of methods and Appendix D Figure D 2a f provides a graphical representation of models The major restriction of the traditional regression based approaches to evaluating the hypothesized population heterog eneity is that it relies on observed variables to determine heterogeneity such as differences due to explicit memberships or groupings such as gender or age groups [62] Because PROs such as PTSS and PTG cannot be observed directly (i.e., latent traits) [181, 182] latent variable approaches are useful to analyze PROs data and generate the memberships or groupings for people who experience variou s types/levels of PTSS and PTG symptoms. Several latent variable models are available in health outcomes including PTSS and PTG. These methods are comprised of latent class or latent prof ile models and factor mixture models [62] Latent class analysis (LCA) and latent profile analysis (LPA) are used to create categorical classes of individuals based on the unobserved population heterogeneity while a ssuming within class homogeneity [183] Specifically, latent class is informed by items with categorical characteristics and latent profile is informed by items with continuous characteristics. Factor mixture model (FMM) is another latent

PAGE 112

112 variable model that accommodates the features of common factor model and latent class model Figure 4 1 displays the general form of the factor mixture model The common factor model (i.e. factor analysis) allow s for modeli ng of a continuous latent factor that represents the underlying latent trait (Appendix D, Figure D 2A) The latent factor is assumed to describe all correlations among the items measuring that specific concept [62] On the one hand, FMM is similar to the common factor model with respect to item dependency, or the items being influenced by an underlyi ng factor or latent trait (i.e. local dependence) [62] It is similar to LCA by creating heterogeneous latent classes of the population based on the unobserved latent trait. On the other hand, the FMM is different from the common factor and latent class models because it can relax the assumption of local or conditional independenc e across the range of the factor scores (or levels of latent traits), thus leading to within class heterogeneity [62] These generalizations of the common factor model and LCA provide more flexible approach to analy zing PRO s with respect to statistical assumptions and the results provide more information regarding the characteristics of the subjects. This study had three main objectives. The first objective was to determine the mixed classes of PTSS and PTG in HSC T survivors using an FMM approach. Second, we examined whether demographic, clinical, and physical symptom factors are associated with the class membership. The third objective was to evaluate whether class membership was associated with HRQOL as measure d by the Medical Outcomes Survey 36 item short f orm health survey (MOS SF 36) Physical Component Summary (PCS) and Mental Component Summary (MCS) [64]

PAGE 113

113 Methods Participants and Data Collection This study is a s econdary data analysis of a multi site study (40 centers) of HSCT recipients at participating transplantation centers with records identified from International Bone Marrow Transplant Registry/Autologous Blood and Marrow Transplant Registry (IBMTR/ABMTR ). Participant eligibility included age at least 18 years or older, a single allogeneic or autologous HSCT, at least 12 months post HSCT, specific diagnosis of cancer (chronic myelogous leukemia, acute leukemi a, lymphoma, or breast cancer), continuous remiss ion since HSCT, and able to read and understand English Participants were randomly selected to be contacted for eligibility confirmation. The flow chart i n Chapter 1 Figure 1 1 demonstrates the process of subject recruitment Participants completed a series of questionnaires through written and phone interview. Clinical information was extracted from medical records from the IBMTR/ABMTR to determine the type of initial diagnosis, type of HSCT, and the nature of donor relationship (allogeneic) Study M easures The Impact of Events Scale (IES) was used to measure PTSS [130] The IES was created prior to the inclusion of post traumatic stress disorder in the DSM IV The ological response to stress and empirical evidence from clinical studies [130, 132] The IES contains two (intrusive thoughts and avoidance ) of the three (hyperarousal) sub scales used to diagn ose posttraumatic disorder (PTSD) [49] The revised IES (IES R) was made available after study implementation [184] The IES used in this study has been widely used for all types of traumatic events (e.g., disasters, assaults ) [130] The IES can discriminate between

PAGE 114

114 individuals with mild to severe stress responses and retains good reliability over time despite the lack of hyperarousal domain [132] I t was not designed as a diagnostic measure but may act as a screening tool for PTSD [132] The IES is comprised of 15 items measuring two concepts/domains: intrusion (7 items) and avoidance (8 items) [130] Response options are on a Likert type with four categories (0= not at all, 1=sometimes, 3=rarely, and 5=often). The scoring guidelines for PTSS are based on the overall score and include high (>=19) medium (8.5 to19) or low (1 to 8.5) PTSS. PTG is measured using the Post Traumatic Growth Inventory (PTGI) [171] The PTGI is a 21 item scale measuring fiv e concepts/domains: relating to others, new possibilities, personal strength, spiritual change, and appreciation for life. A L ikert scale was used from 0 to 5, where 0 indicat ed and 5 indicat ed To apply the items to HSCT as a result HSCT/cancer survivor population [28, 71] Measurement properties of each instrument are available in Appendix E Table E 1. The MOS SF 36 is a generic HRQOL measurement with 36 items designed to measure 8 domains (physical f unctioning, role limitations due to physical problems role limitations due to emotional problems social functioning, bodily pain, mental health, energy vitality, and general health perceptions. The PCS and MCS are component measures of the physical and mental aspects of HRQOL. For this study, PCS is composed of physical functioning, role functioning physical and bodily pain items. Role functioning emotional, social functioning, mental health, and energy/vitality made up the

PAGE 115

115 MCS. The PCS and MCS were treated as the distal outcome variables for physical HRQOL and mental HRQOL. The scores of the PCS and MCS scale are normalized with a mean of 50 and a SD of 10 [64] The PCS and MCS are scored so higher scor es indicate better health. Hypothesis It was hypothesized that multiple classes will be generated, with varied levels of both PTSS and PTG, such as high PTSS/ high PTG, low PTSS/ low PTG, etc. For predictors of class membership, it was hypoth esized that younger current age, lower education, and greater s everity of treatment experience are associated with the class representing higher PTSS and higher PTG. Older age, lower severity of treatment experience, and no comorbid conditions are anticipated to be predictors of the class representing lower PTSS and lower PTG. It was hypothesized that the classes are associated with higher physical HRQOL and mental HRQOL compared to c lasses comprised of su rvi vors with high PTSS and low PTG. Analytic Strategy The goals of this analysis were to use latent variable mixture models for classifying sub populations based on status of PTSS and PTG, determine whether important covariates are associated with class me mbership, and then examine class associations with the distal outcome of HRQOL. Descriptive statistics for the measures used in this analysis (IES, PTGI and MOS SF 36) and demographic and clinical characteristics were calculated. Five steps outlined belo w are conducted to guide statistical analyses: 1) identifying number of factors, 2) identifying optimal number of classes, 3) fitting best factor mixture model variation, 4) examining predictors of class

PAGE 116

116 membership in the factor mixture model identified in step three, and 5) evaluating the association of class membership with the distal outcomes of physical HRQOL and mental HRQOL. Appendix D provides detailed technical information on the model building and Figures D 2A F provide graphical representation s o f these models. Model Building Step one : The first step was to determine the ideal number of factors through confirmatory factor analysis (CFA). As the sub domains of each instrument are not of interest in the analysis, a one factor model and a two fact or model for PTSS and PTG were tested. A one factor model specifies that all items (both PTSS and PTG items) load onto a single latent factor (Figure D 2A). A two factor model specifies that all PTSS items load onto a single latent factor and all PTG ite ms load onto a separate latent factor Standard fit criteria were used for evaluating model performance (Comparative Fit Index (CFI) >0.95; Root Mean Square Error of Approximation (RMSEA) <0.06) [87] Step two: The se cond step was to identify the optimal number of classes for PTSS/PTG without any latent factor model specifications from step one (Appendix D Figure D 2B). Latent class models with PTSS and PTG obs erved variables (items or 2B) were iterati vely tested, beginning with one class to determine the maximum number of classes that fit the data and produce interpretable results While there is no universal agreement upon the set of criteria for evaluating class solutions, several recommended criter ia for assessing model accuracy and model fit were used [185] Model a ccuracy was evaluated using the entropy statistic (higher val ue indicates better accuracy; 0 to 1). M odel fit was evaluated using th e log likelihood ( LL), Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Lo Mendel Rubin

PAGE 117

117 likelihood ratio test ( LMR LRT), Vuong Lo Mendel Rubin LRT (VLRT) and Bootstrap LRT (BLRT). Lower values in LL, AIC, and BIC indicate better fit. LMR LRT, VLR T, and BLRT were used to compare the estimated model to a model with one less class ; a p >0.05 indicates the model with one less class fits significantly better and p <0.05 indicates the model with more classes fits significantly better. VLRT and BLRT wer e used in less complex models (3 classes or fewer) due to the heavy computation and difficulty in convergence in complex models [185] Step three: After determining the best fitting factor and class the mixture model vari ations were tested. There are many variations of the FMM, ranging from highly restrictive models that constrain several parameters across classes, to less restrictive model that allow multiple parameters to be freely estimated across classes (Appendix D, Table D 1). Multiple random start values were estimated to ensure the stability of the estimated model parameters (Appendix D). Appropriate mixture models should have application to the research problem in question ( e.g., s In the first mixture model (FM 1), t he means of the latent factors were conditioned on latent class membership (i.e., freely estimated across classes) [62] Factor loadings, item thresholds, and the factor variance/covariance matrix were invariant across classes (Appendix D, Figure D 2C). Factor loadings represent correlations between the latent factor and the items. For categorical it ems, thresholds are the expected values of the item at which an individual transitions from one response category to the next [83] For the second model (FM 2), the factor loadings and means were conditioned on latent class membership (Appendix D, Figure D 2D) Third (FM 3),

PAGE 118

118 the factor loadings, item thresholds, and means were conditioned on latent class membership (Appendix D, Figure D 2E). Finally, for the fourth model (FM 4), the factor loadings, item thresh olds, and the factor variance/covariance were conditioned on latent class membership (Appendix D, Figure D 2F). However, for modeling purposes, t he means of the latent factors were conditioned to zero across latent classes. The accuracy and fit of each m odel were evaluated using entropy, lowest log likelihood, AIC, BIC, and ABIC in addition to the substantive interpretation and utility of the class results [61] Item threshold values were used to est imate the conditional item response probabilities that are provided in standard latent class models, but are not available in mixture models requiring numerical integration to obtain factor variances/covariance. Item response probabilities provide informat ion on the probability of responding to a particular item response category given the respondent is in a certain class. This information helps to define and describe the population in each class. Item response probabilities are not available in complex m ixture models because the values are estimated probabilities computed at the mean of the factors and not just on class membership (Mplus Discussion Board Latent Variable Mixture Modeling, 2005). Item threshold values can be used to obtain relative estimat es of conditional item response probabilities using the equation p= 1/(1+exp(threshold)) [145] Model comparison checks for two and three class models were conducted to verify whether the four class model wa s optimal. Predictors and Distal Outcomes Associated with Class Membership Predictors of class membership ( step four ): Using the selected FMM from the previous step, factors of demographic, clinical, and physical symptoms were estimated

PAGE 119

119 to associate with t he class membership. These factors include gender, age at survey, race/ ethnicity, marital status, occupational status, education, comorbid conditions, severity of HSCT experience, years since diagnosis, and physical symptoms as measured by the modified F ACT BMT (Chapter 2). Step four requires several stages to assess predictors of latent class membership. The two main approaches for analyzing factors influencing class membership in latent class/mixture models are a one stage and three stage approach. T he one stage method is more efficient than the three stage approach since it integrates predictive factors into the latent class models simultaneously. However, the disadvantages are the inclusion of large numbers of predictive factors (models must be re estimated with specific predictive factors removed or added), difficulty in choosing the appropriate number of classes as determined with or without predictive factors, and the potential change in the intended meaning of the classes after the inclusion of predictive factors [186, 187] In contrast, a three stage approach estimates the latent class model independent of predictive factors and then using a multinomial logistic regression w ithout accounting for classification errors. However, this approach has shown to consistently underestimate the relationship between the predictive factors and class membership, particularly if the classification error is large (entropy <0.60) [186, 188] Vermunt (2010) [186] created and validated a modified three stage approach to address the disadvantages and this method was incorporat ed into Mplus version 7 by Asparouhov and Muthen (2012) [189] This modified approach takes into account the classification error rates. The three stage method was used in step four to evaluate factors of the la tent class membership and subsequently the distal outcomes in this

PAGE 120

120 study (detailed technical methods in Appendix D). To identify factors associated with latent class membership, the Mplus AUXILIARY command (r3step) for automatic three step estimation was used. The three stages in this approach are: estimating latent class membership without accounting for the predictive factors (the FM 3 mode l from step three), 2) adjusting for classification error and 3) conducting m ultinomial logistic regression to iden tify significant predictive factors related to class membership without changing the latent classes The classification error is determined by using the average latent class probabilities of latent class membership and then adjusting for this this error w Class membership and distal outcomes ( step five ): Step five also requires separate stages to assess for the association of class membership with distal outcomes. The latent class comparisons of the unadjusted distal outcome means were conducted using the automatic three stage approaches. The AUXILIARY du3step function uses a Wald 2 test of statistical significance to test the equality of outcome means across t he various classes while also accounting for classification error [190] In this study, t he HRQOL outcomes under investigation were the Physical Component Summary (PCS) and the Mental Component Summary (MCS). This study also examined the adjusted distal outcome means across classes by including the predictors of class membership. The automatic command cannot be used for this; therefor create a manual approach in Mplus to implement this improved method [189] The technical procedure and calculations are described in details in Appe ndix D and derived from the webnote by Asparouhov and Muthen [189] Briefly, using the average latent

PAGE 121

121 class probabilities for most likely latent class membership matrix information, the classification error was manually calculated by taking the log ratio of each class probability against the last class and including this information in the subsequent analysis. We then conducted Wald 2 test for each class comparison to evaluate adjusted distal outcome means by adding the Model TEST command and testing each class individually. STATA version 9 was used for all data management [191] and Mplus Versio n 7 was used for all modeling procedures [100] Results Study Sample Characteristics P articipants mean age at the time of survey was 42 years (SD=11) (Table 4 1) O ver half of the sample was female (62%) and the maj ority was White (92%) One third of the sample (32%) reported some college or technical education and another 30% reported a high school education or less. Approximately 73% of the survivors reported a working or studen t occupational status. The median numbe r of years since diagnosis was 6.5 years with a range of 2 to 22 years. The majority reported low severity of treatment experience (60%) and 16% reported the presence of comorbid condit ions at the time of the survey. The median score on the modified FACT BMT physical symptom scale was 21, with a range of 13 to 39. The sample had 23 (3.6%) cases with missing values on all variables, leaving a sample of 639 to build the latent class mixture model. Model Building Class enumeration : A two factor model (PTSS factor and PTG factor) was confirmed with a CFI = 0.94, RMSEA = 0.05, and no residual correlations among items

PAGE 122

122 above 0.20. Table 4 2 displays the results of the two three and four class model comparisons. The four class model was selected as the best fitting model for the mixture model. A four class model demonstrated the lowest log likelihood value, AIC, BIC, and ABIC and highest entropy (0.95). The LRT and VLRT support four class enumeration with significant p values (p<0.01). The BLRT did n ot converge with the highest number of starts allowed, but this has been noted in more complex models. The four classes of FM 3 were interpretable as heterogeneous groups. Mixture model variations were based on the two factor, four class model. Results of the model comparison checks for two and three class models were conducted to verify that the four class model was optimal and more parsimonious models were not available (Appendix E, Table E 2). Mixture model: Table 4 2 also displays the results of the different four class mixture model variations. FM 1 demonstrated worse log likelihood compared to FM 3 (LL= 19309 vs. 17146) as well as worse AIC, BIC, and ABIC. FM 2 was worse in all criteria compared to FM 1 and FM 3. For FM 4, the lowest log lik elihood could not be replicated at the maximum number of random starts allowed. FM 3 demonstrated the lowest log likelihood as well as lowest AIC, BIC, and ABIC, but lower entropy than FM 1 and FM 2. Based on interpretation and utility of each model, FM 3 was selected as the most appropriate and informative mixture model. Class specific item threshold values provided information on the response probabilities within each class (Table 4 3). Threshold values allow for examining how easy or difficult it was for respondents to endorse an item response category [192] The IES instrument used to measure PTSS was scored negatively, so greater negative

PAGE 123

123 values for the higher response categories suggest easier endorsement of the response higher response categories. Because the PTGI was scored positively, gre ater negative values for the higher response categories suggest easier endorsement of these categories (greater PTG). Based on the proportions from the most likely latent class membership, class one named (n=104; 16.3%) had low endor sement of both PTSS and PTG. Table 4 4 provides a description of the classes and the presence of PTSS and PTG in each class. Item thresholds (Table 4 3) showed for the majority of items, class members had the lowest threshold or highest probability of en dorsing the first response of PTG. Class two named (n=227; 35.5%) was comprised of those with little to no reports of PTSS and h igh PTG. category to indicate never experiencing the symptom in each item (threshold values < 15, or floor effects). Thresholds for PTG indicated this class responded mostl y in the (lower thresholds in higher categories). Class three (n=170; 26.6%) was comprised of those with ems did not demonstrate ceiling effects to indicate very high PTSS, the item thresholds show that higher response categories of PTSS were endorsed more compared to the other classes. Compared to class one, two and four, class three had comparatively lower

PAGE 124

124 reports of PTG (higher thresholds for the greater PTG categories). Class four (n=138; 21.6%) was comprised of ; those reporting some degree of PTSS, less than class three and more than class one and two, and a moderate amoun t of PTG. Factor loadings: The items with significant factor loadings varied across classes (Table 4 5). The PTSS item with the strongest significant factor loadings lo adings across all classes, four items significantly loaded onto their respective factors Predictors and Distal Outcomes Associated with Class Membership Predictors of latent class membership : Table 4 6 displays the results from the multinomial logistic regression of covariates on class membership using the low PTSS/ high PTG (class 2) as the reference class. Compared to individuals in the low PTSS/ high PTG class, those with greater years since diagnosis were more likely to be classified into the low PTSS/ low PTG class (class 1) than those with less years sinc e diagnosis ( b =0.62, p<0.05) Compared to individuals in the positive psychological impact class, those with a high school degree or less education ( b =0.64, p<0.05) and with greater symptoms ( b =0.88, p<0.001) were more likely to be classified into the hi gh PTSS/ moderate PTG class (class 3) than those with graduate degree and with less

