In the three preceding communications of this series,1-3we have suggested methods for solving some of the many problems of clinical epidemiology that occur in the evaluation of prognosis and treatment for patients with cancer. We indicated that "retrospective" surveys of medical records were an intellectual necessity in solving these problems, because the diverse therapeutic agents used in cancer cannot always be compared "prospectively," and because the "prospective" studies, even when feasible, cannot be properly designed until better methods have been developed for reviewing and analyzing the complex events that constitute the initial and subsequent states of the treated patients. For performing these analytic reviews, the complexity of each patient's clinical course can be logically arranged according to the data and decisions that occur at specific points of temporal demarcation.1When the clinical courses of patients are to be appraised in a survey of therapy, a precise strategy