An Electronic Health Record “Nudge” to Reduce Antipsychotic Medication Prescribing for Primary Care Patients With Dementia: A Pragmatic RCT

BACKGROUND: We hypothesized that an electronic health record (EHR) alert grounded in behavioral science could reduce new antipsychotic medication prescriptions for older adults with dementia. METHODS: We conducted a pragmatic clinical trial at a large academic health system, randomizing providers to receive the alert (intervention) or not (control) when prescribing new antipsychotic medications to outpatients with dementia. Eligible providers were those who previously signed a new antipsychotic prescription (n = 150). The EHR alert contained: (1) text stating that antipsychotic medications increase mortality risk; (2) a link to an after-visit summary handout describing non-pharmaceutical approaches; (3) a default to a lower dose and pill-days. The primary outcome was mean total pill-days over 19 months. We used provider-level linear regression, controlling for provider characteristics including prior prescribing behavior. Secondary outcomes included order cancellations and exploratory analyses of treatment effects among providers with above-average baseline prescribing patterns. RESULTS: Between 9/15/2021 and 4/11/2023, 28 providers in the intervention and 21 in the control arm initiated prescriptions to eligible patients, resulting in 139 enrolled patients; mean patient age was 83 (SD 9.7); 67% female. After 19 months, raw mean (SD) pill-days were 126 (228) for the intervention and 225 (601) for the control group. Intervention assignment was not significantly associated with total pill-days among enrolled patients (primary outcome, mean difference -113 95% CI: -256, +30, p = 0.12). Orders were canceled in 12/58 (21%) of intervention encounters. The intervention was estimated to reduce pill-days for providers with average baseline prescribing behavior (-126, 95% CI: -3, -248; p = 0.04); for each additional 10 baseline pill-days, this reduction effect increased by -8.8 pill-days (95% CI: -3.9, -13.6; p < 0.001). CONCLUSIONS: This EHR alert did not reduce the primary outcome of antipsychotic pill-days. However, the nonsignificant effect size was substantial, and exploratory analyses revealed that the intervention significantly reduced pill-days for providers with average baseline prescribing. Future alerts should test targeting high prescribers.