External Validation and Updating of the Cardiac Surgery Score for Prediction of Mortality in a Cardiac Surgery Intensive Care Unit.

ABSTRACT

OBJECTIVE: To externally validate the predictive performance of the logistic and additive Cardiac Surgery Score (CASUS), a postoperative severity of illness score designed specifically for prediction of mortality in the cardiac surgery intensive care unit. DESIGN: A retrospective analysis of prospectively collected data between July 1, 2012, and September 30, 2015. SETTING: Single university cardiac surgery intensive care unit in Canada. PARTICIPANTS: Consecutive adult patients (n = 4,519) admitted to the intensive care unit after cardiac surgery. INTERVENTION None. MEASUREMENTS AND MAIN RESULTS: The mortality predicted by logistic CASUS was calculated for each patient on admission day 0 and postoperative days 2, 4, 7, and 10 using the original model equation. The mortality predicted by additive CASUS was determined on each day with separate logistic regression models, using the total score as a single variable. The observed mortality was 1.8%. Logistic CASUS overestimated mortality by 78%, 59%, 51%, 52%, and 29% on days 0, 2, 4, 7, and 10, respectively. After model updating with logistic calibration, logistic CASUS consistently provided estimates of death comparable with the observed mortality, as determined with the Hosmer-Lemeshow goodness-of-fit test. The stability of those estimates was confirmed by bootstrapping. Similar calibration results were obtained with additive CASUS. Logistic and additive CASUS had good discrimination with areas under the receiver operating characteristic curve greater than 0.7 on each study day. CONCLUSIONS: Recalibrated logistic CASUS reliably predicts mortality in the intensive care unit after cardiac surgery. Logistic regression models derived from additive CASUS perform as well as logistic CASUS.