PAGE 125

125 symptoms. Compared to individuals in the low PTSS/ high PTG class, those with a race/ethnicity other than White, a high school or below education, or some college or techn ical education were significantly more likely to be a part of the moderate PTSS/ moderate PTG class (class 4) than White and those with graduate degree ( b= 1.04, p<0.01; b= 1.17, p<0.01; b= 0.78, p<0.05, respectively). Distal outcomes: Table 4 7 shows the results of physical HRQOL and mental HRQOL means comparison across classes. For the unadjusted Wald 2 tests of both physical HRQOL and mental HRQOL means, the differences between the high PTSS/ moderate PTG class and each of the other classes were signi ficant (class one vs. class three, class two vs. class three, class three vs. class four). The low PTSS/ high PTG class had the lowest adjusted physical HRQOL score (48.10) and the high PTSS/ moderate PTG class had the lowest adjusted mental HRQOL score ( 44.77). For adjusted physical HRQOL comparisons, no Wald 2 test of mean equality was statistically significant although low PTSS/ high PTG class versus high PTSS/ low PTG class was the largest difference ( 2 =1.52, p=0.07). Adjusted mental HRQOL means comparisons resulted in a significant difference between th e low PTSS/ high PTG class and high PTSS/ moderate PTG class ( 2 =8.01, p<0.01). Discussion This study explored the latent structure and population heterogeneity of PTSS and PTG using a factor mixture model, examined the association of predictors with la tent class memberships, and latent class memberships associations with health outcomes such as HRQOL. We found that a two factor, four class model (FM 3) with class varying factor loadings, thresholds, and means was the best fitting mixture model. FM 3 p artially relaxed the conditional dependence assumption by allowing for the factor

PAGE 126

126 loadings, item thresholds, and means of latent factors to vary across class membership (freely estimated across latent classes). HSCT survivors were categorized into one of four heterogen e ous groups: low PTSS/ low PTG (class 1), low PTSS/ high PTG (class two), high PTSS/ moderate PTG (class three) and moderate PTSS/ moderate PTG (class four). Predictive factors significantly associated with class membership (low PTSS/ high P TG class as the reference) included race/ethnicity, education, years since diagnosis, and physical symptoms. In adjusted analyses, no significant differences between physical HRQOL means were found across the four classes. For adjusted mental HRQOL a s ignificant difference was noted between the low PTSS/ high PTG clas s and the high PTSS/ moderate PTG class The two factor four class mixture model supported the our hypothesis that multiple heterogeneous classes would be identified. While the proporti ons of classes identified from factor mixture model were similar to the standard four class latent class mode from step one (class 1 n=104, class 2 n=220, class 3 n=146, class 4 n=170), previous literature examining the latent structure and heterogeneity o f PTSS and PTSD supports the use of FM M approach over the standard latent class approach [193, 194] The inclusion of the factor structure accounts for the residual correlation among the item s Comparing model information criterion demonstrates that constraining the factor loadings and thresholds to be the same across classes (FM 1 and FM 2), which assumes measurement invariance, fits substantially worse compared to FM 3. The FM 3 partially r elaxed the conditional independence assumption, thus suggesting measurement non invariance by allowing factor loadings and thresholds to vary across classes [192] The different factor loadings across classes may be i nterpreted as each

PAGE 127

127 class having class experiencing PTSS and PTG [195] Severity is th e extent to which individuals experience PTSS or PTG and is reflected not only by threshold values and factor loadings but also factor means in FM 3. Higher loadings in one class versus another suggest that there is a higher correlation between the item and the factor in that class. survivors of multiple cancer types [175] and 83% to 98% in breast cancer survivors [176] ), it is not surprising that all four classes reported some degree of PTG, although the low PTSS/ low PTG had the lowest re ports comparatively to other three classes. Overall, reports of PTSS were lower than reports of PTG. Theoret ically, PTSS attenuates further in time from the event [196] (or diagnosis and treatment in this study), thus lower reports of PTSS compared to PTG in all classes is expected in a sample of long term survivors. One study suggests that stronger PTSS symptoms, such as those related to intrusion, are more persistent over time while avoidant symptoms decrease over time [197] Interestingly, three of the four significan t factor loadings for PTSS were These issues may be explored through a longitudinal study to better understand the dynamic of PTSS. Predictors of latent class membership : Literature suggests that demographic characteristics associated with PTSS in cancer survivors include younger age at diagnosis [198] Black race/ethnicity [199] and lower educational status [169, 198] For PTG, demographic factors consistently indicative of greater growth include younger age at disease [199, 200] high spirituality or religiosity, and minorities [199] Education level

PAGE 128

128 has mixed findings with PTG, some studies suggest lower education level is associated with greater growth [201] others suggest higher education predicts greater growth. The results of our study indicate those with lower education levels and minorities were significantly associated with being a part of the moderate PTSS/ moderat e PTG class supporting the notion that these characteristics are associated with a greater overall psychological response, both positive and negative. Previous multivariate studies examining racial/ethnic differences in PTSS conclude that it is still unc lear why minorities (primarily those of Black race/ethnicity) have an increased of risk of PTSS [202] and report the most growth compared to other races [199, 203] Some suggestions for higher PTSS include potential previous anxiety, depression, or decrease socioeconomic status [202] For higher PTG, it is hypothesized that religious coping and associated supp ort may encourage positive growth [199] with membership in the high PTSS/ low PTG class and having fewer years since diagnosis was ass ociated with the low PTSS/ low PTG class Previous multivariate suggest that a worse physical state is predictive of PTSS in survivors [27, 169, 204] In the PTSS literature with typically defined traumatic events, PTSS has been strongly linked with physical symptoms [205] Time since diagnosis and PTSS in cancer survivors has mixed findings in the literat ure, with some studies reporting no association between time since diagnosis and PTSS [206] and others report that those closer to diagnosis report greater distress [27] Stud ies have also identified increased time since diagnosis as associated with greater growth [59] These two previous

PAGE 129

129 supported by our st PTG that states that growth develops over time [171] Latent class membership and physical and mental HRQOL : Accounting for predictors of class m embership increased HRQOL scores from the unadjusted in all classes. The change in scores may likely be explained by the inclusion of physical symptoms as a factor that is predictive of class membership. Physical symptoms were strongly associated with bo th physical and mental HRQOL in Chapters 2 and 3. Adjusted p hysical HRQOL for the low PTSS/ low PTG class, high PTSS/ moderate PTG class and moderate PTSS/ moderate PTG were all above the population norm of 50 using a minimal clinically important differen ce (i.e., 5 SD above or below the general population norm of 50 ) [207] However, no adjusted physical HRQOL scores were statistically significantly different between classes One possible explanation may be that P TG has not been consistently associated with HRQOL, even in longitudinal studies [29, 175, 208] and that PTSS does not have strong of an association with physical HRQOL as mental HRQOL, per results in Chapter 3 A meta analytic review of 77 articles suggests the results are mixed on the link between PTG and health outcomes such as HRQOL [208] When associations between PTG and HRQO L are identified in studies, PTG was associated with an increase in mental HRQOL and no t associated with physical HRQOL [32] The high PTSS/ moderate PTG class had the lowest score in mental HRQOL and was statisticall y significantly different from the low PTSS/ high PTG class. Compared to the general population mental HRQOL norms, individuals in both low PTSS/ low PTG class and low PTSS/ high PTG class had mental HRQOL scores that

PAGE 130

130 were greater than 5 SD above the gene ral population norm of 50 [207] The high PTSS/ moderate PTG had a mental HRQOL score 5 SD below the population norm. The finding that the individuals in the moderate PTSS/ moderate PTG class have a mental HRQOL sc ore slightly above the general population norm supports the notion that experiencing a range of emotion might encourage some positive psychological response in survivors [31, 208, 209] To our knowledge, no study has conducted a latent class/ mixture analysis to examine population heterogeneity from PTSS and PTG simultaneously in long term HSCT survivors. A review by L inle y recommended that measures examining growth should allow for inclusion of negatively worded responses [210] We used the factor mixture model to account for the within class covariation that may occur when constructs such as PTSS and PTG are examined together in traditional methods to identify heterogeneous groups within this sample. The four classes identified are some common characteristics and PTG may deflect some of the e ffect of PTSS on HRQOL [31, 32, 177, 210] Of note, Mosher and colleagues [211] found that only 50% of the ir distressed HSCT survivor population sought mental health services, despite availability in medical centers. The authors found barriers to mental health services included knowledge barriers and emotional barriers, but the interventions to increase servi [211] A classification model, such as the one presented in this study, that also incorporates the dimensionality of distress and growth, may be useful i n future studies for intervention development as well as for informing treatment in clinical practice

PAGE 131

131 Limitations Several important limitations should be considered. First, the data is cross sectional and therefore causality of the response variables and covariates cannot be established. Significance was examined without making any causal inference. Second, the population examined is homogeneous, with approximately 92% of the sample being White. This limits the generalizability to other race/ethnicities where differences in psychological outcomes have been noted. Third, only two of the three current domains of PTSS (intrusive and avoidant thoughts) were measured by the IES because the revised IES was not available at study onset. Without the hyperarous al domain, the presence of PTSS is mildly underestimated, although this domain has been more difficult to relate to cancer survivors compared to intrusive and avoidant thoughts [204] Finally, all latent clas s and mixture model analysis are subject to being considered exploratory procedures when classes are not known prior to analysis and future studies are needed to replicate the findings [195] Conclusion Cordova and Andrykowski (2003) hypothesized that cancer and treatment should negative effects [154] Our study supports the hypothesis that PTSS and PTG can co occur with the emergence of four classes of long term survivors. Furthermore, the mixture model selected suggests there are varying levels of PTSS and PTG among classes. F uture studies should consider longitudinal evaluation through growth mixture modeling to track the potential changes in PTSS and/ or PTG reporting as well as the subsequent effect on health outcomes over time

PAGE 132

132 Table 4 1. Study characteristics N % Demogr aphics Age, years Mean (SD) 662 42.1 (11) Median (Range) 42.4(18 71) <35 182 28 35 39 95 14 40 44 106 16 45 49 120 18 >50 159 24 Sex Male 251 38 Female 411 62 Race White 603 92 Other 56 8 Education 658 High school or below 194 30 Some college or technical education 209 32 College degree 122 18 > College degree 133 20 Marital status 659 Married/living with partner/ committed 483 73 Other 176 27 Occupational status Working or student 484 73 Not working 100 15 Retire d 75 11 Annual family income < $20,000 70 11 $20,000 $40,000 141 22 $40,000 $60,000 156 24 $60,000 $80,000 100 15 >$80,000 181 28 Clinical variables Years since diagnosis Mean (SD) 662 7.0 (3.1) Median (Range) 6.6(1.8 22) Se verity of treatment Low autologous no GVHD 390 60 Moderate allogeneic no GVHD 168 26 High allogeneic GVHD 88 14 Comorbid conditions Presence of comorbid conditions at survey 104 16

PAGE 133

133 Table 4 1. Continued N % Patient reported measur es Refin ed FACT BMT score(observed) Mean (SD) 658 37.4 (4.7) Median (Range) 21(13 39) MOS SF 36 Physical Component Summary(PCS) Mean (SD) 658 44.5 (11.6) Median (Range) 48(6.4 64.5) MOS SF 36 Mental component summary (MCS) Mean (S D) 658 50.6 (10.4) Median (Range) 54(10.1 70.2) Refined version of FACT BMT using 13 items selected from previous study ( Chapter 2) Medical Outcomes Survey 36 item short form health survey

PAGE 134

134 Figure 4 1. General latent variable mixt ure model. Y1 through Y3 represent individual items. Factor 1 is a latent factor defined by items Y1 Y3. X represents covariates that can be included in analysis. Item thresholds are conditioned on latent class variable. Factor l oadings are conditioned on latent class variable.

PAGE 135

135 Table 4 2. Latent class and factor mixture model accuracy and fit parameters (n=639) Model LL # parameters AIC BIC ABIC VLRT (p) LRT (p) BLRT (p) Entropy # Class 1 21885 87 43944 4433 2 2 20261 175 40872 41653 41097 0.02 0.02 0.00* 0.90 3 19156 263 38839 40012 39177 0.004 0.01 0.00 0.94 4 18634 351 37971 39536 38422 0.01 0.01 NOCVG 0.95 4C 2F FM 1 19309 129 38877 39452 39042 0.92 FM 2 25288 23 0 51036 52062 51331 0.82 FM 3 1 7146 498 3 5288 37509 35928 0. 83 FM 4 NCVG Note: LL= loglikelihood; AIC= Akaike Information Criterion; BIC=Bayesian Information Criterion; ABIC=Adjust ed BIC; LL= Log Likelihood; VLRT = Vuong, Lo, Mendel & Rubi n Test; BLRT= Bootstrap likelihood ratio test ; NCVG= Best LL not replicated at maximum random starts FM 1: Factor covariance matrix, threshold, and factor loadings fixed across classes; means freely estimated (conditioned on class membership) FM 2: I tem thr esholds and factor covariance matrix invariant ; f reely estimate means and factor loadings; FM 3: Factor variance invariant; f reely estimate means, factor loadings item thresholds FM 4: Factor mean fixed at 0; free ly estimate covariances and variances, ite m thresholds, and factor loadings

PAGE 136

136 Table 4 3. FM 3 four class model item threshold values Class 1: Low PTSS/ Low PTG N=104 Class 2 : Low PTSS/ High PTG N=2 27 Class 3: High PTSS/ Moderate PTG N=170 Class 4: Moderate PTSS/ Moderate PTG N=138 IES1 I thought about it to T1* 1.95 6.08 2.05 0.58 T2 0.07 4.06 3.07 2.65 T3 1.91 1 .35 31.60 7.23 IES4 I have trouble falling asleep or staying asleep because of pictures or thoughts about it that came into my mind T1 0.10 6.02 3.65 2.42 T2 1.49 3.4 8 5.13 3.41 T3 2.81 1.82 7.07 5.46 IES5 I had waves of strong feelings about it T1 2.2 8 7.60 2.25 1.49 T2 0.7 3 5.35 3.48 4.22 T3 1.93 30.25 31.09 5.47 IES6 I had dreams about it T1 0.57 4.17 3.01 2.87 T2 1.21 2.1 8 4.04 4.05 T3 3.48 0.29 28.55 30.29 IES10 Pictures about it popped into my mind T1 2.14 8.50 6.92 0.78 T2 0.74 6.22 9.02 2.59 T3 1.82 2.56 37.56 5.62 IES11 Other things kept making me think about it T1 2.7 9 7.59 2.45 0.68 T2 0.65 5.8 3.69 2.4 3 T3 1.62 2.48 6.07 3.94 IES14 Any reminder brought back feelings about it. T1 2.44 7.47 2.15 1.13 T2 1.00 5.54 2.84 3.27 T3 1.31 1.92 4.93 5.28 IES2 I avoided letting my self get upset when I thought about it or was reminded of it T1 0.78 4.56 1.41 1.22 T2 0.21 3.69 2.32 2.73 T3 1.88 2.84 3.44 3.80 IES3 I tried to remove it from memory T1 0.0 9 7.73 2.58 2.36 T2 0.95 5.98 3.76 3.2 3 T3 4.22 3.43 4.69 4.35 IES7 I stayed away from reminders about it T1 0.19 5.90 2.91 7.8 7 T2 1.43 4.58 4.31 9.35 T3 4.98 2.15 4.9 9 12.27 IES8 happened or it T1 0.68 2.25 1.58 1.74 T2 1.5 5 0.993 2.46 2.07 T3 3.1 1 29.51 3.78 4.22

PAGE 137

137 Table 4 3. Continued Class 1: Low PTSS/ Low PTG N=104 Class 2 : Low PTSS/ High PTG N=2 27 Class 3: High PTSS/ Moderate PTG N=170 Class 4: Moderate PTSS/ Moderate PTG N=138 IES9 I tried not to talk about it T1 2.43 5.38 1.91 2.61 T2 3.9 6 3.59 3.08 5.67 T3 5.92 2.02 4.72 29.75 IES12 I was aware that I still had a lot of feelings deal with them T1 0.81 6.75 2.6 0 2.77 T2 1.1 8 4.21 3.49 4.08 T3 3.60 1.12 30.96 5.72 I ES13 I tried not to think about it T1 0.04 12.13 3.5 5 2.9 3 T2 0.96 9.36 4.74 4.51 T3 4.38 7.17 6.45 6.19 IES15 My feelings about it were kind of numb T1 0.03 5.34 2.45 1.59 T2 1.24 4.03 3.57 2.3 9 T3 3.22 24.0 9 4.73 4.64 PTGI8 Knowing that I can count on people in times of trouble T1 2.52 4.12 2.3 6 0.78 T2 0.50 1.19 0.06 0.44 PTGI7 An increased sense of closeness with others T1 2.48 5.17 1.60 1.26 T2 0.02 1.02 1.35 1.09 PTGI9 An increase d willingness to express my emotions T1 2.45 3.60 1.75 0.09 T2 0.30 0.57 1.24 1.96 PTGI12 Having more compassion for others T1 28.3 6 6.31 2.14 1.09 T2 1.4 4 1.24 0.40 0.4 5 PTGI15 Putting more effort into my relationships T1 1.93 5.59 1.22 0.28 T2 0.02 0.39 1.28 3.48 PTGI18 I learned a great deal about how wonderful people are T1 2.76 5.28 1.95 1.91 T2 0.23 1.81 1.51 0.09 PTGI2 to try to change things which need changing T1 3.02 3.85 1.90 0.45 T2 0.48 1.49 0.74 2.1 0 PTGI13 things with my life T1 1.92 3.027 2.09 0.62 T2 0.29 1.43 2.22 3.22

PAGE 138

138 Table 4 3. Continued Class 1: Low PTSS/ Low PTG N=104 Class 2 : Low PTSS/ High PTG N=2 27 Cl ass 3: High PTSS/ Moderate PTG N=170 Class 4: Moderate PTSS/ Moderate PTG N=138 PTGI14 New opportunities are available which otherwise T1 0.6 7 0.78 0.01 1.13 T2 0.84 2.04 2.34 2.90 PTGI19 I developed new interests T1 0.9 9 2.18 0.67 0.86 T2 0.72 2.20 1.39 3.19 PTGI20 I more readily accept needing others T1 16.00 5.53 0.76 0.22 T2 8.56 0.79 2.22 1.96 PTGI21 I established a new path for my life T1 1.34 1.99 0.65 0.78 T2 0.42 2.21 1.38 2.32 PTGI4 An increased feeling of self reliance T1 1.59 2.71 1.37 0.45 T2 0.05 0.69 0.45 1.75 PTGI6 Knowing I can handle difficulties T1 2.92 29.87 2.23 2.08 T2 0.45 2.93 0.84 0.77 PTGI10 Being better able to accept t he way things work out T1 3 .04 25.95 1.60 0.47 T2 0.28 0.08 0.98 1.35 PTGI17 stronger than I thought I was T1 1.74 4.4 6 1.61 0.64 T2 1.09 1.00 0.13 0.82 PTGI1 My priorities about what is important in life have changed T1 28.52 4.70 2.44 1 .00 T2 2.88 0.76 0.31 0.88 PTGI3 An increased appreciation for the value of my own life T1 27.93 5.14 4.20 1.91 T2 27.8 5 1.81 1.33 0.02 PTGI11 Appreciating each day more fully T1 28. 91 29.51 3.94 2.11 T2 2.68 1.27 0.67 0.69 PTGI5 A better understanding of spiritual matters T1 1.92 4.05 1.85 0.69 T2 0.41 0.95 0.7 9 0.59 PTGI16 I have a stronger religious faith T1 1.0 6 2.82 0.96 0.0 7 T2 0.3 4 0.6 6 1.44 1.01 *T1 T3: threshold values for each item (# of T values is equal to the # of response categories 1. Values are on the logit scale and probabilities can be estimated using the equation P= 1/(1+exp(threshold))

PAGE 139

139 Table 4 4. Description of fo ur classes in FM 3 model PTSS ( )^ PTG (+) Class one Low PTSS/ Low PTG Class two Low PTSS/ High PTG ( ) / (+) ( ) / (+)(+)(+) Class three High PTSS / Moderate PTG Class four Moderate PTSS/ Moderate PTG ( )( )( ) / (+)(+) ( )( ) / (+)(+) ^( ) low, ( )( ) moderate, ( )( )( ) high reports of PTSS; (+) low, (+)(+) moderate, (+)(+)(+) high reports of PTG.

