Customized predictive algorithms for outcomes of cholecystectomy at a high-volume center

Abstract

Objective: The American College of Surgeons NSQIP® Risk Calculator estimates postoperative risk. Using the NSQIP-predicted complication rates to analyze outcomes for patients undergoing cholecystectomy, we identified significant discrepancies compared to data for patients treated at our high-volume HPB-surgery center. The aim of this study was to develop/validate an institution-specific predictive outcomes model for cholecystectomy. Methods: From 2008 to 2015, 143 patients deemed too high risk for acute care surgery (ACS) had cholecystectomies performed by the Division of Hepatopancreatobiliary Surgery (HPB). Outcomes for HPB surgery were matched against 126 cholecystectomies performed by ACS. NSQIP-generated outcome predictions were recorded and analyzed. Based on these data new predictive models for 6 postoperative outcomes were constructed. Brier score and area under the curve (AUC) were used to assess predictive accuracy for both models and internal validation was performed using bootstrap logistic regression. Logistic regression models were constructed using bivariate (p<0.25) and multivariate (p<0.05) analyses. Model accuracy was validated using retrospective and prospective data. Results: HPB surgeons performed cholecystectomy on a higher acuity population with poorer predicted postoperative outcomes compared with ACS. Brier scores showed little difference in the predictive ability of the NSQIP and our models. For ACS cholecystectomy, our model better predicted mortality, surgical site infection, and cardiac complication (AUC: 0.9450.978, p<0.05). For HPB cholecystectomy, our model better predicted 5/6 postoperative outcomes versus NSQIP (NSQIP AUC: 0.5740.764, p>0.05; institute-specific AUC: 0.7790.982, p<0.01). Conclusion: For higher acuity patients undergoing cholecystectomy, customized models are superior for predicting individual perioperative risk and allow more accurate, patient-specific delivery of care.Table 1Comparison of predictive capacity of ACS-NSQIP® and Institute-Specific risk-prediction models for postoperative complications of cholecystectomy performed at a high-volume center by Acute Care Surgery and Hepatopancreatobiliary SurgeryNSQIP CalculatorCMC ModelAUCAUCBrierAUCAUCBrierp valuescorep valuescoreCholecystectomy performed by Acute Care SurgerySerious complication0.61160.15520.03590.97830.00060.023930-day readmission0.67710.04740.05170.58940.18710.051430-day mortality0.89020.02250.01560.94500.01440.0156Cardiac complication0.62900.12480.01540.95970.01300.0153Cholecystectomy performed by Hepatopancreatobiliary SurgerySerious complication0.76350.00000.11740.88680.00000.0941Discharge to nursing facility0.75850.00130.06340.91600.00000.0494Renal Failure0.83330.00830.02280.77400.03100.017130-day readmission0.57400.12000.11030.77910.00000.113030-day mortality0.66060.00020.20200.86070.00000.1205Cardiac complication0.76060.00050.06940.98170.00000.0212Abbreviations: ACS-NSQIP, American College of Surgeons National Surgical Quality Improvement Program; AUC, Area under the curve. Open table in a new tab Abbreviations: ACS-NSQIP, American College of Surgeons National Surgical Quality Improvement Program; AUC, Area under the curve.