Development and Internal Validation of a Pilot Model for Newly Diagnosed in-Hospital Bacterial and Fungal Infections in Decompensated Hepatitis B Cirrhosis: A Single-Center Retrospective Cohort Study
- 1Department of Clinical Laboratory, The First Affiliated Hospital of Anhui Medical University, Hefei, Anhui, People’s Republic of China
Abstract
Background and Aims: Recognizing infection risk during hospitalization is important in decompensated cirrhosis, although routine measurements may reflect infection already developing. We developed and internally validated an admission-time pilot model for newly diagnosed in-hospital bacterial or fungal infection in decompensated hepatitis B virus-related cirrhosis.
Methods: This single-center retrospective cohort included 150 patients admitted in 2023–2025. Admission neutrophil-to-lymphocyte ratio (NLR) and Model for End-Stage Liver Disease score (MELD) were fixed as predictors for the primary Firth/Jeffreys penalized logistic model. Complete-case analysis included 148 patients and 20 events. We refitted the fixed model in 1000 bootstrap samples, assessed calibration and discrimination, and examined early-event exclusions and an admission-only ridge model. The hospital-recorded procedure variable was excluded from prediction. Secondary analyses used Aalen–Johansen cumulative incidence, Fine–Gray regression and cause-specific Cox, with infection-free discharge and death as competing events.
Results: Among 148 modeled patients, the NLR and MELD odds ratios were 1.073 (95% confidence interval [CI], 1.025–1.124) and 1.062 (95% CI, 1.002–1.125). Apparent area under the receiver operating characteristic curve (AUC) was 0.716 versus 0.633 for MELD alone. Optimism-corrected AUC was 0.700 (approximate 95% CI 0.566–0.846), corrected Brier score was 0.103, and corrected calibration slope was 0.981 (approximate bootstrap interval 0.506–1.646). AUCs after separate exclusions of infections recorded by hospital day 2 or day 3 were 0.750 and 0.731, respectively. Fine–Gray subdistribution hazard ratios were 1.069 (95% CI 1.042–1.098) for NLR and 1.064 (1.021–1.109) for MELD.
Conclusion: This admission model showed modest discrimination with imprecise calibration. Retrospective ascertainment and limited validation preclude clinical use. Prospective external validation at a defined prediction time and horizon is needed before the model guides surveillance or treatment.
Related
- Development and Internal Validation of a Retrospective Prediction Model for Fungal Infections in Patients with Chronic Obstructive Pulmonary Disease.
- Development and content validation of a risk-stratified nursing intervention programme for the prevention of healthcare-associated infections
- Development and validation of an interpretable prediction model for high-grade vesicoureteral reflux in children with urinary tract infections
- Explainable machine learning and multivariable logistic regression for identifying prevalent self-reported physician-diagnosed chronic lung disease in postmenopausal women: development and multi-cohort external validation
- Development and internal validation of a predictive model for clinical deterioration in newly diagnosed lung cancer patients based on interpretable machine learning.
BibTeX
@article{jin2026development,
title = {{Development and Internal Validation of a Pilot Model for Newly Diagnosed in-Hospital Bacterial and Fungal Infections in Decompensated Hepatitis B Cirrhosis: A Single-Center Retrospective Cohort Study}},
author = {Yangyang Jin and Ying Huang and Runqiu Jiang},
journal = {Infection and Drug Resistance},
year = {2026},
month = nov,
doi = {10.2147/idr.s638048},
url = {https://doi.org/10.2147/idr.s638048},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - Development and Internal Validation of a Pilot Model for Newly Diagnosed in-Hospital Bacterial and Fungal Infections in Decompensated Hepatitis B Cirrhosis: A Single-Center Retrospective Cohort Study
AU - Yangyang Jin
AU - Ying Huang
AU - Runqiu Jiang
PY - 2026
DA - 2026/11/05
JO - Infection and Drug Resistance
DO - 10.2147/idr.s638048
UR - https://doi.org/10.2147/idr.s638048
AB - Background and Aims: Recognizing infection risk during hospitalization is important in decompensated cirrhosis, although routine measurements may reflect infection already developing. We developed and internally validated an admission-time pilot model for newly diagnosed in-hospital bacterial or fungal infection in decompensated hepatitis B virus-related cirrhosis. Methods: This single-center retrospective cohort included 150 patients admitted in 2023–2025. Admission neutrophil-to-lymphocyte ratio (NLR) and Model for End-Stage Liver Disease score (MELD) were fixed as predictors for the primary Firth/Jeffreys penalized logistic model. Complete-case analysis included 148 patients and 20 events. We refitted the fixed model in 1000 bootstrap samples, assessed calibration and discrimination, and examined early-event exclusions and an admission-only ridge model. The hospital-recorded procedure variable was excluded from prediction. Secondary analyses used Aalen–Johansen cumulative incidence, Fine–Gray regression and cause-specific Cox, with infection-free discharge and death as competing events. Results: Among 148 modeled patients, the NLR and MELD odds ratios were 1.073 (95% confidence interval [CI], 1.025–1.124) and 1.062 (95% CI, 1.002–1.125). Apparent area under the receiver operating characteristic curve (AUC) was 0.716 versus 0.633 for MELD alone. Optimism-corrected AUC was 0.700 (approximate 95% CI 0.566–0.846), corrected Brier score was 0.103, and corrected calibration slope was 0.981 (approximate bootstrap interval 0.506–1.646). AUCs after separate exclusions of infections recorded by hospital day 2 or day 3 were 0.750 and 0.731, respectively. Fine–Gray subdistribution hazard ratios were 1.069 (95% CI 1.042–1.098) for NLR and 1.064 (1.021–1.109) for MELD. Conclusion: This admission model showed modest discrimination with imprecise calibration. Retrospective ascertainment and limited validation preclude clinical use. Prospective external validation at a defined prediction time and horizon is needed before the model guides surveillance or treatment.
SN - 1178-6973
ER -