Few and feasible preoperative variables can identify high-risk surgical patients: derivation and validation of the Ex-Care risk model

Abstract

BackgroundThe development of feasible preoperative risk tools is desirable, especially for low-middle income countries with limited resources and complex surgical settings. This study aimed to derive and validate a preoperative risk model (Ex-Care model) for postoperative mortality and compare its performance with current risk tools. MethodsA multivariable logistic regression model predicting in-hospital mortality was developed using a large Brazilian surgical cohort. Patient and perioperative predictors were considered. Its performance was compared with the Charlson comorbidity index (CCI), Revised Cardiac Risk Index (RCRI), and the Surgical Outcome Risk Tool (SORT). ResultsThe derivation cohort included 16 618 patients. In-hospital death occurred in 465 patients (2.8%). Age, with adjusted splines, degree of procedure (major vs non-major), ASA physical status, and urgency were entered in a final model. It showed high discrimination with an area under the receiver operating characteristic curve (AUROC) of 0.926 (95% confidence interval [CI], 0.91–0.93). It had superior accuracy to the RCRI (AUROC, 0.90 vs 0.76; P<0.01) and similar to the CCI (0.90 vs 0.82; P=0.06) and SORT models (0.90 vs 0.92; P=0.2) in the temporal validation cohort of 1173 patients. Calibration was adequate in both development (Hosmer–Lemeshow, 9.26; P=0.41) and temporal validation cohorts (Hosmer–Lemeshow 5.29; P=0.71). ConclusionsThe Ex-Care risk model proved very efficient at identifying high-risk surgical patients. Although multicentre studies are needed, it should have particular value in low resource settings to better inform perioperative health policy and clinical decision-making.