Development and Validation of a Machine Learning Prediction Model for Textbook Outcome in Liver Surgery: Results From a Multicenter, International Cohort.

Abstract

Objective: This study aimed to (1) develop a machine learning (ML) model that predicts the textbook outcome in liver surgery (TOLS) using preoperative variables and (2) validate the TOLS criteria by determining whether TOLS is associated with long-term survival after hepatectomy.

Background: Textbook outcome is a composite measure that combines several favorable outcomes into a single metric and represents the optimal postoperative course. Recently, an expert panel of surgeons proposed a Delphi consensus-based definition of TOLS.

Methods: Adult patients who underwent hepatectomies were identified from a multicenter, international cohort (2010-2022). After data preprocessing and train-test splitting (80:20), 4 models for predicting TOLS were trained and tested. Following model optimization, the performance of the models was evaluated using receiver operating characteristic curves, and a web-based calculator was developed. In addition, a multivariable Cox proportional hazards analysis was conducted to determine the association between TOLS and overall survival (OS).

Results: A total of 2059 patients were included, with 62.8% meeting the criteria for TOLS. The XGBoost model, which had the best performance with an area under the curve of 0.73, was chosen for the web-based calculator. The most predictive variables for having TOLS were a minimally invasive approach, fewer lesions, lower Charlson Comorbidity Index, lower preoperative creatinine levels, and smaller lesions. In the multivariable analysis, having TOLS was associated with improved OS (hazard ratio = 0.82, P = 0.015).

Conclusions: Our ML model can predict TOLS with acceptable discrimination. We validated the TOLS criteria by demonstrating a significant association with improved OS, thus supporting their use in informing patient care.