SOFTWARE FOR LONGITUDINAL DATA ANALYSIS RELATED TO CANCER ETIOLOGY

LONGITUDINAL STUDIES PLAY AN INCREASINGLY PROMINENT ROLE IN BIOMEDICAL RESEARCH. EXAMPLES FROM THE CANCER FIELD INCLUDE ANALYSIS OF REPEATED TUMOR MARKER MEASUREMENTS, TUMOR GROWTH CURVE ANALYSIS FOR ANIMAL BIOASSAY, AND CHANGESIN BODY WEIGHT FOR PATIENTS ON CANCER CLINICAL TRIALS. IN THE PAST, ANALYSIS OF SUCH DATA WAS BASED ON STRONG, BUT UNVERIFIABLE, PARAMETRIC ASSUMPTIONS ABOUT THE DATA-GENERATING PROCESS. RECENTLY, THERE HAVE BEEN SEVERAL IMPORTANT METHODOLOGICAL ADVANCES THAT ENABLE LONGITUDINAL DATA TO BE ANALYZED BY NONPARAMETRIC AND SEMI-NONPARAMETRIC METHODS. MOREOVER, THESE METHODS REMAIN VALID IN THE PRESENCE OF MISSING DATA AND TIME-VARYING COVARIATES. THE MAIN GOALS OF THIS PHASE I PROPOSAL ARE (1) TO UNIFY THESE METHODOLOGIC ADVANCES BY PROVIDING A SINGLE, SEMI-NONPARAMETRIC (QUASI-LIKELIHOOD) APPROACH THAT CAN BE USED WITH EITHER CONTINUOUS OR CATEGORICAL REPEATED-MEASURESDATA, WITH POSSIBLY MISSING OBSERVATIONS, AND POSSIBLY TIME-VARYING COVARIATES; (2) TO DEMONSTRATE THE FEASIBILITY OF DEVELOPING STATISTICAL SOFTWARE, MEETING COMMERCIAL STANDARDS OF FRIENDLINESS AND RELIABILITY, TO IMPLEMENT THE NEW ANALYTICAL TECHNIQUES; AND (3) TO ANALYZE SOME REAL LONGITUDINAL DATA SETS WITH NEWLY DEVELOPED DEMONSTRATION SOFTWARE. THE EXPERIENCE GAINED IN THIS PHASE I ENDEAVOR WILL BE USED DURING PHASE II TO DEVELOP A COMPREHENSIVE, COMMERCIALLY VIABLE STATISTICAL PACKAGE FOR LONGITUDINAL DATA ANALYSIS ON MAINFRAME, MINI-, AND MICROCOMPUTERS.