Machine learning based preoperative analytics for the prediction of anastomotic insufficiency in colorectal surgery: a single-centre pilot study

Introduction Anastomotic insufficiency (AI) is a relatively common but grave complication after colorectal surgery. This study aims to determine whether AI can be predicted from simple preoperative data using machine learning (ML) algorithms. Methods and analysis In this retrospective analysis, patients undergoing colorectal surgery with creation of a bowel anastomosis from the University Hospital of Basel were included. Data was split into a training set (80%) and a test set (20%). The group of patients with AI was oversampled to a ratio of 50:50 in the training set and missing values were imputed. Known predictors of AI were included as inputs: age, BMI, smoking status, the Charlson Comorbidity Index, the American Society of Anesthesiologists score, type of operation, indication, haemoglobin and albumin levels, and renal function. Results Of the 593 included patients, 88 experienced AI. At internal validation on unseen patients from the test set, area under the curve (AUC) was 0.61 (95% confidence interval [CI]: 0.44-0.79), calibration slope was 0.16 (95% CI: −0.06-0.39) and calibration intercept was 0.06 (95% CI: 0.02-0.11). We observed a specificity of 0.67 (95% CI: 0.58-0.76), sensitivity of 0.36 (95% CI: 0.08-0.67), and accuracy of 0.64 (95% CI: 0.55-0.72). Conclusion By using 10 patient-related risk factors associated with AI, we demonstrate the feasibility of ML-based prediction of AI after colorectal surgery. Nevertheless, it is crucial to include multicenter data and higher sample sizes to develop a robust and generalisable model, which will subsequently allow for deployment of the algorithm in a web-based application. Strengths and limitations of this study To the best of our knowledge, this is the first study to establish a risk prediction model for anastomotic insufficiency in a perioperative setting in colon surgery. Data from all patients that underwent colon surgery within 8 years at University Hospital Basel were included. We evaluated the feasibility of developing a machine learning model that predicts the outcome by using well-known risk factors for anastomotic insufficiency. Although our model showed promising results, it is crucial to validate our findings externally before clinical practice implications are possible.