Development of a Clinical Prediction Model for Anastomotic Leakage in Colorectal Surgery

1. An integrative machine learning predictive model based on individual base models can accurately predict the incidence of anastomotic leakage (AL) after colon resection.

Evidence Rating Level: 2 (Good)

AL after colorectal surgery is associated with higher reoperative rates, longer hospital stays, and increased morbidity and mortality. Although multiple risk factors correlate with the incidence of AL, it is difficult for surgeons to objectively predict this risk. This study sought to develop a machine learning meta-model that combines pre-existing models to improve the assessment of AL incidence. This retrospective prognostic study was conducted across 13 centres in Europe and North America in 9120 patients (mean [SD] age, 61.26 [15.71] years; 50.8% male). Four pre-existing machine learning algorithms (CatBoost, LightGBM, random forest, and bagging classifier) were trained with 34 preoperative AL risk factors. This yielded 144 meta-model combinations, with the most optimal configuration yielding an F1 score of 87% (95% CI, 78%-95%). The model was also evaluated on an external validation test set, achieving an F1 score of 70%. Compared to its individual models, the meta-model was significantly better at predicting AL risk than LightGBM (p=0.0009), random forest (p=0.0005), and bagging classifier (p=0.0035), but not CatBoost(p=0.1692). These findings suggest that machine learning may reduce bias and help risk-stratify patients based on the likelihood of requiring AL after colorectal surgery. However, prospective validation and larger studies assessing the clinical utility of these models are required before they are integrated into standard care.

Click here to read this study in JAMA Network Open

Image: PD

©2025 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.