Development and external validation of a machine learning-based model to predict postoperative recurrence in patients with duodenal adenocarcinoma: a multicenter, retrospective cohort study.

Abstract

Background: Duodenal adenocarcinoma (DA) has a high recurrence rate, making the prediction of recurrence after surgery critically important.

Methods: Our objective is to develop a machine learning-based model to predict the postoperative recurrence of DA. We conducted a multicenter, retrospective cohort study in China. 1830 patients with DA who underwent radical surgery between 2012 and 2023 were included. Wrapper methods were used to select optimal predictors by ten machine learning learners. Subsequently, these ten learners were utilized for model development. The model's performance was validated using three separate cohorts, and assessed by the concordance index (C-index), time-dependent calibration curve, time-dependent receiver operating characteristic curves, and decision curve analysis.

Results: After selecting predictors, ten feature subsets were identified. And ten feature subsets were combined with the ten machine learning learners in a permutation, resulting in the development of 100 predictive models, and the Penalized Regression + Accelerated Oblique Random Survival Forest model (PAM) exhibited the best predictive performance. The C-index for PAM was 0.882 (95% CI 0.860-0.886) in the training cohort, 0.747 (95% CI 0.683-0.798) in the validation cohort 1, 0.736 (95% CI 0.649-0.792) in the validation cohort 2, and 0.734 (95% CI 0.674-0.791) in the validation cohort 3. A publicly accessible web tool was developed for the PAM.

Conclusions: The PAM has the potential to identify postoperative recurrence in DA patients. This can assist clinicians in assessing the severity of the disease, facilitating patient follow-up, and aiding in the formulation of adjuvant treatment strategies.