PAGE 140

140 Table 4 5. FM 3 factor means and factor loadings for posttraumatic stress and posttraumatic growth by class membership Class 1 N=104 C lass 2 N=227 Class 3 N=170 Class 4 N=138 Posttraumatic stress Factor variance 3.32 3.32 3.32 3.32 Factor covariance 0.50 0.50 0.50 0.50 Factor mean 0.14 7.8 0 0.42 0.31 Factor loadings 0.72 0.71 1.22 2.07 I have trouble falling asleep or staying asleep because of pictures or thoughts about it that came into my mind 1.00 1.00 1.00 1.00 I had waves of strong feelings about it 0.99 1.05 1.11 *** 1.47 I had dreams about it 1.04*** 0.78 *** 0.36 1.1 8 Pictures about it popped into my mind 0.45 1.10 2.60 1.53 Other things kept making me think about it 0.51 0.97 ** 1.06 ** 0.99 Any reminder brought back feelings about it. 0.77 0.95 1.19 1.44 I avoided letting myself get upset when I thought about it or was reminded of it 0.42 0.62 *** 0.89 ** 1.34 I tried to remove it from memory 1.23* 1.15 1.64 1.18 I stayed away from reminders about it 1.94 ** 0.97 1.19 3.51 0.86 0.42 0.66 0.26 I tried not to tal k about it 2.38 0.82 1.05 0.83 I was aware that I still had a lot of feelings about it, 1.23 1.08 *** 1.12 1.48 I tried not to think about it 1.58 1.72 2.02 1.47 My feelings about it were kind of numb 0.99 0.82 1.00 0.66 Pos ttraumatic growth Factor variance 2.33 2.33 2.33 2.33 Factor covariance 0.50 0.50 0.50 0.50 Factor mean 0.61 0.41 0.53 0.34 Factor loadings Knowing that I can count on people in times of trouble 0.67 0.73 0.82 1.10 An increased sense of c loseness with others 0.85 0.79 1.05 *** 1.21 *** An increased willingness to express my emotions 0.83 0.77 0.69 0.90 Having more compassion for others 0.70 1.09 ** 1.23 *** 0.58 Putting more effort into my relationships 1.16 1.08 *** 1.18 *** 0.70 I learne d a great deal about how wonderful people are 1.08 0.78 1.47 *** 0.86 ** need changing 0.40 1.28 *** 0.73 0.82 0.86 1.02 2.15 ** 1.10 New opportunities are availab have been otherwise 0.89 0.82 *** 0.98 0.35

PAGE 141

141 Table 4 5. Continued Class 1 N=104 Class 2 N=227 Class 3 N=170 Class 4 N=138 I developed new interests 1.13** 1.03 1.07** 0.28 I more readily accept needing others 13.48 0.90 1.04* 0.39 I established a new path for my life 0.89** 1.42 1.17** 0.52 An increased feeling of self reliance 0.44 0.90** 0.75** 0.99 Knowing I can handle difficulties 1.00 1.00 1.00 1.00 Being better able to accept the way things work out 0.77 0.87 1.10** 0.97 I 0.90* 0.94** 0.88* 0.58 My priorities about what is important in life have changed 0.06 0.77* 0.75 0.94** An increased appreciation for the value of my own life 0.25 0.73** 0.95 1.27 Appreciating each da y more fully 1.10* 0.65** 1.42 1.58 A better understanding of spiritual matters 1.00 1.60** 0.83 1.10 I have a stronger religious faith 0.87 1.28* 0.91*** 1.04 *p<0.05; p<0.01, p<0.001. ^ Class 1= Low PTSS/Low PTG, class 2= Low PTSS/ High PTG, class 3= H igh PTSS/ Moderate PTG, class 4= Moderate PTSS/ Moderate PTG.

PAGE 142

142 Table 4 6 Results from multinomial logistic regression evaluation effects of covariates on class membership (reference= low PTSS/ high PTG (n=2 27 ) Class 1: Low PTSS/ Low PTG N=104 Class 3: High PTSS/ Moderate PTG N=170 Class 4: Moderate PTSS/ Moderate PTG N=138 b (SE) b (SE) b (SE) Gender (r ef=Male) Female 0.20 (0.27) 0.16 (0.25) 0.35 (0.26) Current age (ref=<30) 35 39 years 0.31 (0.67) 0.62 (0.48) 0.07 (0.52) 40 45 years 0.84 (0.56) 0.59 (0.48) 0.18 (0.52) 46 50 years 0.54 (0.51) 0.62 (0.47) 0.39 (0.45) 50+ years 0.78 (0.48) 0.14 (0.41) 0.18 (0.50) Race/ethnicity (ref=White) Other 0.66 (0.60) 0.12 (0.47) 1.04 (0.40)** Occupational status (Ref=Worki ng) Not working 0.62 (0.39) 0.50 (0.36) 0.30 (0.40) Retired 0.60 (0.48) 0.09 (0.41) 0.10 (0.39) Marital status (ref=Y es) No 0.13 (0.29) 0.39 (0.26) 0.19 (0.28) Education (ref= >C ollege degree ) High school or below 0.06 (0.38) 0.64 (0. 33)* 1.17 (0.38)** Some college or technical education 0.08 (0.36) 0.13 (0.33) 0.78 (0.37)* College degree 0.06 (0.39) 0.27 (0.36) 0.21 (0.41) Comorbid conditions (Ref=N o) Yes 0.01 (0.38) 0.05 (0.35) 0.41 (0.34) Severity of treatment experien ce (ref=Low) Moderate 0.35 (0.31) 0.41 (0.29) 0.11 (0.30) High 0.06 (0.40) 0.34 (0.34) 0.10 (0.37) Years since diagnosis (ref= greater than 6.5years ) Less than 6.5 years 0.62 (0.26)* 0.84 (0.24) 0.14 (0.25) Level of physical symptoms (re f= low; below median 21) High (above 21) 0.01 (0.26) 0.88 (0,25)*** 0.23 (0.25) *p<0.05; p<0.01, p<0.001 Some graduate school or graduate degree. Modified FACT BMT scale score: no cutoff available yet, so median value of 21 (range 13 39) split was obtained. Those with a score of 21 or above considered higher symptoms, below 21 lower symptoms

PAGE 143

143 Table 4 7 Results fr om the Wald Chi Square 2 ) tests of mean equality of the auxiliary analyses of outcomes Physical HRQOL Mental HRQOL Physical HRQOL Men tal HRQOL Unadjusted Mean (SE) Unadjusted Mean (SE) Adjusted Mean (SE) Adjusted Mean (SE) Class 1^ 44.11 (1.21) 50.30 (1.47) 56.16 (7.03) 56.05 (7.47) Class 2 46.63 (0.81) 52.61 (0.95) 48.10 (6.25) 62.87 (3.97) Class 3 41.27 (0.96) 46.30 (0.95) 56.12 (5.30) 44.77 (4.62) Class 4 45.05 (0.97) 53.38 (1.14) 59.87 (23.02) 52.35 (6.36) 18.56*** 28.08*** C1 vs C2 2.80 1.26 1.06 0.63 C1 vs C3 3.33* 5.08* 0 1.56 C1 vs C4 0.37 3.03 0.02 0.17 C2 vs C3 18.08*** 21.69*** 1.52 8.01** C 2 vs C4 1.46 1.26 0.18 2.88 C3 vs C4 7.46** 18.09*** 0.02 1.19 *p<0.05; **p<0.01 ; ***p<0.001. Note: C= Class; ^Class 1= Low PTSS/Low PTG class 2 = Low PTSS/ High PTG class 3= High PTSS/ Moderate PTG class 4= Moderate PTSS/ Moderate PTG test was not calculated for adjusted models due to singular covariance matrix across classes because of empty cells in predictor variable values; individual tests comparing classes individually was doing using Model TEST command

PAGE 144

144 CHAPTER 5 IMPROVING PATIENT REPORTED OUTCOMES RESEARCH IN LONG TERM SURVIVORS OF HEMATOPOIETIC STEM CELL TRANSPLANT: DISCUSSION OF RESULTS Review of Findings We utilized a unique study population comprised of long term survivors of hematopoietic stem cell transplant (HSCT) to addre ss several research gaps in the measurement and evaluation of their patient reported outcomes (PROs) These research gaps included 1) a lack of validated scale to measure physical symptoms for long term HSCT survivors 2) a lack of an evidence based model to describe the relationships among physical and psychological symptoms, psychosocial factors, and the joint effects of these factors on HRQO L, and 3) limited evidence is available on the positive psychological effects in long term HSCT survivors. Three specific aims were established to address these research gaps. The first aim was to refine the HSCT specific physical symptom scale for long term survivors ; the second aim was to develop and test a conceptual framework of factors influencing PROs of HSCT survivors ; the third aim was to examine the population heterogeneity based on posttraumatic stress symptoms ( PTSS ) and posttraumatic growth ( PTG ) in HSCT survivors and the impact on HRQOL. The overall findings for the first aim suggest a refined 13 item ph ysical symptom scale for long term survivors of HSCT demonstrated appropriate measurement properties and known groups validity related to other health markers. In the second aim, we found that phy sical symptoms had a stronger direct effect on physical HRQ OL compared to psychological symptoms (depressive symptoms and PTSS). In contrast, psychological symptoms had the stronger direct effect on mental HRQOL compared to

PAGE 145

145 phy sical symptoms. Psychological symptoms, especially depressive symptoms, mediated the r elationship between physical symptoms and physical HRQOL as well as the relationship between physical symptoms and mental HRQOL. We also found that psychosocial variables (optimism, coping, and social constraints) explained the variance in pathway from ph ysical symptoms through depressive symptoms to physical HRQOL; whereas the psychological mediating pathway between physical symptoms and mental HRQOL remained significant. For the third aim, we found the two factor four class model with partially relaxed conditional independence assumption using mixture modeling, provided unique information into the latent structure and population heterogeneity of PTSS and PTG in long term HSCT survivors. Chapter 2: Refine the HSCT Specific Physical Symptom S cale M easuring treatment related symptoms through a PRO s measure help obtain comprehensive information [76] Instrument development, evaluation, and refinement are an integral process toward of measuring health status. The FACT BMT was initially developed using classical test theory approach (CTT) [37] The first objective of Chapter 2 was to use item response theory (IRT) methodology to ref ine the 25 item FACT BMT scale for measuring physical symptoms of long term HSCT survivors. The second objective was to validate the refined scale based on health markers including mental and physical HRQOL as well as other clinical information. Althoug h the majority of the studies on PROs instrument development and refinement were based on CTT in the past 3 decades, the use of m odern measurement methodology (e.g., IRT) has the advantage to address the problems associated with CTT, typically sample and s cale dependence. To accomplish the first objective, we

PAGE 146

146 used several criteria to select the items with the acceptable measurement properties. These criteria included item content, item response distribution, the IRT assumptions of unidimensionality and lo cal independence, the IRT measurement properties of discrimination and difficulty, and differential item functioning (DIF). To evaluate known groups validity, linear regression was conducted to estimate the difference in physical symptom scores between su rvivors with high and low health status as measured by the the mean difference in scores of the symptom scale between different levels of HRQOL and clinical information (severity of treatment experience, comorbid conditions, intensity The final version of the FACT BMT (13 items) shows improved fit indices from the initial version of the FACT BMT (25 items). Item difficulty values were clustered toward the lower end or less severe latent trait, which suggests that the items of the modified FACT BMT mostly capture those survivors with lowest level of severity of the physical symptom trait The majority of th e items had discrimination values between 0.60 and 1.00, suggesting the items had moderate ability to discriminate between participants separated by small differences in the underlying symptom trait [12] The refine d FACT BMT had good known groups validity with the PROs measures of HRQOL, moderate known groups validity with the clinical measures of severity of treatment experience and comorbid conditions, and poor known groups ale (physician report of health). For example, the effect sizes were largest when comparing physical HRQOL scores in the above general population norm group to the poor physical HRQOL score. The lowest effect

PAGE 147

147 sizes were found for KPS rated by clinicians. This is consistent with the previous [107, 108] thus adding value to including PROs as important indicat ors of health that can supplement clinical information. Instrument refinement is a continuous process requiring re evaluation of current measures as characteristics of the population (e.g., aging) or the side effects related to treatments may change. Th is is especially true with symptom measures as treatments and the subsequent long term effects evolve over time. Because the items generated for the FACT BMT were originally developed in 1996, future research should first focus on the identification of ne w symptoms and/or elimination of symptoms no longer related to current treatment methods through both qualitative (i.e. focus groups and interviews) and quantitative approaches. Second, once new (if any) symptoms are identified, specific items should be f urther developed and tested. For example, if subjects consistently reports becoming tired easily, experiencing dizzy spells, and having no appetite, items measuring these symptoms are clustering to represent a physical state that may be useful in directin g treatment [74] And finally, it is important to note that self reported symptom (i.e. pain, fatigue, dizziness) may be more sensitive to changes health condition compared to clinical measures (i. e. blood pressure) [76] Our study could not verify sensitivity to a change in a health condition (i.e., responsiveness) given the cross sectional study design [12] A future goal for evaluating the symptom instrument for HSCT survivors is to test the responsiveness to

PAGE 148

148 Chapter 3: D evelop and Test a Conceptual Framework of Factors Influenci ng PROs of HSCT S urvivors After the physical symptom scale was refined and tested (Chapter 2), it is important to include this scale in an empirical study to test how physical symptoms, alongside other factors, contribute to physical and mental HRQOL. Cha pter 3 explored the influence of psychological symptoms and psychosocial factors on the relationship of physical symptoms with HRQOL using structural equation modeling (SEM). In the primary objective, we constructed pathways examining whether psychologica l symptoms (depressive symptoms and PTSS) mediated the relationship between physical symptoms (measured by the revised scale from Chapter 2) and physical and mental HRQOL. In the secondary objective, we created a comprehensive conceptual framework incorpo rating psychosocial variables (optimism, social constraints, and coping) along with physical symptoms, depressive symptoms, and physical and mental HRQOL. Results of the primary objective indicated significant direct effects of physical symptoms on both p hysical and mental HRQOL in the full model, with physical symptoms having stronger direct effects on physical HRQOL rather than on mental HRQOL PTSS significantly mediated the relationship between physical symptoms and physical HRQOL. For mental HRQOL, psychological symptoms (especially depressive symptoms) had a stronger direct effect on mental HRQOL than did physical symptoms. Depressive symptoms and PTSS also significantly mediated the effects of physical symptoms on mental HRQOL In separate models of depressive symptoms and PTSS, the mediating effects of these two variables remained significant in the nested models when evaluating the relationship between physical symptoms and mental HRQOL.

