Development of Disease Models for Neuropsychiatric Disorders

MISSING IMAGE

Material Information

Title:
Development of Disease Models for Neuropsychiatric Disorders
Physical Description:
1 online resource (115 p.)
Language:
english
Creator:
Sun, Wan
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Pharmaceutical Sciences, Pharmaceutics
Committee Chair:
Hochhaus, Guenther
Committee Members:
Zhu, Haojie
Derendorf, Hartmut C
Khan, Saeed R
Wang, Yaning

Subjects

Subjects / Keywords:
bipolar -- clinical -- disease -- dropout -- huntington's -- model -- observational -- placebo -- progression -- study -- trial
Pharmaceutics -- Dissertations, Academic -- UF
Genre:
Pharmaceutical Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
The aim of my dissertation research is to develop disease models for two neuropsychiatric diseases, bipolar disorder (BD) and Huntington’s disease (HD). BD is a highly recurrent lifelong psychiatric illness with only subjective evaluation methods available. Clinical trials for BD have been considered as challenging due to large variation in placebo effect level and high dropout rate, which may produce biased efficacy comparison analysis and even result in incorrect conclusion. Based on a BD database built from multiple clinical trials under several approved bipolar drugs, an empirical placebo effect model with an exponential decay process plus a linear progression was developed to quantify the time course of Young Mania Rating Scale (YMRS) total score. Meanwhile, a parametric survival model with log-normal distribution was developed to describe the dropout pattern during the trials. Using graphic analyses and also likelihood ratio test, the individual parameters estimated from the placebo effect model were found to significantly affect the dropout probability. Several trial-specific covariates such as trial starting year, baseline score etc were also significant covariates for dropout. Both placebo effect model and dropout model were able to describe the observed data reasonably well based on various diagnostic plots. Combination of placebo model and dropout model was applied to simulate new clinical trials through Monte-Carlo simulation. HD is an inherited and chronic neurodegenerative disorder caused by gene mutation with only one approved drug for symptomatic treatment but no approved disease modifying treatment available. Based on the available models in the literature and the exploratory analysis results, a mixed effect logistic model was developed to quantify the time course of total motor score (MOT) and total functional capacity (TFC) in order to characterize HD disease progression, and then validated using different methods. Accordingly, the half-maximal score, 62 for MOT and 6.5 for TFC was determined as the inflection point where the progression rate reached the maximal level. Patient age and CAG length were identified as significant covariates for model parameters. This disease model provided information about HD natural progression and may also offer suggestions to design strategy of a trial targeting disease-modifying treatment.
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 Wan Sun.
Thesis:
Thesis (Ph.D.)--University of Florida, 2012.
Local:
Adviser: Hochhaus, Guenther.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2013-12-31

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2012
System ID:
UFE0044903:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 DEVELOPMENT OF DISEASE MODELS FOR NEUROPSYCHIATRIC DISORDERS By WAN SUN A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF P HILOSOPHY UNIVERSITY OF FLORIDA 2012

PAGE 2

2 2012 Wan Sun

PAGE 3

3 This document is dedicated to my parents and all the friends who supported me.

PAGE 4

4 ACKNOWLEDGMENTS I would like to first express my appreciation and grateful thanks to my academic advisor, D r. Guenther Hochhaus. It has been such a great honor to be trained by him during my Ph. D study. I am deeply thankful to Dr. Hochhaus who accepted me to his research group, guided, supported and encouraged me in all respect. I would like to thank the membe rs of my supervisory committee, Dr. Hartmut Derendorf, Dr. Haojie Zhu, Dr. Saeed R. Khan and Dr. Yaning Wang. Special thanks go to Dr. Yaning Wang who offered me a valuable opportunity to work with him and helped me as possible as he could. I truly believe the training with Dr. Yaning Wang will be a strong foundation for my future career and will benefit me a lifetime. I would like to also specially thank Dr. Haojie Zhu who gave me many excellent suggestions on my research project and shared his research ex perience and experimental materials with me. Extended special thanks go to Yufei Tang for her technical assistance on my research and encouragement on my personal life. I would like to thank Dr. Sihong Song to provide me convenience for my experiment and a lso Dr. Yuanqing Lu to teach and help me on my experimental skills as well as Dr. Anthony Palmieri for his support. I would also like to take this opportunity to express my thanks for the technical and administrative support by Mrs. Patricia J. Khan, Ms. R obin Keirnan Sanchez and Mrs. Sarah L. Scheckner. I also want to thank Dr. Erica Pierce and Dr. Paolo Vicini at Pfizer La Jolla for their patient guidance and warm help to make my internship experience impressive and rewarding. I enjoyed my Ph. D study in members for their help, support and encouragement. With them, I could have a colorful Ph. D life. And I also want to thank all my Chinese friends for their help and company over years in Florida. Last b ut not least, I deeply thank my parents who gave me the support and strength to complete my Ph. D study.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TA BLES ................................ ................................ ................................ ........................... 7 LIST OF FIGURES ................................ ................................ ................................ ......................... 9 LIST OF ABBREVIATIONS ................................ ................................ ................................ ........ 11 ABSTRACT ................................ ................................ ................................ ................................ ... 12 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 14 Disease Model ................................ ................................ ................................ ........................ 14 Basic Concepts ................................ ................................ ................................ ................ 14 Disease Models for Neuropsychiatric Disorders ................................ ............................. 15 Overview of Bipolar Disorder ................................ ................................ ................................ 17 Epidemiology and Etiology ................................ ................................ ............................. 17 Symptoms, Diagnosis and Impact ................................ ................................ ................... 18 Evaluation and Treatment ................................ ................................ ................................ 19 ................................ ................................ ......................... 20 Cause of HD ................................ ................................ ................................ .................... 21 ale (UHDRS) ................................ ................... 21 Diagnosis ................................ ................................ ................................ ......................... 22 Treatment ................................ ................................ ................................ ......................... 22 CAG Repeat Length and t he Age of Onset in HD ................................ .......................... 23 Challenges in Clinical Trials for Neuropsychiatric Diseases ................................ ................. 24 Placebo Effect ................................ ................................ ................................ .................. 24 Dropout ................................ ................................ ................................ ............................ 25 Lack of Clinical Endpoints ................................ ................................ .............................. 26 2 DEVELOPMENT OF A DISEASE MODEL FOR BIPOLAR DISORDER ........................ 39 Background and Objectives ................................ ................................ ................................ .... 39 Specific Aim 1 ................................ ................................ ................................ ................. 40 Speci fic Aim 2 ................................ ................................ ................................ ................. 40 Specific Aim 3 ................................ ................................ ................................ ................. 40 Specific Aim 4 ................................ ................................ ................................ ................. 40 Methods ................................ ................................ ................................ ................................ .. 40 Disease Database ................................ ................................ ................................ ............. 40 Placebo Effect Model ................................ ................................ ................................ ...... 41 Analysis of Dropout Data ................................ ................................ ................................ 43 Placebo Dropout Model ................................ ................................ ................................ ... 43

PAGE 6

6 Model Based Clinical Trial Simulation ................................ ................................ ........... 44 Results and Discussion ................................ ................................ ................................ ........... 45 Disease Database ................................ ................................ ................................ ............. 45 Placebo Effect Model ................................ ................................ ................................ ...... 46 Analysis for Placebo Dropout ................................ ................................ ......................... 48 Placebo Dropout Model ................................ ................................ ................................ ... 49 Model Based Clinical Trial Simulation ................................ ................................ ........... 52 Summary ................................ ................................ ................................ ................................ 52 3 ............... 66 Background and Objectives ................................ ................................ ................................ .... 66 Specific Aim 1 ................................ ................................ ................................ ................. 67 Specific Aim 2 ................................ ................................ ................................ ................. 67 Specific Aim 3 ................................ ................................ ................................ ................. 67 Methods ................................ ................................ ................................ ................................ .. 67 Disease Database ................................ ................................ ................................ ............. 67 Data Exploration ................................ ................................ ................................ .............. 68 Structural Model Development ................................ ................................ ....................... 69 Selection of model structure ................................ ................................ ..................... 69 Mode l validation ................................ ................................ ................................ ...... 71 Model fitting on observational data ................................ ................................ ......... 73 Covariate Model Development ................................ ................................ ........................ 73 Results and Discussion ................................ ................................ ................................ ........... 74 Disease Database ................................ ................................ ................................ ............. 74 Data Exploration ................................ ................................ ................................ .............. 75 Structural Model Development ................................ ................................ ....................... 76 Logistic model ................................ ................................ ................................ .......... 76 Selection of model structure ................................ ................................ ..................... 77 Model validation ................................ ................................ ................................ ...... 7 8 Model fitting on observational data ................................ ................................ ......... 80 Covariate Model Development ................................ ................................ ........................ 81 Summary ................................ ................................ ................................ ................................ 82 4 CONCLUSIONS ................................ ................................ ................................ .................. 107 LIST OF REFERENCES ................................ ................................ ................................ ............. 109 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 115

PAGE 7

7 LIST OF TABLES Table page 1 1 Summary of rating scales commonly used in brain disorders ................................ ........... 28 1 2 List of items contained in YMRS ................................ ................................ ...................... 29 1 3 Drug therapy for patients with Bipolar Disorders ................................ .............................. 30 1 4 List of items contained in Unified Huntington's Disease Rating Scale (UHDRS) ............ 31 1 5 List of items contained in total motor score (MOT) ................................ .......................... 32 1 6 List of items contained in total functional capacity (TFC) ................................ ................ 34 1 7 Categories in diagnostic confidence rating ................................ ................................ ........ 35 2 1 Summary of drugs and trials in the disease database for bipolar disorder ......................... 54 2 2 Demographic and Baseline Characteristics in included bipolar trials ............................... 54 2 3 Parameter estimates of placebo effect model for pooled dataset ................................ ....... 55 2 4 Parameter estimates of placebo drop out model for pooled dataset ................................ .... 55 3 1 List of candidate model structures for total motor score ................................ ................... 83 3 2 List of candidate model stru ctures for total functional capacity ................................ ........ 83 3 3 List of included studies in Huntington's disease database ................................ ................. 84 3 4 Demographic and Bas eline Characteristics in included studies ................................ ........ 84 3 5 Results of structural model search for total motor score ................................ ................... 85 3 6 Results of stru ctural model search for total functional capacity ................................ ........ 85 3 7 List of population estimates and estimate precision of model parameters ........................ 85 3 8 List of literature with reported progression rate for TFC ................................ .................. 86 3 9 Summary of all the groups in clinical trial CARE ................................ ............................. 87 3 1 0 Summary of model parameters for all the groups in clinical trial ................................ ..... 88 3 11 Summary of model parameters for MOT in clinical trial and observational studies ......... 88 3 12 Comparison of structural model and covariate model ................................ ....................... 89

PAGE 8

8 3 13 Summary of model parameters in covariate model for observational data ....................... 89

PAGE 9

9 LIST OF FIGURES Figure page 1 1 Clinical manifestations and molecular cause of HD. ................................ ......................... 36 1 2 Mechanisms for the neuroprotective therapeutic targets and strategies. .......................... 37 1 3 Trend of the placebo effect in clinical trials for Schizophrenia. ................................ ....... 38 2 1 Schematic representation of so me empirical models available in literature used in neuropsychiatric diseases ................................ ................................ ................................ .. 56 2 2 Time profile of mean observed YMRS total score in placebo arm. ................................ .. 57 2 3 Individual time profile of observed YMRS total score for a representative set of subjects ................................ ................................ ................................ ............................... 58 2 4 Comparison of individual observation, individual prediction and population prediction on time profile of YMRS total score for a representative set of subjects ......... 59 2 5 Scatter plot of the individual observed and predicted YMRS total scores. ....................... 60 2 6 Dropout rate up to 3 weeks in placebo group for all the studies ................................ ........ 61 2 7 Summary and comparison of percentage of dropout reasons in placebo and active treatment groups for each study ................................ ................................ ......................... 62 2 8 Comparison of mean observ ations between dropout patients and non dropout patients by visit. ................................ ................................ ................................ ................ 63 2 9 Comparison of survival curves between simulation and observation. ............................... 64 2 10 Comparison of YMRS total score time profiles between simulation and observation for all the studies ................................ ................................ ................................ ................ 65 3 1 Individual time profile of observed rating scores for a representative set of subjects. ..... 90 3 2 Exploratory analysi s on relationship of calculated slope of MOT increase with observed score based on linear regression in placebo data of the clinical trial. ................ 91 3 3 Exploratory analysis on relationship of calculated slope of TFC decline with observed score based on linear regression in placebo data of the clinical trial. ................ 92 3 4 Exploratory analysis on relationship of calculated slope of MOT increase w ith observed score based on linear regression in observational studies. ................................ 93 3 5 Exploratory analysis on relationship of calculated slope of TFC decline with observed score based on linear regression in observational studies. ................................ 94

PAGE 10

10 3 6 Schematic representation of logistic model. ................................ ................................ ...... 95 3 7 Comparison of individual observation, individual prediction and population prediction on time profile of total motor score for a representative set of subjects in placebo group of clinical trial. ................................ ................................ ........................... 96 3 8 Comparison of individual observation, individual prediction and population prediction on time profile of total functional capacity for a representative set of subjects in placebo group of clinical trial. ................................ ................................ ......... 97 3 9 Comparison of mean observed progression rate, mean predicted progression rate and simulated population p rogression rate of MOT for subjects in placebo group of clinical trial. ................................ ................................ ................................ ....................... 98 3 10 Comparison of mean observed progression rate, mean predicted progression rate and simulated population p rogression rate of TFC for subjects in placebo group of clinical trial. ................................ ................................ ................................ ....................... 99 3 11 Visual predictive check for MOT in active treatment groups of clinical trial ................. 100 3 12 Comparison of individual observation, individual prediction and population prediction on time profile of total motor score for a representative set of subjects in observational studies. ................................ ................................ ................................ ...... 101 3 13 Scatter plots of between subject variability of parameters over patient age. ................... 102 3 14 Scatter plots of between subject variability of parameters over CAG length. ................. 103 3 15 Compar ison of scatter plots of between subject variability on baseline score over patient age between structural model and covariate model. ................................ ............ 104 3 16 Comparison of scatter plots of between subject variability on baseline score over CAG length between structural model and covariate model. ................................ ........ 105 3 17 Comparison of scatter plots of between subject variability on rate constant over CAG length between structural model and covariate model. ................................ .................. 106

PAGE 11

11 LIST OF ABBREVIATION S BD Bipol ar Disorder HD LOCF Last Observation Carry Forward MOT Total Motor Score TFC Total Functional Capacity UHDRS YMRS Young Mania Rating Scale

PAGE 12

12 Abstract of Dissertation Presented to the Graduat e School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy DEVELOPMENT OF DISEASE MODELS FOR NEUROPSYCHIATRIC DISORDERS By Wan Sun December 2012 Chair: Guenther Hochhaus Major : Pharmaceutical Sciences The aim of my dissertation research is to develop disease models for two neuropsychiatric BD is a highly recurrent lifelong psychiatric illness with only subjective evaluation method s available. Clinical trials for BD have been considered as challenging due to large variation in placebo effect level and high dropout rate, which may produce biased efficacy comparison analysis and even result in incorrect conclusion. Ba sed on a BD database built from multiple clinical trials under several approved bipolar drugs, an empirical placebo effect model with an exponential decay process plus a linear progression was developed to quantify the time course of Young Mania Rating Sca le (YMRS) total score. Meanwhile, a parametric survival model with log normal distribution was developed to describe the dropout pattern during the trials. Using graphic analyses and also likelihood ratio test, the individual parameters estimated from the placebo effect model were found to significantly affect the dropout probability. Several trial specific covariates such as trial starting year, baseline score etc were also significant covariates for dropout. Both placebo effect model and dropout model wer e able to describe the observed data reasonably well based on various diagnostic plots. Combination of placebo model and dropout model was applied to simulate new clinical trials through Monte Carlo simulation.

