Statistical Methods for Recurrent Event Data in RSV Immunoprophylaxis Studies



 DESCRIPTION (provided by applicant): Chronic diseases relapse and remit over time. With the widespread use of electronic medical records, we are able to capture multiple occurrences of the disease over time in the same patient. It is therefore logical to account for disease histor when assessing the treatment effect on disease exacerbation or recurrence. The current proposal will develop methods based on a non-homogeneous Poisson process (Ross, 1996, pp 78-81) and on sums of dependent binary indicators to achieve this goal. For events that follow the "cumulative hazard over time" mechanism, we will use a non-homogeneous Poisson process to explicitly model the impact of the current episode of the event and time-varying respiratory syncytial virus (RSV) immunoprophylaxis status on the chance of the next episode of the event. Neither the current recurrent event survival analysis nor the traditional Poisson regression would enable us to do so. For recurrent events that are really binary indicators observed over time, we reformat the recurrent events as a time series of dependent binary indicator variables and model their sum. We will develop a new class of generalized linear models using distributions we developed (Yu and Zelterman, 2002a and 2002b) for the dependent variable, sums of dependent binary indicators. These new generalized linear models compliment the generalized estimating equation (GEE) method by allowing us to explicitly model sequential dependence among the event indicators. All developed models will be compared with traditional recurrent event survival models or the GEE method through comprehensive simulation studies. These methods will be further applied to the parent study, the Prevention of RSV: Impact on Morbidity and Asthma study (5 R01 HS 018454), to better evaluate the effect of RSV immunoprophylaxis on RSV related morbidity and childhood asthma by age 6 years.