Incorporating death into the statistical analysis of categorical longitudinal health status data

Quality of life measures collected longitudinally in the chronic disease setting may be used to evaluate clinical trial data and help patients and clinicians develop and evaluate a patient's prognosis. In the chronic disease setting a substantial portion of patients in trials may die, and this combined with missing data issues makes the analysis of longitudinal measures difficult. This dissertation developed joint probability models to evaluate trajectories of the probability of being healthy and alive (PAH) while avoiding the imputation of data after death. The model allows comparisons between two groups (D-PAH). A two-dimensional graph has been developed to describe both survival and the joint probability of being alive and healthy over time. Examples are given using a general health question from the Cardiovascular Health Study and the physical component score from the Short Form 36-Item Health Survey from the VA Ambulatory Care Quality Improvement Project. This work has the potential to provide important information to investigators on ways to analyze trial data and in turn to provide better information on changes in quality of life to clinicians and patients. The methods developed in this dissertation address the need for statistical methods that include quality of life information as the outcome and include patients until their time of death while allowing the deaths to affect the statistical analysis.