Effect of Targeted Behavioral Science Messages on COVID-19 Vaccination Registration Among Employees of a Large Health System

This randomized trial evaluates whether individually addressed emails designed with behaviorally informed features increase COVID-19 vaccination rates.


Participants
Data were collected from Geisinger Health System employees, most of whom work remotely or in-person in central and northeastern Pennsylvania (location was not a criterion for eligibility). Sample size was determined by the number of eligible employees (had our study not occurred, all would have received one email from the health system promoting vaccination). Employees have received at least 36 mass emails from the initial announcement of a vaccination plan to the start of the intervention (with two more sent after the intervention), which are detailed in the table below. We used a 1 x 3 factorial design, and Dr. Santos randomly assigned participants with a random number generator in R providing the seed value and the randomizr package, which evenly distributed participants into 3 groups. Geisinger's Human Resources Department only provided individual information on job roles, and the balance between groups is shown in the table below. Participants were enrolled by virtue of being sent the email. Emails were sent with Microsoft Dynamics. Participants did not know to which group they were assigned (for the purposes of reporting masking procedures, there were no providers involved in the intervention). From an original list of 9,723 employees, 157 were not able to be reached with the email address on record, and 1 research team member was removed because they were accidentally included in the list (see the CONSORT flowchart below).

Materials
Both emails were written in the form of a personal message from a medical expert and authority figure on COVID-19 in the health system (the Division Chief of Infectious Diseases). The first paragraph of both emails emphasized the upcoming scarcity of the vaccines, and the second paragraph encouraged employees to make an active choice of whether to get the vaccine or not. The social norms email used an action-oriented subject line: "Indicate your COVID-19 vaccine plan." The reframing risk email used a matter-of-fact subject line: "A message from Dr. Stan Martin." The online survey (reached by clicking a link in the email to decline to schedule a vaccination appointment) presented the following question: "What is the main reason you're declining to receive an FDA-authorized COVID-19 vaccine? (Your responses are anonymous and cannot be linked to your name or email.)" It then presented 15 possible options: (1) Concerns about known side effects (e.g., headache, fatigue); (2) Concerns about unknown risks; (3) I don't feel I need to be vaccinated; I'm not at high risk for infection; (4) I don't feel I need to be vaccinated; I'm not at high risk for serious complications from COVID if I am infected; (5) It depends on when it's offered to me; I'd like to wait to take it until I see how it goes with others; (6) It depends on which vaccine I'm offered; (7) I've had COVID so I don't think I need a vaccine; (8) Privacy concerns about Geisinger or state government tracking; (9) I don't trust the FDA approval process; (10) I am pregnant, breastfeeding or planning to become pregnant; (11) I have a preexisting medical condition that would prevent me from getting the vaccine; (12) I have allergies that could react to the vaccine; (12) Concerns about the vaccine's ingredients (e.g., mRNA); (13) Concerns about the vaccine's efficacy; (14) I work out of state; I plan to receive a vaccine when one is available to me in my state; (15) Other [free response item].
After respondents selected their main reason, an explanation was displayed that addressed that particular concern. These explanations were largely drawn from the health system's online materials on the Pfizer/BioNTech and Moderna vaccines. After reading this explanation, respondents were given a second opportunity to make an appointment, but only three clicked on the registration site link.

Intervention Methods
The first set of emails was sent to employees in the social norms and reframing risks groups at 16:55 EST, January 15, 2021 (see Email Versions A and B in the Annotated Materials, below). An analysis of the primary outcome was planned for January 18, 2021, but the data was not able to be reliably gathered at that time. Without any reliable data on the performance of the two groups from which to choose a clearly superior email to send to the delayed control group, we decided to evenly randomize that group's members to the two email versions (see Email Versions A2 and B2, below). Of this group, 63 were not able to be reached with the email address on record. This second set of the two treatment emails was sent to the delayed control group at 16:40 EST, January 18, 2021. A corrected version of each of these emails was sent at 12:49 EST, January 19, mentioning a change in state guidelines on the next stage of vaccine distribution (see Email Versions A2-Corrected and B2-Corrected, below). The correction did not change any information about the distribution plan for employees, registration method, or upcoming deadline.
The primary outcome was the number of employees who registered for a vaccine appointment during the first three days (i.e., from January 15 to 18), but additional data was gathered until 15:19 EST on January 19, which was when our provided link no longer forwarded employees to the employee appointment system closed. Vaccine registration involves visiting the online scheduling portal on Quickbase, where employees first entered their employee ID and email address, then on a second screen completed an attestation form and screening questionnaire (11 items), acknowledged reading FAQs, and clicked a button labelled submit. Pressing submit on this second page recorded the registration in Quickbase and constitute a registration for purposes of our results, regardless of eligibility for vaccination or the availability of vaccination appointments. Merely entering the employee's ID and email address and proceeding to the second page, without completing these further steps, did not constitute a vaccination registration. Data were initially delivered to the authors on February 3, 2021. Additional data were delivered on May 13, 2021, to enable the research team to more precisely determine how far in the registration process each employee had gone.

Analysis Methods
In our ClinicalTrials.gov record, we preregistered that the study would examine vaccination registration data after two days, a plan we had created on January 15. When it became apparent that the delayed control group would not be emailed until the third day, and without having received any data, we changed the primary outcome period to three days. Additional "engagement metric" data-email opens and link clicks-were also obtained, and will be uploaded on the Open Science Framework (institutional approval is pending). Email opens (depending on the platform, this was recorded by either opening the email or viewing the preview pane) and clicks were reported through Microsoft Dynamics. We preregistered these as secondary outcomes, but with no hypotheses, and they were not analyzed as outcomes in this study.
For all analyses, odds ratios (ORs) from logistic regressions were calculated, along with asymptotic 95% confidence intervals (CIs); two-tailed p-values < 0.05 were used to determine statistical significance. Two logistic regression models were run in parallel. One compared the email conditions against the delayed control group (reference group), and the other compared the social norms email against the reframing risks email (reference group).
There were no harms or unintended effects reported. 3178. Included in analysis.

Not included in analysis (research team member included in error)
13,977. Approximate number excluded for not meeting inclusion criteria