Validation of an online risk calculator for the prediction of anastomotic leak after colon cancer surgery and preliminary exploration of artificial intelligence-based analytics

摘要

Abstract Background Recently published data support the use of a web-based risk calculator ( www.anastomoticleak.com ) for the prediction of anastomotic leak after colectomy. The aim of this study was to externally validate this calculator on a larger dataset. Methods Consecutive adult patients undergoing elective or emergency colectomy for colon cancer at a single institution over a 9-year period were identified using the Binational Colorectal Cancer Audit database. Patients with a rectosigmoid cancer, an R2 resection, or a diverting ostomy were excluded. The primary outcome was anastomotic leak within 90?days as defined by previously published criteria. Area under receiver operating characteristic curve (AUROC) was derived and compared with that of the American College of Surgeons National Surgical Quality Improvement Program ? (ACS NSQIP) calculator and the colon leakage score (CLS) calculator for left colectomy. Commercially available artificial intelligence-based analytics software was used to further interrogate the prediction algorithm. Results A total of 626 patients were identified. Four hundred and fifty-six patients met the inclusion criteria, and 402 had complete data available for all the calculator variables (126 had a left colectomy). Laparoscopic surgery was performed in 39.6% and emergency surgery in 14.7%. The anastomotic leak rate was 7.2%, with 31.0% requiring reoperation. The anastomoticleak.com calculator was significantly predictive of leak and performed better than the ACS NSQIP calculator (AUROC 0.73 vs 0.58) and the CLS calculator (AUROC 0.96 vs 0.80) for left colectomy. Artificial intelligence-predictive analysis supported these findings and identified an improved prediction model. Conclusions The anastomotic leak risk calculator is significantly predictive of anastomotic leak after colon cancer resection. Wider investigation of artificial intelligence-based analytics for risk prediction is warranted.