Why Clinicians Prognosticate Stroke Patients Poorly: Results from the Clinician Judgment versus Risk Score to Predict Stroke Outcomes Randomized Study.

Background: Prognostication after a stroke has important implications for care and for decisions made by patients and their families. It is not clear why clinicians, even experienced stroke neurologists, poorly estimate the risk of disability and death following stroke.

Methods: We analyzed the results from the Clinician Judgment versus Risk Score to predict Stroke Outcomes study in which each clinician estimated the risk of death and the risk of death or disability in 5 case-based ischemic stroke scenarios. We employed a mixed-effect linear model to disentangle the ability of clinicians to discriminate between poor and good prognosis cases (slope) from the calibration of quantitative estimates (intercept), and to assess for any effect of anchoring in the death or disability condition (through a comparison with the death condition).

Results: One hundred eleven clinicians made 1665 predictions. Clinicians were able to discriminate between cases with low and high risks of death (slope of .81, 95% confidence interval [CI] .70-.93), but the quantitative estimates were not well calibrated (intercept of 5.14, 95% CI 3.97-6.33). The discrimination was poorer (slope of .67, 95% CI .60-.75), but the calibration was better (intercept of -.34, 95% CI -5.43 to 4.98) in the death or disability estimates.

Conclusion: Poor stroke prognostication can be explained by poor calibration and an anchoring effect, which are both amenable to specific training interventions.