Original Investigation 

MySurgeryRisk Model Predictions of Postoperative Complications and Mortality

Yuanfang Ren, Esra Adiyeke, Ziyuan Guan, et al

JAMA Surg Published Online: April 29, 2026

doi: 10.1001/jamasurg.2026.1112

Key Points

Question  Will applying the MySurgeryRisk framework, originally developed and prospectively validated on a single-center dataset, to a large multicenter dataset enhance generalizability without degrading performance?

Findings  The models predicted postoperative complications in this cohort analysis of 508 097 encounters from 366 875 patients who underwent major operations at 14 health care institutions. Model performance was comparable with the previously validated MySurgeryRisk models, with procedure type and clinician-specific factors identified as the most influential contributors to predicted risk.

Meaning  In this study, incorporating routinely collected variables from a multicenter cohort enabled highly generalizable, accurate prediction of postoperative complications.

Abstract

Importance  Despite advances in surgical techniques and care, postoperative complications are prevalent and affect up to 15% of patients who undergo inpatient surgery. Analytic models for surgical risk estimation can benefit from incorporating complex electronic health record data across multiple, diverse settings.

Objective  To test the hypothesis that applying the MySurgeryRisk framework, developed and prospectively validated on a single-center dataset, to a large multicenter dataset will enhance generalizability without degrading the predictive performance achieved by the original, single-center model.

Design, Setting, and Participants  This retrospective, longitudinal, multicenter cohort analysis included 508 097 encounters from 366 875 adult patients admitted and underwent major inpatient operations at 14 health care institutions within the OneFlorida + network from 2012 through 2023. Models were trained on a development set (2012-2020; n = 358 216 encounters) and evaluated on a validation set (2020-2023; n = 149 881 encounters). These data were analyzed from January 2025 to February 2026.

Main Outcomes and Measures  Using the feature selection and transformation methods validated in the MySurgeryRisk framework, eXtreme Gradient Boosting models were developed and validated to predict the postoperative risk of intensive care unit (ICU) admission, postoperative mechanical ventilation (MV), postoperative acute kidney injury (AKI), and in-hospital mortality. Performance was primarily evaluated using area under the receiver operating characteristics curve (AUROC) values.

Results  Among 366 875 total patients who underwent 508 097 major inpatient operations, the mean (SD) age was 59 (18) years; 190 799 (52%) were women and 176 076 were men (48%). The prevalence of complications was 8% (n = 42 302) for ICU admission, 4% (n = 20 435) for MV, 7% (n = 36 027) for AKI, and 1% (n = 5131) for in-hospital mortality. AUROC values were as follows: ICU admission, 0.93 (95% CI, 0.93-0.93); MV, 0.94 (95% CI, 0.94-0.94); AKI, 0.92 (95% CI, 0.92-0.92); and in-hospital mortality, 0.95 (95% CI, 0.94-0.95). Model predictive performance was comparable with previously validated MySurgeryRisk models. Primary procedure code and clinician-specific factors were consistently the most influential variables.

Conclusions and Relevance  A model using routinely collected variables from a large multicenter cohort was developed and validated, demonstrating accurate prediction of postoperative complications and death with generalizability across a large network of health care institutions. Procedure type and clinician-specific factors were the most influential contributors to predicted risk, providing insights into factors influencing surgical outcomes.