Age-dependent association between BUN/Cr ratio and in-hospital mortality in older patients with atrial fibrillation: a multicenter explainable machine learning study

The prognostic significance of the blood urea nitrogen-to-creatinine ratio (BUN/Cr) across older age groups remains unclear. We investigated its association with in-hospital mortality in older critically ill patients with atrial fibrillation (AF) and developed an explainable machine learning prediction model. This retrospective study included patients aged ≥ 65 years with AF who were alive and remained in the intensive care unit (ICU) at 24 h after admission. Multivariable logistic regression and restricted cubic splines assessed associations between BUN/Cr and all-cause in-hospital mortality in the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Age-related effect modification was evaluated using formal interaction tests with multiple-testing correction. Five machine learning algorithms were developed using a 70% training set and compared in a 30% internal validation set. The random forest model underwent external validation in the eICU Collaborative Research Database (eICU) and interpretation using SHapley Additive exPlanations (SHAP). The MIMIC-IV cohort included 12,887 patients, of whom 1,892 (14.68%) died during hospitalization; the external cohort included 6,879 patients. Compared with BUN/Cr 10–20, ratios < 10 and > 20 were associated with higher adjusted mortality odds: odds ratios (ORs) of 1.45 (95% confidence interval [CI], 1.11–1.90) and 1.56 (1.39–1.76), respectively. The overall association was U-shaped (P for nonlinearity < 0.001). Age interactions remained significant after Holm correction. Nonlinearity was evident at ages 65–74 and 75–84 years, but not ≥ 85 years ( P = 0.338). Random forest achieved areas under the receiver operating characteristic curve (AUCs) of 0.802 internally and 0.754 externally, with Brier scores of 0.103 and 0.119. Its internal discrimination exceeded Sequential Organ Failure Assessment (SOFA), Acute Physiology Score III (APS III), and an exploratory CHADS2-like model (heart failure, hypertension, age ≥ 75 years, diabetes mellitus, and stroke or transient ischemic attack) (all Holm-adjusted P < 0.001). A web-based calculator was developed. BUN/Cr showed a nonlinear association with in-hospital mortality that varied with age. The random forest model demonstrated discrimination across two databases, supporting further evaluation for individualized risk assessment.

Authors

Institutions

Publication Details

Journal
BMC Cardiovascular Disorders
Published
2026-11-14
DOI
https://doi.org/10.1186/s12872-026-07018-7
Primary Topic
Sepsis Diagnosis and Treatment
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Age-dependent association between BUN/Cr ratio and in-hospital mortality in older patients with atrial fibrillation: a multicenter explainable machine learning study

薛莉, Hongshi Li, Heng Cai, Hongmei Zheng et al.
BMC Cardiovascular Disorders
Sepsis Diagnosis and Treatment
article

Age-dependent association between BUN/Cr ratio and in-hospital mortality in older patients with atrial fibrillation: a multicenter explainable machine learning study

薛莉, Hongshi Li, Heng Cai, Hongmei Zheng, Yuxin Zhu, Chongzhe Pei
article en

Abstract

The prognostic significance of the blood urea nitrogen-to-creatinine ratio (BUN/Cr) across older age groups remains unclear. We investigated its association with in-hospital mortality in older critically ill patients with atrial fibrillation (AF) and developed an explainable machine learning prediction model. This retrospective study included patients aged ≥ 65 years with AF who were alive and remained in the intensive care unit (ICU) at 24 h after admission. Multivariable logistic regression and restricted cubic splines assessed associations between BUN/Cr and all-cause in-hospital mortality in the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Age-related effect modification was evaluated using formal interaction tests with multiple-testing correction. Five machine learning algorithms were developed using a 70% training set and compared in a 30% internal validation set. The random forest model underwent external validation in the eICU Collaborative Research Database (eICU) and interpretation using SHapley Additive exPlanations (SHAP). The MIMIC-IV cohort included 12,887 patients, of whom 1,892 (14.68%) died during hospitalization; the external cohort included 6,879 patients. Compared with BUN/Cr 10–20, ratios < 10 and > 20 were associated with higher adjusted mortality odds: odds ratios (ORs) of 1.45 (95% confidence interval [CI], 1.11–1.90) and 1.56 (1.39–1.76), respectively. The overall association was U-shaped (P for nonlinearity < 0.001). Age interactions remained significant after Holm correction. Nonlinearity was evident at ages 65–74 and 75–84 years, but not ≥ 85 years ( P = 0.338). Random forest achieved areas under the receiver operating characteristic curve (AUCs) of 0.802 internally and 0.754 externally, with Brier scores of 0.103 and 0.119. Its internal discrimination exceeded Sequential Organ Failure Assessment (SOFA), Acute Physiology Score III (APS III), and an exploratory CHADS2-like model (heart failure, hypertension, age ≥ 75 years, diabetes mellitus, and stroke or transient ischemic attack) (all Holm-adjusted P < 0.001). A web-based calculator was developed. BUN/Cr showed a nonlinear association with in-hospital mortality that varied with age. The random forest model demonstrated discrimination across two databases, supporting further evaluation for individualized risk assessment.

BMC Cardiovascular Disorders
Tianjin Medical University General Hospital (CN)
Sepsis Diagnosis and Treatment
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon