Background: Acute heart failure (AHF) poses significant diagnostic challenges in emergency room (ER) settings due to its varied clinical presentations and the limitations of traditional diagnostic methods. This study aims to develop and evaluate a deep learning model that uses electrocardiogram (ECG) data to enhance the identification of AHF in ER patients. Methods: In this retrospective cohort study, we analyzed ECG data from 19,285 patients who visited the ERs of three hospitals between 2016 and 2020. Among the patients, 9,119 patients with available left ventricular ejection fraction and NT-proBNP levels, who were diagnosed with AHF were included. We extracted morphological and clinical parameters from the ECG data to train and validate four machine learning models: a baseline linear regression and more advanced models including XGBoost, Light GBM, and CatBoost. Results: The CatBoost algorithm outperformed other models, showing superior AUROC and AUPRC diagnostic accuracy across both internal (0.89±0.01 and 0.89±0.01) and external validation datasets (0.90 and 0.89). In addition, the model demonstrated high metrics in accuracy, precision, recall, and f1-score, indicating robust performance in the identification of AHF. Conclusion: The developed machine learning model significantly enhances the detection of AHF in ER patients using conventional 12-lead ECGs combined with clinical data. These findings suggest that ECGs, a commonplace tool in ERs, can effectively screen for AHF.

Background: Acute heart failure (AHF) poses significant diagnostic challenges in emergency room (ER) settings due to its varied clinical presentations and the limitations of traditional diagnostic methods. This study aims to develop and evaluate a deep learning model that uses electrocardiogram (ECG) data to enhance the identification of AHF in ER patients. Methods: In this retrospective cohort study, we analyzed ECG data from 19,285 patients who visited the ERs of three hospitals between 2016 and 2020. Among the patients, 9,119 patients with available left ventricular ejection fraction and NT-proBNP levels, who were diagnosed with AHF were included. We extracted morphological and clinical parameters from the ECG data to train and validate four machine learning models: a baseline linear regression and more advanced models including XGBoost, Light GBM, and CatBoost. Results: The CatBoost algorithm outperformed other models, showing superior AUROC and AUPRC diagnostic accuracy across both internal (0.89±0.01 and 0.89±0.01) and external validation datasets (0.90 and 0.89). In addition, the model demonstrated high metrics in accuracy, precision, recall, and f1-score, indicating robust performance in the identification of AHF. Conclusion: The developed machine learning model significantly enhances the detection of AHF in ER patients using conventional 12-lead ECGs combined with clinical data. These findings suggest that ECGs, a commonplace tool in ERs, can effectively screen for AHF.

Myocardial damage after exposure to high-voltage electrical current has been reported to result in a serious and often life-threatening situation. However, because the diagnosis of myocardial injur...

The prognostic implications of BMI in AHF differ according to HF phenotype and sex. Lower BMI is a consistent adverse marker in HFrEF in both sexes and shows a borderline adverse association in women with HFpEF, whereas BMI is not prognostic in men with HFpEF. These findings highlight the importance of sex- and phenotype-specific interpretation of BMI in risk assessment.

In this nationwide secondary prevention cohort, the risk of NODM differed across individual statins despite comparable cardiovascular outcomes. These findings suggest that the diabetogenic effect of statins may be agent-specific rather than a uniform class effect, highlighting the importance of individualized statin selection balancing metabolic and cardiovascular benefits.

ClinicalTrials.gov Identifier: NCT01861080.