"Mortality Prediction in Heart Failure using Explainable AI: Development and Validation of the EXACT-HF Model in a Nationwide Multi-Ethnic Asian Cohort"
- 1National University of Singapore
Abstract
BackgroundEstablished heart-failure (HF) risk scores were developed largely in Western cohorts, assume linear predictor effects, and stop at prognosis without indicating action. We developed and externally validated machine-learning (ML) models for 30-day and 1-year mortality in a multi-ethnic Asian HF population, benchmarked against three established scores, and tested whether they localised correctable guideline-directed medical therapy (GDMT) gaps.
MethodsWe analysed 9,050 adults hospitalised with HF across eight Singapore hospitals (November 2016-March 2023). Gradient-boosted trees (XGBoost, primary) and elastic-net logistic regression were developed on three hospitals (n=4,034) and validated on five independent hospitals (n=5,016), without patient or site overlap. Comparators (MAGGIC, Singapore HF Score, OPTIMIZE-HF-Asia) were logistically recalibrated on development data.
ResultsOn external validation XGBoost achieved AUROCs of 0.862 (95% CI 0.833-0.890) for 30-day and 0.749 (0.734-0.764) for 1-year mortality, versus 0.711-0.754 and 0.682-0.699 for the three recalibrated scores; logistic regression performed near-identically (0.864, 0.757) and was better calibrated at 1 year. Superiority held under temporal validation, multiple imputation and optimism correction, although the 30-day advantage was no longer distinguishable from MAGGIC once medications were removed. Among predicted-high-risk patients with reduced ejection fraction, mean GDMT exposure was 0.92 of four pillars versus 1.87 across all HFrEF patients, and 36.3% received none of the four versus 2.0% of low-risk patients. Ten variables reproduced full-model 30-day discrimination (AUROC 0.865).
ConclusionsIn external validation, interpretable models substantially outperformed established HF risk scores for short- and long-term mortality and simultaneously localised correctable therapeutic gaps, supporting a prediction-to-action tool deployable with ten bedside variables.
Clinical PerspectiveO_ST_ABSWhat Is New?C_ST_ABSO_LIIn 9,050 patients hospitalised for heart failure across eight Singapore hospitals, prediction models estimated 30-day and 1-year all-cause mortality substantially better than three established risk scores (30-day AUROC 0.86 versus 0.71-0.75), validated in five hospitals independent of model development. C_LIO_LITen routinely recorded bedside variables reproduced the full 85-variable models 30-day discrimination. This enables bedside risk estimation without electronic health record integration. C_LI
What Are the Clinical Implications?O_LIThe same model identifies both who is at risk and which guideline-directed therapies they lack: flagged patients were receiving a mean of 0.9 of the four pillars, against 1.9 across all patients with reduced ejection fraction, although most non-prescription had a documented renal or haemodynamic contraindication, so this correctable fraction is an upper bound requiring prospective validation. C_LI
Related
- Development and validation of an explainable machine learning-based risk prediction model for macrosomia in han Chinese women with singleton cephalic pregnancies
- Development and validation of an interpretable machine learning model for early prediction of mortality in ICU patients with chronic obstructive pulmonary disease and heart failure
- Admission-time prediction of 7-day mortality in clinician-attributed hypertensive intracerebral hemorrhage: development and external validation against the ICH score
- Prediction of in-hospital mortality at a 72-hour intensive care landmark in patients with sepsis and pre-existing hypertension: development and external validation
- Development and external validation of a preprocedural prediction model for 1-year mortality after transcatheter aortic valve implantation
BibTeX
@misc{jauhari2026mortality,
