Recent progress in the understanding of HIV disease has suggested new research directions that pose complex challenges for design and analysis of HIV clinical trials. We propose to develop new statistical methods for both novel and recurring such problems. Although the pretest-posttest comparison is ubiquitous in HIV and other research, there is still no consensus on the relative merits of various methods of analysis, so that inefficient approaches are often used. The first part of this research focuses on developing a general theory for inference in this setting that leads to methods that improve over current approaches and provide the data-analyst with a clear strategy for choosing the most suitable method for a given application. Because missing responses, particularly at follow-up, are common in this context, the third part of the research involves extensions to take appropriate account of missing data. Increasingly, HIV clinical trials are focused on strategies where subjects may be treated through a series of stages; e.g., studies of structured treatment interruption involve similar considerations. However, ad hoc approaches to design and analysis are used due to absence of an appropriate methodology. The second part of the research is devoted to development of formal statistical framework for such complex studies that allows unambiguous statement of and valid inference on scientific questions of interest. Considerable interest in HIV research has focused on characterizing the relationship between longitudinal patterns of biomarkers, such as viral load or CD4, and measures of clinical progression, so much recent attention has focused on so-called joint models for longitudinal and event time data. In the final part of the research, we will carry out a detailed study of assumptions underlying this enterprise, clarify the practical implications of such assumptions, evaluate robustness to departures from assumptions, and develop a new approach for joint models that may improve over existing methods.
Showing the most recent 10 out of 53 publications
