D. Asuzu1, G. Chao1, K. Y. Pei1 1Yale University School Of Medicine,Department Of General Surgery,New Haven, CT, USA
Introduction:
The American College of Surgeons (ACS) National Surgical Quality Improvement Program (NSQIP) universal surgical risk calculator estimates risk of postoperative cardiovascular complications with a good to excellent overall accuracy in the NSQIP dataset (c-statistic 0.895). However, the NSQIP calculator requires up to 21 variables for prediction including intraoperative details, and its retrospective use is limited by unavailability of individual regression coefficients. The Revised Cardiac Risk Index (RCRI) can be readily estimated prospectively or retrospectively using only six clinical variables, however its accuracy in large surgical datasets has not been tested. Our objective was to determine accuracy of the RCRI for predicting cardiovascular complications after adhesiolysis for small bowel obstruction (SBO) using the NSQIP dataset.
Methods:
8,196 cases of open or laparoscopic adhesiolysis (Current Procedural Terminology codes 44005 and 44180) for SBO (International Classification of Diseases, tenth edition 560.81 and 560.9) were analyzed from 2005 to 2013 using the NSQIP dataset. RCRI estimates were calculated for each case and compared to reported cardiovascular complications (myocardial infarction or cardiac arrest) using univariable logistic regression. Overall predictive accuracy was assessed by measuring model discrimination (c-statistics) and model calibration (Hosmer-Lemeshow Chi-squared statistics).
Results:
RCRI predicted cardiovascular complications with odds ratio 1.96, 95% confidence interval (CI) 1.54 – 2.50, P < 0.001. However, c-statistic was poor (0.64, 95% CI 0.59 – 0.68), and Hosmer-Lemeshow Chi-square did not reach statistical significance (X2 35.49, P < 0.001, 3 groups) indicating low model discrimination and calibration.
Conclusion:
Despite its relative computational simplicity, the RCRI performed poorly as a predictor of cardiovascular complications after adhesiolysis for SBO. These findings call into question the utility of the RCRI in this patient population. Future studies should aim to develop models that are computationally simple while retaining predictive accuracy.