Cross-study Hierarchical Modeling of Stratiied Clinical Trial Data Cross-study Hierarchical Modeling of Stratiied Clinical Trial Data

Hierarchical random e ects models are steadily increasing in popularity as statistical tools in clinical trial research They can be used to estimate treatment or other covariate e ects in single study analyses coordinated over multiple clinical units and can also be extended to a wide variety of cross study applications After reviewing the single study case we use data from ve di erent protocols of the Terry Beirn Community Programs for Clinical Research on AIDS CPCRA to look for units that tend to have treatment e ects consistently above or below the study speci c grand mean across several studies As a rst step we summarize the patient level data as study and unit speci c estimated treatment e ects and standard errors using independent Cox regression models We then compare the results of a hierarchical model using these data summaries as input to those produced by a more fully Bayesian method which uses the actual patient level survival data Both models are shown to be easily t using the BUGS software suggesting the possibility of their routine implementation We also compare various di erent models using a Deviance Information Criterion DIC a recent extension of the Akaike Information Criterion AIC designed for hierarchical models Our procedure appears e ective at answering the question of whether certain CPCRA units are better than others at identifying treatment e ects where they exist