Development and validation of a post-operative delirium prediction model for patients undergoing abdominal surgery: A retrospective, observational, single-center study

Abstract

Abstract Background Postoperative delirium (POD) is considered as a relatively common and serious problem after major abdominal surgery procedures. It is presumed to be preventable in most of cases. The purpose of this study was to develop and evaluate a POD prediction model for patients undergoing abdominal surgery. Methods From July 2019 to December 2019, patients underwent elective abdominal surgery in our hospital were retrospectively analyzed, and their demographics, pre-operative evaluation, intra-operative and anesthesiologic factors were recorded. Based on the results of the multivariate regression analyzes using P < 0.05 and P < 0.001 as two significance level, we obtained 2 different prediction models comprising of 10 and 4 factors respectively. After factorizing the risk of overfitting and cross-validation, we proposed a final POD prediction model consisting of 4 predictors. From January 2021 to December 2021, 346 more qualifying patients were enrolled for the external validation of the 4-factor model. The study was retrospectively registered on the World Health Organization International Clinical Trials Registry Platform (WHO-ICTRP) with ID ChiCTR2100047405. Results After screening, 838 patients were included as the training cohort and 10.9% (91/838) of the patients manifested POD. Those patients who developed POD were more likely to be aged more than 60 years (OR = 1.345, P =0.005), with history of diazepam usage (OR = 3.622, P =0.003), history of cerebrovascular disease (OR = 2.150, P = 0.012) and intraoperative positive fluid balance (OR = 1.41, P <0 .001). The optimum cut-off point of the predicted probability that maximized the sum of sensitivity and specificity was 0.12. The fitting set AUC was 0.703 (95%CI: 0.637–0.753). The cross validation set AUC was 0.684 (SD = 0.068) and the external validation AUC of the model was 0.63 (95%CI: 0.511–0.758), quite closed to that of the fitting set, which indicated that the selected model was robust. Conclusions The 4-factor POD prediction model shows good prediction efficiency and can prompt for prophylactic intervention in patients at risk for POD. Trial registration : A retrospective registration of the study has been submitted to the World Health Organization International Clinical Trials Registry Platform (WHO-ICTRP) with Registration ID ChiCTR2100047405 (18/06/2021).