A registry-based algorithm to predict ejection fraction in patients with heart failure.

Area of Science:

• Cardiology
• Medical Informatics
• Epidemiology

Background:

• Left ventricular ejection fraction (EF) is crucial for classifying heart failure (HF) into HF with preserved (HFpEF), mid-range (HFmrEF), and reduced (HFrEF) EF categories.
• EF is frequently unavailable in large population-based cohorts and non-HF registries, hindering comprehensive HF research.

Purpose of the Study:

• To develop and validate an algorithm for identifying EF subphenotypes (HFpEF, HFmrEF, HFrEF) using routinely collected clinical characteristics.
• To enable robust research on HF within big data settings where EF measurements are often absent.

Main Methods:

• Utilized logistic regression and multivariable multinomial analysis on 42,061 HF patients from the Swedish Heart Failure Registry.
• Included 22 clinical variables to predict EF categories: EF ≥50% vs. <50%, EF ≥40% vs. <40%, and the three distinct subphenotypes (HFrEF, HFmrEF, HFpEF).
• Validated the models using data from the CHECK-HF study (10,627 patients) in the Netherlands.

Main Results:

• The algorithm demonstrated good discriminative ability for predicting HFpEF (C-statistic 0.78) and HFrEF (C-statistic 0.76).
• Accuracy for identifying HFmrEF was lower (C-statistic 0.63).
• External validation confirmed similar predictive performance in an independent cohort.

Conclusions:

• Routine clinical data can effectively identify HF subphenotypes, particularly HFpEF and HFrEF, in large datasets lacking direct EF measurements.
• The developed algorithm facilitates large-scale research on HF epidemiology and outcomes.
• Further refinement may be needed to improve the accuracy of HFmrEF classification.