Key result
Predictive models like the Seattle Heart Failure Model (C statistic 0.76) provided more accurate 1-year prognostic estimates than physicians, who substantially overestimated absolute risk.
Why the study?
Physicians often err in estimating prognosis, and no studies had directly compared physician versus model predictive performance for mortality in heart failure.
Do predictive models outperform physician intuitive estimates in predicting 1-year mortality in outpatients with HFrEF?
Cohort (n=1,643)
Blinded to model predictions
Yes
Do predictive models outperform physician intuitive estimates in predicting 1-year mortality in outpatients with HFrEF?
Effect estimate: C statistic 0.76 (SHFM)
Predictive models like the Seattle Heart Failure Model provide better calibration and risk reclassification for 1-year mortality in HFrEF outpatients compared to physician estimates, which tend to overestimate risk.
Model-based estimates may improve 1-year mortality accuracy over physician judgment in HFrEF outpatients; leaves open effects on decisions or outcomes.
BACKGROUND: Many studies have demonstrated that physicians often err in estimating patient prognosis. No studies have directly compared physician to model predictive performance in heart failure (HF). We aimed to compare the accuracy of physician versus model predictions of 1-year mortality. METHODS: This multicenter prospective cohort study on 11 HF clinics in 5 provinces in Canada included consecutive consented outpatients with HF with reduced left ventricular ejection fraction (<40%). By collecting clinical data, we calculated predicted 1-year mortality using the Seattle HF Model (SHFM), the Meta-Analysis Global Group in Chronic HF score, and the HF Meta-Score. HF cardiologists and family doctors, blinded to model predictions, estimated patient 1-year mortality. During 1-year follow-up, we recorded the composite end point of mortality, urgent ventricular assist device implant, or heart transplant. We compared physicians and model discrimination (C statistic), calibration (observed versus predicted event rate), and risk reclassification. RESULTS: The study included 1643 patients with ambulatory HF with a mean age of 65 years, 24% female, and mean left ventricular ejection fraction of 28%. Over 1-year follow-up, 9% had an event. The SHFM had the best discrimination (SHFM C statistic 0.76; HF Meta-Score 0.73; Meta-Analysis Global Group in Chronic Heart Failure 0.70) and calibration. Physicians' discrimination differed little (0.75 for HF cardiologists and 0.73 for family doctors) but both physician groups substantially overestimated risk by >10% in both low- and high-risk patients (poor calibration). In risk reclassification analysis, among patients without events, the SHFM better classified 51% in comparison to HF cardiologists and 43% in comparison to family doctors. In patients with events, the SHFM erroneously assigned lower risk to 44% in comparison to HF cardiologists and 34% in comparison to family doctors. CONCLUSIONS: Family doctors and HF cardiologists showed adequate risk discrimination, with however substantial overestimation of absolute risk. Predictive models showed higher accuracy. Incorporating models in family and HF cardiology practices may improve patient care and resource use in HF with reduced left ventricular ejection fraction. REGISTRATION: URL: https://www. CLINICALTRIALS: gov; Unique identifier: NCT04009798.
No takes yet. Share an insight, caveat, or question.
Alba et al. (2023) conducted a cohort in Heart failure with reduced left ventricular ejection fraction (n=1,643). Predictive models (Seattle HF Model, MAGGIC, HF Meta-Score) vs. Physician intuitive estimates (HF cardiologists and family doctors) was evaluated on Composite end point of mortality, urgent ventricular assist device implant, or heart transplant (C statistic 0.76 (SHFM)). Predictive models like the Seattle Heart Failure Model (C statistic 0.76) provided more accurate 1-year prognostic estimates than physicians, who substantially overestimated absolute risk.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: