Copyright: ©Author(s) 2026. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution-NonCommercial (CC BY-NC 4.0) license. No commercial re-use. See permissions. Published by Baishideng Publishing Group Inc.
World J Gastroenterol. Jan 21, 2027; 33(3): 123339
Published online Jan 21, 2027. doi: 10.3748/wjg.123339
Development and validation of a machine learning model for hospital-acquired bacterial infection prediction in acute-on-chronic liver failure
Jing-Yi Chen, Xin-Yi Chen, Yi-Wen Xie, Yu-Wei Chen, Xiao-Wei Shi, Ning Lin, Chen-Jie Huang, Rui Luo, Xiao-Qing Lu, Xiao-Xiao Chen, Jian Wu, Hai-Jun Huang, Qiang Zhu, Lan-Juan Li, Jiong Yu, Hong-Cui Cao
Jing-Yi Chen, Xin-Yi Chen, Yi-Wen Xie, Xiao-Wei Shi, Chen-Jie Huang, Rui Luo, Xiao-Qing Lu, Xiao-Xiao Chen, Lan-Juan Li, Jiong Yu, Hong-Cui Cao, State Key Laboratory for the Diagnosis and Treatment of Infectious Diseases, National Clinical Research Center for Infectious Diseases, National Medical Center for Infectious Diseases, Collaborative Innovation Center for Diagnosis and Treatment of Infectious Diseases, The First Affiliated Hospital, Zhejiang University School of Medicine, Hangzhou 310003, Zhejiang Province, China
Yu-Wei Chen, Department of Cardiovascular Medicine, The Second Affiliated Hospital, Fujian Medical University, Quanzhou 362000, Fujian Province, China
Ning Lin, Department of Clinical Laboratory, The Affiliated Huai’an No. 1 People’s Hospital, Nanjing Medical University, Huai’an 223300, Jiangsu Province, China
Jian Wu, Department of Clinical Laboratory, The Affiliated Suzhou Hospital of Nanjing Medical University, Suzhou Municipal Hospital, Gusu School, Nanjing Medical University, Suzhou 215001, Jiangsu Province, China
Hai-Jun Huang, Department of Infectious Diseases, Zhejiang Provincial People’s Hospital & People’s Hospital affiliated of Hangzhou Medical College, Hangzhou 310014, Zhejiang Province, China
Qiang Zhu, Department of Infectious Disease, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan 250021, Shandong Province, China
Co-first authors: Jing-Yi Chen and Xin-Yi Chen.
Co-corresponding authors: Jiong Yu and Hong-Cui Cao.
Author contributions: Chen JY contributed to conceptualization, data curation, methodology, formal analysis, visualization, and drafting of the manuscript; Chen XY contributed to data curation, formal analysis, and investigation; Xie YW, Chen YW, Shi XW and Lin N contributed to data curation, supervision, and validation; Huang CJ, Luo R, Lu XQ, Chen XX, Wu J, Huang HJ and Zhu Q contributed to data curation; Li LJ provided academic guidance and critically revised the manuscript; Yu J and Cao HC contributed to conceptualization, supervision, and critical revision of the manuscript. Chen JY and Chen XY contributed equally to this work as co-first authors. We respectfully request that Cao HC and Yu J be designated as co-corresponding authors of this manuscript, with Cao HC serving as the lead corresponding author. As reflected in the author contributions, both took primary responsibility for the conceptualization, overall supervision, and critical revision of the manuscript, providing the senior academic oversight that guided the study from design through completion. Their leadership encompassed complementary aspects of the work, and their joint supervision was essential to ensuring its scientific rigor and intellectual integrity. Designating both as corresponding authors allows the editorial office and readers to reach the appropriate senior author for questions concerning either the conceptual or the methodological dimensions of the study, both before and after publication. Cao HC, as lead corresponding author, will manage the primary editorial correspondence and serve as the principal point of contact. Both authors have approved the final manuscript and agree to be accountable for all aspects of the work.
AI contribution statement: Portions of this manuscript were edited using AI tools solely for language refinement and readability improvement. All AI-assisted edits were reviewed and verified by the authors, who take full responsibility for the integrity, accuracy, originality, and scientific validity of the manuscript and all submitted materials.
