A Prospective Randomized Trial of Personalized Nudges to Increase Influenza Vaccinations

Personalized messages that use machine learning to tailor reminders that are predicted to be most effective have the power to substantially increase healthy behaviors, and there is a need for more randomized evaluations to validate the approach. In two previous randomized trials, including the flu-shot "megastudy" (Milkman et al., 2022), we tested 18 different flu-shot nudge messages sent directly to patients with upcoming appointments. A subsequent analysis of the megastudy used a machine learning model to retrospectively identify nudge messages most effective for different patient subpopulations. Results suggested that overall flu vaccination rates would have been substantially higher had patients been sent the most effective message for their subpopulation. In the present study, we will test, prospectively, whether personalizing nudges according to a decision policy derived from those results increases flu vaccination relative to other approaches. Over 60,000 patients with scheduled flu shot-eligible appointments at Geisinger will be randomized to one of several study arms, including a passive control arm, a "best nudge" that had the highest effectiveness overall, a random nudge, and a personalized nudge arm applying the megastudy-derived algorithm. Our primary outcome will be flu vaccination on the day of the appointment or in the three days prior to the appointment, as recorded in the electronic health system (EHR). We will also examine the timing of flu shot receipt over the flu season.