title = {{"Mortality Prediction in Heart Failure using Explainable AI: Development and Validation of the EXACT-HF Model in a Nationwide Multi-Ethnic Asian Cohort"}},
author = {S. M. Jauhari and W. L. Yeung and T. Li and J. Chua and A. A. Zhang and E.-H. Lim and R. Cherian and D. Sim and S. Lee Shao Guang and D. K. N. Soon and P. F. Chan and S. Y. Loh and F. R. Jaufeerally and S.-P. Chan and M. Y.-Y. Chan and W. Lin and R. C. C. Wong},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.10.26362800},
url = {https://www.medrxiv.org/content/10.64898/2026.09.10.26362800v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - "Mortality Prediction in Heart Failure using Explainable AI: Development and Validation of the EXACT-HF Model in a Nationwide Multi-Ethnic Asian Cohort"
AU - S. M. Jauhari
AU - W. L. Yeung
AU - T. Li
AU - J. Chua
AU - A. A. Zhang
AU - E.-H. Lim
AU - R. Cherian
AU - D. Sim
AU - S. Lee Shao Guang
AU - D. K. N. Soon
AU - P. F. Chan
AU - S. Y. Loh
AU - F. R. Jaufeerally
AU - S.-P. Chan
AU - M. Y.-Y. Chan
AU - W. Lin
AU - R. C. C. Wong
PY - 2026
DA - 2026/09/10
PB - medRxiv
DO - 10.64898/2026.09.10.26362800
UR - https://www.medrxiv.org/content/10.64898/2026.09.10.26362800v1
AB - BackgroundEstablished heart-failure (HF) risk scores were developed largely in Western cohorts, assume linear predictor effects, and stop at prognosis without indicating action. We developed and externally validated machine-learning (ML) models for 30-day and 1-year mortality in a multi-ethnic Asian HF population, benchmarked against three established scores, and tested whether they localised correctable guideline-directed medical therapy (GDMT) gaps. MethodsWe analysed 9,050 adults hospitalised with HF across eight Singapore hospitals (November 2016-March 2023). Gradient-boosted trees (XGBoost, primary) and elastic-net logistic regression were developed on three hospitals (n=4,034) and validated on five independent hospitals (n=5,016), without patient or site overlap. Comparators (MAGGIC, Singapore HF Score, OPTIMIZE-HF-Asia) were logistically recalibrated on development data. ResultsOn external validation XGBoost achieved AUROCs of 0.862 (95% CI 0.833-0.890) for 30-day and 0.749 (0.734-0.764) for 1-year mortality, versus 0.711-0.754 and 0.682-0.699 for the three recalibrated scores; logistic regression performed near-identically (0.864, 0.757) and was better calibrated at 1 year. Superiority held under temporal validation, multiple imputation and optimism correction, although the 30-day advantage was no longer distinguishable from MAGGIC once medications were removed. Among predicted-high-risk patients with reduced ejection fraction, mean GDMT exposure was 0.92 of four pillars versus 1.87 across all HFrEF patients, and 36.3% received none of the four versus 2.0% of low-risk patients. Ten variables reproduced full-model 30-day discrimination (AUROC 0.865). ConclusionsIn external validation, interpretable models substantially outperformed established HF risk scores for short- and long-term mortality and simultaneously localised correctable therapeutic gaps, supporting a prediction-to-action tool deployable with ten bedside variables. Clinical PerspectiveO_ST_ABSWhat Is New?C_ST_ABSO_LIIn 9,050 patients hospitalised for heart failure across eight Singapore hospitals, prediction models estimated 30-day and 1-year all-cause mortality substantially better than three established risk scores (30-day AUROC 0.86 versus 0.71-0.75), validated in five hospitals independent of model development. C_LIO_LITen routinely recorded bedside variables reproduced the full 85-variable models 30-day discrimination. This enables bedside risk estimation without electronic health record integration. C_LI What Are the Clinical Implications?O_LIThe same model identifies both who is at risk and which guideline-directed therapies they lack: flagged patients were receiving a mean of 0.9 of the four pillars, against 1.9 across all patients with reduced ejection fraction, although most non-prescription had a documented renal or haemodynamic contraindication, so this correctable fraction is an upper bound requiring prospective validation. C_LI
ER -