Modeling surgeon belief updating under bias: a Bayesian simulation in shoulder arthroplasty

Shoulder surgeons often offer markedly different treatment recommendations for a given patient and pathophysiology within the context of a common body of evidence. Prior beliefs and cognitive bias may contribute to this variability. Using Bayesian decision theory informed by principles of behavioral science, we modeled how surgeons with different initial beliefs and degrees of bias update their treatment preferences in response to new evidence comparing anatomic total shoulder arthroplasty (aTSA) and reverse total shoulder arthroplasty (rTSA) when making surgery recommendations for advanced primary glenohumeral osteoarthritis.

We developed a Bayesian simulation involving 3 hypothetical surgeons with distinct initial beliefs: aTSA Loyalist (90% belief aTSA is superior), Neutral Thinker (50%), and rTSA Advocate (10% belief aTSA is superior) and varying degrees of confirmation bias (eg, the selective discounting of evidence that contradicts one's current belief). Each surgeon was sequentially exposed to 10 simulated randomized trials modestly favoring rTSA, with belief trajectories updated after each trial under 2 conditions: (1) an unbiased scenario, in which all new evidence was weighted at face value, and (2) a biased scenario, in which disconfirming evidence was systematically downweighted.

Under a condition simulating no confirmation bias, all surgeons gradually shifted toward lower belief in aTSA superiority as they reviewed the 10 rTSA-favoring trials: the aTSA Loyalist moved from about 90% to 25% confidence in aTSA, the Neutral Thinker from 50% to 2.0%, and the rTSA Advocate from 10% to 1.2%. Under biased conditions, belief change was markedly reduced for the aTSA Loyalist, who remained 64% confident in aTSA superiority despite consistent rTSA-favoring evidence. Changes for the rTSA Advocate (10%-1.2%) and Neutral Thinker (50%-2.3%) were largely unchanged.

This Bayesian simulation provides a practical framework to demonstrate how prior beliefs and cognitive bias can markedly influence the way shoulder surgeons interpret and act upon new evidence, contributing to unwarranted variation in care. When it comes to treatment recommendations, what surgeons believe at the outset may matter as much or more than the data itself. Implementing targeted strategies such as foundational principles based in behavioral ethics, evidence-based decision and debiasing aids, structured peer review, and routine performance feedback may help align treatment decisions more closely with a patient's values based on the best available evidence. The use of rTSA-favoring evidence in this simulation is solely for illustrative purposes and should not be interpreted as an endorsement of increased rTSA use in clinical practice.