(CCSP067) FACTORS IMPACTING PHYSICIAN PROGNOSTIC ACCURACY IN HEART FAILURE PATIENTS WITH REDUCED LEFT VENTRICULAR EJECTION FRACTION

Background: In a recent multicenter Canadian prospective cohort study we found that model predictions were significantly more accurate than heart failure (HF) cardiologists and family doctors to estimate 1-year mortality in ambulatory HF patients. In this study, we evaluate patient- and physician-related factors associated with physician accuracy.

METHODS AND RESULTS: We included consented consecutive HF outpatients (LVEF < 40%) followed at 11 HF clinics in Canada. HF cardiologists and family doctors estimated patient 1-year mortality using their clinical judgment in a closed-question survey. We collected clinical information to calculate model predicted mortality using the Seattle HF Model (SHFM). We followed patients for at least a year and recorded their composite end point of urgent heart transplant, ventricular assist device (VAD) implant or mortality. By defining discordancy as a >10% difference between model and physician predictions, we evaluated factors, including physician experience and confidence in estimates, duration of patient-physician relationship, patient-physician sex concordance, and predicted risk, associated with discordant results using multivariable logistic regression. In a sensitivity analysis including high-confidence estimates, we compared the accuracy of model and physician predictions evaluating discrimination (c-statistic) and calibration (observed vs predicted event rate).
We included 1,643 patients, with a 1-year event rate of 10% (8%-12%). Half of the estimates showed discrepant results between model and physician predictions, mainly due to physician risk overestimation. Discrepancies were more frequent with increasing patient 1-year mortality from 38% in low-risk patients (1-year predicted mortality < 5%) to ~75% in high-risk patients (1-year predicted mortality >30%) (Figure 1). Low confidence in predictions was associated with discrepant results. In a sensitivity analysis including only physician estimates with high confidence, the SHFM showed excellent discrimination (c-statistic 0.77) like HF cardiologists (c-statistic 0.79) but superior to family doctors (c-statistic 0.65); in comparison to the SHFM, family doctors and HF cardiologists overestimated risk by >5% throughout the risk spectrum (Figure 2). Female HF cardiologists were 27% more likely to have discordant predictions (OR 0.73, 95%CI:0.59-0.90) than their male counterparts when making predictions on male patients. Duration of patient-physician relationship, years from graduation, or status of therapy optimization were not associated with discrepant estimates.

Conclusion: Higher patient risk and low confidence in estimates were associated with increased discrepancies between physician and model estimates. Among highly confident estimates, however, physician accuracy was inferior to the SHFM with significant risk overestimation from low to high-risk patients. Integrating model estimates into practice may guide informed decision making.