Herding-like behaviour in medical decision making: An experimental study investigating general practitioners' prescription behaviour.

Previous observational studies have indicated that social influences, such as arising from herding-like behaviour, can contribute to medical errors. In this study, we experimentally examined whether general practitioners (GPs) would follow incorrect prescription recommendations from fellow GP or specialists. To investigate this, we conducted an online survey with 475 GPs practicing in England that included two case vignettes. Case vignette 1 focused on sleeping tablets, and case vignette 2 was centred around antibiotics. The vignettes were presented in random order, and within each vignette, study participants were assigned to one of three experimental conditions: control condition (lacking peer recommendation), fellow GP condition (including a recommendation from a fellow GP not aligned with best practice clinical guidelines), or specialist condition (including a recommendation from a specialist not aligned with best practice clinical guidelines). The primary outcome measure was the proportion of GPs who prescribed medication that deviated from best practice clinical guidelines. We found that, in both case vignettes, the percentage of respondents prescribing such medication was highest in those assigned to the specialist condition, followed by those assigned to the control condition. It was lowest in those assigned to the fellow GP condition (case vignette 1: 73.8% vs. 55.6% vs. 36.6% and case vignette 2: 24.0% vs. 12.4% vs. 10.1%). In the case of vignette 1, the difference between the fellow GP condition and the control condition is statistically significant, suggesting that GPs are less likely to prescribe sleeping tablets when recommended by a fellow GP. This implies that GPs are more inclined to prescribe non-guideline-recommended medication when advised by specialists. This study is the first to experimentally demonstrate that physician herding behaviour can result in prescription errors. Future research could extend this inquiry to diverse contexts, including diagnosis.