PAGE 149

149 In the secondary objective, optimism, social constraints, and coping played a larger role to change the association of physical symptoms with physical HRQOL than with mental HRQOL. P sychosocial variables likely explained the mediating effect of PTSS on the relationship between physical symptom and physical HRQO L identified in objective (although not directly tested in the full framework). For mental HRQOL, only coping was significantly associated with mental HRQOL and the magnitude of its effect of the pathway from physical symptoms through depression to mental HRQOL remained unchanged. Chapter 3 contributes to the literature by building the relationships of physical symptoms, psychological symptoms, psychosocial factors, and HRQOL together in a larger conceptual framework, and testing specific pathways regarding physical symptoms, psychological symptoms, and HRQOL. Previous studies have evaluated some combination of the important factors for influencing HRQOL in HSCT, but have not included all of the factors in the same model [26, 42, 42, 161] SEM is a useful technique to test the conceptual framework and specific pathways [25] Through the evaluation of the pathways that included symptoms, psychosocial factors, and HRQOL, we provided evidence that psychosocial factors do play role in physical HRQOL. SEM analyses based on our framework identified two possible points of interven tion for physical HRQOL: physical symptoms and psychosocial factors such as social constraints. The importance of physical symptoms on physical HRQOL is consistent with findings of symptom burden in cancer survivorship [212] Additionally, the findings

PAGE 150

150 HRQOL, with an emphasis on the stronger impact of depressive symptoms compared to PTSS. Chapter 3 provides some foundati on for future studies regarding PROs research in HSCT survivors. First, future studies may test our proposed conceptual framework in a longitudinal design for HSCT sample populations. Although no causal implications can be made in our study, SEM analysis hypothesized temporal relationships between variables and demonstrated how the inclusion of certain factors may affect previously established relationships (e.g. inclusion of psychosocial factors explaining the pathway from physical symptoms through depre ssive symptoms to physical HRQOL). Second, this study may be extended by applying the same framework to alternative outcomes such as mortality/survival or secondary cancer. Better HRQOL has been suggested to be a significant predictor of survival in canc er patient and survivor populations [213, 214] Third, our model represents one of several plausible conceptual frameworks. For example, PTSS has been theorized to result in physical sy mptoms or have physical manifestations of stress [205] A future study attempting to differentiate physical symptoms due to disease and treatment and symptoms related to PTSS such as back pain, headaches, or gas trointestinal distress may be of interest to fully explore this possibility. Chapter 4: E xamine the Population Heterogeneity Based on PTSS and PTG in HSCT S urvivors and the I mpact on HRQOL In Chapter 3, we identified significant relationships between phy sical symptoms psychological symptoms, psychosocial variables, and HRQOL. As previously discussed in Chapter 4, the experience of cancer, including diagnosis and treatment, may lead to a trauma related and negative and positive psychological outcomes such as PTSS and

PAGE 151

151 posttraumatic growth (PTG) from cancer may co exist in cancer patients and survivors and the impact on HRQOL is still mixed [208] Chapter 4 further explored the self reported psychological impact of HSCT on long term survivors through two objectives. In the first objective, we used a factor mixture modeling (FMM) approach to classify long term HSCT survivors based on their responses to PTSS and PTG. In the second objective, we evaluated the specific predictors of class membership and examined the extent to which the class membership was associated with HRQOL outcomes. We incorporated categorical (latent class model) and continuous approache s (common factor model) to generate class membership and examine the heterogeneity introduced by PTSS and PTG between different classes; we also allowed for severity of PTSS and PTG to be varied within each class (i.e., relaxing conditional independence as sumption of latent class models) [62] The FMM approach resulted in a two factor (PTSS and PTG) model and a four class model, which was further used for testing the mixture model variations. The best fitting mixtu re model partially relaxed the conditional independence assumption by allowing for factor means of PTG and PTSS, item thresholds, and factor loadings to vary across classes. The four classes were defined as low PTSS/ low PTG, low PTSS/ high PTG, high PTSS / moderate PTG, and moderate PTSS/ moderate PTG. Survivors in t he low PTSS/ low PTG class had the lowest level of PTG compared to other three classes (Table 4 4) Survivors in t he high PTSS/ moderate PTG class had the highest level of PTSS and survivo rs in the low PTSS/ high PTG class had the lowest level of PTSS compared to other classes. The presence of some

PAGE 152

152 level of PTG in all classes was not unanticipated based on the hypothesis that the oscillation, or perhaps circular process, of intrusion and a voidance of PTSS may act as response [215] Essentially, the adaptation to stress may be part of cognitive processing which does not necessarily result in distress and may lead the survivor to establish a positive reinterpretation of the trauma or posttraumatic growth [216] Our findings that lower education and a race/ethnicity other than Wh ite were significant predictors of membership in the moderate PTSS/moderate PTG class associations with reports of PTSS and PTG [199, 201] Also, previous studies suggested that worse physical symptoms were associated with higher levels of PTSS [27, 169, 204] This is supported by our finding that worse physical symptoms predicted membership of high PTSS/ low PTG class. Class membership was not significantly associated with physical HRQOL compared to mental HRQOL in the predictor adjusted comparisons. No statistically signifi cant differences were found between classes in adjusted physical HRQOL. Physical HRQOL of survivors from all classes (except for the low PTSS/high PTG class) were above the general population norm. The below norm physical HRQOL in the low PTSS/ high PTG discrepancy did no t reach the minimal clinically important difference Adjusted mental HRQOL was statistically significantly between the low PTSS/ high PTG class and the high PTSS/ moderate PTG class. Mental HRQOL of survivors in the high PTSS/ moderate PTG class was greater than the norm and reached the clinically minimally important difference.

PAGE 153

153 It is very challenging to establish diagnostic classification systems or profiles for individuals based on PROs measures. Previous studies have attempted to inv estigate the negative and positive psychological states (e.g., PTSS and PTG) experienced by long term HSCT survivors [29] but no studies have examined the population heterogeneity of PTG with PTSS simultaneously. Suma lla [55] and Park [31, 32] warn ed that evaluating mixed response (i.e., positive and negative) measures creates difficulty in scoring and interpretation. In cluding PTSS and PTG in typical regression models as separate predictors for HRQOL merely help understand the common variance of PTSS and PTG and inform on the additive effects and multiplicative effects, given an interaction term is included However, th e traditional approach neglects the population heterogeneity. This study was able to demonstrate that the use of FMM provided useful information for describing the population heterogeneity. However, future research should focus on tracking on the experie nce of PTSS and PTG over time because theoretically PTSS will decrease and PTG will increase over time [171] Analytically, growth mixture modeling incorporates the element of time into the model [179] allowing for evaluation of the fluctuations of reports of PTSS and PTG, and response. Using a classification model for psychological impact m ay provide insight into how to identify at risk clusters of individuals using on their reports of PTSS and PTG to better tailor interventions, such as cognitive behavioral therapy, psychoeducation, or complementary and alternative medicine, for patients ex periencing trauma [29, 153] The importance of cognitive processing is highlighted by McCullough and colleagues for

PAGE 154

154 traumatized individuals to express the traumatic features of their event and the p ersonal benefits [209] Findings suggest that positive growth occurred when the individual acknowledges the severity of the trauma and tr ies to make sense of the trauma [209] The authors argue that balancing mechanism exists that is associated with experiencing the trauma, cognitive processing, and moving forward [209] Evaluating Disease and Health Outcomes In clinical e pidemiologic research, tracking disease and treatment outcomes has focused on clinical information and physician report The results of each aim (Chapter 2 through Chapter 4) will contribute to epidemiological health outcomes research in the long term HSC T survivors. Symptoms and HRQOL have become an important endpoint of clinical trials and other regulatory agencies when evaluating different treatment regimens [ 23, 76] The results in Chapters 2 and Chapter 3 show that symptoms, both physical and psychological, are not just clinical endpoints but are factors strongly associated with mental and physical HRQOL outcomes. Additionally, Chapter 4 demonstrates the c omplexities of evaluating two interrelated psychological states (e.g., PTSS and PTG) and the relation of class membership with HRQOL. Exploring the use of classification techniques such as FMM may be useful for future studies examining patient factors tha t may produce population heterogeneity. Physicians and patients often face several choices regarding treatment plans with equal survival rates, and the final decision or preference of patients may be influenced by physical and psychological symptoms, func tional status, and HRQOL [217] Obtaining the patient or participant perspective on their health outcome is an invaluable approach to tracking disease and treatment [12]

PAGE 155

155 Limitations The findings presented should be interpreted with the knowledge of several limitations. First, t he study population is largely homogenous (91.3% White). Results may not be generalized to populations comprised of different demographic and clin ical background Second, t he cross sectional nature of the data does not allow for estimating the causal relationship between variables of interests. Specifically, for the conceptual framework and path analysis, the findings can be interpreted as signif icant associations but not as a causal pathway. Future studies may test the validity of the conceptual fr amework with a longitudinal design. Third, this study may be subject to selection bias with respect to health status of survivors. Study participat ion may indicate that the survivor is functioning well enough to complete the process compared to those who decline due to poor health. This selection bias of a healthier sample indicates that we may underestimate the reporting of symptoms or overestimate the overall HRQOL status. L ongitudinal studies have noted that patients who continue d to engage in the entire study period compared to those discontinued were in better health, leading to over estimates of functi onal status [109] Evidence suggests a significant association between HRQOL trajectory and disease progression, thus concluding that the HRQOL in those who drop out of study is worse tha n those who remain in the study [119, 218, 219] Finally t he IES used in this study only contains 2 (intrusive thoughts and avoidance) of 3 (hyperarousal) domain s that are now used to diagnose posttraumatic stress disorder The original IES in this study contains the intrusive thoughts and avoidance domain but does not include the hyper arousal domain. The revised IES

PAGE 156

156 was not readily available at the time of the study. Given not measuring the third domain of hyperarousal, we may miss some components of PTSS and may underestimate the number of symptoms reported. Sub syndromal PTSS has received attention in the literature, but a standard definition has yet to be determined [115] Some d efinitions require one symptom in at least two domains and others require one symptom in each of the three domains [220] Conclusion This dissertation addressed some of the measurement and assessment issues related to the complexities of PROs in long term HSCT survivors. Several main points are important to note for this study and in the broader context of epidemiology outcomes research. First, symptoms, both physical and psychological, are important factor s for influencing physical and mental HRQOL in this population. Physical symptoms are a clinical measures. Secondly, the refinement of the FACT BMT is a first step of an ongoing, iterative process to generating physical symptom measures in this population. Careful development or selection of appropriate instruments should be paramount when designing epidemiologic studies to minimize measurement errors and maximize accura cy of the results. Third, mapping the relationships between physical symptoms, psychological symptoms, psychosocial factors, and HRQOL through SEM allowed for identifying important relationships that influence HRQOL. Conceptual frameworks provide a found ation for interpreting findings and comparing results across studies. Finally, population heterogeneity as well as within class heterogeneity due to PTSS and PTG can be identified through a mixture modeling approach. While clinical and biological measure

PAGE 157

157 status affecting their day to day functioning, and they are important to consider when determining methods for measuring health outcomes.

PAGE 158

158 APPENDIX A CHAPTER 2 DETAILED METHODS FOR DIF AND IT EM SELECTION Differential Item Functioning An important part of the instrument validation or evaluation process is to determine whether each item within a multi item scale operates the same way for different groups of study respondents [221] In IRT, the discrimination and difficulty parameters are assumed to be invariant over samples drawn from the same population [92] Differential item functioning ( DIF ) analysis is a method that can be used to identify items intended to measure the trait of interest (physical symptoms) which are responded differently by sub groups (e.g. gender or treatment type) after controlling for the underlying latent trait of symptom [92, 93] For example, do males and females underlying physical symptom trait? Gender differences are often reported in ph ysical symptoms across diseases, so this particular sub group is important to evaluate [222 224] For treatment type, the sub groups tested were autolog ous and allogeneic. This scale is intended for survivors of both treatment types and the items need to be evaluated for invariance in both groups. Technically, two types of DIF can be identified: uniform and non uniform In uniform DIF, an item shows the same level of DIF regardless of the underlying trait level (analogous to the concept of confounding). Non uniform DIF is a difference in discrimination parameters between groups. In non un iform DIF the magnitude of DIF varies based on the underlying trait level (analogous to the concept of effect modification). The presence of DIF by gender or treatment type can threaten the validity of an instrument and is a useful criterion for item removal. However, not all DIF

PAGE 159

159 is clinically meaningful and the magnitude of DIF should be considered [94] DIF analysis is not the primary aim of this analysis and the purpose is to help identify biased items rega rdless of the direction or type of DIF. Several statistical methods are available to evaluate DIF, with the major categories being CTT and modern test approaches like IRT. An IRT approach on uniform DIF [101, 225] A DIF analysis specifying gender as a group and then treatment type as a group was conducted for each stage. An EM algorithm was us ed and starting values for threshold and slopes were applied to facilitate analysis [101] The chi square tests of item location and item slope contrasts were evaluated for significant DIF in each item with chi squar e values >10 considered a minimal clinically important difference. Item evaluation For stage one, items were evaluated for at least three measurement issues with the primary criteria acting as the initial marker for further evaluation. In addition to t he primary criteria, slope, location, ICC, and DIF were evaluated for items that were flagged for the primary criteria of content issues, ceiling effects, or local dependency (residual correlation). At this stage, if an item had discrimination (slope, a ) and difficulty (location b ) parameter values that were below desired values (discrimination <0.5 or difficulty outside of 2 standard deviations from mean) [91] it was retained if this was the only violation identif ied for the item. In stage two, we collapse d the categories from 5 to 3 by combining the lowest two categories as well as the middle categories to create the three categories T he criteria for item removal focused on the discrimination, difficulty, ICC, and DIF criteria.

PAGE 160

160 Items were considered for removal if they met two or more of the following criteria: discrimination values less 0.5, difficulty value greater than 2 standard deviations from the mean, poor ICC, and significant DIF with large magnitude ( chi square >25) in either gender or treatment type. For stage three it ems were considered for removal if they met one or more of the criteria that was used in stage two. The same criteria for consideration of removal included discrimination values less 0 .5, difficulty values greater than 2 standard deviations from the mean, poor ICC, and significant DIF. In the final stage (stage four), IRT parameters and ICCs were evaluated for the previously stated criteria and were also considered with respect to thei r performance in the previous stages. Additionally, the TIF was compared to preliminary stages to evaluate for final measurement reliability. Items with DIF were evaluated to determine if significant enough (chi square greater than 10) to require adjustm ent in later analyses. The latent factor scores based on the stage four item set was then exported for the known groups validity analysis. The latent trait or ability parameters were estimated using expected a posteriori (EAP) estimations.

PAGE 161

161 APPENDIX B CH APTER 2 DETAILED RESULTS FOR ITEM SELECTION urinating difficult Content issues were raised for BMT1 due to its lack of assessing an actual symptom related to the disease or treatment Additionally the discrimination value of BMT1 was less 0.5 and difficulty was grea ter than 2 standard deviations from the mean ( a =0.455 b =2.131 ). The item response distributions of the initial set of 25 items were skewed left with 16 of 25 items having at least 40% of responses in the highest category. We found ceiling effects with >80% of re sponses in the highest category for three items Additionally, BMT5 was flagged for ceiling effects (CE=84.5%). The frequency distribution of BMT5 also showed that none of the participants responded in the lowest/worst category. The difficulty parameter was shifted almost 3 standard deviations from the mean ( b= 2.919). BMT10 was identified with local dependency based on a residual correlation BMT10 also demons trated worse discrimination and difficulty parameters ( a= 0.135, b= 8.75) compared to BR1 In addition, this item was identified with DIF by gender, both uniform and non uniform. P7 also was identified to have both ceiling effects (CE=87.5%) as well as l ocal issues or ceiling effects, thus P7 was selected for removal. SCL5 was also identified

PAGE 162

162 with ceiling effects (CE=83.1%) as well as a difficulty greater than 3 standard deviati ons from the mean ( b =3.247). For stage two, four additional items were selected for removal. These included BL4 had a poor discrimination value in stage two ( a =0.344) as well as a significant item fit p value that indicates it does not fit well with the other items in the IRT analysis. In addition, BL4 was identified with large magnitude of DIF by gender. BMT7 had low discrimination as well as a highly skewed difficulty parameter towards the lower end of the latent trait ( a= 0.120, b = 10.687). The ICC was approximately horizontal suggesting that the probability of endorsing the item did not fluctuate with levels underlying latent trait Additionally, IIF for BMT7 indicated almost no information was provided from this item. BMT14 and SCL2 both had difficulty values greater than 2 standard deviations from the mean ( b = 2.62 7 b= 2.030, respectively). T he BMT14 ICC was skewed towards the lower end of the latent trait with limited threshold range SCL2 had a discrimination value less 0.500 ( a =0.450), a skewed ICC curve, and a low IIF supp lementing the rational for removal. BMT11 had a discrimination value of 0.497 and was also iden tified with DIF by treatment type. BMT15 had a discrimination and difficulty value within the criteria ( a= 1.393, b= 1.398), but the threshold values indicated it covered a limited level of the latent trait resulting in less than desirable ICC. Additiona lly, BMT15 was identified with

PAGE 163

163 sign ificant DIF by treatment type. At this stage, the discrimination and difficulty of SCL4 fell below the acceptable criteria ( a =0.450, b =2.03). Finally in stage four, the 13 i tem difficult y values ranged betwe en 1.149 for C7 and 2.194 for SCL3. SCL3 also demonstrated a lower discrimination at this stage but had good item properties in each of the previous stages. The item is retained due to clinical relevance, item properties are only marginally outside of t he desired range, the fit indices are acceptable, and the item fit p value does not indicate removal (p=0.973). SCL6 showed mild DIF by gender, but the magnitude was low (location contrast of 0.610; chi square <10).

