Do clinicians always maximize patient outcomes? A conjoint analysis of preferences for carotid artery testing

The value clinicians place on diagnostic information is subject to psychological influences and systematic biases, but there is limited evidence of how these factors may affect patients' health outcomes. We assessed the relative value attached by experienced clinicians to different diagnostic test characteristics and how their preferences relate to patient outcomes, focusing on strategies for testing symptomatic patients for carotid artery stenosis.

Using conjoint analysis, experienced neurologists and vascular surgeons ranked 10 diagnostic strategies defined in terms of four characteristics. Clinicians' preferences were analysed using an ordered probit model and compared with those obtained using a risk neutral expected value (EV) model developed to predict the consequences of each strategy as if the clinicians' sole goal were to optimize patient outcome. Results were tested for internal consistency and robustness to key model assumptions.

Preferences for positive predictive value (PPV), relative to negative predictive value (NPV), elicited from the clinicians diverged substantially from those estimated by the EV model based on 5-year stroke-free survival (ratios of -0.8 and -32.8, respectively). Conversely, preferences for NPV, relative to test morbidity, from the two models matched closely.

Clinicians attached substantially more importance to the PPV of carotid artery tests than would be justified by their impact on patient outcomes. Cognitive errors and attitudes to risk are likely to play an important role in explaining this finding. This study casts doubts on the validity of common assumptions made in the evaluation of health interventions, and in clinical and policy decisions.