PAGE 13

13 HD is an inherited and chronic neurodegenerat ive disorder caused by gene mutation with only one approved drug for symptomatic treatment but no approved disease modifying treatment available. Based on the available models in the literature and the exploratory analysis results, a mixed effect logistic model was developed to quantify the time course of total motor score (MOT) and total functional capacity (TFC) in order to characterize HD disease progression, and then validated using different methods. Accordingly, the half maximal score, 62 for MOT and 6.5 for TFC was determined as the inflection point where the progression rate reached the maximal level. Patient age and CAG length were identified as significant covariates for model parameters. This disease model provided information about HD natural pro gression and may also offer suggestions to design strategy of a trial targeting disease modifying treatment.

PAGE 14

14 CHAPTER 1 INTRODUCTION A modeling and simulation based approach is increasingly applied to enhance the efficiency of clinical trials in neuropsy chiatric disorders that are generally considered to be challenging given disease characteristics as well as problems in diagnosis, evaluation methods and therapeutics. In order to generate quantitative information and guide the design of clinical trials, w e intended to develop disease models for neuropsychiatric disorders to quantify the disease progression as well as placebo effect and also describe the dropout pattern during the trials. In this introduction section, the basic concept and current literatur e on disease model especially for neuropsychiatric dise ases, the general information about two of neuropsychiatric clinical trials for these two diseases will be d iscussed. Disease Model Basic Concepts The value of quantitative thinking in drug development and regulatory review is increasingly being appreciated (Bhattaram, et al., 2005) Modeling and simulation of clinical data associated to pharmacokinetic, pharmacodynamic and disease p rogression, has been applied in multiple therapeutic areas throughout the whole process of drug development. Disease models are mathematical representations of longitudinal time course data on clinical endpoints, either categorical or continuous variables after placebo or drug treatment for a specific disease based on individual study or pooled dataset across various trials. They are developed to summarize the information about epidemiology, prevention, diagnosis and treatment, providing great benefit to re search planning, clinical trial design and analysis, decision making, regulatory as well as education.

PAGE 15

15 Disease models can be generally divided into two groups, empirical models and mechanistic models. Empirical models describe the time course of clinical efficacy endpoints (e.g. objective biomarkers or subjective rating scales) based on fitting a mathematical function to the observed data without considering the underlying mechanisms of action (Pilla V, et al., 2011) They are simple and descriptive but may have less predictive power than mechanism based models, especially in terms of scaling and extrapolation (Post, et al., 2010) Mechanism based models describe the time course of clinical endpoints based on the parameters assessed in vitro and physiological values according to the assumption and information about mechanism of action, so most of them are built in a pharmacological sense for a drug treatment rather than placebo response unless the mechanism of placebo response is known. Considering the relevance to my research, only e mpirical models will be discussed in detail. It is essential to first develop a disease progression model and a placebo model, and then with the disease progression and placebo effect model parameters fixed, the different drug effect models will be explor ed. Subsequently, all the parameters of the final placebo and drug effect models are simultaneously estimated. But for a specific type of disease, progression model and placebo model for same clinical endpoints is more valuable to be shared to design diffe rent trials for different drugs and to provide information for comparison analysis with active treatment groups. Disease M odels for N europsychiatric D isorders Most of neuropsychiatric disorders have in common that they lack objective measures (laborator y tests or biomarkers) to evaluate the outcome of treatment effects (Kane and Leucht, 2008) Therefore, the evaluations of disease status as well as therapeutic effect rely on some well built and validat ed rating scales that were usually designed to measure psychopathological symptoms and assessed by physicians or professional raters. Normally these rating scales consist

PAGE 16

16 of some questionnaires (items) where each item is scored based on the severity of sym ptoms and total score is constrained within a theoretical range. Table 1 1 summarized several rating scales commonly used to assess the disease symptoms and treatment effects in brain disorders. So the outcomes on these rating scales are typically treated as categorical variables, but when the possible maximal number of categories in a rating scale is sufficiently large, they can be considered as continuous variables instead. In neuropsychiatry, disease progression and placebo response can be modeled separ ately (Bhattaram, et al., 2009; Ploeger and Holford, 2009) However in many other cases, it is very hard to distin guish between placebo response and disease progression because of the episodic nature of the disease, such as schizophrenia, depression and bipolar disorder et al. In those cases, disease progression and placebo response can be considered as a single entit y. When the little data in controlled studies is available, clinical measurements obtained during observational studies might provide useful information about disease progression. Various disease models have been published to quantify the time profile of rating scores in neuropsychiatric diseases. Gomeni and Merio Pich developed a mixed Weibull and linear model to describe the time course of HAMD 17 total score and could also predict the response at end point given previous measurements in patients with major depressive disorders in placebo group based on data from seven clinical trials. Their work provided a methodological framework to implement a population enrichment strategy in the design of clinical trials for the assessment of novel antidepressant d rugs (Gomeni and Merlo Pich, 2007) e course of total PANSS score was characterized for placebo by using Weibull model and for asenapine treatments by an Emax model in clinical trials for acute schizophrenia (Friberg, et al.,

PAGE 17

17 2009) group in US Food and Drug Administration (US FDA) developed models to describe the longitudi nal course of total UPDRS change in placebo group by applying simple linear model to characterize the natural disease progression and adding exponential function to catch the symptomatic effect produced by placebo. Their work focused on understanding the f eatures of delayed start design using prior knowledge from published data and data submitted to FDA, and the outcome was part of ongoing discussion between FDA and the pharmaceutical industry on the standards required for demonstrating disease modifying ef disease (Bhattaram, et al., 2009) Ito and colleagues developed a model based on comprehensive literature data to describe the longitudinal response in the ADASC in patients with mild to placebo treated patients and acetylcholinesterase (AChE) inhibitor treated patients, and to identify factors that aff ected disease progression. Overview of Bipolar Disorder Epidemiology and E tiology Bipolar disorder (BD) is now recognized as a highly recurrent, potentially treatable but frequently misdiagnosed and under diagnosed lifelong psychiatric illness, but every aspect of its definition, boundaries, mechanisms and treatment is still subject to debate (Swann, 2006) In 2004, the World Health Organization ranked BD collectively as the 12 th most common moderately to severely disabling condition in the work for any age group. In a recent representative, US based population survey, the lifetime preva lence of BD was estimated as 4% (Merikangas, et al., 2007) BD has no predilection for race, sex, or ethnicity. Although they can occur at any age, BD is more common in persons younger than 25. As far as etiolo gy, inheritance pattern has been related to the higher risk of being affected BD (Barnett and Smoller, 2009) And

PAGE 18

18 environmental factors, such as stressful life events, disruptions in the sleep, family members with high expressed emotion et al are also strongly associated with the inheritance pattern (Goldstein, et al., 2010) as the Genetic Association Database and identified glycogen synthase kinase plausible candidate ge ne for BD from a pharmacological and a genetic perspective (Luykx, et al., 2010) Symptoms, Diagnosis and I mpact Symptoms of patients suffering from bipolar include both a lowering of mood (depression) and an exaggerated elevation of mood (mania), and these changes occur in cycles, referred to as commonly divided into 2 sub categories, BD I and BD II. BD I is characterized by episodes of mania along with major depression, while in BD II, the episodes of elevated mood never exceed the criteria for hypomania (Grunze, 2011; Rosen, et al., 1984) But actually the clinical course of bi polar disorders varies much and patients rarely experience a single episode, with relapse rates reported at more than 70 percent over five years (Gitlin, et al., 1995) So the cro ss sectional symptomatology in this dynamic disorder is only of little value for the longitudinal diagnosis. This may be one of the main reasons why BD is frequently misdiagnosed or at least delayed. Most of BD patients are depressed most of time, and BD s hared some similar symptoms as several other mental disorders. So many BD patients were initially diagnosed with a different condition. Ghaemi et al reported that in those patients who were ever diagnosed as having BD, it took an average of 8 10 years befo re the correct diagnosis was established, and even longer in BD II patients (Ghaemi, et al., 1999) BD is considered as a devastating disease (Morselli, et al., 2004) Based on literature data, suicide rates were 20 times higher in patients with bipolar disorders than in the general

PAGE 19

19 population (Novick, et al., 2010) ; and one third of patients with bipolar disorders attempted suicide, a rate that was among the highest of any psychiatric diagnosis (Cassidy, 2011) The socioeconomic burden of BD is considerable, and it ranked 6 th in the Global Burden of Disease ranking (Grunze, 2011) Patients with BD are often at increased risk of having one or more additional psychiatric disorders. And general medical conditions, including diabetes mellitus, obesity, and cardiovascular diseas e, are more common in patients with BD compared with age matched cohorts, and cardiovascular risk is higher in BD patients than those with other mental disease (Fiedorowicz, et al., 2009) With unrecognized or m isdiagnosed problem, and accordingly delayed treatment, BD could produce even worse results. Shi and colleagues pointed out that the risk of suicide attempts in people with unrecognized BD was significantly higher than in people with recognized BD (Shi, et al., 2004) Evaluation and T reatment BD lacks objective measures (laboratory tests or biomarkers) to evaluate the treatment effects. Thus, the evaluation of the severity of illness and the treatment effects is based on the rating scales as assessed by a physician or by a trained rater. A standard and well validated rating scale usually consists of several questionnaires (items), where each item is scored based on the severity of the symptoms. The Young Mania Ra ting Scale (YMRS) is the most commonly used symptom rating scale to assess changes in mania (Young, et al., 1978) and identified as the primary efficacy endpoint in most of clinical trials for BD. As Table 1 2 showed, it consists of 11 items, where each item is scored from 0 to 4 or 0 to 8 (0 indicating the absence of the symptom and the highest score indicating extremely suffering from the symptom), so the range of the total score is 0 to 60 The mainstay of treatment for bipolar disorder includes atypical antipsychotics, anticonvulsant and m ood stabilizer. Approved drug therapy for patients with BD was listed in

PAGE 20

20 Table 1 3. Typical antipsychotics and several atypical antipsychotics had some efficacy against acute mania but not patients with bipolar depression or for maintenance treatment. And majority of BD patients did not do well on monotherapy due to risk of relapse, which occurs in one third of patients in the first year after presentation and in more than 70 percent of patients within five years (Gitlin, et al., 1995) risk of treatment resistance, co morbid psychiatric conditions or others. So several combination therapies and add on agents have been tested and proved to be superior to monotherapy. Combination the rapy with lithium or valproate plus an antipsychotic was superior to either single agent alone in the treatment of acute mania (Scherk, et al., 2007) Quetiapine combined with lithium or valproate was more effe ctive than lithium or valproate alone (Price and Marzani Nissen, 2012) But there was no evidence yet to support any combination therapy or the addition of an antidepressant in the acute phase of depression (Van Lieshout and MacQueen, 2010) The treatment of BD is complex since treatment needs to be considered separately depending on different episodes (m anic, mixed or depression) as well as severity of illness. But any intervention during different episodes cannot be considered detached from long term goal of treatment in BD which is long term stability, and regaining and maintaining full functionality an d a good quality of life. that is characterized by progressive motor dysfunction, psychiatric symptoms, and cognitive decline et al. H D has been reported with an overall prevalence of about 10 per 100,000 in Caucasian population (Langbehn, et al., 2004) But it occurs worldwide in all races and ethnic groups and the incidence is as high as 1 3% (Goldberg, et al., 1993) The average age of onset is

PAGE 21

21 37 years old, but the range spans f rom infancy into the ninth decade of life (Krobitsch and Kazantsev, 2011) Cause of HD HD is the first human disease to successfully use reverse genetics to identify the affected gene after extensive research (1993) The disease was cause by mutation of HD gene IT15 where expanding beyond the normal polymorphism length that varies from 7 to 36 repeats normally. The IT15 gene product, huntingtin (Htt), is a large, highly conserved cytoplasmic protein that is ubiquitously expressed and widely distributed throughout the brain (DiFiglia, et al ., 1995) The pathologically extended polyglutamines cause Htt to acquire a non native structural conformation and get misfolded, allowing it to involve in diverse aberrant interactions with multiple cellular components, consequently disturbing many cellu lar functions essential for neuronal homeostasis and resulting in a prolonged process of neuronal dysfunction and ultimately in cell death (Kaltenbach, et al., 2007; Kazantsev, 2007) So the patients experience functional decline, psychiatric problem. And eventually, disease can result in disability and death. Figure 1 1 shows clinical manifestations and molecular basis of HD (Krobitsch and Kazantsev, 2011) Considering multiple domains of symptoms (motor dysfunction, psychiatric disorder and cognitive as w ell as functional decline), the investigators of the Huntington Study Group (HSG) four domains of clinical symptoms and performances of patients with HD as well a s individuals at risk for HD (1996) instead of relyin g on various rating scales to evaluate different features of HD separately. Based on longitudinal database from HSG, there was a high degree of internal

