A New Risk Score Based on Machine Learning in Patients with Acute Heart Failure: The ML-HF score.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Traditional prognostic scores often inadequately predict mortality in acute heart failure (AHF) patients.
• There is a need for more accurate tools to assess AHF patient mortality risk.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based prognostic score for in-hospital death in AHF patients.
• To compare the performance of the novel ML score against established traditional scores (ADHERE, GWTG-HF).

Main Methods:

• Utilized data from 1157 AHF patients admitted to Brazilian hospitals (2016-2022).
• Collected clinical data, laboratory results, and WHO Quality of Life (WHOQOL-Bref) scores.
• Trained an ML model on 70% of data and validated on 30%, comparing Area Under the ROC Curve (AUC) with traditional scores.

Main Results:

• The ML-HF score identified Physical Health Domain Quality (WHOQOL-BREF), serum sodium, urea, creatinine, and systolic blood pressure as key predictors.
• The ML-HF score demonstrated superior discrimination in the test set (AUC=0.722) compared to GWTG-HF (AUC=0.616) and ADHERE (AUC=0.601).
• Model calibration was adequate (Hosmer-Lemeshow p=0.056).

Conclusions:

• A novel machine learning-based score (ML-HF) was successfully developed and validated for predicting in-hospital mortality in AHF.
• The ML-HF score significantly outperformed conventional prognostic scores, offering improved predictive accuracy for AHF patients.