An interpretable deep learning framework for predictive modeling of postoperative infections in ICU patients

Xiaoyu Wu, Isaac Luria, Meisheng Xiao, Patrick Tighe, Fei Zou, Baiming Zou

Abstract

A significant proportion of intensive care unit (ICU) patients undergo surgical procedures, and some may develop postoperative infections. Accurately predicting postoperative infection risk and identifying key contributing factors is crucial for improving postoperative management and understanding infection mechanisms.

Introduction

Millions of patients admitted to intensive care units (ICUs) may need to undergo different types of surgical procedures, facing a heightened risk of developing postoperative infections [1]. The postoperative infections have a prevalence rate ranging from 3.0% to 20.7% and an incidence rate of 5–10% in tertiary care hospitals, with rates rising to 28% among ICU patients [2,3].

Methods

EHRs for ICU Patient Postoperative Infection Risk Predictions In this study, we utilized data extracted from the Medical Information Mart for Intensive Care (MIMIC-III) database [41], which includes EHRs of patients admitted to the ICUs of Beth Israel Deaconess Medical Center from 2001 to 2012, encompassing over 40,000 individuals.

Results

Under the permutation feature importance test framework, we evaluated the importance of each of the 25 features using the aforementioned feature importance statistics for each of the four commonly used machine learning models: DNN, RF, SVM, and XGBoost.

Discussion

Developing robust models for predicting postoperative infections in ICU surgical patients and identifying critical factors associated with these infections is crucial for effective postoperative management and improved patient outcomes. In this study, we employed several machine learning models, including SVM, XGBoost, RF, and DNN, to predict postoperative infection risk using large-scale electronic health records.

Conclusion

Our study demonstrates the effectiveness of the stable DNN model in conjunction with the permutation feature importance test framework in evaluating each input factor’s impact on predicting postoperative infection risk.

Citation: Wu X, Luria I, Xiao M, Tighe P, Zou F, Zou B (2026) An interpretable deep learning framework for predictive modeling of postoperative infections in ICU patients. PLoS One 21(4): e0346896. https://doi.org/10.1371/journal.pone.0346896

Editor: Young-Seob Jeong, Chungbuk National University, KOREA, REPUBLIC OF

Received: March 11, 2025; Accepted: March 25, 2026; Published: April 9, 2026

Copyright: © 2026 Wu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The analytic R codes to duplicate the analysis results are available on GitHub (https://github.com/BZou-lab/DeepInfection). The data set used in the analysis is publicly available and extracted from the MIMIC-III database at https://physionet.org/content/mimiciii/1.4/.

Funding: This study was partially supported by NIH (National Institutes of Health) R56 (1R56LM013784) and R01 (R01LM014407 and 1R01HL173044) grants.

Competing interests: The authors have declared that no competing interests exist.