Musculoskeletal surgeons use mixed reasoning rather than pure Bayesian strategies in clinical practice

OBJECTIVES: To inform efforts to promote regular and normalized Bayesian reasoning, we studied factors associated with the degree to which surgeons use Bayesian reasoning to navigate uncertainty across different clinical scenarios. METHODS: Science of Variation Group members (153; 58% North America, 30% Europe, 69% over 15 years of experience) completed an online survey reading 8 scenarios of test and treatment decisions and chose one of 4 answer options with higher scores indicating more Bayesian reasoning. Internal consistency of the survey was assessed using Cronbach alpha. RESULTS: The average Bayesian reasoning score across all scenarios was 3.0 (IQR 2.7-3.2) on a 4-point scale, indicating a relative context-dependent variability. Completely non-Bayesian reasoning was selected least often (8.6%, 90 of 1,044) and fully Bayesian reasoning represented 29% (301 of 1,044) of responses. Most surgeons showed mixed patterns (defined as reasoning in which prior probability is acknowledged but underweighted, without explicit probabilistic updating): 85% (121 of 142) used fully Bayesian reasoning at least once (121 of 142) while 42% (60 of 142) used completely non-Bayesian reasoning at least once. The Cronbach alpha was 0.43 suggesting the scenarios measured different aspects of clinical reasoning rather a unified construct. CONCLUSIONS: The finding that surgeons use relatively context-dependent reasoning suggests an opportunity for surgeons to develop and practice Bayesian reasoning strategies both in training programs and in practice.