TTP3.11 Role of machine learning in predicting postoperative complications in colorectal surgery: A systematic review and meta-analysis

Introduction This systematic review and meta-analysis aimed to investigate machine learning techniques to predict postoperative complications in colorectal surgery patients. Methods A systematic literature search was conducted using PubMed, MEDLINE, Embase, and Google Scholar. Clinical studies investigating machine learning models' roles in postoperative complications after colorectal surgery were included. The outcome measure was the area under the curve (AUC) for the model under investigation. AUC and standard error (SE) were pooled using a random effect model to estimate the overall effect. Data analysis was performed with Medcalc version 23 software, and results were presented in forest plots. Results The systematic search yielded 18 eligible articles reporting the following postoperative complications: anastomosis leak, mortality, prolonged length of hospitalisation, re-admission, bleeding, ileus, and surgical site infection (SSI). The pooled AUC for anastomosis leak was 0.813 [Standard error =0.0305, 95% CI (0.753 to 0.873)], for mortality was 0.867 [Standard error=0.0147, 95% CI (0.838 to 0.896)], Prolonged LOS was 0.810 [standard error=0.0418, 95% CI (0.728 to 0.892)], and SSI was 0.802 [standard error=0.0306, 95% CI (0.742 to 0.862)]. Conclusion Machine learning shows promising clinical utility and applicability in accurately predicting patient risk of developing complications following colorectal surgery. Large multicentre studies are needed to enhance the generalisability of these models.