Supported by Prevention and Control of Emerging and Major Infectious Diseases-National Science and Technology Major Project, No. 2025ZD01906501; Fundamental Research Funds for the Central Universities, No. 2025ZFJH03; Central Guidance Fund for Local Science and Technology Development, No. 2024ZY01054; and High-level Personnel Cultivating Project of Zhejiang Province, No. 2023R5243.
Institutional review board statement: Patient data were anonymized and retrieved from electronic medical records. The study complied with the Declaration of Helsinki and was approved by the institutional ethics committees of all participating centers, including The First Affiliated Hospital, Zhejiang University School of Medicine (No. 2024-1044), Taizhou Hospital of Zhejiang Province (No. K20230227), the First Affiliated Hospital of Wenzhou Medical University (No. KY2023-158), Zhejiang Provincial People’s Hospital (No. KY2024182), and Shulan (Hangzhou) Hospital (No. KY2025052).
Informed consent statement: Written informed consent was waived because of the retrospective design.
Conflict-of-interest statement: The authors report there are no competing interests to declare.
STROBE statement: The authors have read the STROBE Statement-checklist of items, and the manuscript was prepared and revised according to the STROBE Statement-checklist of items.
Data sharing statement: The datasets generated during the current study are available from the corresponding author upon reasonable request.
Corresponding author: Hong-Cui Cao, MD, PhD, State Key Laboratory for the Diagnosis and Treatment of Infectious Diseases, National Clinical Research Center for Infectious Diseases, National Medical Center for Infectious Diseases, Collaborative Innovation Center for Diagnosis and Treatment of Infectious Diseases, The First Affiliated Hospital, Zhejiang University School of Medicine, No. 79 Qingchun Road, Hangzhou 310003, Zhejiang Province, China.
hccao@zju.edu.cn
Received: May 18, 2026
Revised: June 25, 2026
Accepted: July 8, 2026
Published online: January 21, 2027
Processing time: 181 Days and 14.7 Hours
BACKGROUND
Hospital-acquired (HA) bacterial infections are a frequent and potentially preventable driver of early deterioration in acute-on-chronic liver failure (ACLF), yet no validated tool exists to identify high-risk patients at the time of hospital admission.
AIM
To develop and externally validate an interpretable admission-time machine learning (ML) model for predicting HA bacterial infection in ACLF.
METHODS
In this multicenter retrospective study, 3103 hospitalized patients with cirrhosis were screened across five tertiary centers in China (2016-2025), and 874 eligible patients with ACLF were included. Admission-time variables were incorporated into ML models, which were evaluated for discrimination, calibration, and clinical utility. Model robustness was further assessed using time-dependent discrimination, cumulative incidence, and competing-risk analyses accounting for in-hospital death, transfer, or liver transplantation.
RESULTS
Among 644 patients in the derivation cohort, 259 (40.2%) developed HA bacterial infections. A 12-variable Gradient Boosting Decision Tree model was selected as the final model, achieving an internally validated area under the curve (AUC) of 0.906 [95% confidence interval (CI): 0.879-0.930]. Discrimination remained good in the external validation cohort (n = 230; 99 events), with an AUC of 0.890 (95%CI: 0.843-0.931). The model outperformed conventional liver disease severity scores in predicting HA bacterial infections. Risk stratification remained consistent across time-dependent, cumulative incidence, and competing-risk analyses.
CONCLUSION
An interpretable admission-time ML model was developed to identify patients at high risk for HA bacterial infections in ACLF. This framework may support early risk stratification and targeted infection surveillance.
Core Tip: This multicenter retrospective study developed and externally validated an interpretable admission-time machine learning model to predict hospital-acquired bacterial infections in patients with acute-on-chronic liver failure. A 12-variable Gradient Boosting Decision Tree model demonstrated robust discrimination in both derivation and external validation cohorts, outperformed conventional liver disease severity scores, and remained stable across calibration, decision-curve, time-dependent, cumulative incidence, and competing-risk analyses. By providing individualized risk estimates and SHapley Additive exPlanations-based interpretations through a web application, this model may support early risk stratification and targeted infection surveillance during hospitalization.