PAGE 22

22 consistency within each of the domains of UHDRS and there were significant inter correlations between t he domains of the UHDRS except for total behavior score. As Table 1 4 showed, UHDRS includes four main domains, motor, behavioral, cognitive and functional component to combine all the important elements of disease symptoms. This comprehensive instrument h as been applied widely to track changes in the clinical features of HD over time in clinical trials and observational studies. Among all the rating scales involved in UHDRS, total motor score (MOT) and total functional capacity (TFC) are two of the most f requently used outcome variables. MOT is the sum of all the individual motor ratings and it was found that MOT had excellent degree of inter rater reliability (1996) TFC has established psychometric properties including inter rater reliability and validity based on radiographic measures of disease progressi on (Young, et al., 1986) The detailed information about items contained in these two scales was listed in Table 1 5 and Table 1 6. Diagnosis Although HD is a disease composed of multiple domains of clinical symptoms, its diagnosis historically relies on the emergence of chorea (abnormal body movement) which belongs to motor section in UHDRS. At present, di agnostic confidence rating is responded by movement disorder specialist to convey his or her belief that the participant at risk has developed manifest HD. The categories of diagnostic confidence rating were shown in Table 1 7. Treatment The current medica l care, especially at clinical edge or industry sponsored studies has focused on symptom management and assistance as disability worsens. Previous studies of interventions designed to improve symptoms and slow functional decline have included the antioxida nts vitamin E (Peyser, et al., 1995) idebenone (Ranen, et al., 1996) baclofen

PAGE 23

23 (Shoulson, et al., 1989) the glutamate antagonists lamotrigine (Kremer, et al., 1999) remacemide and coenzyme Q10 (2001) et al, but none of them showed significant impact on functional decline. The only drug to date approved by the US Food and Drug Administration (FDA) for HD, tetrabenazine, is indicated to suppress involuntary movements (chorea), but does nothing to delay disease progression (Poon, et al., 2010) Since the neurodegenerative processes in the brain begin long before symptom of HD is recognizable, neuroprotective strategies has potential to delay the onset and slow the progression of HD (Hersch and Rosas, 2008) Therefore, significant efforts are also ongoing to search therapeutic targets and strategies at pre clinical studies as well as provide disease modification biomarkers that might allow the tracking of disease progress before the onset of clinical symptoms. Based on disease cause, mutant gene and its product Htt became attractive therapeutic targets, which was approved in a HD mous e model that disease progression could be stopped or even reversed when expression of mutant Htt was suppressed (Diaz Hernandez, et al., 2005; Yamamoto, et al., 2000) Current studies on neuroprotective strategi es include application of RNA interference (RNAi) and anti sense DNA oligonucleotides to reduce expression levels of both wild type and mutant alleles, or to selectively target on mutant Htt RNA as well as small molecule drugs to inhibit abnormal structura l conformation of Htt protein and fragment. The detailed mechanisms for above therapeutic strategies were described in Figure 1 2. CAG R epeat L ength and the A ge of O nset in HD As described earlier, HD is an inherited neurodegenerative disease caused by a pathological CAG repeat. People with positive gene testing results are identified as HD patients although some of them do not have any clinical manifestations yet, and people with family disease history (at least one parent with HD) can be identified as pa tients at risk for HD. And it was already confirmed in many studies that there was an inverse correlation between the CAG

PAGE 24

24 repeat length and the age of HD onset, which meant the longer the CAG repeat length was, the earlier the patients will have disease on set in their life (Penney, Jr., et al., 1997) So for the people having CAG expansion, a major unanswered qu relevance for drug aimed at disease prevention or onset delay. Langbehn DR and his colleagues developed a parametric survival model to predict the probability of neurological disease onset characterized by motor symptoms based on CAG repeat length at different ages for individual patients and also defined the variability in HD onset that was contributed from modifiers (b oth genetic and environmental) other than CAG size (Langbehn, et al., 2004) Their predicti ve model will be helpful to identify early markers of pathogenetic progression that may be useful in the development and monitoring of preventive treatment and also allow the clinician to provide detailed prognostic information to patients to plan their fu ture. Challenges in Clinical Trials for Neuropsychiatric Diseases Clinical trials for neuropsychiatric diseases have been considered as challenging due to unique disease characterization and limited measurement methods. Three of the main challenges will b e discussed detailed below. Placebo Effect Placebo is a substance or procedure that is considered pharmacologically inactive for the treatment arm as the product of placebo biological phenomenon combined with other potential factors contributing to symptom improvement such as natural history, life style, biases, phenomeno n and can be studied in dedicated experimental protocols.

PAGE 25

25 The inclusion of placebo treatment group has became the typical trial design to facilitate comparison with new therapeutics relating to clinical efficacy as well as adverse event by discriminating them from nonspecific effects that may be caused by trial design, expectation of patients or physicians, demographic differences and medical or environmental intervention factors et al. However, in recent years there has been a trend towards increasing pla cebo effects in clinical trials (Kemp, et al., 2010) And for central nervous system diseases, approximately 50% of recent trials failed to show superiority of the new drug over placebo statistically due to larg e variation in placebo effects (Fava, et al., 2003) Kemp et al summarized the mean change in Positive and Negative Syndrome Scal e (PANSS) score from baseline for patients in placebo arm and compared those numbers across different trials conducted during 1993 to 2006. The larger score reduction from baseline shown in Figure 1 3 indicated the increasing trend of placebo effect in ant ipsychotic trials during the last twenty years. Increasing and variable placebo effect caused negative impact significantly by increasing costs for clinical trials, more inconclusive and failed trials and even delays in the development of new antipsychotic s since the poor signal detection is sometimes incorrectly interpreted as less potent for new therapeutics relative to older treatment. Dropout High dropout rate is another significant challenge in clinical trials for neuropsychiatric diseases and it hap pens in placebo arm as well as active treatment arm. The common reasons for patients to leave from the trials earlier included lack of efficacy, adverse event, protocol noncompliance, patient lost to follow up and some other reasons unrelated to the trials Except withdrawal due to reasons unrelated to the trials, it was highly possible to produce biased

PAGE 26

26 treatment comparisons and reduce the overall statistical power if missing response values were ignored. There are a couple of methods to handle the dropou t problem during the trials. Last observation carry forward (LOCF) is one of the missing data imputation methods applied widely in practice. In this method, the last observed value is retained and carried forward to following time points until the last tim e point with missing data under assumption of constant profile after dropout, and the comparisons between placebo and active treatment arms are conducted at single time points without considering within subject correlation among different time points. LOCF may give biased results when the dropout mechanism depends on any observed or unobserved variables, or if unequal dropout rates between treatment arms result from certain reasons. And it was also doubted that potential bias resulted from LOCF imputation c ould lead to a more conservative analysis outcome and decision with less risk, because LOCF is done on both placebo and active treatment arms, so the conservative imputation would not result in conservative conclusion due to deduction when different groups were compared to each other. But on the other side, the reason that LOCF is still commonly used is because it is straightforward and convenient to researchers, clinicians and regulatory agency. Lack of C linical E ndpoints Most of neuropsychiatric disease s are chronic illness and have very slow disease at trials for symptom management drug, but it is much more difficult to evaluate treatment effect on disease progression within the regular duration time of clinical trials because it is almost impossible to follow up patients for a couple of years. Thus it became significant to find early and also reliable clinical endpoints or biomarkers to describe disease progression.

PAGE 27

27 Based on challenges existing in clinical trials for neuropsychiatric diseases, placebo model is needed to describe disease progression and provide an ins ight into the parameters that contribute to differences in treatment response among patients or studies as well as high dropout rate during the trials.

PAGE 28

28 Table 1 1. Summary of rating scales commonly used in brain disorders Brain Disorder Rating Scale I ndividual Items (n) Range Reference Schizophrenia PANSS 30 30 120 (Leucht, et al., 2005) BPRS 18 18 126 (Gorham and Overall, 1960) CGI 7 1 7 NA Depression HAMD 17 17 0 54 (HAMILTON, 1960) MADRS 10 0 60 (Mont gomery and Asberg, 1979) Alzheimer's disease ADASC 11 0 70 (Rosen, et al., 1984) Parkinson's disease UPDRS 44 0 176 NA Cognitive; BPRS=Brief Psychiatric Rating Scale; CGI=Clinical Global Impr ession; HAMD 17=17 item Hamilton Depression Rating Scale; MADRS=Montgomery Asberg Depression Rating Scale; MMSE=Mini Mental State Examination; Table Adapted from P illa et al (Pilla, V, et al., 2011)

PAGE 29

29 Table 1 2. List of items contained in YMRS Item Scale Increased Motor Activity 0 4 Appearance 0 4 Content 0 4 Disruptive/Aggressive Behavior 0 4 Irritabi lity 0 8 Insight 0 8 Language/Thought Disorder 0 4 Elevated Mood 0 8 Sexual Interest 0 8 Sleep 0 4 Speech (Rate and Amount) 0 4 Total 0 60

PAGE 30

30 Table 1 3. Drug therapy for patients with Bipolar Disorders Medication Indication Acute mania Mainte nance Biplar depression Atypical antipsychotics: Aripiprazole (Abilify) Yes No No Olanzapine (Zyprexa) Yes Yes Yes Quetiapine (Seroquel) Yes Yes Yes Risperidone (Risperdal) Yes Yes No Ziprasidone (Geodon) Yes No No Typical antipsychotics: Haloperidol lactate (Haldol) Yes No No Anticonvulsant: Carbamazepine (Equetro) Yes Yes Yes Lamotrigine (Lamictal) No Yes Yes Mood stabilizer: Divalproex (Depakote) Yes Yes Yes Valproate acid (Depakene) Yes Yes Yes Lithium Yes Yes Yes Table ad apted from Price et al (Price and Marzani Nissen, 2012)

PAGE 31

31 Table 1 4. List of items contained in Unified Huntington's Disease Rating Scale (UHDRS) Component Assessment Scale No. of items Possible scores Motor oculomotor function, dysarthria, chorea, dystonia, gait, and postural stab ility Total Motor Score (MOT) 31 0 (best) to 124 (worst) Behavioral Mood, behavior, psychosis and obsessiveness Total Behavioral Score 16 0 (best) to 93 (worst) Cognitive Verbal fluency, Symbol digit and Stroop test Total Cognitive Assessment 5 0 (worst) to >300 (best) Functional Capability to continue normal life and perform the certain tasks Total Functional Capacity (TFC) 5 0 (worst) to 13 (best) Functional Checklist Score 25 0 (worst) to 25 (best) Independence Scale 10 10 (worst) to 100 (best )

PAGE 32

32 Table 1 5. List of items contained in total motor score (MOT) Items 0 1 2 3 4 OCULAR PURSUIT (horizontal and vertical) complete (normal) jerky movement interrupted pursuits/full range incomplete range cannot pursue SACCADE INITIATION (hor izontal and vertical) normal increased latency only suppressable blinks or head movements to initiate unsuppressable head movements cannot initiate saccades SACCADE VELOCITY (horizontal and vertical) normal mild slowing moderate slowing severely slow, fu ll range incomplete range DYSARTHRIA normal unclear, no need to repeat must repear to be understood mostly incomprehensible mute TONGUE PROTRUSION can keep fully protruded for 10 s cannot keep fully protruded for 10 s cannot keep fully protruded for 5 s cannot fully protrude tongue cannot protrude tongue MAXIMAL DYSTONIA (trunk and extremities) absent slight/ intermittent mild/common or moderate/ intermittent moderate/ common marked/ prolonged MAXIMAL CHOREA (face, mouth, trunk and extremities) absen t slight/intermittent mild/common or moderate/intermittent moderate/common marked/ prolonged RETROPULSION PULL TEST normal recovers spontaneously would fall if not caught tends to fall spontaneously cannot stand FINGER TAPS (right and left) normal mild slowing and/or reduction in amplitude moderately impaired severely impaired can barely perform the task PRONATE/SUPINATE HANDS (right and left) normal mild slowing and/or irregular moderate slowing and irregular severe slowing and irregular cannot perform LURIA (fist hand palm test) no cue <4 in 10 seconds, no cue with cues <4 in 10 seconds, with cue cannot perform RIGIDITY ARMS (right and left) absent slight or present only with activation mild to moderate severe, full range of motion severe, with limited range

PAGE 33

33 Table 1 5. Continued Items 0 1 2 3 4 BRADYKINESIA BODY normal minimally slow mildly but clearly slow moderately slow, some hesitation markedly slow, long delays in initiation GAIT normal gait, narrow base wide base and/or slow wide base and walks with difficulty walks only with assistance cannot attempt TANDEM WALKING normal for 10 steps 1 to 3 deviations from straight line >3 deviations cannot complete cannot attempt

PAGE 34

34 Table 1 6. List of items contained in total functional capacity (TFC) Items 0 1 2 3 OCCUPATION unable marginal work only reduced capacity for usual job normal DOMESTIC CHORES unable impaired normal CARE LEVEL full time skilled nursing home or chronic care home F INANCES unable major assistance slight assistance normal ADL total care gross tasks only minimal impairment normal

PAGE 35

35 Table 1 7. Categories in diagnostic confidence rating Category Diagnosis Confidence level 0=Normal No abnormalities 0 1=Soft signs Non specific motor abnormalities <50% 2=Probable HD May be signs of HD 50% to 89% 3=Likely HD Likely to be signs of HD 90% 98% 4= Definite HD Unequivocal signs of HD

PAGE 36

36 Figure 1 1. Clinical manifestations and molecula r cause of HD. A) Progressive disease conditions in HD. B) Multiple pathophysiological changes mediated by mutant Htt protein and its fragments, eventually leading to neuronal dysfunction and degeneration (Krobitsch and Kazantsev, 2011) A B

PAGE 37

37 Figure 1 2. Mechanisms for the neuroprotective therapeutic targets and strategies A) Applications of RNA interference (RNAi) to silence the mutant HD gene. B) Application of small molecule inhibitors to reduce Htt aggregation. C) Application of small molecule histone deacetylases (HDAC) inhibitors to ameliorate chromatin repression an d transcription (Krobitsch and Kazantsev, 2011) A B C

PAGE 38

38 Figure 1 3. Tre nd of the placebo effect in clinical trials for Schizophrenia. The mean reduction in the PANSS score from baseline for patients receiving placebo treatment has increased in the trials conducted between 1993 and 2006 (Kemp, et al., 2010)

PAGE 39

39 CHAPTER 2 DEVELOPMENT OF A DISEASE MODEL FOR BIPOLAR DISORDER Background and Objectives As other neuropsychiatric disorders, clinical trials for bipolar disorder (BD) have similar challenges given disease characteristics as wel l as problems in diagnosis and treatment. One of the major challenges is large variation in placebo effect within and across clinical trials because of subjective evaluation method. This has been associated with diminishing drug placebo differences in clin ical trials, which significantly interferes signal detection for new medications and even results in failure of trials (Pilla, V, et al., 2011) Thus it is important to develop a placebo model to ch aracterize time course of placebo effect that can be shared in multiple clinical trials for different bipolar drugs, accordingly facilitates the quantification of drug efficacy. In the area of antipsychotic placebo effect modeling, various empirical models (such as linear model, weibull model, inverse Bateman function etc.) have been used to describe the placebo effect in schizophrenia (Kimko, et al., 2000; Friberg, et al., 2009) depression (Shang, et al., 2009) (Holford and Peace, 1992) (Bhattaram, et al., 2009) However, the placebo effect model for BD is not available in literature yet. High dropout ra te is another common problem at all treatment groups in bipolar trials, and last observation carry forward (LOCF) was applied in most of those trials to impute missing data and then carry out the comparison on primary efficacy endpoint. Development of drop out models can be utilized to describe missing data pattern and identify the significant covariates that affect dropout probability of patients. The specific aims were designed to provide valuable and sharable information to optimize the clinical trial de sign for BD:

PAGE 40

40 Specific Aim 1 To build a disease database for bipolar disorder based on multiple clinical trials under several approved bipolar drugs. Specific Aim 2 To create an empirical placebo effect model to quantify the time course of Young Mania Rati ng Scale (YMRS) total score. Specific Aim 3 To investigate the missing data mechanism using statistical analysis and to describe the dropout pattern during the trials and explore influence of potential covariates on probability of dropout in placebo group based on a parametric dropout model. Specific Aim 4 Combination of placebo effect model and dropout model was applied to simulate new clinical trials through Monte Carlo simulation approach. Methods Disease Database The disease database included 11 c linical trials on 5 approved drugs for acute treatment of bipolar disorder with manic/mixed episode. Trial selection was driven by availability of electronic datasets containing longitudinal YMRS score measurements and same primary efficacy endpoint in tri al design and analysis. All the data was organized into a pooled database in a uniform structure and format with common naming convention. All these clinical trials were randomized, double blind, placebo controlled, parallel group studies with same primar y efficacy endpoint (change from baseline in YMRS total score at day 21) and similar inclusion criteria. But they were different at trial duration time (3 or 12 weeks), number of treatment groups (2 or

PAGE 41

41 3), trial locations (only in US, mainly in US or non U S) etc. SAS 9.2 was mainly applied to organize data and build the dataset. Placebo Effect Model The exploratory data analysis was carried out first to present YMRS time profile in placebo arm graphically to facilitate the selection and creation of structu ral model. Time course of mean observed YMRS total score for each study and individual YMRS score profile were checked respectively. Empirical models are relatively simple and descriptive to fit the rating scale profile in neuropsychiatric disorders (Pilla, V, et al., 2011) Although the time course of mean YMRS total score in placebo arm of clinical trials for bipolar disorder had similar shape as other neuropsychiatric disorders, none of empirical models been used in other mental diseases was flexible enough to fit various individual profiles. The final structural model we decided to use was a combination of an exponential decay process and a linear progression, and these two components could switc h to be dominant depending on the shape of individual profiles, as Equation 2 1 showed: E=S 0 e ( Kt) +Slopet +Error (2 1) Where E and S 0 were the score at any time point and observed baseline score at original scale. This model only included two fixed effect parameters, K and Slope, needed to be estimated. K was the rate constant of exponential disease improvement, describing how fast the score dropped from baseline, and Slope was the rate constant of linear disease relapse which characterizes how fast the score grew again. Both K and slope were restricted to be non negative. Error represented the difference between the observed and predicted values.

PAGE 42

42 In order to constrain the future simulated score within the range of 0 to 60 given the possible range of YMRS total score, model fitting was performed in the logit transformation domain by using Equation 2 2: (2 2) Random variability was attributed to two sources: inter subject variability and residual variability. The inter subject variability was taken into account on both two parameters K and Slope, and exponential models were used to evaluate inter subject variability, such that value of parameter for each individual (Pi) was a function of the param eter value for the typical population was assumed to follow normal distribution with zero median and variance that was estimated as part of model estimation: Pi=EXP(Log( i ) (2 3) Where Log(Ppop) was the population median log transformed parameter. The residual variability came from error caused by model mis specification and/ or the evaluation of YMRS score and additive error structure was selected. One of the main objectives of the placebo effect model was to predict the YMRS total score of patients at certain visit time in placebo group, so the match between individual predicted values and observed data was considered as key diagnostic criterion for this model. Since there was some variation among these 11 studies, data was fitted separately for each study by the same model structure with minor adjustments on initial val ues and boundaries to obtain better individual predictions and yield more accurate parameter estimates. And the pooled dataset was fitted together as well to provide one set of parameter estimates including baseline score being also estimated, so that the information could be shared to trials for different bipolar drugs.

PAGE 43

43 Analysis of Dropout D ata Dropout rate up to 3 weeks in placebo arm across all the studies was calculated to assess the incidence of dropout event during clinical trials for bipolar drugs Then the percentage of main single dropout reasons, lack of efficacy and adverse event, documented by trial conductors in dropout patients were summarized and compared across different treatment groups for each study to investigate whether dropout was a completely random issue during the trials. In order to explore further the relationship between dropping out happened in placebo group and efficacy endpoint measurement, the mean value of observed YMRS total score for the patients who dropped out at the ne xt visit with those who did not was compared for each study. Placebo D ropout M odel In order to establish a quantitative link between duration of time until dropout and efficacy measurements as well as factors associated with trial design for use in the des ign of future trials, a parametric survival model was utilized to describe the dropout pattern during the trials and relate these potential predictors with time to dropout. For determining the relationship between efficacy measurements and time to dropout, the observed baseline score and individual parameters K, Slope estimated from our placebo effect model were used as predict factors. Besides, several trial specific covariates thought to be associated with the survival time variable, such as trial duratio n, trial starting year and number of treatment arms were also included. Several widely used survival functions, including exponential, log normal, weibull, and gamma, were tested. Likelihood ratio tests and diagnostic plots were applied for selection of an appropriate survival function. This model was fitted to studies separately in order to assess whether it would be appropriate to pool these studies together, and final parameters were updated with a pooled analysis to provide sharable information to trial design for BD.

PAGE 44

4 4 Log normal was found to be the best survival function to describe the relationship between time to dropout and a set of predictors, as shown in Equation (2 4). Log(T)=intercept+coefficientcov +error (2 4) Where T was the time to dropout (day), cov was a vector of possible covariate values; coefficient was a vector of unknown slopes for each of those covariates; error was the residual variability following a norm al distribution with a mean of zero and vari 2 T Because trial duration of 12 weeks, trials with 3 treatment arms and trials starting after 2002. Intercept, slopes 2 T were estimated from the model and chi square tests were carried out on all the slope values at same time, which allowed us to identify significant factors to dropout and predict dro pout probability for a new clinical trial. Evaluation of the dropout model was conducted through internal and external validation. Data from 5 of 11 trials was used to estimate parameters from built dropout model. And then based on these parameters and th eir variance covariance matrix as well as baseline score and individual parameters estimated from our placebo effect model, two hundred replicates of dropout data were simulated for each of all 11 trials. The median survival time and 95% confidence interva l of simulation was summarized across 200 replicates for each study and then the simulated survival curves were compared to the observed survival curves. Simulation for the 5 trials and the rest 6 trials were considered as internal and external validation respectively. Model Based C linical T rial S imulation Combination of placebo effect model and dropout model was applied to simulate new clinical trials through Monte Carlo simulation. The detailed procedure was as follows:

PAGE 45

45 A total of 200 replicates of dropo ut parameters, including intercept, slope for each predictor and error term, were simulated based on their typical value, variance and covariance obtained from built dropout model. For each study, 200 replicates of new dataset were produced by selecting sa me number of random samples with replacement from original dataset with individual parameters estimated from placebo effect model. 200 replicates of dropout parameters and 200 replicates of placebo effect parameters produced from the two previous steps wer e merged together. According to Equation 2 4, dropout time was predicted based on individual dropout parameters simulated from the first step and individual efficacy related parameters randomly produced from the second step as well as trial specific covar iates for each study; If predicted dropout time was not earlier than end of trial (3 weeks or 12 weeks), then the patient was counted as completer; The YMRS total score was predicted at each planned visit time based on individual efficacy related paramet ers and residual error estimated from placebo effect model for each study according to Equation 2 5. Predicted dropout time for each patient was compared to planned visit time and all the predicted score at time points that were later than predicted dropo ut time was deleted since those predictions should be considered as missing observations in the simulated trials; 90% confidence interval was summarized on median value, 5% and 95% quantile of predicted scores and visualized the closeness of it to the summ ary of observed score at all planned visit time points for each study to evaluate the predictive capability of our models. Results and Discussion Disease D atabase Individual longitudinal data from over 3000 subjects enrolled in 11 clinical trials for 5 approved bipolar disorder drugs were included in the database. As Table 2 1 showed, most of these 11 trials lasted for 3 weeks and only two trials for drug D had longer duration of 12 weeks to assess maintenance of the drug effect. It was also found that a ll the studies with two treatment arms were conducted only within US except study B3; while three arm trials were carried out at multiple international centers. Study A1, B3, C1 and C2 had several international centers, but

PAGE 46

46 more than 50% of trial centers w ere at US. All of 11 studies were started after year 2000, and study C1, C2 and E1 were more recent. The demographic information and baseline YMRS total score were summarized in Table 2 2 and compared across trials conducted at non US sites, mainly US site s and only US sites. Most of information was consistent across the three groups except race and baseline score. Different percentage of races resulted from different trial sites and higher baseline score in Non US trials were due to the different enrollmen t criteria in these two trials that had further requirement on score of individual items in YMRS in additional to total score when recruiting patients. Placebo Effect Model At the beginning of placebo model development, several empirical model structures available in literature (Pilla, V, et al., 2011) shown in Figure 2 1 were tried to fit pooled dataset of bipolar trials since the overall trend and shape of mean YMRS total score time profile (Figur e 2 2) was similar as other neuropsychiatric disorders. However, it was found out later that none of these models (linear, Emax, Exponential, Weibull, inverse Bateman etc) could fit the profiles with the total score rising from the baseline directly, which was one of the typical profile shapes happening in every study, especially for subjects who dropped out of study early. And we also found other various shapes when we checked YMRS time profile at individual level as Figure 2 3 showed. At the end it turne d out that the mixed model with linear function added on the basic exponential model could provide flexible enough fitting to cover almost all the types of profile shape, and was selected as our structural model. The comparison of individual observations, individual predictions and population predictions of time course of YMRS total score for some representative subjects was shown in Figure 2 4 to demonstrate this model was flexible enough to describe various types of individual profiles, although the scor e time profile in an individual patient could differ substantially from

PAGE 47

47 the population prediction. In Figure 2 5, scatter plot of the individual observed and predicted YMRS total scores for each study, completed by the reference identity line was shown to illustrate the same structural model could fit all the studies reasonably well. It was interesting that this model structure had been utilized for modeling tumor size (Wang, et al., 2009) which implied this si mple model structure might be applicable in multiple areas, since it was physiologically reasonable to describe the process of disease improvement initially by drug intervention or placebo response and following disease progression or relapse. The final e stimates of parameters baseline score, K and Slope and estimate precision obtained from the placebo effect model for the overall population in pooled dataset was listed in Table 2 3. There was high variability on K and slope among different subjects, which partly resulted from variability across different studies; while baseline scores were relatively close. The narrow confidence intervals on typical value of parameters as well as small relative standard error (RSE %) indicated parameter estimation was reas onably precise in our model based on large dataset. Based on the clinical data we collected in database, demographic factors such as gender, age, race etc were not significant covariates to placebo effect (data not shown). Thus no covariates were added in to our structural model yet. But it might be possible that some other factors regarding disease feature and treatment history of patients affect change of YMRS total score over time. Lipkovich and colleagues found out from one clinical study the presence o f some features including individual YMRS item, type of episode at baseline, treatment history, number of previous manic episodes etc as potential risk factors for relapse in patients who had responded to treatment (Lipkovich, et al., 2008) Since in our dataset different studies collected the information differently or some studies did not include the related information, it was hard to

PAGE 48

48 explore the relationship be tween those potential factors and change of YMRS total score for pooled dataset. Thus at this stage we did not further investigate other potential covariates other than demographic factors. Analysis for P lacebo D ropout Most of 11 studies had dropout rate of more than 30% or even beyond 50% in placebo arm (Figure 2 6) which yielded reasonable sample size and enough data source to us to evaluate the missing data mechanism and build placebo dropout model with sufficient precision. The missing data mechanism r elates to why observations are missing and the connection of those reasons with treatment outcomes. In all 11 trials, the reasons for discontinuation were recorded in study report and two of the single main reasons for discontinuation in overall population were lack of efficacy and adverse event. As Figure 2 7 showed, our analysis on these two reasons in all treatment arms for each study demonstrated that dropout in placebo arm was mainly due to the lack of efficacy and in active treatment group dropout eve nt was more likely to result from adverse event. This result ruled out the mechanism that dropout was a completely random event independent of the observed values of outcome variables and any other covariates. In this case, the dropout could not be ignore d for the proper analysis to the clinical response profiles. In order to explore further the relationship between dropping out happened in placebo group and observed efficacy endpoints, we compared mean observed YMRS total score of patients who did not d rop out at the next visit (shown as circles) with last observation of patients who left from the trial by the next visit (shown as crosses) at each planned visit day in placebo arm in Figure 2 8. It was obvious that at almost all the time points the observ ed score from dropout patients were higher than those from non dropout patients, which meant that they suffered from more severe symptoms than non dropout patients. Therefore, occurrence of

PAGE 49

49 dropout event in placebo arm depended on the observed efficacy mea surements and parameters related to placebo effect should be important predictors for dropout events. Placebo D ropout M odel Since parametric models can be used for clinical trial simulation and might be able to extrapolate model predictions under differe nt design condition, we applied a parametric approach with sequential estimation steps where efficacy related parameters were obtained from placebo effect model first and then included into survival dropout model as potential predictors on time to dropout. It was not surprising that two efficacy related parameters were the most significant predictors of time to dropout in survival analysis based on chi square test as shown in Table 2 4. The coefficient value associated with the natural logarithm of K was po sitive, so larger the K value implying disease symptoms improved faster, the longer time patient would stay in the trial. The other parameter slope, describing the velocity of disease deterioration, was the top factor to drive patients to leave from the tr ials prematurely, and the larger the value was meaning disease relapse faster, the more likely patient would drop out of the trial. It was also reasonable to indicate from the results that higher baseline score was one of the reasons for early dropout alth ough the relative influence was not significant as K and slope. Thus it should be kept in mind that treatment effect might be easier to be detected on severe patients, but more severe disease symptoms was also likely to increase probability of dropout. Although there are already a couple of methods to handle the dropout problem during the trials, such as LOCF and mixed effect model repeated measure (MMRM) model, the best way is still to reduce the dropout rate by optimizing the trial design, since any me thod is only valid under certain assumptions which might be hard to justify in most clinical trial settings. Our dropout model also incorporated design related covariates and may shed light on the design accounting for reduction of early dropouts.

