Key Points
Question How are behavioral interventions targeting physician performance associated with clinical decision-making?
Findings In this process evaluation of a randomized clinical trial, exposure to an educational trauma triage video game was associated with a moderate increase in the willingness of emergency department physicians to transfer injured patients to trauma centers and a smaller improvement in the recognition of severely injured patients. There was limited heterogeneity of the estimated treatment effect.
Meaning These findings suggest the potential of educational video games to change physicians’ willingness to adhere to clinical practice guidelines, and therefore may serve as useful adjuncts to existing continuing medical education efforts.
Abstract 
Importance Understanding how behavioral interventions work and whom they help can increase their effectiveness.
Objective To evaluate the mechanism of action and heterogeneity of the estimated treatment effect of a customized video game (Night Shift) designed to recalibrate physician heuristics (pattern recognition) in trauma triage.
Design, Setting, and Participants This process evaluation of a randomized clinical trial was performed with a national sample of 800 physicians responsible for the triage of patients with trauma at the emergency departments (EDs) of levels III, IV, and V trauma centers and nontrauma centers in the US. Data were collected online from November 27, 2023, to March 11, 2024. Data were analyzed based on intention to treat.
Interventions Usual education or customized video game played for 2 hours. All participants completed a virtual simulation, mimicking 3 ED shifts.
Main Measures and Outcomes The intervention’s mechanism of action was analyzed using signal detection theory, which describes decision-making as the product of perceptual sensitivity (the ability to recognize signal [severe injuries] and noise [minor injuries]) and decisional threshold (tolerance for false-positive or false-negative decisions). The heterogeneity of the estimated treatment effect was evaluated using prespecified subgroup analyses to test moderation by participant characteristics (ie, sex, age, and clinical volume). Findings were validated using a data-driven approach with bayesian additive regression trees.
Results The 800 participants (566 [71%] male; mean [SD] age, 43.8 [9.4] years) had mean (SD) professional experience of 12.0 (8.4) years, worked at nontrauma centers (488 [61%]) or at level III, IV, or V trauma centers (312 [39%]), and were board-certified in emergency medicine (673 [84%]). Most intervention participants (339 [85%]) played the customized video game for at least 2 hours or until they completed the content, and most (345 of 398 [87%] for the intervention and 231 of 397 [58%] for the control) used the simulation. Assignment to the intervention arm was associated with a reduction in undertriage (22% vs 38%; percentage point difference, 16 [95% CI, 15-18]; P < .001). The intervention was associated with a moderate increase in tolerance for false-positive decisions (intervention 0.14 SD units [95% CI, 0.07-0.22]; control 0.53 SD units [95% CI, 0.43-0.63]; Cohen d = 0.6) and a small improvement in the ability to recognize severely injured patients (intervention 1.00 SD units [95% CI, 0.94-1.07]; control 0.87 SD units [95% CI, 0.79-0.94]; Cohen d = 0.2). Limited heterogeneity of the estimated treatment effect was observed, although participants’ clinical volume was associated with moderation.
Conclusions In this process evaluation of a randomized clinical trial, exposure to a theory-based video game was associated with liberalized thresholds for transfer and limited heterogeneity of the estimated treatment effect.
