Key points are not available for this paper at this time.
INTRODUCTION: Oesophageal resection carries significant morbidity and mortality. Artificial intelligence (AI) advances in medical research enable enhanced predictions, flexibility, and interpretability, especially for complex interactions and nonlinear relationships. MATERIAL AND METHODS: We used a register-based case-control design nested within prospectively collected data from the Swedish National Quality Register for Oesophageal and Gastric Cancer (NREV) to perform traditional logistic regression (LR) and machine learning (ML) with explainable AI (XAI) to predict 90-day mortality and anastomotic leakage in 1846 patients who underwent oesophageal resection between November 2005 and February 2018. RESULTS: The 90-day mortality was 6.0 % and anastomotic leakage was 12.4 %. XAI models yielded an area under the curve (AUC) of 0.95 for 90-day mortality, compared to 0.88 for LR. For anastomotic leakage, the AUC was 0.84 with XAI versus 0.74 with LR. LR identified significant odds ratios for 90-day mortality associated with age, ASA 2-3, BMI, and anastomotic leakage. ML models identified the same variables plus year of surgery as significant. For anastomotic leakage, LR was significant only for ASA 3, whereas ML found all examined variables to be significant predictors. XAI showed age and perioperative bleeding as important survival factors, while high BMI and age were significant risk factors for anastomotic leakage. All factors demonstrated nonlinear associations. XAI also visualises individual risk assessments for each procedure. CONCLUSIONS: By applying XAI, we advance surgical understanding of anastomotic leakage and mortality after oesophagectomy. Our data contain significant nonlinear relationships that cannot be visualised LR. With XAI, we extract personalised risk assessments, bringing oesophageal surgery closer to personalised medicine.
Djerf et al. (Fri,) studied this question.