Predictive performance of the Modified Frailty Index and ASA-PS classification for 30-day mortality after esophagectomy in patients with esophageal cancer

Abstract

Background Esophageal cancer is commonly diagnosed at a locoregional stage and is typically treated with neoadjuvant chemoradiotherapy followed by esophagectomy. Despite advances in perioperative care, esophagectomy remains associated with substantial morbidity and mortality. Frailty assessment has been proposed as a tool for preoperative risk stratification, but its predictive value in patients undergoing esophagectomy remains uncertain. Methods We conducted a retrospective cohort study of patients who underwent curative-intent esophagectomy between January 2012 and December 2024. The predictive performance of the 5-item Modified Frailty Index (mFI-5) was compared with the American Society of Anesthesiologists Physical Status (ASA-PS) classification. Receiver operating characteristic (ROC) curve analysis and exploratory multivariable logistic regression were performed to evaluate the predictive performance of mFI-5 and ASA-PS classification for 30-day mortality. Results A total of 123 patients underwent esophagectomy, with a 30-day mortality rate of 9.8%. Patients who died had significantly longer intensive care unit and hospital stays and a higher incidence of severe postoperative complications. ASA-PS demonstrated better discrimination than mFI-5 for predicting 30-day mortality (AUC 0.67 vs. 0.55), although the discriminatory performance of both tools remained limited. In the exploratory multivariable analysis, ASA-PS classification was independently associated with 30-day mortality, whereas mFI-5 was not. Conclusions Although ASA-PS classification was independently associated with 30-day mortality, both ASA-PS and mFI-5 demonstrated limited predictive performance following esophagectomy. These findings suggest that mFI-5 and ASA-PS alone provide limited discrimination for predicting 30-day mortality and support the development of esophagectomy-specific risk prediction models incorporating objective clinical, physiological, functional, and procedure-specific variables.