Text-based Interventions to Promote COVID-19 Vaccinations

This study investigates whether and which type of text-based interventions affect the take-up of the COVID-19 vaccine. The primary research question is whether vaccine takeup can be boosted by a text-message intervention encouraging eligible patients to schedule a vaccination appointment. Patients, when becoming eligible for receiving the COVID-19 vaccine at UCLA Health, will be first notified about their eligibility and encouraged to schedule a vaccination appointment via one of the channels (email, voice call, or snail mail) depending on the contact information available to UCLA Health. Then patients eligible for the study will be randomized at a 1:4 ratio into a holdout control arm that does not receive a text message vs. a text-message arm that receives a text message.

The secondary research question concerns which type of text message is more effective. To study this question, a nested 2x2 factorial design within the text-message arm will be used, where the two factors will be a) whether patients receive the link to an educational video specifically designed to shift patients' intentions to get the COVID-19 vaccine and b) whether the text message contains language aimed at encouraging patients to follow through on their intentions.

• In the Holdout arm: patients will not receive text messages about COVID-vaccine.
• In the text-message arm, all participants will receive a text message that invites them to schedule their vaccination appointment and includes a link to the appointment website
• In the Simple Text sub-arm, participants will not receive any additional information.
• In the Simple Text+Video sub-arm, together with the appointment link, participants will also receive a link to a 2-minute video in the text message. The video contains information about the prevalence of COVID-19 and the effectiveness and safety of the COVID-19 vaccine.
• In the Enhanced Text sub-arm, in addition to the appointment link, the text message will use enhanced language aimed at reducing psychological barriers that prevent patients from scheduling their appointment.
• In the Enhanced Text+Video sub-arm, in addition to the appointment link, the text message will encourage patients to watch a 2-minute video (the same as in the Simple Text+Video sub-arm) and use enhanced language aimed at reducing patients' psychological barriers of following through on scheduling an appointment.

Patients will enter our study on a rolling basis, as they become eligible to get the vaccine (and if they fit our inclusion criteria). For the first batch of patients, those in the text-message arm will receive the text message within 4 -10 days after the initial invitation is sent (because the infrastructure of running this study is not ready until then). For subsequent batches of patients, they will be randomized and receive the text message (if they are not in the holdout arm) the first weekday after the initial communication goes out. We will measure a) whether patients schedule a COVID-19 vaccination appointment for the first dose and b) whether and when patients get the first dose of COVID-19 vaccine.

Analysis:

For the main analysis, we will run ordinary least squares regressions (OLS) with robust standard errors to predict the aforementioned outcome variables, except that we will use a Cox proportional hazards model with administrative censoring to predict time of obtaining the first COVID-19 vaccine. The significance level will be 0.05. Our primary hypothesis is that the text-message arm is significantly better than the holdout arm, so our primary analysis will compare the four text-message sub-arms all together with the holdout group.

Our secondary analysis will investigate whether (1) the Simple Text sub-arm, (2) the two sub-arms containing a video, and (3) the two sub-arms containing enhanced language are better than the holdout arm.

Furthermore, we will test (1) the effect of adding a video (vs. no video) to the text, (2) the effect of adding enhanced language (vs not adding enhanced language), and (3) whether the combination of video and enhanced language will outperform video alone or enhanced language alone.

Our regressions will include the following control variables:

• Participant age
• Indicators for participant race/ethnicity (Black non-Hispanic, Hispanic, Asian non-Hispanic, white non-Hispanic, other/mixed, unknown; white non-Hispanic omitted)
• Whether the patient's preferred language is Spanish (which affects the language of text)
• Indicators for participant gender (male, female, other/unknown)
• Social vulnerability index score
• COVID19 Risk Factors Model
• Indicators for the batches of patients (patients will become eligible and receive initial communications in batches)

As a robustness check, we will re-run the analysis as a logit regression (instead of an OLS regression) for binary outcome variables.

We will explore the following moderators:

• Whether the patient is female or male
• Whether the patient is Black, Caucasian, Hispanic, or other
• Whether the patient's preferred language is Spanish
• Whether the patient is 65+ (including 65) or below 65
• Patient's Social vulnerability index score
• Patient's COVID risk score
• Patient's population risk score
• Whether the patient is married (which is a proxy for whether they live together with family members)
• Whether or not the patient received a flu shot in either the 2019-2020 season or the 2020-2021 flu season prior to receiving our text message according to the patient's medical record
• The day of the week when the text message is sent to a patient. We will compare each day of the week.
• How strongly the participant's neighborhood is in favor of the Republican (vs. Democratic) Party if UCLA Health eventually agrees to provide de-identified address (e.g., zipcode).

Plan for Early and Subsequent Analyses

To inform policy as soon as possible, we plan to first assess the effects of our interventions in the early phase of vaccination outreach at UCLA Health. For this purpose, we plan to first analyze the data from the start of this RCT to the end of February. Given that we are using a 6-day time window for our primary dependent variable, we will examine data from patients who are randomized to either the holdout or text-message arm in this RCT before or on Feb 23, 2021. For this population, we will test:

1. whether the text-message arm is significantly better than the holdout arm;
2. whether the Simple Text sub-arm, the two subarms containing a video, and the two sub-arms containing enhanced language are better than the holdout arm
3. the effect of adding a video (vs. no video) to the text
4. the effect of adding enhanced language (vs not adding enhanced language)
5. whether (1), (2), (3), and (4) varies based on whether patients received the flu vaccination in either the 2019-2020 season or the 2020-2021 season
6. we will report the raw data for each sub-arm without conducting hypothesis testing across conditions that are not pre-registered in (1)-(4).

In our early analysis, we will include controls that are available to us (it is possible that we do not have all of the controls described above at the time of early report).

After all UCLA patients have been invited (or if vaccine distribution plan changes and UCLA Health no longer sends out text messages to patients at some point), we will do the following additional analyses:

• If the additional data collected afterwards exceeds 30K (which gives us 80% power to detect a 2pp difference between the holdout arm and the text message arm, assuming that holdout arm has a 50% baseline), then we will analyze the main effect of sending a text message (vs. holdout) and report the raw data for each sub-arm (to see if the patterns are qualitatively comparable with those in the early data).
• We will use the full sample (including the early data and subsequent data) to analyze (1) whether the combination of video and enhanced language will outperform video alone or enhanced language alone and (2) the aforementioned heterogeneous treatment effect.