(P110) PREDICTING 1-YEAR MORTALITY IN AMBULATORY HEART FAILURE PATIENTS: EMPIRIC MODELS OUTPERFORM PHYSICIAN INTUITIVE ESTIMATES

Background: Accurate prognosis assessment is key in heart failure (HF) care. This multicenter Canadian prospective cohort study compared the accuracy of physician predictions versus model predictions to estimate 1-year mortality in ambulatory patients with HF.

METHODS AND RESULTS: We included consented consecutive ambulatory HF patients (LVEF < 40%) followed at 11 HF clinics in 5 provinces in Canada (2018-2020). Clinical data allowed calculation of model predicted mortality using three models: the Seattle HF Model (SHFM), the MAGGIC score and the HF Meta-score. HF cardiologists and family doctors, blinded to model predictions and using their clinical judgment, estimated patient 1-year mortality. We followed patients for ≥1 year, recorded the composite end point of mortality, urgent heart transplant or ventricular assist device (VAD) implant, and compared the accuracy of model and physician predictions evaluating discrimination (c-statistic), calibration (observed versus predicted event rate), and risk reclassification analysis (absolute net reclassification improvement [NRI]). Using multivariable logistic regression, we evaluated concordance, defined as a < 10% difference in risk estimates, between model and physician predictions. We included 1,563 ambulatory HF patients, aged 65 (IQR 55-74) years, 24% female, and LVEF 28% (23%-31%). The population 1-year event rate was 10% (8%-12%). While the three models overestimated risk in patients with a predicted risk >20%, in lower risk patients, the SHFM showed excellent calibration while the HF Meta-score and the MAGGIC score showed adequate calibration. The SHFM demonstrated the highest discrimination (c-statistic 0.76) in comparison to the HF Meta-score (c-statistic 0.73) and MAGGIC score (c-statistic 0.70). Physicians’ discrimination was similar to that of the models (0.73 for family doctors and 0.75 for HF cardiologists) but calibration was limited by substantial risk overestimation in low- and high-risk patients. In comparison to the SHFM, family doctors and HF cardiologists overestimated risk by >10% throughout the risk spectrum. By risk reclassification, the SHFM better risk-classified 43% of patients in comparison to HF cardiologists and 30% in comparison to family doctors. Discrepancies between model and physician predictions were more frequently due to physician risk overestimation than underestimation and more frequently observed in physicians with fewer years since MD graduation and low confidence in their predictions.

Conclusion: Predictive models showed higher accuracy in predicting 1-year mortality in ambulatory patients with HF. Family doctors and HF cardiologists consistently overestimated risk, potentially exposing patients to more aggressive therapies than necessary. Incorporating predictive models in family and HF cardiology practices may improve patient care and resource use.