Introduction: About 25% of patients undergoing pancreaticoduodenectomy (PD) for right-sided pancreatic ductal adenocarcinoma (PDAC) die within 1 y of diagnosis. These patients carry all the risks of significant morbidity with no survival advantage when compared to nonsurgical options. We aimed to determine if machine learning models have superior accuracy to traditional regression models at predicting futile surgery in patients with PDAC.
Methods: We analyzed data from patients in the National Cancer Database undergoing PD for PDAC between 2004 and 2020. PD was defined as futile if the patient died within 12 mo of cancer diagnosis. We trained predictive models using 80% of the dataset and 16 preoperative input variables. Models included logistic regression, multilayer perceptron, decision tree, random forest, and gradient boosting classifiers. Models were tested on a 20% test set using area under the receiver operating characteristic curve and Brier scores.
Results: Of the 66,331 patients identified, 34,260 (51.7%) were men, with a median age of 67 y (interquartile range, 59 to 74 y). A total of 16,772 (25.3%) patients met the criteria for futile surgery. The gradient boosting model outperformed other models with an area under the receiver operating characteristic curve of 0.689, followed by logistic regression (0.679), random forest (0.675), and decision tree (0.664). Key predictors of futile PD included advanced age (> 79 y), tumor size ≥ 4 cm, and poor differentiation. Neoadjuvant therapy was associated with lower futility risk.
Conclusions: We demonstrated the ability of machine learning models to predict the odds of futile PD with moderate accuracy. Although similar analyses are needed on more granular datasets, our study has important implications for shared decision-making and optimized care for patients with PDAC.
Keywords: Futility prediction; Machine learning; Pancreatic cancer; Pancreaticoduodenectomy; Risk stratification.
Copyright © 2025 Elsevier Inc. All rights reserved.