PAGE 50

50 It was r easonable to imply from the test in our dropout model that newer trials with longer duration may benefit to lower the probability of dropout. The trials conducted in recent years possessed more advanced medical facilities and technologies as well as more c onsiderate care and service to patients compared to before, which improved compliance of patients to the trials, accordingly lengthened the time patients stayed in trials. Dropout rate was often calculated as cumulative, so longer trials would be much more likely to have higher dropout rate than shorter trials at the end of trial, but it made sense to explore the relationship between duration of trial and dropout rate after same time since the trials started. And it was even more relevant in our database si nce the two 12 week trials had same primary efficacy endpoint at 3 weeks with other 3 week trials, so in our analysis trial duration was included as a potential predictor as well. We found out from the test that patients dropped out slowly in the longer tr ials, probably because they would like to wait longer to see the expected treatment effect. Thus it may be helpful to design trial with the duration longer than time to measure the primary efficacy endpoint (as trial D1 and D2), if the more consumed time a nd funds is acceptable. Number of treatment arms has been considered to be one of the potential predictors for both efficacy and dropout, since it corresponded with the probability of patients coming from active treatment groups, either comparator drug or new drug. The more treatment groups, the higher probability that patients came from active treatment groups, probably the higher expectation on disease improvement (lower YMRS total score in case of BD) of both physicians and patients, accordingly the low er score given due to subjective evaluation method, and then it may result in less happening of dropout. However, in our model the influence of number of arms on probability of dropout was the least significant among all the predictors, and the sign of ass ociated coefficient was also opposite to what we had expected. It was very likely that our

PAGE 51

51 dataset could not provide enough comparison power to test the significance since the included trials only had two or three treatment arms, or it did not make differe nce to have two or three treatment groups in terms of effect on dropout rate. It may help to include trials with more treatment groups to investigate further this finding. Normally trial location could affect the analysis results, including quantificati on of placebo effect, especially when the evaluation method was subjective. And it might influence dropout probability due to different characteristics of people in different places and different practical operation of local clinic to the same protocol. Bu t in our dropout model, we did not include it as predictor, because the factor of trial locations, non US, mainly US or only US, was confounding with the other two factors, duration of trial and number of arms in our dataset. As Table 2 1 showed, both two non US trials happened to be the two with longer duration of 12 weeks, and all the only US trials had two treatment arms. As we know, the purpose of requirement or recommendation to conduct trials at different countries or regions was to obtain information about therapeutic responses and safety issues from different races and ethnic groups. Moreover, the selection of trial locations was usually based on the availability of local clinical facilities and resource of drug development companies or institutions. So it might be harder to adjust tr ial location in design compared to the flexibility of treatment arm number and trial duration But it was definitely necessary to explore the effect of trial location on dropout probability if no other co nfounding factors in the analysis If trials conducted at any countries or specific regions yielded obviously higher dropout rate, it would be needed to investigate the possible reasons for that, such as people characteristics, probl ems in operation of trial protocol diff erent analysis, etc; and it should be considered to remove the data from those locations or avoid to conduct trials at

PAGE 52

52 those places in the future without affecting the trial objectives if the problems resu lting in higher dropout rate were difficult to be s olve d The graphic comparison of survival curves with simulated data and those with observed data was shown in Figure 2 9 and demonstrated reasonably good prediction of the survival model for all the studies, including studies for both internal and exter nal validation, which indicated the good adequacy of our dropout model to describe the dropout pattern during clinical trials for bipolar disorder in population treated with placebo. Model Based C linical T rial S imulation Simulation was carried out based o n combination of our placebo effect model and dropout model and the simulated YMRS score time profile was compared to observation for each study. From the graphic comparison shown in Figure 2 10, the predicted median, 5% quantile and 95% quantile were all very close to observed value except Trial A2 and A3 where the observed 95% quantile fell a little out of predicted confidence interval. The reason for that was probably due to high dropout rate during trials accordingly less data available from these studi es for model development, especially for some severe patients with higher baseline score and also patients with score increasing all the way from baseline and dropping out early, and there was also large variability in efficacy measurements of these patien ts, which made model even harder to describe this part of data well. However, the generally reliable predictions indicated reasonable predictive capability of our models and these two models can be used to simulate new trials for bipolar disorder. Summary All the above results suggested that: The compiled bipolar disease database offered a standard data structure and a convenient data source across study comparison for meta analysis.

PAGE 53

53 Both placebo effect model and dropout model described the observation rea sonably well based on various diagnostic plots. The developed models can be combined to simulate the most likely level of placebo effect with dropout events in a new clinical study. The information obtained from our modeling work can be used to help trial design for bipolar disorder. Nevertheless, our current analysis was preliminary and still had limitations that mainly related to the need to confirm our preliminary results in other dataset including more clinical trials with different study design.

PAGE 54

54 Tab le 2 1. Summary of drugs and trials in the disease database for bipolar disorder Drug No. Trial No. Duration (week) No. of Arms Year of Trial Country of Trial Center A A1 3 3 2000 2001 Mainly US A2 3 2 2000 2001 Only US A3 3 2 2002 2003 Only US B B1 3 2 2000 2001 Only US B2 3 2 2000 2001 Only US B3 3 2 2002 2003 Mainly US C C1 3 3 2004 2006 Mainly US C2 3 3 2004 2006 Mainly US D D1 12 3 2001 2002 Non US D2 12 3 2001 2002 Non US E E1 3 2 2006 2007 Only US Table 2 2. Demograp hic and Baseline Characteristics in included bipolar trials Characteristic Non US (n=2) Mainly US (n=4) Only US (n=5) No. of patients (%) a 599 (18.32%) 1584 (48.44%) 1087 (33.24%) Gender, % Female 52.75% 44.26% 49.40% % Race White 62.94% 59.79% 6 6.24% Asian 34.39% 13.07% 1.38% Black <1% b 15.97% 26.68% Age, Mean (range) 41.1 (18 79) 39.0 (18 76) 39.7 (18 76) YMRS baseline Mean (SD) 33.22 (6.53) 28.58 (5.84) 28.35 (5.56) a. All the percentage values showed percentage summarized in each g roup except that for number of patients (the first line) that showed percentage in the whole dataset. b. The exact number for this information was not available in study report.

PAGE 55

55 Table 2 3. Parameter estimates of placebo effect model for pooled dataset M odel parameter Estimate 95% confidence interval Inter subject variability CV% *RSE% Shrinkage% Baseline 28.5 [28.1, 28.9] baseline 20.2% 5.9% ETA baseline 17.2% K 0.0287 [0.0281, 0.0294] K 122.1% 5.6% ETA K 22.8% Slope 0.1381 [0.1368, 0.1393] sl ope 104.9% 1.4% ETA slope 32.4% RSE (relative standard error) was calculated based on between subject variance Table 2 4. Parameter estimates of placebo dropout model for pooled dataset Parameter Coefficient Estimate StdErr ChiSq P Value Ln(baseline) 0.58 0.17 11.05 0.0009 Ln(k) 0.34 0.04 73.11 <.0001 Ln(slope) 0.50 0.04 181.68 <.0001 No. of Arms 0.20 0.08 6.23 0.0126 Duration 0.89 0.12 57.80 <.0001 Starting Year 0.25 0.07 10.96 0.0009

PAGE 56

56 Figure 2 1. Sc hematic representation of some empirical models available in literature used in neuropsychiatric diseases (Pilla, V, et al., 2011)

PAGE 57

57 Figure 2 2. Time profile of mean observed YMRS total score in placebo arm A) 3 week studies B ) 12 week studies A B

PAGE 58

58 Figure 2 3. Individual time profile of observed YMRS total score for a representative set of subjects

PAGE 59

59 Figure 2 4. Comparison of individual observation, individual prediction and population prediction on time profile of YMRS total score for a representative set of subjects

PAGE 60

60 Figure 2 5. Scatter plot of the in dividual observed and predicted YMRS total scores.

PAGE 61

61 Figure 2 6. Dropout rate up to 3 weeks in placebo group for all the studies

PAGE 62

62 Figure 2 7. Summary and comparison of percentage of dropout reason s in placebo and active treatment groups for each study

PAGE 63

63 Figure 2 8. Comparison of mean observations between dropout patients and non dropout patients by visit A) 3 week studies. B ) 12 week studies A B

PAGE 64

64 Figure 2 9. Comparison of survival curves between simulation and observation A) F or internal validation B ) For external validation A B

PAGE 65

65 Figure 2 10. Comparison of YMRS total score time profiles between simulation and o bservation for all the studies

PAGE 66

66 CHAPTER 3 Background and Objectives them aimed to predict age of disease onset as well as increase the predictive precision and accuracy (Maat Kievit, et al., 2002; Falush, et al., 2001; Pilla, V, et al., 2011; Langbehn, et al., 2004) because the confirmed inverse correlation of CAG length and age at onset provided confidence on prediction, and predicting disease onset is to help enrichment design for disease prevention trial. However, much less research effort was made on description of disease progression after disease onset due to lack of enough data source and uncertainty on selection of efficacy endpoint among many rating subscales under UHDRS. Until now only linear model was applied to assess progression rate of total motor score (MOT) and total functional capacity (TFC) in either clin ical trial (Meyer, et al., 2012) or observational studies (Mayeux, et al., 1986 ; Penney, Jr., et al., 1990; Marder, et al., 2000) but the reported value for same outcome scale was variable probably because of different patient populations and different observation time. In research by Marder et al, it was found that lower TFC score at baseline, indicating more severe function impairment, was associated with less rapid annual decline in TFC score (Langbehn, et al., 2004; Marder, et al., 2000) Our long term goal was to develop natural his tory models for disease progression in untreated HD patient population with availability of huge dataset from multiple observational studies and clinical trials. The model might be clinically relevant to drugs aimed at disease treatment and useful to guide trial design for disease modifying drugs and also shed light on the etiology of HD. During the model development, the effects of primary predictor variables CAG length and age can be quantified. Attempts will be also made in the future to find biomarkers for

PAGE 67

67 early detection of disease progression. At present, the most promising candidates for biomarkers are imaging variables, because the earlier and most striking neuropathologic changes in HD are in the neostriatum and cerebral cortex and are detectable ma ny years before clinical manifestations and may mediate the effects of age and CAG length on various outcome variables (Krobitsch and Kazantsev, 2011) The specific aims were designed to provide preliminary analysis, reliable methodology and modeling framework to serve for our long term goal: Specific Aim 1 To create a pooled modeling database for diverse studies and assembles, organizes as well as updates in a common format and using naming conventions. Specific Aim 2 To describe disease progression of HD patients by modeling time course of selected outcome variables total motor score and total functional capacity. Specific Aim 3 To build a covariate model to incorporate the effect of influential covariates on disease progression characterized by selected outcome variables. Methods Disease Database The current modeling database for HD included 5 studies, 1 clinical trial CARE and 4 observational studies COHORT, PREDICT, REGISTRY AND TRACK based on the availability and secured accessibility of electronic datasets collected from multiple collaborative institutions despite of different trial design, patient populations and objectives. All the data was organized into a pooled database in a uniform structure and format with common naming convention. This

PAGE 68

68 database mainly collected information on components Scale (UHDRS), and also incorporated core predictor variables age, CAG length, age at diagnosis, disease burden, gene testing results, family disease history and demographic information gender as well as trial informa tion such as trial site, visit number, age at entry of trial etc. Some other important variable like imaging measurements were not integrated into the current database since only two of studies collected the related information. SAS 9.2 was mainly applied to organize data and build the dataset. Dat a Exploration The exploratory analysis on two selected rating scales MOT and TFC was carried out first to present individual time profile of MOT and TFC graphically and also check roughly the change of annual pr ogression rates of these two variables during observation period. For time profile, the analysis was only conducted on placebo data in clinical trial CARE because of little missing data and less variable visit time during the trial; while assessment of pro gression rate was done for both clinical trial and observational studies. Since only linear model available in literature was used for MOT and TFC, our data exploration also applied linear regression to evaluate mean slope of increase in MOT and decline in TFC. Stepwise description of our analysis method was as follows: In clinical trial CARE, almost all the subjects contributed 10 observation points, so each the sl each part. So in clinical trial, there were two estimated mean slopes for each subject. For observational studies, only one slope was estimated based on all the observation po ints for each subject due to sparse observations and a lot of missing data. Mean score (Xi) was summarized for each part of subjects in clinical trial and each subject in observational studies. The estimated slope (Yi) was paired with the mean score ( Xi).

PAGE 69

69 Based on rank and range of Xi as well as number of patients, subgroups of patients were created. In each subgroup, the mean score of patients was relatively close to each other. Mean slope Y and associated confidence interval as well as mean score X w as calculated based on Yi and Xi in each subgroup. Relationship between Y and X was visualized based on plot of Y over X. Structural Model Development Selection of model structure Based on literature research and data exploration, logistic model was used to characterize the disease progression. Equation 3 1 is the generalized logistic model that represents the rate of disease progression in the form of a differential equation: (3 1) where E is the score at each time point and dE/dt refers to progression rate of rating score; And r progression rate chan ges; Emax is the maximu m possible score (124 for MOT and 13 for TFC); In this nonlinear mixed effect model, fixed effect parameters to be estimated included baseline score that was score of patients at entry of study, rate constant and possible sha pe factors; random effect came from both between subject variability associated with baseline and rate constant and intra subject variability accounted for residual error. Based on objective function value and parameter distribution, exponential models wer e used to evaluate between subject variability for baseline score, such that value of baseline score for each individual (BSi) was a function of the parameter value for the typical individual (BSpop) and a random individual

PAGE 70

70 deviation represented BS i BS i in the population was assumed to follow normal distribution with zero median and variance that was estimated as part of model estimation: BS i ) (3 2) For between subject variability associated with another parameter rate constant, exponential model and additive model (shown in Equation 3 3) were used for TFC and MOT respectively. r i =r pop ri (3 3) W here value of rate constant for each individual (r i ) was a function of the parameter value for the typical individual (r pop ) and a random individual deviation rep ri ri in the population was assumed to follow normal distribution with zero median and variance that was estimated as part of model estimation. The residual variability came from error caused by model mis specification and/ or the eval uation of YMRS score and additive error structure was selected. Differential equation was used because of enough flexibility to handle complex equations when shape factors were introduced in. First order conditional estimation method (FOCE) in NONMEM versi on 7.1.0 was applied to estimate parameters. Table 3 1 and Table 3 2 listed all the candidate model structures for MOT and TFC respectively during our structural model search. Besides differential equations that represented progression rate, the equation f or inflection point was also given for each model in the tables. If disease progression followed logistic model, the progression rate could reach the maximal level determined based on maximal possible score and value of shape factors. For each outcome, the simplest linear model was tested first where progression rate was constant and therefore the inflection point did not exist. Then a series of logistic models was f ollowed to fit data. In Logistic 1, all the shape factors were set to 1, and the inflection point at this model was exactly at

