Machine Learning Model IDs Heart Failure With Reduced Ejection Fraction Using Routine Lab Indicators

FRIDAY, Aug. 28, 2026 (HealthDay News) — A 13-indicator routine laboratory model shows moderate discrimination for distinguishing heart failure with reduced ejection fraction (HFrEF), according to a study published online Aug. 9 in Frontiers in Cardiovascular Medicine.

Zhiping Meng, from the Eighth Affiliated Hospital of Guangxi Medical University in Guigang, China, and colleagues developed and internally evaluated a routine laboratory-based machine-learning model to distinguish HFrEF from HF with mildly reduced (mr) EF/HF with preserved (p) EF. The analysis included 1,480 hospitalized patients with chronic heart failure (377 with HFrEF and 1,103 with HFmrEF/HFpEF) with 13 routine indicators used in six algorithms.

The researchers found that the HFrEF group had higher pro-B-type natriuretic peptide, blood urea nitrogen, total bilirubin, direct bilirubin, gamma-glutamyl transferase, hematocrit, and uric acid levels versus the HFmrEF/HFpEF group. Random forest and XGBoost achieved the highest areas under the receiver operating characteristic curve (AUCs; both 0.789) in the independent test set, whereas logistic regression showed comparable discrimination (AUC, 0.784) and strong calibration (intercept 0.017; slope 1.069). The lowest Brier score (0.152) was seen for random forest, but pairwise bootstrap comparisons showed no significant AUC differences between the models. Sensitivity was limited at the default threshold; however, moving the random-forest threshold to 0.15 increased sensitivity to 0.912 and negative predictive value to 0.935, supporting a rule-out triage strategy for prioritized echocardiography. proBNP was the dominant predictor in a SHapley Additive exPlanations analysis.

“The model should be regarded as an adjunctive triage tool pending external and prospective validation, not as a substitute for echocardiographic phenotyping,” the authors write.

Abstract/Full Text