PAGE 164

164 Figure B 1 Test information function for final 13 item physical symptom scale

PAGE 165

165 APPENDIX C CHAPTER 3 INSTRUMENT MEASUREMENT PROPERTIES Table C 1. Measurement properties of instruments Latent variable (Instrument) Factor loading (Unstandardized) Factor Loading (STDYX) CFI RMSEA Phy sical symptoms ( modified FACT BMT 13 items ) Like the appearance of my body 0.505 0.505 0.96 0.06 Get tired easily 0.743 0.743 Eyesight is blurry 0.584 0.584 Be content with quality of my life right now 0.594 0.594 Have dizzy spells 0. 739 0.739 Have trouble with my bowels 0.527 0.527 Be able to concentrate 0.582 0.582 Feel distant from other people 0.588 0.588 Be bothered by a change in the way food tastes 0.655 0.655 I have a good appetite 0.554 0.554 Have been short of breath 0.677 0.677 Have headaches 0.627 0.627 Have hearing loss 0.391 0.391 PTSS (Impact of Events Scale 2 domains ) Intrusive thoughts domains 1.00 0.844 N/A* Avoidant thoughts domain 1.02 0.848 Depressi ve symptoms (Center for Epid emiological Studies Depression 4 domains ) Depressed affect 3.904 0.913 0.99 0.02 Somatic 3.452 0.801 Interrelationships 0.534 0.539 Low positive affect 1.774 0.703 Optimism (Life Orientation Test, 12 items) In uncertain times I usually expect the best. 0. 610 0. 610 0.88 0.16 0.559 0.559 If something can go wrong for me, it will. 0.721 0.721 I always look on the bright side of things. 0.780 0.780 0.722 0.722 I enjoy my friends a lot. 0.490 0.490 0.067 0.067 I hardly ever expect things to go my way. 0.877 0.877 Things never work out the way I want them to. 0.841 0.841 0.507 0 .507 0.553 0.553 I rarely count on good things happening to me. 0.777 0.777

PAGE 166

166 Table C 1. Continued Latent variable (Instrument) Factor loading (Unstandardized) Factor Load ing (STDYX) CFI RMSEA Social constraints (Social Constraints Scale, 16 items) Change the subject when you tried to discuss your experience with cancer or the blood or marrow transplant (BMT)? 0. 756 0. 756 0.94 0.11 Not seem to understand your s ituation? 0.777 0.777 Avoid you? 0.750 0.750 Minimize your problems? 0.795 0.795 Seem to be hiding their feelings? 0.779 0.779 Act uncomfortable when you talked about your experience with cancer or BMT? 0.844 0.844 Trivialize your problems? 0.837 0.837 Complain about their own problems when you wanted to share yours? 0.729 0.729 Act cheerful around you to hide their true feelings or concerns? 0.699 0.699 Tell you not to worry so much about your health? 0.704 0.704 Tell you to try not to think about cancer or BMT? 0.753 0.753 about your experience with cancer or BMT? 0.911 0.911 Make you feel as though you had to keep your feelings about your experience with cancer or BMT to your self, because your feelings made them feel uncomfortable? 0. 948 0. 948 Make you feel as though you had to keep your feelings about your experience with cancer or BMT to yourself, because your feelings made them feel upset? 0.909 0.909 Let you down by not showing you as much love? 0.741 0.741 Do you still feel the need to talk about your experience with cancer and/or BMT? 0.342 0.342 Coping (Brief COPE) Problem focused or approach coping 1.000 0.309 N/A* Avoidant coping 1.464 0. 767 Physical HRQOL (MOS SF 36 PCS^, 4 domains ) Role functioning physical 31.55 0.781 0.99 0.05 General health 17.03 0.717 Bodily pain 20.35 0.803 Physical functioning 20.78 0.812

PAGE 167

167 Table C 1. Continued Latent variable (Instrume nt) Factor loading (Unstandardized) Factor Loading (STDYX) CFI RMSEA Mental HRQOL (MOS SF 36 MCS^, 4 domains ) Role functioning emotional 25.798 0.735 0.99 0.08 Mental health 13.895 0.747 Vitality 17.681 0.748 Social functioning 19.172 0.77 9 ^Medical Outcomes Study 36 item short form health survey. PCS: Physical Component Summary, MCS: Mental Component Summary

PAGE 168

168 APPENDIX D CHAPTER 4 DETAILED METHODS Factor Mixture Models A factor mixture model is a broad concept encompassing hybrid approa ches to analyzing both categorical and continuous latent variables [61] Factor analysis and Latent class or profile analysis can be thought of as specific cases of FMM with particular parameter restr ictions under the factor mixture model umbrella [62] The common factor model (i.e. factor analysis) is a latent variable approach allow ing for modeling of a continuous latent variable (factor) that represents the un derlying latent trait. The primary advantages over traditional regression models are the allowance of different weights in different items in estimating the factor score, and the ability to handle multidimensional data. However, the restriction of the co mmon factor model includes an assumption of a homogeneous population, meaning the interpretation and metric is equally applied across the range of the factor scores (or levels of latent traits) [6 2, 226] Latent class analysis (LCA) is used to create categorical classes of individuals based on the unobserved population heterogeneity while assuming within class homogeneity [183] Latent profile analys is (LPA) is an alternative to LCA in identifying classes by allowing for handling the continuous observed variables (i.e. items) [62, 179] The ability of LCA to create heterogeneous classes of the population is the key difference between the latent class model and the common factor model approaches [62] LCA allows for identifying sub populations that have similar scores on categorical or ordinal observat ions [62] Similar to common factor model approaches, LCA allows

PAGE 169

169 for different weights in different items in estimating the factor score [61] The assumption of within class homogeneity (i.e., local independence) of the LCA limits its application to this study because this assumption indicates that all survivors within each class are the same with respect to the underlying PTSS and PTG latent trait. In many ca ses, within class heterogeneity meaning the metric of measures is not equal across the range of the factor scores (or levels of latent traits) may exist, which will violate the local independence assumption [227] If present, t he within class variation can provide i n the classes (Figure D 1) The FMM may be useful for examining psychological concepts due to its flexi bility in accommodating the theoretical structure of these concepts [193, 228] One perspective suggests that psychological states can be represented by diagnostic categories or classes. Ano ther alternative view is that these states are dimensional and should be represented by continuous distributions (i.e. individuals can varying levels of accounting f or continuous distribution (factor analysis) a potentially useful analytic method. A FMM analysis can ad o pt a confirmatory or exploratory approach with respect to class enumeration and constructing the measurement model (i.e. number of latent factors). T here are many variations of FMM, with some being more restrictive than data and hypothesis. This study followed the general guidelines suggested by Clark & Muthen to h elp build the FMM through a structured approach [195] First, to establish

PAGE 170

170 the best factor and class solutions, the LCA model and FA model was determined separately for the data. The best fitting FMM will be compared to the previously established LCA and FA to evaluate whether alternative models are plausible and more parsimonious. Comparisons of model fit will be based on information criterion and class interpretation. The most useful and best fitting final model will be used for additional analysis for covariates and distal outcomes. Fitting Factor Model and Latent Class M odel The focus of this study was to examine psychological response to trauma, incl uding specific positive and negative outcomes. The concepts of PTSS and PTG are well defined as independent concepts, thus a CFA approach to confirming the presence of two separate domains was be taken. (Figure D 2A) The latent class model identifies s ubgroups of a study population. Class membership is determined by either the response pattern to items (categorical or continuous) [229] is the local independence assumption. This assumption has two implications. First, all correlation between items is assumed to be explained by the classes. Second, each individual within a class is assumed to have the same conditional probability for each item (Figure D 2B). To determine the appropriate number of classes for the FMM in this study, the LCA model was fit without any defined factors or sub domains first. Information criterion used to determine the best number of classes included the model with the lowest log likelihood (LL), Akaike Information Criterion (AIC), Bayesian information criterion (BIC), and sample size adjusted BIC (ABIC) [185] Additionally, three criteria were evaluated for less complex models due to l engthy computation for larger models: the Lo Mendel

PAGE 171

171 Rubin likelihood ratio test (LRT), the Vuong Lo Mendel Rubin likelihood ratio test (VLRT), and the bootstrapped likelihood ratio test (BLRT). The VLRT and BLRT p values provide information on whether the current model or model with k 1 classes fits significantly better. The LL, AIC, BIC, and ABIC criteria, in addition to class size and proportion, will be examined for the LCA as well as the full mixture models. Multiple random starts will also be used to t est whether the parameter estimates are stable across starting values. Essentially, Mplus automatically generates random parameter starting values to test whether the results are replicated, providing confidence that the solution is not a result of local maxima [145] Fitting the F ac tor Mixture M odel In this study, several variations of the FMM will be tested with different parameter specifications in each of the four models (Table D 1). The first factor model (FM 1, Figure D 2C) is the most restrictive model. The item thresholds, factor loadings, and factor covariance matrix are invariant across classes. The factor means were conditioned on latent class membership (allowed to be freely estimated across classe s). Because the factor covariance matrix, factor loadings, and thresholds were the same across classes, this implies that there is no severity (or variation) in the psychological concept being examined within classes. This model retains the conditional in dependence assumption. In FM 2, the factor loadings and factor means were conditioned on latent class membership (Figure D 2E). Item thresholds and the factor covariance matrix were invariant across classes. This model partially relaxes the conditional in dependence assumption because the items are being influenced by a latent factor as well as the categorical class outcome.

PAGE 172

172 In FM 3 the factor loadings, item thresholds, and factor means were conditioned on class membership. The factor covariance matrix rema ined invariant across classes. This model further relaxes the conditional independence assumption. It suggests that classes are based on item responses in addition to the factors. Each latent class has different factors and different interpretation of fact ors. The fourth model variation (FM 4, Figure D 2F) also allows the factor loadings and item thresholds to be freely estimated across classes while also permitting the factor covariance matrix to vary across classes. The factor mean is held invariant. T his model also suggests that classes are based on item responses. This model completely relaxes the conditional independence assumption by allowing the factor covariance matrix to be freely estimated as well as the factor loadings and thresholds. It is i mportant to note that substantive interpretation of results in important in evaluating the best fitting model. Sample code for modeling variations is available at the end of the Appendix D. Adding Predictors of Class Membership to Final M odel Two main app roaches are utilized for dealing with covariates in latent class/mixture models: one step and three step approaches. The one step method is a simultaneous estimation of covariates with the latent class models. The one step method has been described extens ively for categorical covariates [230 232] and continuous covariates [233, 234] Essentially, the one step method estimates the latent class model simultaneously with the regression portion of the model that related the classes to covariates. Several disadvantages reviewed by Vermunt (2010) include practical issues with large numbers of covariates (m odels must be re estimated with covariate removed or added), a question of whether to determine the number of classes

PAGE 173

173 with or without covariates, and the inclusion of covariates changes the intended meaning of the classes [186] The standard three step approach is not without disadvantage either [186, 188] This approach involves first determining the classes without covariates, assigning subjects to classes based on their posterior class membership probabilities (through one of several classification methods), and then using standard multinomial logistic regression to examine class membership probability per covariate category. A major dr awback of this estimation method is the consistent underestimation of the relationship between covariates and class membership, particularly if the classification error is large (entropy <0.60) [188] Vermunt (2 010) [186] proposed and simulated an alternative three step method expanding on a method by Bolck and colleagues [188] that takes into account the classification err or that occurs in step two of the original three step approach. Mplus Version 7 has created two automatic three step commands to accommodate the modified method to explore significant covariates associated with class membership and compare unadjusted dist al outcome means. The auxiliary commands are included outcomes, respectively, are as follows, where x1 x10 represent covariates and y1 and y2 are distal outcomes: Covari ate: auxiliary= (r3step) x1 x10; Distal outcome: auxiliary= (du3step) y1 y2;. In this study for the identification of significant predictors of class membership and the HRQOL outcomes means comparison for the unadjusted models, the two

PAGE 174

174 automatic 3 ste p approaches were used and are considered steps four and five (see code below). However, this study also required examining the distal outcome means across classes while accounting for significant predictors of class membership. Asparouhov and Muthen (201 by step approach in Mplus to implement this improved method [189] The first three steps presented below are done automatically in the r3step and dus3step. The modified manual approach conducts the first three steps separately and then incorporates steps four and five. 1. The appropriate mixture model with ideal number of classes and all model specifications (i.e. relaxing conditional independence) is identifi ed. 2. The most likely class variable is created using the posterior probabilities for latent class distribution. Using Mplus, this variable is saved as an output data file using Class Proba the classification uncertainty is calculated (example for a 4 class model below). The log ratio of each class probability against the last class is calculated. If an error equals zero it is replaced by 15, following the logit threshold value logic [145] A sample matrix and calculation is shown below: 1 2 3 4 1 0.939 0.037 0.011 0.013 2 0.027 0.899 0.049 0.026 3 0.009 0.064 0.88 0.046 4 0.006 0.026 0.038 0.931 Ln(0.939/0.013)=4.28 Ln(0.037/0.013)=1.05 Ln(0.011/0.013)= 0.17 Ln(0.027/0.026)=0.04 Ln(0.899/0.026)=3.54 Ln(0.049/0.026)= 0.633 Ln(0.009/0.046)= 1.63 Ln(0.064/0.046)= 0.33 Ln(0.88/0.046)=2.95 Ln(0.006/0.931)= 5.04 Ln(0.026/0.93 1)= 3.60 Ln(0.038/0.931)= 3.20 3. nominal variable from the CPROB dataset representing class membership. The

PAGE 175

175 step to account for the measurement error in the final step of the model. The significance of covariates in class membership was tested using a multinomial logistic regression. A class specific Model statement example is provided below for class 1: %C#1% F1 by u1 u21 ; F2 by b1 b15 ; [ u1$1 u 21$1] ; [ u1 $2 u 21$2] ; [ b 1$1 b 15$1]; [ b 1$2 b 15$2] ; [ b 1$3 b 15$3] ; [ f1 f2 ]; [N#1@ 4.28 ]; [N#2@1.05 ]; [N#3@0. 17 ]; Evaluating Class Membership and Distal O utcomes For this study, we included analysis (a fourth and fifth s tep) to examine the compare unadjusted distal outcomes means and adjusted distal outcome means across classes. 4. As previously stated, the auxiliary command specifying testing distal outcome variables (Mplus auxiliary option du3step) was used to conduct a Wa ld test of the unadjusted distal outcome means across classes on the primary dataset Auxiliary = du3step (y1) 5. Using the dataset generated in step 1 and the classification error rates calculated for each class, the distal outcome variable is regressed on the predictors within each class. For the Wald test of adjusted means across classes, the previously output dataset from Step 1 (CPROB data) and the accompanying code for accounting for classification error were used to conduct individual class vs. class means comparisons while controlling for predictors using the Mplus Model TEST command. Sample code for a 2 class model is shown below (covariates x1 x10; and distal outcome Y1): %C#1% F1 by u1 u21 ; F2 by b1 b15 ; [ u1$1 u 21$1] ;

PAGE 176

176 [ u1 $2 u 21$2] ; [ b 1$1 b 15$1 ]; [ b 1$2 b 15$2] ; [ b 1$3 b 15$3] ; [ f1 f2 ]; [N#1@ 4.28 ]; [N#2@1.05 ]; [N#3@0. 17 ]; Y1 on x1 x10; regress distal outcome variable on predictors in each class [Y1] (m1); defines the class 1 distal outcome mean as m1 %C#2 % F1 by u1 u21 ; F2 by b1 b15 ; [ u1 $1 u 21$1] ; [ u1 $2 u 21$2] ; [ b 1$1 b 15$1]; [ b 1$2 b 15$2] ; [ b 1$3 b 15$3] ; [ f1 f2 ]; [N#1@0.04 ]; [N#2@3.54 ]; [N#3@0.63 ]; Y2 on x1 x10; [Y2] (m2); defines the class 2 distal outcome mean as m2 Model TEST: m1=m2; asks for a Wald test between class1 and cl ass2 outcome means Mplus Sample Code f or 2 Class 2 Factor Mixture Model V ariations FM 1: Factor means conditioned on class membership (factor loadings, variance, and thresholds invariant) %overall% ptg by rptgi1 rptgi21; !specifies measurement model ies by ies1 ies15; %C#1% [ptg ]; factor means free estimated [ies ]; %C#2% [ptg ]; [ies ]; FM 2: Factor means and factor loadings conditioned on class membership (variance and thresholds invariant)

PAGE 177

177 %overall% ptg by rptgi1 rptgi21; ies by ies1 ies15; %C# 1% [rptgi1$1 rptgi21$1] (1 21); parantheses defines each threshold parameter and !constrains it across classes [rptgi1$2 rptgi21$2] (22 42); [ies1$1 ies15$1] (43 57); [ies1$2 ies15$2] (58 72); [ies1$3 ies15$3] (73 87); ptg by rptgi1 rptgi21; factor loadings freely estimated ies by ies1 ies15; [ptg ies]; %C#2% [rptgi1$1 rptgi21$1] (1 21); [rptgi1$2 rptgi21$2] (22 42); [ies1$1 ies15$1] (43 57); [ies1$2 ies15$2] (58 72); [ies1$3 ies15$3] (73 87); ptg by rptgi1 rptgi21; ies by ies1 ies15; [ptg ies]; FM 3: Factor loadings, item thresholds, and factor means conditioned on class membership (variance invariant) %overall% ptg by rptgi1 rptgi21; ies by ies1 ies15; %C#1% [rptgi1$1 rptgi21$1]; !removing parantheses allows thresholds to be freely estimated [rptgi1$2 rptgi21$2] ; [ies1$1 ies15$1] ; [ies1$2 ies15$2] ; [ies1$3 ies15$3] ; ptg by rptgi1 rptgi21; ies by ies1 ies15; [ptg ies]; %C#2% [rptgi1$1 rptgi21$1]; [rptgi1$2 rptgi21$2] ; [ies1$1 ies15$1] ;

PAGE 178

178 [ies1$2 ies15$2] ; [ies1$3 ies15$3] ; ptg by rptgi1 rptgi21; ies by ies1 ies15; [ptg ies]; FM 4: Factor loadings, item thresholds, factor means, and variance conditioned on class membership %overall% ptg by rptgi1 rptgi21; ies by ies1 ies15; %C#1% [rptgi1$1 rptgi21$1]; [rptgi1$2 rptgi21$2] ; [ies1$1 i es15$1] ; [ies1$2 ies15$2] ; [ies1$3 ies15$3] ; ptg by rptgi1 rptgi21; ies by ies1 ies15; ptg ies; !freely estimates variance across classes %C#2% [rptgi1$1 rptgi21$1] ; [rptgi1$2 rptgi21$2] ; [ies1$1 ies15$1] ; [ies1$2 ies15$2] ; [ies1$3 ies15$3] ; ptg by rptgi1 rptgi21; ies by ies1 ies15; ptg ies;

PAGE 179

179 A B Figure D 1. Graphic representation of allowing variation within classes A) No variation within c condition B ) Each individual can have different levels of severity within each class

PAGE 180

180 Table D 1. Parameter specifications for mixture model variations Latent class only Factor analysis o nly FM1 FM2 FM3 FM4 Item parameters Threshold FE FE IV IV FE FE Factor loading FE IV FE FE FE Factor parameters Variance C1 IV IV IV FE Covariance FE IV IV IV FE Mean FE FE FE C0 Note: IV= invariant across cl asses; FE=freely estimated across classes or conditioned on class membership; C0=constrained to zero across classes; C1= constrained to one across classes

PAGE 181

181 A B C D

PAGE 182

18 2 E F Figure D 2. Common factor model, latent class model, and mixture model variations. ( Y1 through Y3 represent items defining Factor 1 and Y4 through Y6 define Factor 2. X represents covariates that can be included in analysis. ) A) Preliminary factor analysis model B) Preliminary latent class model C) FM 1 Factor means vary across classes D) FM 2 Factor loadings and means conditioned o n class membership E) FM 3 Thresholds, factor loading, and means conditioned on latent class membership F) Thresholds, factor loading, and variance conditioned o n class membe rship; Item thresholds are conditioned on latent class variable. Factor loadings are conditioned on latent class variable.

