Machine learning insight: Unveiling overlooked risk factors for postoperative complications in gastric cancer

Abstract

Abstract Since postoperative complications after gastrectomy for gastric cancer are associated with poor clinical outcomes, it is crucial to predict and prepare for the occurrence of complications preoperatively. We evaluated machine learning for predicting complications after gastric cancer surgery, emphasizing its advantage in uncovering unnoticed risk factors and improving preoperative strategies over linear regression models. We retrospectively reviewed cohort data from 865 patients who underwent gastrectomy for gastric cancer from 2018–2022. A total of 83 variables including demographics, clinical features, laboratory values, intraoperative parameters, and pathologic results were used to conduct the machine learning model. The data set was partitioned into 80% for training and 20% for validation. Utilizing the backward elimination method and a moderate strategy for handling missing data, machine learning models achieved an impressive area under the curve value of 0.744, outshining linear regression in performance. We pinpointed 15 significant variables linked to postoperative complications. Among these, operation time emerged as the most impactful, with pre-operative albumin levels and Mean Corpuscular Hemoglobin (MCH) trailing closely. This research underscores the capabilities of machine learning in refining predictions of post-gastric cancer surgery complications. It highlights previously overlooked risk factors, emphasizing the nuanced role of Complete Blood Count (CBC) parameters.