PAGE 71

71 the half maximum score (62 for MOT and 6.5 for TFC). In all the other three logistic models, the shape factors were added at diff erent sites of the first logistic equation to adjust the model fitting and then the location of inflection point might leave from the center of score range depending on the estimate of shape factors. In term of number of estimated parameters, the last thre e logistic models needed one additional parameter compared the first two due to adding shape factor. The first logistic model was selected as our structural model based on objective function value and likelihood ratio test for significance of the parameter which differed between two nested models. The match between individual predicted values and observed data was considered as key diagnostic criterion for this model, so graphic comparisons of score time profiles between population prediction, individual prediction and observation was made for subjects in placebo group of clinical trial to evaluate predictive ability of selected structural model. Model v alidation Our application of logistic model was based on literature and data exploration, so it was re asonable to compare the relationship between model predicted progression rate and observed score with that done by linear regression on observed data in exploratory analysis. The model predicted individual progression rates were obtained from individual es timates of rate constant and observed score at each time point according to the Equation 3 4. Rate ij =r i E ij (1 E ij /Emax) (3 4) Where Rateij was the predicted progr ession rate for individual i at time point j and ri was individual estimate of rate constant from our structural model. Eij represented measured score for individual i at time point j and Emax was still the maximal possible score. Next the similar method as used for observed placebo data was applied to summarize simulated rates and then graphic comparison between observation and simulation was carried out:

PAGE 72

72 After calculation on simulated data, each patient had predicted progression rates at all observation points. Then two mean progression rates and two mean observed scores were obtained for each subject based on the first 5 points and the last 5 points respectively. part of subject in clinical trial. Same subgroups of patients were created as we did for observed data in the previous section. group. Population progression rates Rate pop were also calculated according to the Equation 3 4 but using population estimate of rate constant r pop and all the possible scores in the whole range of observed score at this population so that the population pr ogression rates yielded a smooth line when it was plotted over score. pop over all the scores together to visualize the closeness among these three. Besides graphic comparison based on our own data, we also did numerical comparisons with mean annual progression rate available in literature. Several publications were found to report mean rate of change in MOT and TFC per year using linear model based on data from either clinical trial or observational study. Then based on our typical value of progression rate constant and the mean MOT and TFC scores in their studies, we calculated mean predicted progression rates and compared to their published values. At the end external validation was conducted using data from active treatment groups in the same clinical trial CARE since the tested interventions failed to significantly alter the change in TFC, MOT and other measures (Krobitsch and Kazantsev, 2011; 2001) Firstly we used parameter estimates from our model built based on only placebo data to simulate 50 replicates of MOT for population in the other three active treatment groups and then visual predictive check was done on time profile of 5% quartile, median and 95% quartile score between prediction and

PAGE 73

73 observation. Moreover, three sets of parameters were also compared after model fitting on only placebo data, only active treatment data and all the data together in CARE. Model fitting on observational data After se lection of model structure based on clinical trial data, model fitting was moved to data in observational studies. Among all these four observational studies, the pooled dataset from COHORT and REGISTRY was used to do analysis and estimate parameters. And only subjects with at least 3 visits and without missing values for age were included into our current analysis Normal people and patients at risk were removed from analysis as well. The selected logistic model was applied to fit pooled dataset of COHORT and REGISTRY, and the goodness of fitting was checked at individual level. The parameter estimates were also compared with those from fitting clinical trial data to get comprehensive information about disease progression of HD. Covariate Model Developme nt Preliminary exploration on potential covariates and development of covariate model was carried out on observational data from COHORT and REGISTRY due to much larger data size. The most interested covariates age and CAG length were explored first by visu ally checking the trend of change in between subject variability of parameters baseline and rate constant as change of age and CAG respectively on plots, and based on the observable trend, the relationships between covariates and the individual parameters were quantified according to Equation 3 5 and Equation 3 6: baseline (CAG/45) (AGE/50) (3 5) r pop = rate constant (CAG/45) 5 (3 6) baseline was the typical value of the estimated baseline for a subject with the value of the covariates CAG and AGE scaled to the typical values 45 and 50; rate constant was the typic al

PAGE 74

74 value of the estimated rate constant for a subject with the value of the covariates CAG scaled to 3 4 5 were the scaling parameters for the range of the covariates. Likelihood ratio test was used to compare the covariate m odel with structural model to identify covariates with statistically significant effect on parameters, and visual inspections were made again to check trend observed previously in structural model. Results and Discussion Disease D atabase Individual longit udinal data from 347 subjects in 1 clinical trial and over 10,000 subjects enrolled in 4 observational studies were included in the disease database for HD. As Table 3 3 showed, in clinical trial CARE almost every patient contributed 10 observation points with more frequent measurements while in observational studies the visiting was less frequent and there was a lot of missing data due to lost following up that was very common in observational studies, although the duration of observation period was longer in observational studies. As a result, only maximal number of observation points per subject was listed in Table 3 3. In REGISTRY, number of observation points per subject could be from 1 up to 30 because some patients were followed up for 10 years, but a bove 80% of subjects in this study provided only 1 to 3 observations. Observational studies enrolled not only HD patients, but also small number of normal people and patients at risk. Since we only focused on HD patients for our analysis and model develop ment, demographic information and baseline MOT and TFC score were also summarized in only HD patient population for each study as shown in Table 3 4. Demographic information age, percentage of female were relatively consistent across different studies exce pt that pediatric and adolescent patients were also enrolled in COHORT and REGISTRY and PREDICT had a little more female patients. And in terms of disease related variables, it was obvious that COHORT

PAGE 75

75 and REGISTRY also recruited and followed up patients wi th longer CAG length, higher MOT and TFC baseline score while patients in other studies, especially PREDICT and TRACK were at early disease stage, because the study objectives were to examine the prognostic significance of early clinical symptoms or biomar kers. Thus it might be reasonable to look into patients in COHORT and REGISTRY if we desired to explore the characteristic of disease progression during the entire disease process. Dat a Exploration As individual time profiles of MOT and TFC shown in Figur e 3 1, MOT score increased and TFC score declined as observation time went. Based on graphic checking, the relationship of MOT over time was approximately linear in some of subjects, but the patients were not followed up for a long time due to the short tr ial duration. And for TFC, the decline already showed nonlinear trend in a short observation time. Exploratory analysis on relationship of assessed progression rate with observed score was carried out based on linear regression. And results of MOT and TFC in the clinical trial were shown in Figure 3 2 and Figure 3 3; results in observational studies were shown in Figure 3 4 and Figure 3 5. In the clinical trial there were smaller population size and much smaller number of patients at late disease stage, so most of observed MOT score fell into the lower half of the whole range and most of observed TFC score were at higher half of the entire range. As a result, only 4 subgroups were created. For MOT, the trend was observable that as MOT increased, mean slop e increased as well and when score reached at around 50 to 70, the slope was very large. And then as the MOT continued to increase, slope started to drop from the largest level. For TFC, there

PAGE 76

76 was similar trend but in opposite direction since lower the TFC score the worse of functional capacity. Then we used same method to analyze observational data. More subgroups, 11 for MOT and 7 for TFC were divided because of much more data available and more patients with higher MOT and lower TFC in observational stu dies. As Figure 3 4 and Figure 3 5 showed, similar but more obvious trend was observed for both MOT and TFC in observational studies, and the associated confidence intervals turned to be narrow because of more subjects in each subgroup. Structural Model De velopment Logistic model In order to model biological systems, including natural progression of disease, numerous models have been introduced. Linear model becomes one of the most frequently applied models especially when disease progression is relatively slow. But the assumption that the progression rate keeps constant was usually invalid during longer observation period and the assessment of slope at time zero was considerately affected by baseline disease status. Another common model, simple exponential growth model can provide an adequate approximation to biological growth or disease progression for the initial period, but the unrestricted growth or progression is not realistic generally in the cases of intraspecific competition, functional maturity, or due to saturation level or the limit of evaluation method (Tsoularis and Wallace, 2002) Taken this characteristic into account, logistic model was introduced. The logistic growth model was successfully applied to pharma codynamic analysis of in vitro bactericidal kinetics (Ya no, et al., 1998) and in disease model area, it has traditionally been used in pharmacodynamic models of tumor growth (Mager, et al., 2003) disease progression (Samtani, et al., 2012; Ashford and Schmitt, 2001)

PAGE 77

77 Figure 3 6A showed the typical relationship between disease progression rate and observed score if disease progression was assumed to fol low logistic model where progression rate depended on the score at each time point through a rate constant. As score increased, progression rate increased as well until it reached the maximal level at inflection point where the progression rate was the fas test and progression rate was slower at scores above and below this point. The location of inflection point was not fixed and would be driven by data. Then the progression rate slowed down as it approached the maximal limit. The time profile of observed sc ore under assumption of logistic model was also simulated and shown in Figure 3 6B. During the early phase of the disease, the logistic model assumed that the rise in disease scores increased exponentially and then approximately linear within a short time period. After that score continued to increase but with a very slow rate until it achieved plateau of score due to floor effects that made it hard to measure the change of disease status in severe patients. nic illness and disease scores are constrained to lie within a certain theoretical range. Therefore with inspiration from application we decided to use logisti c model to characterize the disease progression. Selection of model structure As shown in Table 3 5 and Table 3 6, the simplest logistic model was selected as the best structural model for both MOT and TFC based on likelihood ratio test because it fitte d data better than linear model and additional shape factors did not improve the fitting significantly, accordingly the inflection point was determined at half maximal score. The goodness of fitting plots were shown in Figure 3 7 and Figure 3 8 for MOT and TFC respectively, which indicated selected structural model described observation reasonably well. Compared to fitting plots for MOT, the decline in TFC tended to show more nonlinear trend during the same duration of

PAGE 78

78 observation. Population estimates was well as estimate precision of model parameters were listed in Table 3 7. Model validation It was shown in Figure 3 9 and Figure 3 10 that mean of individual observed progression rate, mean of individual predicted progression rate and simulated population p rogression rate for MOT and TFC were relatively close to each other, which indicated that our structural model could describe the overall trend of observed data reasonably well. In this clinical trial, almost all the patients were early HD patients, so the ir MOT score focused on the lower part of score range (0 to 70) and TFC score distributed in the range of 13 to 6. As a result, only based on data in clinical trial it was hard to observe the approximate bell shape for relationship of change rate of score and observed score. Besides our own data, it was also necessary to compare our model predicted progression rate to number reported in literature. Meyer et al evaluated the progression of several UHDRS subscales, including MOT and TFC by linear regress ion and estimated slope based on data of eight measurements from screening to study end in a 36 month clinical trial, and reported the mean rate of change in MOT was 4.748 units per year (Meyer, et al., 2012) And the mean MOT score was estimated as around 31 from the mean MOT time profile shown in their publication. Then according to Equation 3 4 and typical value of progress ion rate constant in our model 0.205 per year, our model predicted progression rate for their patient population was 4.766 units per year that was very close to their reported value. It could be implied that during relatively short observation period (30 m the progression of MOT score was approximately linear, and logistic model was flexible enough to also catch this approximate linear part. For TFC, more publications with progression rate reported were found and same as for MOT, all these studies estimated TFC progression rate

PAGE 79

79 based on linear regression. The comparison of their reported value and our model prediction as well as some information was listed in Table 3 8. From numerical comp arison, the second literature based on shorter duration study had the closest progression rate to our model prediction. The possible reason was the disease progression described by decline in TFC was approximately linear within short term, but during longe r observation time, the progression showed more nonlinear trend, so the progression rate might be underestimated or at least evaluated inadequately if using linear model. The clinical trial CARE was the first controlled trial in HD to test the benefit of potential drug treatment on HD progression that was quantified by functional decline, improvement of motor features, et al (2001) Although the current findings failed to pr ovide sufficient power for slower disease progression by drug treatment, the design and conduction of controlled trial offered valuable data source to explore HD disease progression and contribute information for future trial design. Because of insignifica nt difference between active treatment groups and placebo, data in drug treatment groups was used as external validation to our structural model. In the visual predictive check for MOT score in active treatment groups shown in Figure 3 11, the prediction w as quite well for median MOT score time profile and reasonable at 5% and 95% quantiles. The difference of prediction from observation at 5% quantile was probably due to larger variability when the score was very low, and another reason was the size of plac ebo data was mush smaller than that of pooled dataset from the three active treatment groups (Table 3 9) so model based on only placebo data might not describe variability of data in drug treatment groups adequately. The estimates of model parameter were a lso compared in Table 3 10 after separately fitting placebo data, drug treatment data and all the data together for MOT and TFC respectively. The typical value of baseline score and rate constant as well as between subject

PAGE 80

80 variability on rate constant were very close among different groups for both MOT and TFC while the between subject variability on baseline was larger in active treatment groups due to larger data size. In summary, based on both diagnostic plots and model validation, our structural model was demonstrated to adequately describe observed data in controlled trial and qualified for clinical trial simulation. Model fitting on observational data Compared with clinical trial, observational studies possessed much more data from more patients at b oth early and late disease stages, and also followed up patients for longer time. And more importantly, placebo data in clinical trials may carry over placebo response in additional to disease progression, so analysis of observational data was an important supplement to characterize the natural disease progression in untreated patient population. Therefore, our logistic model was also used to fit observational data. As shown in Table 3 4, COHORT and REGISTRY possessed larger sample size, patients at both e arly and later disease stages accordingly larger range of observed rating scores as well as similar design mainly on enrollment criteria, and the main difference between them was that COHORT was conducted in the US, Canada and Australia while REGISTRY was conducted in Europe (Dorsey, 2012; Orth, et al., 2011) They are currently being merged into a single study in some research groups. Therefore our data from these two studies was also merged together to do analy sis. And only HD patients with at least 3 visits and without missing values for age were included into our current analysis because it was hard for model to fit data based on many patients with only one or two observations and would produce biased estimate s. From goodness of fitting plots in Figure 3 12, our logistic model still provided reasonable prediction to observational data although there was obviously larger variability. And it also could

PAGE 81

81 be observed from more flat prediction lines that the increas ing of motor score in observational studies was slower than that in clinical trial, and this observation was confirmed by parameter estimates listed in Table 3 11. The typical value of baseline score and associated between subject variability was very clos e between clinical trial and observational studies, but progression rate constant of MOT in observational studies was much smaller with much larger between subject variability. It was very likely due to much less restrictive control on design of observatio nal studies where the patients might take drugs or have different levels of care. Large number of missing data due to lost of follow up was another common reason for this difference. So because of the large data source and long observation time, it deserve d to do further data cleaning to detect, correct or remove corrupt or inaccurate records from dataset based on study design and case report forms in order to characterize the disease progression more adequately. Covariate Model Development Exploration of potential covariates based on observational studies was started with visualization of the relationship between inter subject variability patient age and CAG length (Figure 3 13 and Figure 3 14). Positive relationship was observed baseline rate constant with CAG length rate con stant with patient age. After incorporation of these two covariates into our structural model, the same scatter plots were made again using parameter estimates in covariate model and compared with plots in structural model in Figure 3 15, Figure 3 16 and F igure 3 17. From graphic comparison, the trend observed in structural model disappeared in covariate model. Meanwhile, the model fitting was improved demonstrated by reduced objective function value, and between subject variability associated with baseline score and rate constant were smaller because these two covariates explained part of the variability (Table 3 12). The estimate for all the parameters in covariate model was listed in Table 3 13.

