Purpose of review: To illustrate the limitations of predictive risk models in cardiac surgery, highlight the difficulty in interpreting risk-adjusted outcome analysis and discuss the challenges of making clinical decisions based on risk predictions, particularly in high-risk patients.
Recent findings: Predictive risk models developed after logistic regression or other complex statistical analysis are commonly perceived as rigorous means to determine risk-adjusted mortality in cardiac surgery. However, the discrimination provided by those predictive models is barely better than clinical judgment. Moreover, validation studies of those models show that their calibration is inconsistent, limiting their application for comparisons between different patient cohorts. Recent data also show that, without a reasonable overlap of case-mix distributions, apparently calibrated models used for risk-adjusted outcome analysis may lead to inaccurate side-by-side comparisons of provider performance. Finally, most predictive models overestimate risk, particularly in the high-risk patients.
Summary: Failure to account for many biological and procedural variables and for the constantly evolving practice of surgery and perioperative medicine likely contributes to the modest predictive performance of risk models in cardiac surgery. Consequently, those models should have limited input in the analysis of provider performance and in the decision to accept or deny surgery to the high-risk patients.