STATISTICAL METHODOLOGY

THE TECHNICAL PROBLEM ADDRESSED BY THIS PROJECT IS THE DEVELOPMENT OF SAS-COMPATIBLE COMPUTER SOFTWARE IMPLEMENTINGADVANCED STATISTICAL METHODOLOGY. THE SPECIFIC OBJECTIVES ARE TO (1) DEVELOP SOFTWARE FOR REGRESSION DIAGNOSTICS, GOODNESS-OF-FIT ASSESSMENTS, AND MODEL VALIDATION (E.G., BOOTSTRAPPING) OF THE COX PROPORTIONAL HAZARDS AND LOGISTIC REGRESSION MODELS; (2) DEVELOP SOFTWARE TO ACCOMODATE TIME- DEPENDENT COVARIATES WITHIN THE COX MODEL; (3) EVALUATE THE BASIS FOR IMPLEMENTING COMPUTATIONAL PROCEDURES FOR RECURSIVE PARTITIONING AND DEVELOP THE PLANS FOR THIS IMPLEMENTATION; AND (4) DEVELOP THE STRUCTURE FOR SHORT COURSES TO FACILITATE SOFTWARE DISSEMINATION AND TO INSTRUCTINVESTIGATORS CONCERNING THE PROPER USE OF THESE STATISTICALAND COMPUTING TOOLS. TO ACHIEVE THESE GOALS, RESEARCHERS WILL CRITICALLY EVALUATE THE ANALYTIC TECHNIQUES CURRENTLY AVAILABLE FOR USE WITH THE LOGISTIC AND COX MODELS AND SELECT FOR IMPLEMENTATION METHODS THAT ARE STATISTICALLY EFFICIENT AND COMPUTATIONALLY PRACTICAL. THE SOFTWARE WILL BE DESIGNED TO EXTEND AND COMPLEMENT RATHER THAN DUPLICATE CONCURRENT DEVELOPMENTS BY SAS INSTITUTE. A CRITICAL APPRAISAL WILL BE COMPLETED, THE IMPORTANT TECHNICAL QUESTIONS WILL BE ADDRESSED, AND OVERALL PLANS FOR THE SOFTWARE ITSELF WILL BE DEFINED. RESULTS OF THIS PROJECT WILL ADD SIGNIFICANTLY TO THE BATTERY OF TOOLS AVAILABLE FORPROPERLY ANALYZING MEDICAL DATA.