The methodology used by the Department of Veterans Affairs for data collection and analysis to derive observed/expected mortality ratios in cardiac surgical patients is reviewed. The Department of Veterans Affairs' use of univariate and multivariate analysis to develop risk ratios for individual risk factors is described. Its experience with tracking observed/expected mortality and morbidity associated with cardiac surgery and length of hospital stays is reviewed. Results of the Department of Veterans Affairs study of the relationship between hospital surgical volume and observed/expected ratios are reported. Feasible goals for the improvement of the predictive capability of database models and the limitations affecting model accuracy are discussed.