EGS P05 Evaluating the predictive value of NSQIP surgical risk calculator in patients undergoing emergency laparoscopic cholecystectomy at a single UK centre

Abstract

Abstract Background Machine learning algorithms for surgical risk predictions have been utilised as an informed consent tool for surgeons and patients. The American College of Surgeons National Surgical Quality Improvement Program (NSQIP) is a surgical risk calculator utilising machine learning to predict adverse postoperative outcomes with surgeon adjustment of risk. However, there is a paucity of validation studies for laparoscopic cholecystectomy data. This study aimed to evaluate the predictive value of the NSQIP calculator in patients undergoing emergency laparoscopic cholecystectomy at a single UK-based centre. Methods Preoperative patient characteristics and postoperative outcome data was collected from patients who underwent emergency laparoscopic cholecystectomy between January 2022 and January 2023. NSQIP surgical risk scores were retrospectively calculated using preoperative characteristics and risk factor data. Surgeon adjustment of risk was set at the default for all patients. The predicted rates of postoperative complications, serious complications (definition as per the NSQIP calculator), and hospital readmission were compared to the observed rates using Brier scores. Results Eighty-five patients were analysed. The median age was 50 (range 19-91) and BMI was 29 (range 21-44). The NSQIP predicted an overall complication rate of 4.2%. Overall observed complication rate was 3.7% and included hospital-acquired pneumonia (n=1), pancreatitis (n=1), and postoperative atrial fibrillation (n=1). The predicted and observed complication rates were comparable (4.2% vs 3.7% respectively, Brier score 0.066). Likewise, the observed rate was also lower than predicted rate for serious complications (2.4% vs. 3.4% respectively, Brier score 0.047). The observed rate of hospital readmission was similar to predicted risk (3.7% vs. 4.0% respectively, Brier score 0.033). Conclusions In this cohort, the NSQIP surgical risk calculator demonstrated acceptable predictive ability for the risk of postoperative complications after emergency laparoscopic cholecystectomy. There was a propensity of the NSQIP calculator to overestimate the risk of complications, within the limitations of the small sample size and retrospective nature of this study. Further studies are warranted to assess the ability of the NSQIP tool to predict the risk of surgery-specific complications.