Daily self-reports (often called diaries) are an important data collection modality for understanding of STD/HIV related behaviors. The resulting data contains large amounts of information on subjects' sexual activities in their original time sequence. Lack of well-established bio-statistical methods for the analysis of behavioral diary data has hindered more extensive use of diaries in STD/HIV studies. The proposed research is designed to reduce methodological barriers to efficient analysis of diary data. Two classes of methods will be developed: 1) Re-sampling based estimation and inference procedures for time-to-infection analysis. The new method will expand upon the existing techniques in survival analysis by re-sampling for the unobserved infection time from the coital episode times recorded in the diary. Along this line, the proposed research develops a bootstrap procedure for the estimation of the survival function in the one sample situation, a testing procedure for the two-sample case, and a model-fitting algorithm for the Cox regression model setting. 2) A unified class of mixed effect autoregressive models for data following the exponential family of distributions. This class of models is designed to fill the current methodological gap in the modeling of various sexual activities (such as coitus, condom use in coitus, etc), and certain biological measurements (such as the viral load in the daily shedding of the Herpes simplex viruses). The new models combine the strengths of traditional autoregressive models in accounting for the effects of recent behavioral events, with the flexibility of random effect models in accommodating subject-specific effects. A likelihood based estimation procedure is proposed for model fitting. Software to implement the new procedures will be developed for both classes of methods. Asymptotic and finite sample performance of these procedures will also be evaluated. Finally, the new methods will be tested using the real data collected in four studies. It is hoped that this research will help future STD/HIV investigations and enhance our understanding of sexually transmitted infections by providing more detailed models for the behavioral and temporal antecedents of infection, as well as the necessary computational programs that facilitate the use of the new methods. ? ?
Showing the most recent 10 out of 18 publications