PAGE 82

82 Thus, CAG length and patient age were implied to play impor tant role on baseline score and progression rate constant. Summary All the above results suggested that: The application of logistic model to HD disease progression is appropriate based on both exploratory analysis and model validation. For two of importa nt efficacy variables, total motor score and total functional capacity, disease progress relatively fast at round half maximal score. CAG length and patient age have marked effect on disease progression from the preliminary covariate analysis. The result s discussed here are only based on preliminary analysis and more work is needed to confirm these findings and also conduct further exploration toward our long term research goal. As this project continues, our database will be updated with more available d ata, and more subscales of UHDRS will be modeled, such as functional checklist score, behavior score, verbal fluency test, etc. With larger database, structure of covariate model is expected to be refined and more covariates, including disease burden, dise ase stages as well as many gene testing variables other than CAG length will be searched comprehensively.

PAGE 83

83 Table 3 1. List of candidate model structures for total motor score Model Progression Rate Inflection point No. of parameters Linear dE/dt=r NA 5 Logistic 1 dE/dt=rE(1 E/124) 124/2 5 Logistic 2 dE/dt=rE(1 E/124) 6 Logistic 3 dE/dt=rE[1 (E/124) ] [124 6 Logistic 4 dE/dt=rE (1 E/124) 6 Table 3 2. List of candidate model structures for total functiona l capacity Model Progression Rate Inflection point No. of parameters Linear dE/dt=r NA 5 Logistic 1 dE/dt=rE(1 E/13) 13/2 5 Logistic 2 dE/dt=rE(1 E/13) 6 Logistic 3 dE/dt=rE[1 (E/13) ] [13 6 Logistic 4 dE/dt=rE (1 E/13) 6

PAGE 84

84 Table 3 3. List of included studies in Hun tington's disease database Study Type No. of Subj No. of obser per subj No. of Observation Population Duration Interval CARE Controlled 347 10 3313 HD 30 months 4 to 5 months COHORT Observati onal 3199 <=5 7336 HD+ At_Risk +Control up to 55 months Yearly PREDICT Observational 1055 <=7 3934 HD+ At_Risk +Control up to 75 months Yearly REGISTRY Observational 6195 1~30 (1,2,3) 15502 HD+ At_Risk +Control up to 10 years Yearly TRACK Observatio nal 366 <=3 1045 HD+Control 24 months Yearly Table 3 4. Demographic and Baseline Characteristics in included studies Characteristic CARE COHORT PREDICT REGISTRY TRACK % of HD patients 100.0% 56.7% 69.8% 92.5% 66.4% Gender, %F 49.3% 55.3% 62.8% 5 2.5% 54.7% Age, Mean (range) 47.4 (18 75) 49.3 (12.5 89.5) 41.4 (18.1 75.9) 48.9 (6.8 96.1) 44.9 (18.6 64.1) CAG, Mean (range) 45.0 (37 69) 43.7 (36 100) 42.4 (38 61) 4 4.3 (36 90) 43.4 (39 59) Baseline score MOT, Mean (range) 31.1 (3 71) 30.7 (0 100) 5.5 (0 44) 33.4 (0 124) 13.24 (0 52) TFC, Mean (range) 10.2 (7 13) 9.2 (0 13) 12.8 (7 13) 8.2 (0 13) 11.8 (7 13)

PAGE 85

85 Table 3 5. Results of structural model search for total motor score Model Progression Rate Inflection po int Inflection point estimate No. of parameters Objective function value Linear dE/dt=r NA NA 5 4091.12 Logistic 1 dE/dt=rE(1 E/124) 124/2 62 5 4080.49 Logistic 2 dE/dt=rE(1 E/124) 60.19 6 4080.46 Logistic 3 dE/dt=rE[1 (E/124) ] [124 64.06 6 4080.40 Logistic 4 dE/dt=rE (1 E/124) 64.67 6 4080.31 Table 3 6. Results of structural model search for total functional capacity Model Progression Rate Inflection point Inflection point estimate No. of parameters Objective function value Linear dE/dt=r NA NA 5 1165.28 Logistic 1 dE/dt=rE(1 E/13) 13/2 6.5 5 1146.58 Logistic 2 dE/dt=rE(1 E/13) 6.57 6 1146.54 Logistic 3 dE/dt=rE[1 ( E/13) ] [13 6.44 6 1146.57 Logistic 4 dE/dt=rE (1 E/13) 6.35 6 1146.52 Table 3 7. List of population estimates and estimate precision of model parameters Outcome Parameter Population estimate (RSE%) Between subject variability (RSE%) MOT Rate constant (1/year) 0.205 (8.1%) 63.22% (27.5%) Baseline (units) 28.1 (4.4%) 39.75% (14.0%) TFC Rate constant (1/year) 0.401 (8.9%) 62.05% (26.8%) Baseline (units) 10.1 (1.7%) 14.83% (13.1%)

PAGE 86

86 Table 3 8. List of litera ture with reported progression rate for TFC No. of literature No. of subjects Duration (year) Mean TFC baseline Mean TFC Progression rate (units/year) Model predicted rate (units/year) Reference 1 379 3.0 11 10.34 0.44 0.85 (Meyer, et al., 2012) 2 575 1.5 10 9.27 0.97 1.07 (Marder, et al., 2000) 3 129 3.6 10 8.87 0.63 1. 13 (Mayeux, et al., 1986) 4 593 3.7 10 7.41 1.40 1.28 (Penney, Jr. et al., 1990)

PAGE 87

87 Table 3 9. Summary of all the groups in clinical trial CARE Treatment groups No. of subject No. of observation Total Placebo 87 837 837 Treatment 1 87 832 2474 Treatment 2 86 811 Treatment 3 87 831

PAGE 88

88 Table 3 10. Summary of mo del parameters for all the groups in clinical trial Efficacy outcomes Treatment groups Population estimate (RSE%) Between subject variability (RSE%) Rate constant (1/year) Baseline (units) Rate constant Baseline MOT Placebo 0.205 (8.1%) 28.1 (4 .4%) 63.22% (27.6%) 39.75% (14.0%) Treatment 0.212 (4.8%) 27.0 (3.3%) 62.75% (14.6%) 51.58% (10.0%) ALL 0.209 (4.1%) 27.3 (2.7%) 63.11% (13.2%) 48.79% (8.6%) TFC Placebo 0.401 (8.9%) 10.1 (1.7%) 62.05% (26.8%) 14.83% (13.1%) Treatment 0.376 (5 .6%) 10.0 (1.1%) 65.27% (14.0%) 17.20% (7.2%) ALL 0.382 (4.7%) 10.0 (1.0%) 64.50% (12.4%) 16.64% (6.4%) Table 3 11. Summary of model parameters for MOT in clinical trial and observational studies Study Population estimate (RSE%) Between subjec t variability (RSE%) Rate constant (1/year) Baseline (units) Rate constant Baseline CARE 0.209 (4.1%) 27.3 (2.7%) 63.11% (13.2%) 48.79% (8.6%) COHORT+REGISTRY 0.0815 (7.6%) 28.0 (1.5%) 228.23% (10.0%) 55.59% (4.1%)

PAGE 89

89 Table 3 12. Comparison of st ructural model and covariate model Parameters Structural model Covariate model OFV* 83930.1 83149.1 CV% (rate constant) 228.22% 171.02% CV% (baseline) 55.59% 47.96% R epresents objective function value Table 3 13. Summary of mod el parameters in covariate model for observational data Population estimate (RSE%) Between subject variability (RSE%) Scaling parameter (RSE%) Rate constant (1/year) Baseline (units) Rate constant Baseline CAG on baseline Age on baseline CAG on rate constant 0.0918 32 171.02% 47.96% 3.19 1.2 2.8 6.50% 1.30% 11.30% 4.70% 4.50% 4.80% 9.20%

PAGE 90

90 Figure 3 1. Individual time profile of observed rating scores for a representative set of subjects. A ) MOT. B) TFC. A B

PAGE 91

91 Figure 3 2. Exploratory analysis on relationship of calculated slope of MOT increase with observed score based on linear regression in placebo data of the clinical trial. A) Plot of mean slope value in each sub group. B) Plot of mean slope value and its confidence interval in each subgroup B A

PAGE 92

92 Figure 3 3. Exploratory analysis on relationship of calculated slope of TFC decline with observed score based on linear regression in placebo da ta of the clinical trial. A) Plot of mean slope value in each subgroup. B) Plot of mean slope value and its confidence interval in each subgroup A B

PAGE 93

93 Figure 3 4. Exploratory analysis on relationship of calculated slope of MOT incr ease with observed score based on linear regression in observational studies. A) Plot of mean slope value in each subgroup. B) Plot of mean slope value and its confidence interval in each subgroup A B

PAGE 94

94 Figure 3 5. Exploratory an alysis on relationship of calculated slope of TFC decline with observed score based on linear regression in observational studies. A) Plot of mean slope value in each subgroup. B) Plot of mean slope value and its confidence interval in each subgroup B A

PAGE 95

95 Figure 3 6. Schematic representation of logistic model. A) Relationship between progression rate and observed score under logistic model. B) Time profile of observed score under logistic model. A B

PAGE 96

96 Fig ure 3 7. Comparison of individual observation, individual prediction and population prediction on time profile of total motor score for a representative set of subjects in placebo group of clinical trial.

PAGE 97

97 Figure 3 8. Comparis on of individual observation, individual prediction and population prediction on time profile of total functional capacity for a representative set of subjects in placebo group of clinical trial.

PAGE 98

98 Figure 3 9. Comparison of me an observed progression rate, mean predicted progression rate and simulated population progression rate of MOT for subjects in placebo group of clinical trial.

PAGE 99

99 Figure 3 10. Comparison of mean observed progression rate, mean predicted progression rate and simulated population progression rate of TFC for subjects in placebo group of clinical trial.

PAGE 100

100 Figure 3 11. Visual predictive check for MOT in active treatment groups of clinical trial

PAGE 101

101 Figure 3 12. Comparison of individual observation, individual prediction and population prediction on time profile of total motor score for a representative set of subjects in observational studies. A) COHORT. B) REGISTRY. A B

PAGE 102

102 Figure 3 13. Scatter plots of between subject variability of parameters over patient age. A) Baseline score. B) Progression rate constant. A B

PAGE 103

103 Figure 3 14. Scatter plots of between subject variability o f parameters over CAG length. A) Baseline score. B) Progression rate constant. A B

PAGE 104

104 Figure 3 15. Comparison of scatter plots of between subject variability on baseline score over patient age between structural model and covariat e model. A) Structural model. B) Covariate model. A B

PAGE 105

10 5 Figure 3 16. Comparison of scatter plots of between subject variability on baseline score over CAG length between structural model and covariate model. A) Structural model B) Covariate model. A B

PAGE 106

106 Figure 3 17. Comparison of scatter plots of between subject variability on rate constant over CAG length between structural model and covariate model. A) Structural model. B) Covariate model. A B

PAGE 107

107 CHAPTE R 4 CONCLUSIONS The overall objective of my dissertation project is to facilitate and optimize the clinical trial design for neuropsychiatric diseases based on current challenges and problems existing in the clinical trials by development of disease models Neuropsychiatric diseases possess common characteristics such as multiple sectional symptoms, subjective evaluation methods, misdiagnos is or delayed diagnosis etc, and also have different pathological causes and disease properties. Accordingly considerab le problems were produced in clinical trials. Bipolar Disorder (BD) and A large number of clinical trials for BD have been conducted and there is already regular trial design for different study objectives, and at same time common challenges and issues were recognized, including high placebo response and high dropout rate. Our developed placebo effect model and dropout model reasonably characterized the level of placebo effect and dropout pattern, which could be shared with different trials for different drugs. Besides, significant design factors were identified as well to help reduce missing data. All these information obtained from disease models could be useful to optimize th e current trial design. For HD, there is still lack of commonly accepted clinical trial design, and no disease modifying treatment available although some promising therapeutic targets are identified and related drug development is ongoing. As a result, t he aim of our disease model for HD was to describe the natural disease progression based on some primary efficacy variables to provide information to trial design, including reliable clinical endpoint, appropriate trial duration and target patient populati on. Our preliminary results showed a logistic model properly characterized HD disease progression and suggested that enrichment strategy can be applied to increase the success chance of a trial targeting disease modifying regimens.

PAGE 108

108 Disease models have been developed and utilized in multiple therapeutic areas, and my dissertation project demonstrates the benefit of disease models to clinical trial design and provides modeling framework for neuropsychiatric diseases.

