Predicting 1-Year Mortality in Outpatients With Heart Failure With Reduced Left Ventricular Ejection Fraction: Do...

Area of Science:

• Cardiology
• Medical Prognostics
• Health Informatics

Background:

• Physician prognosis estimation accuracy is frequently debated.
• Direct comparisons between physician and model-based predictions in heart failure (HF) are lacking.
• This study addresses the need to compare physician versus model predictive performance for 1-year mortality in HF.

Purpose of the Study:

• To compare the accuracy of physician predictions against established models for 1-year mortality in patients with heart failure (HF).
• To evaluate discrimination, calibration, and risk reclassification capabilities of both physicians and predictive models.
• To inform potential improvements in patient care and resource allocation within HF management.

Main Methods:

• A multicenter prospective cohort study involving 1643 outpatients with HF with reduced ejection fraction across 11 Canadian clinics.
• Clinical data were used to calculate 1-year mortality predictions using the Seattle HF Model (SHFM), HF Meta-Score, and Meta-Analysis Global Group in Chronic HF score.
• HF cardiologists and family physicians, blinded to model outputs, provided independent 1-year mortality predictions. Outcomes included mortality, urgent ventricular assist device implant, or heart transplant over 1 year.

Main Results:

• The Seattle HF Model (SHFM) demonstrated the best discrimination (C-statistic 0.76) and calibration.
• Physicians exhibited similar discrimination (0.75 for cardiologists, 0.73 for family doctors) but significantly overestimated risk (>10%) in both low- and high-risk groups.
• The SHFM showed superior risk reclassification, correctly identifying more patients without events and more accurately assessing risk in patients who experienced events compared to physicians.

Conclusions:

• While physicians demonstrated adequate risk discrimination, they substantially overestimated absolute risk.
• Predictive models, particularly the SHFM, exhibited higher accuracy in predicting 1-year mortality for HF patients.
• Integrating predictive models into clinical practice may enhance patient care and optimize resource utilization in HF management.