PAGE 183

183 APPENDIX E CHAPTER 4 DETAILED RESULTS TABLES Table E 1. Measurement properties of Impact of Events Scale (IES ) and Posttraumatic Growth Inventory (PTGI) Items Factor Loading (STDYX) CFI RMSEA Impact of Events Scale (15 items) 0.97 0.08 0.79 I have trouble falling asleep or staying asleep because of pictures or t houghts about it that came into my mind 0.84 I had waves of strong feelings about it 0.87 I had dreams about it 0.80 Pictures about it popped into my mind 0.83 Other things kept making me think about it 0.79 Any reminder brought back feeling s about it. 0.85 I avoided letting myself get upset when I thought about it or was reminded of it I tried to remove it from memory 0.81 I stayed away from reminders about it 0.87 0.55 I trie d not to talk about it 0.75 I was aware that I still had a lot of feelings about it, but I 0.85 I tried not to think about it 0.91 My feelings about it were kind of numb 0.73 Posttraumatic Growth Inventory (21 items) 0.94 0. 07 Knowing that I can count on people in times of trouble 0.72 An increased sense of closeness with others 0.77 An increased willingness to express my emotions 0.71 Ha ving more compassion for others 0.76 Putting more effort into my relation ships 0.82 I learned a great deal about how wonderful people are 0.73 changing 0.73 o do better things with my life 0.85 otherwise 0.72 I developed new interests 0.75 I more readily accept needing others 0.74 I established a new path for my life 0.78 An increased feeling of self reliance 0.74 Knowing I can handle difficulties 0.78 Being better able to accep t the way things work out 0.82 0.76 My priorities about what is important in life have changed 0.79 An increased appreciation for the value of my own life 0.85 Appreciating each day more fully 0.89 A better understanding of spiritual matters 0.94 I have a stronger religious faith 0.91

PAGE 184

184 Table E 2. Two and three class factor mixture model comparison LL # par ameters AIC BIC ABIC Entropy 2 C 2F FM 1 20406 126 41065 416 27 41226 0.92 FM 2 20322 164 40973 41704 41184 0.89 FM 3 20261 251 41024 42144 41347 0.90 FM 4 NCVG 3C 2F FM 1 19309 129 38877 39452 39042 0.92 FM 2 19290 203 38986 39891 39247 0.92 FM 3 19156 377 39067 40749 39552 0.94 FM 4 NCVG Note: LL= loglikelihood; AIC= Akaike Information Criterion; BIC=Bayesian Information Criterion; ABIC= sample size a djusted BIC; NCVG= Best LL not replicated at maximum random starts FM 1: Factor covariance matrix, threshold, and factor loadings fixed across classes; means freely estimated (conditioned on class membership) FM 2: I tem thresholds and factor covariance matrix invariant ; f reely estimate means and factor loadings; FM 3: Factor variance invariant; f reely estimate means, factor loadings item t hresholds FM 4: Factor mean fixed at 0; free ly estimate covariances and variances, item thresholds, and factor loadings

PAGE 185

185 LIST OF REFERENCES 1. Gratwohl, A., Baldomero, H., Aljurf, M., Pasquini, M.C., Bouzas, L.F., Yoshimi, A., et al. (2010 ). Hematopoietic stem cell transplantation: A global perspective Journal of the American Medical Association, 303 (16), 1617 1624. 2. Copelan, E.A. (2006). Hematopoietic stem cell transplantation The New England Journal of Medicine, 354 (17), 1813 1826. 3. Horowitz, M. M. (2009). Uses and growth of hematopoeitic cell transplantation. In F. Appelbaum, & E. D. Thomas (Eds.), Thomas' H ematopoietic C ell T ransplantation. (pp. 15 21). Oxford: Wiley Blackwell. 4. Wingard, J.R., Majhail, N.S., Brazauskas, R., Wa ng, Z., Sobocinski, K.A., Jacobsohn, D., et al. (2011). Long term survival and late deaths after allogeneic hematopoietic cell transplantation Journal of Clinical Oncology, 29 (16), 2230 2239. 5. Martin, P.J., Counts, G.W.,Jr, Appelbaum, F.R., Lee, S.J., Sanders, J.E., Deeg, H.J., et al. (2010). Life expectancy in patients surviving more than 5 years after hematopoietic cell transplantation Journal of Clinical Oncology, 28 (6), 1011 1016. 6. Mohty, M., & Apperley, J.F. (2010). Long term physiological side effects after allogeneic bone marrow transplantation Hematology, 2010 229 236. 7. Ferrara, J.L., Levine, J.E., Reddy, P., & Holler, E. (2009). Graft versus host disease The Lancet, 373 (9674), 1550 1561. 8. Arora, M., Nagaraj, S., Witte, J., DeFor, T. E., MacMillan, M., Burns, L.J., et al. (2009). New classification of chronic GVHD: Added clarity from the consensus diagnoses Bone Marrow Transplantation, 43 (2), 149 153. 9. Pavletic, S., & Vogelsand, G.B. (2008). Treatment of high risk chronic GVHD Bio logy of Blood and Marrow Transplantation, 14 (12), 1436 1437. 10. Sun, C.L., Francisco, L., Kawashima, T., Leisenring, W., Robison, L.L., Baker, K.S., et al. (2010). Prevalence and predictors of chronic health conditions after hematopoietic cell transplant ation: A report from the bone marrow transplant survivor study Blood, 116 (17), 3129 3139. 11. Bhatia, S. (2011). Long term health impacts of hematopoietic stem cell transplantation inform recommendations for follow up Expert Review of Hematology, 4 (4), 437 52; quiz 453 4.

PAGE 186

186 12. Fayers, P. M., & Machin, D. (2007) Quality of Life: T he A ssessment, A nalysis, and I nterpretation of P atient R eported O utcomes. Chichester; Hoboken, NJ: John Wiley & Sons. 13. Guyatt, G.H., Feeny, D.H., & Patrick, D.L. (1993). M easuring health related quality of life Annals of Internal Medicine, 118 (8), 622 629. 14. Wilson, I.B., & Cleary, P.D. (1995). Linking clinical variables with health related quality of life. A conceptual model of patient outcomes Journal of the American Medical Association, 273 (1), 59 65. 15. Hjermstad, M.J., Knobel, H., Brinch, L., Fayers, P.M., Loge, J.H., Holte, H., et al. (2004). A prospective study of health related quality of life, fatigue, anxiety and depression 3 5 years after stem cell transpla ntation Bone Marrow Transplantation, 34 (3), 257 266. 16. Kiss, T.L., Abdolell, M., Jamal, N., Minden, M.D., Lipton, J.H., & Messner, H.A. (2002). Long term medical outcomes and quality of life assessment of patients with chronic myeloid leukemia followed at least 10 years after allogeneic bone marrow transplantation Journal of Clinical Oncology, 20 (9), 2334 2343. 17. Le, R.Q., Bevans, M., Savani, B.N., Mitchell, S.A., Stringaris, K., Koklanaris, E., et al. (2010). Favorable outcomes in patients survivin g 5 or more years after allogeneic hematopoietic stem cell transplantation for hematologic malignancies Biology of Blood and Marrow Transplantation, 16 (8), 1162 1170. 18. Bush, N.E., Donaldson, G.W., Haberman, M.H., Dacanay, R., & Sullivan, K.M. (2000). Conditional and unconditional estimation of multidimensional quality of life after hematopoietic stem cell transplantation: A longitudinal follow up of 415 patients Biology of Blood and Marrow Transplantation, 6 (5, Part A), 576 591. 19. Heinonen, H., Vol in, L., Uutela, A., Zevon, M., Barrick, C., & Ruutu, T. (2001). Gender associated differences in the quality of life after allogeneic BMT Bone Marrow Transplantation, 28 (5), 503 509. 20. Lee, S.J., Fairclough, D., Parsons, S.K., Soiffer, R.J., Fisher, D. C., Schlossman, R.L., et al. (2001). Recovery after stem cell transplantation for hematologic diseases Journal of Clinical Oncology, 19 (1), 242 252. 21. Rusiewicz, A., DuHamel, K.N., Burkhalter, J., Ostroff, J., Winkel, G., Scigliano, E., et al. (2008). Psychological distress in long term survivors of hematopoietic stem cell transplantation Psycho Oncology, 17 (4), 329 337. 22. Hjermstad, M.J., Loge, J.H., Evensen, S.A., Kvaloy, S.O., Fayers, P.M., & Kaasa, S. (1999). The course of anxiety and depression during the first year after allogeneic or autologous stem cell transplantation Bone Marrow Transplantation, 24 (11), 1219 1228.

PAGE 187

187 23. Food and Drug Administration. (2006). Draft guidance for industry on patient reported outcome measures: Use in medicinal p roduct development to support labeling claims. Federal Register, 71 5862 5863. 24. Behavioral Risk Factor Surveillance System. (2012). Health Related Quality of Life: Surveillance Data. Centers for Disease Control and Prevention. http://www.cdc.gov/hrqol/surveillance.htm Accessed 02/10 2013. 25. Institute of Medicine [IOM]. (2006) From C ancer P atient to C ancer S urvivor: Lost in T ransition. Washington, D.C.: National Academies Press. 26. Wingard, J.R., Huang, I., Sobocinski, K.A., Andrykowski, M.A., Cella, D., Rizzo, J.D., et al. (2010). Factors associated with self reported physical and Mental health after hematopoietic cell transplantation Biology of Blood and Marrow Transplantatio n, 16 (12), 1682 1692. 27. Kangas, M., Henry, J.L., & Bryant, R.A. (2002). Posttraumatic stress disorder following cancer. A conceptual and empirical review Clinical Psychology Review, 22 (4), 499 524. 28. Widows, M.R., Jacobsen, P.B., Booth Jones, M., & Fields, K.K. (2005). Predictors of posttraumatic growth following bone marrow transplantation for cancer Health Psychology, 24 (3), 266 273. 29. Jim, H.S., & Jacobsen, P.B. (2008). Posttraumatic stress and posttraumatic growth in cancer survivorship: A re view Cancer Journal, 14 (6), 414 419. 30. Lai, J.S., Garcia, S.F., Salsman, J.M., Rosenbloom, S., & Cella, D. (2012). The psychosocial impact of cancer: Evidence in support of independent general positive and negative components Quality of Life Research, 21 (2), 195 207. 31. Park, C.L., & Helgeson, V.S. (2006). Introduction to the special section: Growth following highly stressful life events -current status and future directions Journal of Consulting and Clinical Psychology, 74 (5), 791 796. 32. Park, C .L., Chmielewski, J., & Blank, T.O. (2010). Post traumatic growth: Finding positive meaning in cancer survivorship moderates the impact of intrusive thoughts on adjustment in younger adults Psycho Oncology, 19 (11), 1139 1147. 33. Andrykowski, M.A., Bisho p, M.M., Hahn, E.A., Cella, D.F., Beaumont, J.L., Brady, M.J., et al. (2005). Long term health related quality of life, growth, and spiritual well being after hematopoietic stem cell transplantation Journal of Clinical Oncology, 23 (3), 599 608.

PAGE 188

188 34. Kopp M., Holzner, B., Meraner, V., Sperner Unterweger, B., Kemmler, G., Nguyen Van Tam, D.P., et al. (2005). Quality of life in adult hematopoietic cell transplant patients at least 5 yr after treatment: A comparison with healthy controls European Journal of Haematology, 74 (4), 304 308. 35. Fife, B.L., Huster, G.A., Cornetta, K.G., Kennedy, V.N., Akard, L.P., & Broun, E.R. (2000). Longitudinal study of adaptation to the stress of bone marrow transplantation Journal of Clinical Oncology, 18 (7), 1539 1549. 3 6. McQuellon, R.P., Russell, G.B., Rambo, T.D., Craven, B.L., Radford, J., Perry, J.J., et al. (1998). Quality of life and psychological distress of bone marrow transplant recipients: The 'time trajectory' to recovery over the first year Bone Marrow Trans plantation, 21 (5), 477 486. 37. McQuellon, R.P., Russell, G.B., Cella, D.F., Craven, B.L., Brady, M., Bonomi, A., et al. (1997). Quality of life measurement in bone marrow transplantation: Development of the functional assessment of cancer therapy bone ma rrow transplant (FACT BMT) scale Bone Marrow Transplantation, 19 (4), 357 368. 38. Stein, K.D., Syrjala, K.L., & Andrykowski, M.A. (2008). Physical and psychological long term and late effects of cancer Cancer, 112 (S upp 11), 2577 2592. 39. Rothrock, N.E., K aiser, K.A., & Cella, D. (2011). Developing a valid patient reported outcome measure Clinical Pharmacology and Therapeutics, 90 (5), 737 742. 40. Pidala, J., Anasetti, C., & Jim, H. (2010). Health related quality of life following haematopoietic cell tran splantation: Patient education, evaluation and intervention British Journal of Haematology, 148 (3), 373 385. 41. Lee, S. J. (2008). Chapter 23: Quality of life issues post transplantation. In R. J. Soiffer (Ed.), Hematopoietic Cell Transplantation (pp. 5 37 559). Totowa, NJ: Humana Press. 42. Mosher, C.E., Redd, W.H., Rini, C.M., Burkhalter, J.E., & DuHamel, K.N. (2009). Physical, psychological, and social sequelae following hematopoietic stem cell transplantation: A review of the literature Psycho Oncol ogy, 18 (2), 113 127. 43. Slovacek, L., & Slovackova, B. (2007). Quality of life in oncological and hematooncological patients after hematopoietic stem cell transplantation: The effect of selected psychosocial and health aspects on quality of life: A revie w of the literature Reports of Practical Oncology & Radiotherapy, 12 (1), 53 59. 44. Sun, C.L., Francisco, L., Baker, K.S., Weisdorf, D.J., Forman, S.J., & Bhatia, S. (2011). Adverse psychological outcomes in long term survivors of hematopoietic cell tran splantation: A report from the bone marrow transplant survivor study (BMTSS) Blood, 118 (17), 4723 4731.

PAGE 189

189 45. Bevans, M.F., Marden, S., Leidy, N.K., Soeken, K., Cusack, G., Rivera, P., et al. (2006). Health related quality of life in patients receiving red uced intensity conditioning allogeneic hematopoietic stem cell transplantation Bone Marrow Transplantation, 38 (2), 101 109. 46. Khera, N., Storer, B., Flowers, M.E.D., Carpenter, P.A., Inamoto, Y., Sandmaier, B.M., et al. (2012). Nonmalignant late effect s and compromised functional status in survivors of hematopoietic cell transplantation Journal of Clinical Oncology, 30 (1), 71 77. 47. Fromm, K., Andrykowski, M.A., & Hunt, J. (1996). Positive and negative psychosocial sequelae of bone marrow transplanta tion: Implications for quality of life assessment Journal of Behavioral Medicine, 19 (3), 221 240. 48. Haase, J., & Braden, C. J. (1998). Guidelines for achieving clarity of concepts related to quality of life. In C. R. King, & P. S. Hinds (Eds.), Quality of life: From Nursing and Patient Perspectives. Theory, Research, and Practice. (pp. 54 -73). Sudbury, MA: Jones and Bartlett. 49. American Psychiatric Association. (2000) Diagnostic and Statistical Manual of Mental Disorders. Revised 4th ed Washington D.C: American Psychiatric Association. 50. Widows, M.R., Jacobsen, P.B., & Fields, K.K. (2000). Relation of psychological vulnerability factors to posttraumatic stress disorder symptomatology in bone marrow transplant recipients Psychosomatic Medicine, 62 (6), 873 882. 51. Galea, S., Nandi, A., & Vlahov, D. (July 2005). The epidemiology of post traumatic stress disorder after disasters Epidemiologic Reviews, 27 (1), 78 91. 52. Hobfoll, S.E., Tracy, M., & Galea, S. (2006). The impact of resource loss and traumatic growth on probable PTSD and depression following terrorist attacks Journal of Traumatic Stress, 19 (6), 867 878. 53. Xu, J., & Liao, Q. (2011). Prevalence and predictors of posttraumatic growth among adult survivors one year following 2008 sich uan earthquake Journal of Affective Disorders, 133 (1 2), 274 280. 54. Joseph, S., & Linley, A. (2008) Trauma, R ecovery, and G rowth: P ositive P sychological P erspectives on P osttraumatic S tress Hoboken, NJ: John Wiley & Sons. 55. Sumalla, E.C., Ochoa, C ., & Blanco, I. (2009). Posttraumatic growth in cancer: Reality or illusion ? Clinical Psychology Review, 29 (1), 24 33. 56. Horowitz, M. (1986) Stress Response Syndrome New York: Jason Aronson.

PAGE 190

190 57. Calhoun, L.G., Cann, A., Tedeschi, R.G., & McMillan, J (2000). A correlational test of the relationship between posttraumatic growth, religion, and cognitive processing Journal of Traumatic Stress, 13 (3), 521 527. 58. Tedeschi, R.G., & Calhoun, L.G. (2004). Target article: "posttraumatic growth: Conceptual foundations and empirical evidence" Psychological Inquiry, 15 (1), 1 18. 59. Cordova, M.J., Cunningham, L.C., Carlson, C.R., & Andrykowski, M.A. (2001). Posttraumatic growth following breast cancer: A controlled comparison study Health Psychology, 20 (3) 176 185. 60. Lechner, S.C., Carver, C.S., Antoni, M.H., Weaver, K.E., & Phillips, K.M. (2006). Curvilinear associations between benefit finding and psychosocial adjustment to breast cancer Journal of Consulting and Clinical Psychology, 74 (5), 828 840. 61. McLachlan, G. J., & Peel, D. (2001) Finite mixture models New York: Wiley. 62. Lubke, G.H., & Muthen, B. (2005). Investigating population heterogeneity with factor mixture models Psychological Methods, 10 (1), 21 39. 63. Patrick, D.L., & Deyo, R.A (1989). Generic and disease specific measures in assessing health status and quality of life Medical Care, 27 (3 Suppl), S217 S232. 64. Ware, J.E.,Jr., & Sherbourne, C.D. (1992). The MOS 36 item short form health survey (SF 36): I. conceptual framework and item selection Medical Care, 30 (6), 473 483. 65. Schag, C.C., Heinrich, R.L., & Ganz, P.A. (1984). Karnofsky performance status revisited: Reliability, validity, and guidelines. Journal of Clinical Oncology, 2 (3), 187 193. 66. Hays, R.D., Morales, L .S., & Reise, S.P. (2000). Item response theory and health outcomes measurement in the 21st century Medical Care, 38 (9 Suppl), II28 42. 67. Reise, S.P., & Waller, N.G. (2009). Item response theory and clinical measurement Annual Review of Clinical Psych ology, 5 27 48. 68. Weinrib, A.Z., Rothrock, N.E., Johnsen, E.L., & Lutgendorf, S.K. (2006). The assessment and validity of stress related growth in a community based sample Journal of Consulting and Clinical Psychology, 74 (5), 851 858. 69. Muthen, B., & Asparouhov, T. (2006). Item response mixture modeling: Application to tobacco dependence criteria Addictive Behaviors, 31 (6), 1050 1066.