PAGE 109

109 LIST OF REFERENCES (1993) A novel gene containing a trinucleotide repeat that is expanded and unstable on Huntington's disease chromosomes. The Huntington's Disease Collaborative Research Group. Cell 72 :971 983. (1996) Unified Huntington's Disease Rating Scale: reliability and consistency. Huntington Study Group. Mov Disord 11 :136 142. (2001) A randomized, placebo controlled trial of coenzyme Q10 and remacemide in Huntington's disease. Neurology 57 :397 404. Ashford JW and Schmitt FA (2001) Modeling the time cour se of Alzheimer dementia. Curr Psychiatry Rep 3 :20 28. Barnett JH and Smoller JW (2009) The genetics of bipolar disorder. Neuroscience 164 :331 343. Bhattaram VA, Booth BP, Ramchandani RP, Beasley BN, Wang Y, Tandon V, Duan JZ, Baweja RK, Marroum PJ, Uppoor RS, Rahman NA, Sahajwalla CG, Powell JR, Mehta MU and Gobburu JV (2005) Impact of pharmacometrics on drug approval and labeling decisions: a survey of 42 new drug applications. AAPS J 7 :E503 E512. Bhattaram VA, Siddiqui O, Kapcala LP and Gobburu JV (2009) Endpoints and analyses to discern disease modifying drug effects in early Parkinson's disease. AAPS J 11 :456 464. Cassidy F (2011) Risk factors of attempted suicide in bipolar disorder. Suicide Life Threat Behav 41 :6 11. Diaz Hernandez M, Torres Peraza J, Salvatori Abarca A, Moran MA, Gomez Ramos P, Alberch J and Lucas JJ (2005) Full motor recovery despite striatal neuron loss and formation of irreversible amyloid like inclusions in a conditional mouse model of Huntington's disease. J Neurosci 25 :9773 9781 DiFiglia M, Sapp E, Chase K, Schwarz C, Meloni A, Young C, Martin E, Vonsattel JP, Carraway R, Reeves SA and (1995) Huntingtin is a cytoplasmic protein associated with vesicles in human and rat brain neurons. Neuron 14 :1075 1081. Dorsey ER (2012) Chara cterization of a large group of individuals with huntington disease and their relatives enrolled in the COHORT study. PLoS One 7 :e29522. Falush D, Almqvist EW, Brinkmann RR, Iwasa Y and Hayden MR (2001) Measurement of mutational flow implies both a high ne w mutation rate for Huntington disease and substantial underascertainment of late onset cases. Am J Hum Genet 68 :373 385. Fava M, Evins AE, Dorer DJ and Schoenfeld DA (2003) The problem of the placebo response in clinical trials for psychiatric disorders: culprits, possible remedies, and a novel study design approach. Psychother Psychosom 72 :115 127.

PAGE 110

110 Fiedorowicz JG, Solomon DA, Endicott J, Leon AC, Li C, Rice JP and Coryell WH (2009) Manic/hypomanic symptom burden and cardiovascular mortality in bipolar dis order. Psychosom Med 71 :598 606. Friberg LE, de GR, Kerbusch T and Karlsson MO (2009) Modeling and simulation of the time course of asenapine exposure response and dropout patterns in acute schizophrenia. Clin Pharmacol Ther 86 :84 91. Ghaemi SN, Sachs GS, Chiou AM, Pandurangi AK and Goodwin K (1999) Is bipolar disorder still underdiagnosed? Are antidepressants overutilized? J Affect Disord 52 :135 144. Gitlin MJ, Swendsen J, Heller TL and Hammen C (1995) Relapse and impairment in bipolar disorder. Am J Psych iatry 152 :1635 1640. Goldberg YP, Kremer B, Andrew SE, Theilmann J, Graham RK, Squitieri F, Telenius H, Adam S, Sajoo A, Starr E and (1993) Molecular analysis of new mutations for Huntington's disease: intermediate alleles and sex of origin effects. Nat Genet 5 :174 179. Goldstein BI, Shamseddeen W, Axelson DA, Kalas C, Monk K, Brent DA, Kupfer DJ and Birmaher B (2010) Clinical, demographic, and familial correlates of bipolar spectrum disorders among offspring of parents with bipolar disorder. J Am Acad Ch ild Adolesc Psychiatry 49 :388 396. Gomeni R and Merlo Pich E (2007) Bayesian modelling and ROC analysis to predict placebo responders using clinical score measured in the initial weeks of treatment in depression trials. Br J Clin Pharmacol 63 :595 613. Gorh am DR and Overall JE (1960) Drug action profiles based on an abbreviated psychiatric rating scale. J Nerv Ment Dis 131 :528 535. Grunze H (2011) The clinical side of bipolar disorders. Pharmacopsychiatry 44 Suppl 1 :S43 S48. Hamilton M (1960) A rating scale for depression. J Neurol Neurosurg Psychiatry 23 :56 62. Hersch SM and Rosas HD (2008) Neuroprotection for Huntington's disease: ready, set, slow. Neurotherapeutics 5 :226 236. Holford NH and Peace KE (1992) Results and validation of a population pharmacodyn amic model for cognitive effects in Alzheimer patients treated with tacrine. Proc Natl Acad Sci U S A 89 :11471 11475. Kaltenbach LS, Romero E, Becklin RR, Chettier R, Bell R, Phansalkar A, Strand A, Torcassi C, Savage J, Hurlburt A, Cha GH, Ukani L, Chepan oske CL, Zhen Y, Sahasrabudhe S, Olson J, Kurschner C, Ellerby LM, Peltier JM, Botas J and Hughes RE (2007) Huntingtin interacting proteins are genetic modifiers of neurodegeneration. PLoS Genet 3 :e82.

PAGE 111

111 Kane JM and Leucht S (2008) Unanswered questions in sc hizophrenia clinical trials. Schizophr Bull 34 :302 309. Kazantsev AG (2007) Cellular pathways leading to neuronal dysfunction and degeneration. Drug News Perspect 20 :501 509. Kemp AS, Schooler NR, Kalali AH, Alphs L, Anand R, Awad G, Davidson M, Dube S, Er eshefsky L, Gharabawi G, Leon AC, Lepine JP, Potkin SG and Vermeulen A (2010) What is causing the reduced drug placebo difference in recent schizophrenia clinical trials and what can be done about it? Schizophr Bull 36 :504 509. Kimko HC, Reele SS, Holford NH and Peck CC (2000) Prediction of the outcome of a phase 3 clinical trial of an antischizophrenic agent (quetiapine fumarate) by simulation with a population pharmacokinetic and pharmacodynamic model. Clin Pharmacol Ther 68 :568 577. Kremer B, Clark CM, A lmqvist EW, Raymond LA, Graf P, Jacova C, Mezei M, Hardy MA, Snow B, Martin W and Hayden MR (1999) Influence of lamotrigine on progression of early Huntington disease: a randomized clinical trial. Neurology 53 :1000 1011. Krobitsch S and Kazantsev AG (2011) Huntington's disease: From molecular basis to therapeutic advances. Int J Biochem Cell Biol 43 :20 24. Langbehn DR, Brinkman RR, Falush D, Paulsen JS and Hayden MR (2004) A new model for prediction of the age of onset and penetrance for Huntington's diseas e based on CAG length. Clin Genet 65 :267 277. Leucht S, Kane JM, Kissling W, Hamann J, Etschel E and Engel RR (2005) What does the PANSS mean? Schizophr Res 79 :231 238. Lipkovich IA, Houston JP and Ahl J (2008) Identifying patterns in treatment response pr ofiles in acute bipolar mania: a cluster analysis approach. BMC Psychiatry 8 :65. Luykx JJ, Boks MP, Terwindt AP, Bakker S, Kahn RS and Ophoff RA (2010) The involvement of GSK3beta in bipolar disorder: integrating evidence from multiple types of genetic stu dies. Eur Neuropsychopharmacol 20 :357 368. Maat Kievit A, Losekoot M, Zwinderman K, Vegter van der Vlis M, Belfroid R, Lopez F, Van Ommen GJ, Breuning M and Roos R (2002) Predictability of age at onset in Huntington disease in the Dutch population. Medicin e (Baltimore) 81 :251 259. Mager DE, Wyska E and Jusko WJ (2003) Diversity of mechanism based pharmacodynamic models. Drug Metab Dispos 31 :510 518. Marder K, Zhao H, Myers RH, Cudkowicz M, Kayson E, Kieburtz K, Orme C, Paulsen J, Penney JB, Jr., Siemers E a nd Shoulson I (2000) Rate of functional decline in Huntington's disease. Huntington Study Group. Neurology 54 :452 458.

PAGE 112

112 Mayeux R, Stern Y, Herman A, Greenbaum L and Fahn S (1986) Correlates of early disability in Huntington's disease. Ann Neurol 20 :727 731. Merikangas KR, Akiskal HS, Angst J, Greenberg PE, Hirschfeld RM, Petukhova M and Kessler RC (2007) Lifetime and 12 month prevalence of bipolar spectrum disorder in the National Comorbidity Survey replication. Arch Gen Psychiatry 64 :543 552. Meyer C, Landwehrmeyer B, Schwenke C, Doble A, Orth M and Ludolph AC (2012) Rate of change in early Huntington's disease: a clinicometric analysis. Mov Disord 27 :118 124. Montgomery SA and Asberg M (1979) A new depression scale designed to be sensitive to chang e. Br J Psychiatry 134 :382 389. Morselli PL, Elgie R and Cesana BM (2004) GAMIAN Europe/BEAM survey II: cross national analysis of unemployment, family history, treatment satisfaction and impact of the bipolar disorder on life style. Bipolar Disord 6 :487 4 97. Novick DM, Swartz HA and Frank E (2010) Suicide attempts in bipolar I and bipolar II disorder: a review and meta analysis of the evidence. Bipolar Disord 12 :1 9. Orth M, Handley OJ, Schwenke C, Dunnett S, Wild EJ, Tabrizi SJ and Landwehrmeyer GB (2011) Observing Huntington's disease: the European Huntington's Disease Network's REGISTRY. J Neurol Neurosurg Psychiatry 82 :1409 1412. Penney JB, Jr., Vonsattel JP, MacDonald ME, Gusella JF and Myers RH (1997) CAG repeat number governs the development rate of pathology in Huntington's disease. Ann Neurol 41 :689 692. Penney JB, Jr., Young AB, Shoulson I, Starosta Rubenstein S, Snodgrass SR, Sanchez Ramos J, Ramos Arroyo M, Gomez F, Penchaszadeh G, Alvir J and (1990) Huntington's disease in Venezuela: 7 years o f follow up on symptomatic and asymptomatic individuals. Mov Disord 5 :93 99. Peyser CE, Folstein M, Chase GA, Starkstein S, Brandt J, Cockrell JR, Bylsma F, Coyle JT, McHugh PR and Folstein SE (1995) Trial of d alpha tocopherol in Huntington's disease. Am J Psychiatry 152 :1771 1775. Pilla R, V, Kozielska M, Johnson M, Vermeulen A, de GR, Liu J, Groothuis GM, Danhof M and Proost JH (2011) Structural models describing placebo treatment effects in schizophrenia and other neuropsychiatric disorders. Clin Pharma cokinet 50 :429 450. Ploeger BA and Holford NH (2009) Washout and delayed start designs for identifying disease modifying effects in slowly progressive diseases using disease progression analysis. Pharm Stat 8 :225 238. Poon LH, Kang GA and Lee AJ (2010) Rol e of tetrabenazine for Huntington's disease associated chorea. Ann Pharmacother 44 :1080 1089.

PAGE 113

113 Post TM, Cremers SC, Kerbusch T and Danhof M (2010) Bone physiology, disease and treatment: towards disease system analysis in osteoporosis. Clin Pharmacokinet 49 :89 118. Price AL and Marzani Nissen GR (2012) Bipolar disorders: a review. Am Fam Physician 85 :483 493. Ranen NG, Peyser CE, Coyle JT, Bylsma FW, Sherr M, Day L, Folstein MF, Brandt J, Ross CA and Folstein SE (1996) A controlled trial of idebenone in Hunt ington's disease. Mov Disord 11 :549 554. Rosen WG, Mohs RC and Davis KL (1984) A new rating scale for Alzheimer's disease. Am J Psychiatry 141 :1356 1364. Samtani MN, Farnum M, Lobanov V, Yang E, Raghavan N, DiBernardo A and Narayan V (2012) An improved mod el for disease progression in patients from the Alzheimer's disease neuroimaging initiative. J Clin Pharmacol 52 :629 644. Scherk H, Pajonk FG and Leucht S (2007) Second generation antipsychotic agents in the treatment of acute mania: a systematic review an d meta analysis of randomized controlled trials. Arch Gen Psychiatry 64 :442 455. Shang EY, Gibbs MA, Landen JW, Krams M, Russell T, Denman NG and Mould DR (2009) Evaluation of structural models to describe the effect of placebo upon the time course of majo r depressive disorder. J Pharmacokinet Pharmacodyn 36 :63 80. Shi L, Thiebaud P and McCombs JS (2004) The impact of unrecognized bipolar disorders for patients treated for depression with antidepressants in the fee for services California Medicaid (Medi Cal ) program. J Affect Disord 82 :373 383. Shoulson I, Odoroff C, Oakes D, Behr J, Goldblatt D, Caine E, Kennedy J, Miller C, Bamford K, Rubin A and (1989) A controlled clinical trial of baclofen as protective therapy in early Huntington's disease. Ann Neuro l 25 :252 259. Swann AC (2006) What is bipolar disorder? Am J Psychiatry 163 :177 179. Tsoularis A and Wallace J (2002) Analysis of logistic growth models. Math Biosci 179 :21 55. Van Lieshout RJ and MacQueen GM (2010) Efficacy and acceptability of mood stabi lisers in the treatment of acute bipolar depression: systematic review. Br J Psychiatry 196 :266 273. Wang Y, Sung C, Dartois C, Ramchandani R, Booth BP, Rock E and Gobburu J (2009) Elucidation of relationship between tumor size and survival in non small ce ll lung cancer patients can aid early decision making in clinical drug development. Clin Pharmacol Ther 86 :167 174. Yamamoto A, Lucas JJ and Hen R (2000) Reversal of neuropathology and motor dysfunction in a conditional model of Huntington's disease. Cell 101 :57 66.

PAGE 114

114 Yano Y, Oguma T, Nagata H and Sasaki S (1998) Application of logistic growth model to pharmacodynamic analysis of in vitro bactericidal kinetics. J Pharm Sci 87 :1177 1183. Young AB, Shoulson I, Penney JB, Starosta Rubinstein S, Gomez F, Travers H, Ramos Arroyo MA, Snodgrass SR, Bonilla E, Moreno H and (1986) Huntington's disease in Venezuela: neurologic features and functional decline. Neurology 36 :244 249. Young RC, Biggs JT, Ziegler VE and Meyer DA (1978) A rating scale for mania: reliability validity and sensitivity. Br J Psychiatry 133 :429 435.

PAGE 115

115 BIOGRAPHICAL SKETCH Wan Sun was born in Beijing, P. R. China. Wan rec p harmaceutical s cience from Peking University in July 2005, and she entered the graduate school p harmacology from Peking University in July 2007. The same year, she joined the graduate program in Department of Pharmaceutics at The Ohio State University, and then transferr ed to University of Florida as a Ph. D student in Department of Pharmaceutics under the supervision of Dr. Guenther Hochhaus in August, 2008. Wan has finished two internships during her Ph. D study in the Department of Pharmacokinetics, Dynamics and Metabo lism at Pfizer La Jolla, CA and Division of Pharmacometrics, Office of Clinical Pharmacology, at the US Food and Drug Administration, Silver Spring, MD. After she passed the qualify exam to become a Ph. D candidate in October 2011, she worked as a research fellow in Division of Pharmacometrics, Office of Clinical Pharmacology at FDA with Dr. Yaning Wang to continue her Ph. D dissertation work. Wan received her Doctor of Philosophy degree in p harmaceutics in December 2012.