Abstract
On average, 8% of the US population gets sick from influenza each flu season. Since 2010, the annual disease burden of influenza in the U.S. has included 9-41 million illnesses, 140,000-710,000 hospitalizations, and 12,000-52,000 deaths. The CDC recommends flu vaccination to everyone aged 6 months and older, with rare exceptions; almost anyone can benefit from the vaccine, which can reduce illnesses, missed work, hospitalizations, and death.
Successful efforts to get patients vaccinated against influenza have included text message reminders timed to precede upcoming flu shot-eligible appointments by up to 3 days. For example, the Roybal-funded flu shot megastudy conducted with Penn Medicine and Geisinger patients (Milkman et al. 2021, Patel et al., 2023) assessed the effectiveness of numerous types of messages in increasing vaccination, relative to standard communications by the respective health systems; an average 2.1 percentage point absolute increase (or 5% relative increase) in flu shots occurred due to the messages. Similarly, follow-up analysis of the megastudy using machine learning revealed that interventions differed in relative effectiveness across groups of patients as a function of overlapping covariates (e.g., age, sex, insurance type, and comorbidities). This analysis found that nudges optimally targeted to subgroups who responded most strongly to those nudges in the megastudy would have resulted in up to three times the increases in vaccination observed when simply randomly assigning patients to messages.
The present study aims to prospectively test the efficacy of a patient-facing, message-based nudge that is predicted by this retrospective machine learning algorithm to be most effective for them (Personalized Nudge) relative to Passive Control (no messages), Active Control (simple reminder message), and Best Nudge (best performing message from the 2020 megastudy). Patients will be randomized to one of these four arms.
Of the 19 original messages from the megastudy, 12 can be carried out at Geisinger in 2024; the other 7 are either no longer relevant (e.g., those that refer to an ongoing COVID-19 pandemic) or cannot be carried out due to a technical limitation (e.g., Geisinger is unable to send pictures, so nudges with pictures are excluded). A treatment assignment tree based on the algorithm described above will be applied to this subset of nudges to generate rules for assigning patients to personalized messages based on observed covariates.
The last patients will be enrolled on December 28th for appointments scheduled on December 31st. At least 90,000 patients are expected to be enrolled.