PAGE 191

191 70. Bishop, M.M., Lee, S.J., Beaumont, J.L., Andrykowski, M.A., Rizzo, J.D., Sobocinski, K.A., et al. (2010). The preventive health behaviors of long term survivors of cancer and hematopoietic stem cell transplantation compared with matched controls Biology of Blood and Marrow Transplantation, 16 (2), 207 214. 71. Bishop, M.M., Beaumont, J.L., Hahn, E.A., Cella, D., Andrykowski, M.A., Brady, M.J., et al. (2007). Late effects of cancer and hematopoietic stem cell transplantation on spouses or partners compared with survivors and survivor matched controls Journal of Clinical Oncology, 25 (11), 1403 1411. 72. Norkin, M ., Hsu, J.W., & Wingard, J.R. (2012). Quality of life, social challenges, and psychosocial support for long term survivors after allogeneic hematopoietic stem cell transplantation Seminars in Hematology, 49 (1), 104 109. 73. Aziz, N.M. (2007). Cancer surv ivorship research: State of knowledge, challenges and opportunities Acta Oncologica, 46 (4), 417 432. 74. Burkett, V.S., & Cleeland, C.S. (2007). Symptom burden in cancer survivorship Journal of Cancer Survivorship, 1 (2), 167 175. 75. Hacker, E., Ferran s, C., Verlen, E., Ravandi, F., van Besien, K., Gelms, J., et al. (2006). Fatigue and physical activity in patients undergoing hematopoietic stem cell transplant Oncology Nursing Forum, 33 (3), 614 624. 76. Gotay, C.C., Kawamoto, C.T., Bottomley, A., & Ef ficace, F. (March 10, 2008). The prognostic significance of patient reported outcomes in cancer clinical trials Journal of Clinical Oncology, 26 (8), 1355 1363. 77. Bevans, M.F., Mitchell, S.A., Barrett, A.J., Bishop, M., Childs, R., Fowler, D., et al. (2 011). Function, adjustment, quality of life and symptoms (FAQS) in allogeneic hematopoietic stem cell transplantation (HSCT) survivors: A study protocol Health and Quality of Life Outcomes, 9 24. 78. Jim, H.S., Andrykowski, M.A., Munster, P.N., & Jacobs en, P.B. (2007). Physical symptoms/side effects during breast cancer treatment predict posttreatment distress Annals of Behavioral Medicine, 34 (2), 200 208. 79. Syrjala, K.L., & Chapko, M.E. (1995). Evidence for a biopsychosocial model of cancer treatmen t related pain Pain, 61 (1), 69 79. 80. Pidala, J., Kurland, B.F., Chai, X., Vogelsang, G., Weisdorf, D.J., Pavletic, S., et al. (2011). Sensitivity of changes in chronic graft versus host disease activity to changes in patient reported quality of life: R esults from the chronic graft versus host disease consortium Haematologica, 96 (10), 1528 1535.

PAGE 192

192 81. Prieto, J.M., Atala, J., Blanch, J., Carreras, E., Rovira, M., Cirera, E., et al. (2005). Patient rated emotional and physical functioning among hematologi c cancer patients during hospitalization for stem cell transplantation Bone Marrow Transplantation, 35 (3), 307 314. 82. Allen, M., & Yen, W. (1979/2002) Introduction to M easurement T heory Long Grove, IL: Waveland Press. 83. Hambleton, R. K., Swaminath an, H., & Rogers, H. (1991) Fundamentals of Item Response Theory. Newbury Park, CA: SAGE Publications. 84. Huang, I., Rizzo, D., Andrykowski, M.A., & Wingard, J.R. (2009). Development of the Functional Assessment of Cancer Therapy Bone Marrow Transplant Survivor (FACT BMTS) scale. Abstract 1738. 85. Hambleton, R. K., & Swaminathan, H. (1985) Item R esponse T heory: Principles and A pplications. Boston, MA: Kluwer Nijhoff. 86. Lord, F. (1980). Study of item bias. Applications of item Response Theory to Pra ctical Testing Problems (pp. 212 224). Hillsdale, NJ.: Lawrence Erlbaum Associates, Publishers. 87. Hu, L., & Bentler, P.M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives Structural Equation Modeling, 6 (1), 1 55. 88. Wainer, H.H. (1996). How is reliability related to the quality of test scores? what is the effect of local dependence on reliability? Educational Measurement, Issues and Practice, 15 (1), 22 29. 89. Yen, W.M. (1993). S caling performance assessments: Strategies for managing local item dependence Journal of Educational Measurement, 30 (3), 187 213. 90. Samejima, F. (1969). Estimation of latent ability using a response pattern of graded scores Psychometrika Monograph Sup plement, 34 (4), 100. 91. Reeve, B. B., & Fayers, P. (2005). Chapter 1 Applying item response theory modelling for evaluating questionnaire item and scale properties. In P. Fayers, & R. Hays (Eds.), Assessing Q uality of L ife in C linical T rials: M ethods and P ractice. (pp. 55 73). New York: Oxford University Press. 92. Teresi, J.A. (2006). Overview of quantitative measurement methods. equivalence, invariance, and differential item functioning in health applications Medical Care, 44 (11 Suppl 3), S39 49.

PAGE 193

193 93. Teresi, J.A. (2006). Different approaches to differential item functioning in health applications. advantages, disadvantages and some neglected topics Medical Care, 44 (11 Suppl 3), S152 70. 94. Teresi, J.A., Ramirez, M., Lai, J.S., & Silver, S. (2008). Occurrences and sources of differential item functioning (DIF) in patient reported outcome measures: Description of DIF methods, and review of measures of depression, quality of life and general health Psychology Science Quarterly, 50 (4), 538. 95. Reise, S.P., Ainsworth, A.T., & Haviland, M.G. (2005). Item response theory Current Directions in Psychological Science, 14 (2), 95 101. 96. Maglinte, G.A., Hays, R.D., & Kaplan, R.M. (2012). US general population norms for telephone administration of the SF 36 v2 Journal of Clinical Epidemiology, 65 (5), 497 502. 97. Cohen, J. (1988) Statistical P ower A nalysis for the B ehavioral S ciences Hillsdale, NJ: L. Erlbaum Associates. 98. Norman, G.R., Sloan, J.A., & Wyrwich, K.W. (2003). Interpretation of changes in health related quality of life: The remarkable universality of half a standard deviation Medical Care, 41 (5), 582 592. 99. SAS Institute. (2008; 2009). SAS 9.2 9.2 100. Muthen, B., & Muthen, L.K. (2012). Mplus Version 7.1 7 101. Muraki, E., & Bock, R.D. (1997). PARSCALE: IRT item analysis and test scoring for rating scale data 4 102. Aggen, S.H., Neale, M.C., & Kendler, K.S. (2005). DSM criteria for major depression: Evaluating symptom patterns using latent trait item response models Psychologic al Medicine, 35 (4), 475 487. 103. Chan, K.S., Orlando, M., Ghosh Dastidar, B., Duan, N., & Sherbourne, C.D. (2004). The interview mode effect on the center for epidemiological studies depression (CES D) scale: An item response theory analysis Medical Car e, 42 (3), 281 289. 104. Gomez, R., Cooper, A., & Gomez, A. (2005). An item response theory analysis of the carver and white (1994) BIS/BAS scales Personality and Individual Differences, 39 (6), 1093 1103. 105. Hays, R.D., Liu, H., Spritzer, K., & Cella, D. (2007). Item response theory analyses of physical functioning items in the medical outcomes study Medical Care, 45 (5 Suppl 1), S32 8.

PAGE 194

194 106. Cleeland, C.S., & Reyes Gibby, C.C. (2002). When is it justified to treat symptoms? measuring symptom burden On cology, 16 (9 Suppl 10), 64 70. 107. Woodend, A.K., Nair, R.C., & Tang, A.S. (1997). Definition of life quality from a patient versus health care professional perspective. International Journal of Rehabilitation Research, 20 (1), 71 80. 108. Undn, A., & E lofsson, S. (2001). Health from the patient's point of view. how does it relate to the physician's judgement? Family Practice, 18 (2), 174 180. 109. Pidala, J., Anasetti, C., & Jim, H. (2009). Quality of life after allogeneic hematopoietic cell transplant ation Blood, 114 (1), 7 19. 110. Jacobsen, P.B., & Jim, H.S.L. (2011). Consideration of quality of life in cancer survivorship research Cancer Epidemiology Biomarkers & Prevention, 20 (10), 2035 2041. 111. Syrjala, K.L., Martin, P.J., & Lee, S.J. (2012). Delivering care to long term adult survivors of hematopoietic cell transplantation Journal of Clinical Oncology, 30 (30), 3746 3751. 112. Wingard, J.R. (1994). Functional ability and quality of life of patients after allogeneic bone marrow transplantatio n Bone Marrow Transplantation, 14 Suppl 4 S29 33. 113. Syrjala, K.L., Langer, S.L., Abrams, J.R., Storer, B., Sanders, J.E., Flowers, M.E., et al. (2004). Recovery and long term function after hematopoietic cell transplantation for leukemia or lymphoma Journal of the American Medical Association, 291 (19), 2335 2343. 114. Gurevich, M., Devins, G.M., & Rodin, G.M. (2002). Stress response syndromes and cancer: Conceptual and assessment issues Psychosomatics, 43 (4), 259 281. 115. Shelby, R.A., Golden Kre utz, D.M., & Andersen, B.L. (2008). PTSD diagnoses, subsyndromal symptoms, and comorbidities contribute to impairments for breast cancer survivors Journal of Traumatic Stress, 21 (2), 165 172. 116. Pirl, W.F., Greer, J., Temel, J.S., Yeap, B.Y., & Gilman, S.E. (2009). Major depressive disorder in long term cancer survivors: Analysis of the national comorbidity survey replication Journal of Clinical Oncology, 27 (25), 4130 4134. 117. Smith, S.K., Williams, C.S., Zimmer, C.R., & Zimmerman, S. (2011). An exp loratory model of the relationships between cancer related trauma outcomes on quality of life in non hodgkin lymphoma survivors Journal of Psychosocial Oncology, 29 (1), 19 34.

PAGE 195

195 118. Braamse, A.M., Gerrits, M.M., van Meijel, B., Visser, O., van Oppen, P., Boenink, A.D., et al. (2012). Predictors of health related quality of life in patients treated with auto and allo SCT for hematological malignancies Bone Marrow Transplantation, 47 (6), 757 769. 119. Wong, F.L., Francisco, L., Togawa, K., Bosworth, A., G onzales, M., Hanby, C., et al. (2010). Long term recovery after hematopoietic cell transplantation: Predictors of quality of life concerns Blood, 115 (12), 2508 2519. 120. Jensen, R.E., Arora, N.K., Bellizzi, K.M., Rowland, J.H., Hamilton, A.S., Aziz, N.M ., et al. (2013). Health related quality of life among survivors of aggressive non hodgkin lymphoma Cancer, 119 (3), 672 680. 121. Mosher, C.E., Lepore, S.J., Wu, L., Austin, J., Valdimarsdottir, H., Rowley, S., et al. (2012). Social correlates of distres s following hematopoietic stem cell transplantation: Exploring the role of loneliness and cognitive processing Journal of Health Psychology, 17 (7), 1022 1032. 122. Byar, K.L., Eilers, J.E., & Nuss, S.L. (2005). Quality of life 5 or more years post autolo gous hematopoietic stem cell transplant Cancer Nursing, 28 (2), 148 157. 123. Kline, R. B. (2011) Principles and Practice of Structural Equation Modeling New York, NY: Guilford Publications. 124. Curbow, B., Somerfield, M.R., Baker, F., Wingard, J.R., & Legro, M.W. (1993). Personal changes, dispositional optimism, and psychological adjustment to bone marrow transplantation Journal of Behavioral Medicine, 16 (5), 423 443. 125. Somerfield, M.R., Curbow, B., Wingard, J.R., Baker, F., & Fogarty, L.A. (1996 ). Coping with the physical and psychosocial sequelae of bone marrow transplantation among long term survivors Journal of Behavioral Medicine, 19 (2), 163 184. 126. Schoulte, J.C., Lohnberg, J.A., Tallman, B., & Altmaier, E.M. (2011). Influence of coping style on symptom interference among adult recipients of hematopoietic stem cell transplantation Oncology Nursing Forum, 38 (5), 582 586. 127. Scheier, M.F., & Carver, C.S. (1985). Optimism, coping, and health: Assessment and implications of generalized ou tcome expectancies Health Psychology, 4 (3), 219 247. 128. Hochhausen, N., Altmaier, E.M., McQuellon, R., Davies, S.M., Papadopolous, E., Carter, S., et al. (2007). Social support, optimism, and self efficacy predict physical and emotional well being afte r bone marrow transplantation Journal of Psychosocial Oncology, 25 (1), 87 101.

PAGE 196

196 129. Goetzmann, L., Klaghofer, R., Wagner Huber, R., Halter, J., Boehler, A., Muellhaupt, B., et al. (2007). Quality of life and psychosocial situation before and after a lung liver or an allogeneic bone marrow transplant. results from a prospective study Swiss Medical Weekly, 137 Suppl 155 115S 124S. 130. Horowitz, M., Wilner, N., & Alvarez, W. (1979). Impact of event scale: A measure of subjective stress Psychosomatic Me dicine, 41 (3), 209 218. 131. American Psychiatric Association. (1994) Diagnostic and S tatistical M anual of M ental D isorders Washington, D.C.: American Psychiatric Association. 132. Sundin, E.C., & Horowitz, M.J. (2002). Impact of event scale: Psychomet ric properties The British Journal of Psychiatry, 180 205 209. 133. Radloff, L.S. (1977). The CES D scale Applied Psychological Measurement, 1 (3), 385 401. 134. Lepore, S.J., Silver, R.C., Wortman, C.B., & Wayment, H.A. (1996). Social constraints, int rusive thoughts, and depressive symptoms among bereaved mothers Journal of Personality and Social Psychology, 70 (2), 271 282. 135. Carver, C.S. (1997). You want to measure coping but your protocol's too long: Consider the brief COPE International Journa l of Behavioral Medicine, 4 (1), 92 100. 136. Carver, C.S., Scheier, M.F., & Weintraub, J.K. (1989). Assessing coping strategies: A theoretically based approach Journal of Personality and Social Psychology, 56 (2), 267 283. 137. Krageloh, C.U. (2011). A s ystematic review of studies using the brief COPE: Religious coping in factor analyses Religions, 2 216 246. 138. Knowles, S.R., Wilson, J.L., Connell, W.R., & Kamm, M.A. (2011). Preliminary examination of the relations between disease activity, illness perceptions, coping strategies, and psychological morbidity in crohn's disease guided by the common sense model of illness Inflammatory Bowel Diseases, 17 (12), 2551 2557. 139. Moore, B.C., Biegel, D.E., & McMahon, T.J. (2011). Maladaptive coping as a med iator of family stress Journal of Social Work Practice in the Addictions, 11 (1), 17 39. 140. Maurice Stam, H., Oort, F.J., Last, B.F., & Grootenhuis, M.A. (2009). A predictive model of health related quality of life in young adult survivors of childhood cancer European Journal of Cancer Care, 18 (4), 339 349. 141. Ullman, J. B., & Bentler, P. M. (2003). Structural Equation Modeling. Handbook of Psychology (pp. 607 634). New York: John Wiley & Sons, Inc.

PAGE 197

197 142. Bentler, P. (1990). Comparative fit indices i n structural m odels Psychological Bulletin, 107 (2), 238 246. 143. Browne, M.W., & Cudeck, R. (1992). Alternative ways of assessing model fit Sociological Methods & Research, 21 (2), 230 258. 144. Lepore, S.J., & Revenson, T.A. (2007). Social constraints on disclosure and adjustment to cancer Social and Personality Psychology Compass, 1 (1), 313 333. 145. Muthen, L.K., & Muthen, B. (1998 2012). Statistical analysis with latent variables. User's guide. Mplus, 7 146. Kav, S., Aslan, O., Tekin, F., Yesil, H., Meral, C., Ozturk, U., et al. (2009). Quality of life and difficulties of patients encountered after autologous stem cell transplantation Journal of B.U.ON., 14 (4), 673 680. 147. Mosher, C.E., DuHamel, K.N., Rini, C., Corner, G., Lam, J., & Redd, W. H. (2011). Quality of life concerns and depression among hematopoietic stem cell transplant survivors Supportive Care in Cancer, 19 (9), 1357 1365. 148. Kessler, R. (1997). The effects of stressful life events on depression Annual Review of Psychology, 4 8 191 214. 149. Ashing Giwa, K.T., & Lim, J.W. (2011). Health related quality of life outcomes among cervical cancer survivors: Examining ethnic and linguistic differences Cancer Epidemiology, 35 (2), 194 201. 150. Reyes Gibby, C.C., Anderson, K.O., Mor row, P.K., Shete, S., & Hassan, S. (2012). Depressive symptoms and health related quality of life in breast cancer survivors Journal of Women's Health (2002), 21 (3), 311 318. 151. Hoodin, F., & Weber, S. (2003). A systematic review of psychosocial factor s affecting survival after bone marrow transplantation Psychosomatics, 44 (3), 181 195. 152. Rodriguez, J.R., Pearman, T.P., & Moreb, J. (1999). Morbidity and mortality following bone marrow transplantation: Predictive utility of pre BMT affective functio ning, compliance, and social support stability International Journal of Behavioral Medicine, 6 (3), 241 254. 153. Cooke, L., Gemmill, R., Kravits, K., & Grant, M. (2009). Psychological issues of stem cell transplant Seminars in Oncology Nursing, 25 (2), 1 39 150. 154. Cordova, M.J., & Andrykowski, M.A. (2003). Responses to cancer diagnosis and treatment: Posttraumatic stress and posttraumatic growth Seminars in Clinical Neuropsychiatry, 8 (4), 286 296.

