Objective: To test prospectively the unsubstantiated claim that patient-specific predictions of time-related outcome after coronary artery bypass grafting (CABG) from multivariable parametric equations are reliable for medical decision making and for intra- and interdepartmental quality control in surgical training and practice.
Methods: 3720 survival curves were generated prospectively for all primary, isolated CABG patients operated upon at the Katholieke Universiteit (KU) Leuven between July, 1987 and January, 1992 using the published AHA/ACC guidelines multivariable equation derived from prior KU Leuven experience. The average of these curves (risk-adjusted predicted survival) was compared to the Kaplan-Meier (actual) estimates, overall and for patient subsets. Variables associated with systematic deviation of actual from predicted number of deaths were sought by multivariable residual risk analysis.
Results: Actual overall survival was less good than predicted (P = 0.03) and the excess risk was distributed uniformly across time. The excess risk was not attributable to substantial changes in prevalence of known risk factors. It was attributable largely to a small subset of patients (n = 292) with low-prevalence, but important risk factors not accounted for by the equation (P = 0.7, for difference in survival among the remaining 3428 patients).
Conclusions: Within the confines of a single institution, patient-specific predictions of outcome after CABG can be made reliably in most patients using multivariable equations developed from a heterogeneous experience, despite changes in prevalence of risk factors. New subsets of high-risk patients, failure or inability to account for important rare risk factors or for institutional changes, may lead to systematic errors of prediction. Under these limitations it is an excellent tool for medical decision making and audit of surgical training and practice.