PAGE 198

198 155. Kleim, B., & Ehlers, A. (2009). Evidence for a c urvilinear relationship between posttraumatic growth and posttrauma depression and PTSD in assault survivors Journal of Traumatic Stress, 22 (1), 45 52. 156. Deimling, G.T., Kahana, B., Bowman, K.F., & Schaefer, M.L. (2002). Cancer survivorship and psycho logical distress in later life Psycho Oncology, 11 (6), 479 494. 157. Nuamah, I.F., Cooley, M.E., Fawcett, J., & McCorkle, R. (1999). Testing a theory for health related quality of life in cancer patients: A structural equation approach Research in Nursi ng & Health, 22 (3), 231 242. 158. Molassiotis, A. (1997). A conceptual model of adaptation to illness and quality of life for cancer patients treated with bone marrow transplants Journal of Advanced Nursing, 26 (3), 572 579. 159. Somjaivong, B., Thanasil p, S., Preechawong, S., & Sloan, R. (2011). The influence of symptoms, social support, uncertainty, and coping on health related quality of life among cholangiocarcinoma patients in northeast thailand Cancer Nursing, 34 (6), 434 442. 160. Zhou, E.S., Pene do, F.J., Bustillo, N.E., Benedict, C., Rasheed, M., Lechner, S., et al. (2010). Longitudinal effects of social support and adaptive coping on the emotional well being of survivors of localized prostate cancer The Journal of Supportive Oncology, 8 (5), 196 201. 161. Anderson, M., Wingard, J., Shenkman, E., & Huang, I.,. (2012). Investigating the mechanism between psychosocial factors and health outcomes in hematopoietic cell transplantation survivors. (ISOQOL Abstract 459) Quality of Life Research, 20 (S1) 1 106. 162. Dunn, J., Occhipinti, S., Campbell, A., Ferguson, M., & Chambers, S.K. (2011). Benefit finding after cancer: The role of optimism, intrusive thinking and social environment Journal of Health Psychology, 16 (1), 169 177. 163. Friedman, L.C., Kalidas, M., Elledge, R., Chang, J., Romero, C., Husain, I., et al. (2006). Optimism, social support and psychosocial functioning among women with breast cancer Psycho Oncology, 15 (7), 595 603. 164. Wanchai, A., Armer, J.M., & Stewart, B.R. (2011). Nonp harmacologic supportive strategies to promote quality of life in patients experiencing cancer related fatigue: A systematic review Clinical Journal of Oncology Nursing, 15 (2), 203 214. 165. Osborn, R.L., Demoncada, A.C., & Feuerstein, M. (2006). Psychoso cial interventions for depression, anxiety, and quality of life in cancer survivors: Meta analyses International Journal of Psychiatry in Medicine, 36 (1), 13 34.

PAGE 199

199 166. Ganz, P.A., Earle, C.C., & Goodwin, P.J. (2012). Journal of clinical oncology update on progress in cancer survivorship care and research Journal of Clinical Oncology, 30 (30), 3655 3656. 167. Jarden, M., Nelausen, K., Hovgaard, D., Boesen, E., & Adamsen, L. (2009). The effect of a multimodal intervention on treatment related symptoms in pa tients undergoing hematopoietic stem cell transplantation: A randomized controlled trial Journal of Pain and Symptom Management, 38 (2), 174 190. 168. Wu, L.M., Austin, J., Hamilton, J.G., Valdimarsdottir, H., Isola, L., Rowley, S., et al. (2012). Self ef ficacy beliefs mediate the relationship between subjective cognitive functioning and physical and mental well being after hematopoietic stem cell transplant Psycho Oncology, 21 (11), 1175 1184. 169. Jacobsen, P., Widows, M., Hann, D., Andrykowski, M., Kro nish, L., & Fields, K. (1998). Posttraumatic stress disorder symptoms after bone marrow transplantation for breast cancer Psychosomatic Medicine, 60 (3), 366 371. 170. Andrykowski, M.A., Cordova, M.J., Hann, D.M., Jacobsen, P.B., Fields, K.K., & Phillips, G. (1999). Patients' psychosocial concerns following stem cell transplantation Bone Marrow Transplantation, 24 (10), 1121 1129. 171. Tedeschi, R.G., & Calhoun, L.G. (1996). The posttraumatic growth inventory: Measuring the positive legacy of trauma Jour nal of Traumatic Stress, 9 (3), 455 471. 172. Affleck, G., & Tennen, H. (1996). Construing benefits from adversity: Adaptational significance and dispositional underpinnings Journal of Personality, 64 (4), 899 922. 173. Park, C.L., Cohen, L.H., & Murch, R .L. (1996). Assessment and prediction of stress related growth Journal of Personality, 64 (1), 71 105. 174. Zebrack, B.J., Stuber, M.L., Meeske, K.A., Phipps, S., Krull, K.R., Liu, Q., et al. (2012). Perceived positive impact of cancer among long term sur vivors of childhood cancer: A report from the childhood cancer survivor study Psycho Oncology, 21 (6), 630 639. 175. Stanton, A. L., Bower, J. E., & Low, C. A. (2006). Posttraumatic growth after cancer. In L. G. Calhoun, & R. G. Tedeschi (Eds.), Handbook of Posttraumatic Growth (pp. 138 175). Mahwah, N.J.: Lawrence Erlbaum Associates, Inc. 176. Koutrouli, N., Anagnostopoulos, F., & Potamianos, G. (2012). Posttraumatic stress disorder and posttraumatic growth in breast cancer patients: A systematic review Women & Health, 52 (5), 503 516.

PAGE 200

200 177. Morrill, E.F., Brewer, N.T., O'Neill, S.C., Lillie, S.E., Dees, E.C., Carey, L.A., et al. (2008). The interaction of post traumatic growth and post traumatic stress symptoms in predicting depressive symptoms and quali ty of life Psycho Oncology, 17 (9), 948 953. 178. Calhoun, T. G., & Tedeschi, R. G. (1998). Posttraumatic growth: future directions. In R. G. Tedeschi, C. L. Park, & L. G. Calhoun (Eds.), Posttraumatic Growth: Positive Changes in the Aftermath of Crisis ( pp. 215 238). Mahwah, NJ: Lawrence Erlbaum Associates. 179. Muthen, B., & Muthen, L.K. (2000). Integrating person centered and variable centered analyses: Growth mixture modeling with latent trajectory classes Alcoholism, Clinical and Experimental Resear ch, 24 (6), 882 891. 180. Sawatzky, R., Ratner, P., Kopec, J., & Zumbo, B. (2012). Latent variable mixture models: A promising approach for the validation of patient reported outcomes Quality of Life Research, 21 (4), 637 650. 181. Zinbarg, R.E., & Barlow D.H. (1996). Structure of anxiety and the anxiety disorders: A hierarchical model. Journal of Abnormal Psychology (105), 181 193. 182. Taku, K., Cann, A., Calhoun, L.G., & Tedeschi, R.G. (2008). The factor structure of the posttraumatic growth inventor y: A comparison of five models using confirmatory factor analysis Journal of Traumatic Stress, 21 (2), 158 164. 183. McCutcheon, A. L. (1987) Latent C ass A nalysis Beverly Hills, CA: Sage Publications. 184. Weiss, D. S., & Marmar, C. R. (1997). The Imp act of Event Scale Revised. In J. P. Wilson, & T. M. Keane (Eds.), Assessing Psychological Trauma and PTSD: A (pp. 399 411). New York: Guilford Press. 185. Nylund, K., Asparouhov, T., & Muthen, B. (2007). Deciding on the Number of Classes in Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo Simulation Study Structural Equation Modeling, 14 (4), 535 569. 186. Vermunt, J.K. (2010). Latent class modeling with covariates: Two improved three step approaches Political Ana lysis, 18 (4), 450 469. 187. Tofighi, D., & Enders, C. K. (2008). Identifying the correct number of classes in growth mixture models. In G. R. Hancock, & K. M. Samuelsen (Eds.), Advances in L atent V ariable M ixture M odels (pp. 317 342). Charlotte, NC: Infor mation Age Publishing. 188. Bolck, A., Croon, M., & Hagenaars, J. (2004). Estimating latent structure models with categorical variables: One step versus three step estimators Political Analysis, 12 (1), 3 27.

PAGE 201

201 189. Asparouhov, T., & Muthn, B. (2012). Aux iliary variables in mixture modeling: A 3 Step approach using Mplus. Mplus Web Notes 15, Version 5. 190. Petras, H., & Masyn, K. (2010). General Growth Mixture Analysis with Antecedents and Consequences of Change. In A. R. Piquero, & D. Weisburd (Eds.), H andbook of Quantitative Criminology (pp. 69 100): Springer New York. 191. StataCorp. (2005). Stata Statistical Software: Release 9 9 192. Lubke, G., & Neale, M. (2008). Distinguishing between latent classes and continuous factors with categorical outco mes: Class invariance of parameters of factor mixture models Multivariate Behavioral Research, 43 (4), 592 620. 193. Elhai, J.D., Naifeh, J.A., Forbes, D., Ractliffe, K.C., & Tamburrino, M. (2011). Heterogeneity in clinical presentations of posttraumatic stress disorder among medical patients: Testing factor structure variation using factor mixture modeling Journal of Traumatic Stress, 24 (4), 435 443. 194. Naifeh, J.A., Richardson, J.D., Del Ben, K.S., & Elhai, J.D. (2010). Heterogeneity in the latent st ructure of PTSD symptoms among canadian veterans Psychological Assessment, 22 (3), 666 674. 195. Clark, S.L., Muthen, B., Kaprio, J., D'Onofrio, B., Viken, R., & Rose, R.J. (2010). Models and strategies for factor mixture analysis: Two examples concerning the structure underlying psychological disorders Mplus. Http://www.statmodel2.com/download/FMA%20Paper_v142.Pdf; 196. Figley, C. (1978) Stress Disorders Among Vietnam Veterans: Theory, Research, Treatment New York, NY: Brunner Routledge. 197. Andrykowski, M.A., Cordova, M.J., Mcgrath, P.C., Sloan, D.A., & Kenady, D.E. (2000). Stability and change in posttraumatic stress disorder symptoms following breast cancer treat ment: A 1 year follow up Psycho Oncology, 9 (1), 69 78. 198. Cordova, M.J., Andrykowski, M.A., Kenady, D.E., McGrath, P.C., Sloan, D.A., & Redd, W.H. (1995). Frequency and correlates of posttraumatic stress disorder like symptoms after treatment for breas t cancer Journal of Consulting and Clinical Psychology, 63 (6), 981 986. 199. Bellizzi, K.M., Smith, A.W., Reeve, B.B., Alfano, C.M., Bernstein, L., Meeske, K., et al. (2010). Posttraumatic growth and health related quality of life in a racially diverse c ohort of breast cancer survivors Journal of Health Psychology, 15 (4), 615 626. 200. Bellizzi, K.M., & Blank, T.O. (2006). Predicting posttraumatic growth in breast cancer survivors Health Psychology, 25 (1), 47 56.

PAGE 202

202 201. Weiss, T. (2004). Correlates of p osttraumatic growth in husbands of breast cancer survivors Psycho Oncology, 13 (4), 260 268. 202. Roberts, A.L., Gilman, S.E., Breslau, J., Breslau, N., & Koenen, K.C. (2011). Race/ethnic differences in exposure to traumatic events, development of post tr aumatic stress disorder, and treatment seeking for post traumatic stress disorder in the united states Psychological Medicine, 41 (01), 71. 203. Tomich, P.L., & Helgeson, V.S. (2004). Is finding somethin g good in the bad always good? B enefit finding among women with breast cancer Health Psychology, 23 (1), 16 23. 204. Smith, M.Y., Redd, W.H., Peyser, C., & Vogl, D. (1999). Post traumatic stress disorder in cancer: A review Psycho Oncology, 8 (6), 521 537. 205. Mcfarlane, A.C., Atchison, M., Rafalowicz, E ., & Papay, P. (1994). Physical symptoms in post traumatic stress disorder Journal of Psychosomatic Research, 38 (7), 715 726. 206. Levine, E.G., Eckhardt, J., & Targ, E. (2005). Change in post traumatic stress symptoms following psychosocial treatment fo r breast cancer Psycho Oncology 14 (8), 618 635. 207. Ware, J. E., Kosinski, M., Bjorner, J., Turner Bowker, B., & Gandek, M. E. (2007) 36v2 Health Survey Lincoln, RI: Quality Metric Incorporated. 208. Helgeson, V.S., Reynolds K.A., & Tomich, P.L. (2006). A meta analytic review of benefit finding and growth Journal of Consulting and Clinical Psychology, 74 (5), 797 816. 209. McCullough, M.E., Root, L.M., & Cohen, A.D. (2006). Writing about the benefits of an interpersonal tra nsgression facilitates forgiveness Journal of Consulting and Clinical Psychology, 74 (5), 887 897. 210. Linley, P.A., & Joseph, S. (2004). Positive change following trauma and adversity: A review Journal of Traumatic Stress, 17 (1), 11 21. 211. Mosher, C .E., DuHamel, K.N., Rini, C.M., Li, Y., Isola, L., Labay, L., et al. (2010). Barriers to mental health service use among hematopoietic SCT survivors Bone Marrow Transplantation, 45 (3), 570 579. 212. Cleeland, C.S. (2007). Symptom burden: Multiple symptom s and their impact as patient reported outcomes Journal of the National Cancer Institute Monographs, 2007 (37), 16 21.

PAGE 203

203 213. Montazeri, A. (2009). Quality of life data as prognostic indicators of survival in cancer patients: An overview of the literature f rom 1982 to 2008 Health and Quality of Life Outcomes, 7 102 7525 7 102. 214. Quinten, C., Coens, C., Mauer, M., Comte, S., Sprangers, M.A., Cleeland, C., et al. (2009). Baseline quality of life as a prognostic indicator of survival: A meta analysis of i ndividual patient data from EORTC clinical trials The Lancet Oncology, 10 (9), 865 871. 215. Dalgleish, T. (2004). Cognitive approaches to posttraumatic stress disorder: The evolution of multirepresentational theorizing Psychological Bulletin, 130 (2), 22 8 260. 216. Salsman, J.M., Segerstrom, S.C., Brechting, E.H., Carlson, C.R., & Andrykowski, M.A. (2009). Posttraumatic growth and PTSD symptomatology among colorectal cancer survivors: A 3 month longitudinal examination of cognitive processing Psycho Onc ology, 18 (1), 30 41. 217. Efficace, F., Kemmler, G., Vignetti, M., Mandelli, F., Molica, S., & Holzner, B. (2008). Health r elated quality of life assessment and reported outcomes in leukaemia randomised controlled trials a systematic review to evaluate the added value in supporting clinical decision making European Journal of Cancer, 44 (11), 1497 1506. 218. Altmaier, E.M., Ewell, M., McQuellon, R., Geller, N., Carter, S.L., Henslee Downey, J., et al. (2006). The effect of unrelated donor marrow transplantation on health related quality of life: A report of the unrelated donor marrow transplantation trial (T cell depletion t rial) Biology of Blood and Marrow Transplantation, 12 (6), 648 655. 219. Broers, S., Kaptein, A.A., Le Cessie, S., Fibbe, W., & Hengeveld, M.W. (2000). Psychological functioning and quality of life following bone marrow transplantation: A 3 year follow up study Journal of Psychosomatic Research, 48 (1), 11 21. 220. Blanchard, E.B., Hickling, E.J., Barton, K.A., Taylor, A.E., Loos, W.R., & Jones Alexander, J. (1996). One year prospective follow up of motor vehicle accident victims Behaviour Research and T herapy, 34 (10), 775 786. 221. Scott, N.W., Fayers, P.M., Aaronson, N.K., Bottomley, A., de Graeff, A., Groenvold, M., et al. (2010). Differential item functioning (DIF) analyses of health related quality of life instruments using logistic regression Heal th and Quality of Life Outcomes, 8 81. 222. Brant, J.M., Beck, S., Dudley, W.N., Cobb, P., Pepper, G., & Miaskowski, C. (2011). Symptom trajectories in posttreatment cancer survivors Cancer Nursing, 34 (1), 67 77. 223. Kirkova, J., Rybicki, L., Walsh, D ., & Aktas, A. (2012). Symptom prevalence in advanced cancer: Age, gender, and performance status interactions The American Journal of Hospice & Palliative Care, 29 (2), 139 145.

PAGE 204

204 224. Gijsbers van Wijk, C.M.T., Huisman, H., & Kolk, A.M. (1999). Gender dif ferences in physical symptoms and illness behavior: A health diary study Social Science & Medicine, 49 (8), 1061 1074. 225. Cohen, A.S., Kim, S., & Baker, F.B. (1993). Detection of differential item functioning in the graded response model Applied Psycho logical Measurement, 17 (4), 335 350. 226. Muthen, B., & Asparaouhov, T. (2002) Latent Variable Analysis w ith Categorical Outcomes: Multiple Group and Growth Modeling in Mplus. Los Angeles: University of California and Muthn & Muthn. 227. Morin, A.J.S. Morizot, J., Boudrias, J., & Madore, I. (2011). A multifoci person centered perspective on workplace affective commitment: A latent Profile/Factor mixture analysis Organizational Research Methods, 14 (1), 58 90. 228. Conway, C., Hammen, C., & Brennan, P (2012). A comparison of latent class, latent trait, and factor mixture models of DSM IV borderline personality disorder criteria in a community setting: Implications for DSM 5 Journal of Personality Disorders, 26 (5), 793 803. 229. Lazarsfeld, P. F., & Henry, N. W. (1968) Latent S tructure A nalysis : Houghton, Mifflin. 230. Clogg, C.C., & Goodman, L.A. (1984). Latent structure analysis of a set of multidimensional contingency tables Journal of the American Statistical Association, 79 (388), 762 771. 231 Goodman, L.A. (1974). The analysis of systems of qualitative variables when some of the variables are unobservable. part I A modified latent structure approach American Journal of Sociology, 79 (5), 1179 1259. 232. Vermunt, J. (1997) Log linear M odels for E vent H istories, A dvanced Q uantitative T echniques in the S ocial S ciences S eries : Thousand Oaks, CA: SAGE Publications, Inc. 233. Bandeen Roche, K., Miglioretti, D., Zeger, S., & Rathouz, P. (1997). Latent variable regression for multiple discrete outc omes Journal of the American Statistical Association, 92 (440), 1375 1386. 234. Dayton, C., & Macready, G. (1988). Concomitant variable latent class models Journal of the American Statistical Association, 83 (401), 173 178.

PAGE 205

205 BIOGRAPHICAL SKETCH Kelly Kenzik received her Bachelor of Science degree in health education and Department of Health Education and Behavior in December of 2007. After graduating Summa cum laude, s he continued into the Master of Science degree program (January 2008) for health behavior in the Department of Health Education and Behavior in the College of Health and Human Performance at the University of Florida. In May 2008 she accepted a graduate r esearch assistant position for GatorWell Health Promotion Services. In August of 2009, she transitioned to a graduate teaching assistant position for an undergraduate course in Personal and Family Health in the College of Health and Human Performance. Kel ly received her Master of Science Degree in health behavior from the University of Florida in December of 2008. In January 2009, Kelly continued her graduate teaching assistantship and began the doctoral program in the College of Health and Human Performa nce. In May 2009, Kelly transferred into the Epidemiology PhD program through both the College of Medicine and College of Public Health and Health Professions. In addition to the PhD program, she accepted a pre doctoral research fellowship with the Insti tute for Child Health Policy in the Department of Health Outcomes and Policy in the College of Medicine. She received her PhD from the University of Florida in May of 2013.