Nudging Preventive Screening Via Message Framing and Bundling

The investigators will conduct a randomized controlled trial in the domain of preventive care engagement. Participants with two exams pending (a cancer screening test and a chronic disease test for diabetes and dyslipidemia) will be randomly assigned to a 3 × 3 factorial design. The first factor manipulates message framing, randomly assigning participants to receive either (i) a standard Whatsapp message, (ii) a Whatsapp message emphasizing age-related health risks and the importance of early detection, or (iii) a message emphasizing the potential peace of mind associated with completing screening. The second factor manipulates the type of exam highlighted in the outreach, randomly assigning participants to receive outreach focused on (i) cancer screening only, (ii) chronic disease testing only, or (iii) both cancer screening and chronic disease testing. Patients who have only one pending cancer exam are randomly assigned across the three message framing conditions only and are encouraged to get the cancer screening.

The experiment will test two sets of research questions.

Question 1. How does emphasizing different motivations for completing cancer screening affect patient engagement?

H1a. This study will test how messages emphasizing age-based health risks and the importance of early detection affect cancer screening engagement, relative to a basic message. Emphasizing health risks may increase perceived urgency and boost engagement; however, prior work on information avoidance suggests this framing may instead prompt avoidance of screening-related information and reduce engagement.

H1b. This study will test whether messages emphasizing that completing screening can bring peace of mind increase engagement relative to a basic message.

H1c. This study will test whether the effects of the two framing manipulations on cancer screening differ when the same messaging strategies are applied to chronic disease testing.

Question 2. How does combining recommendations for multiple preventive tests affect patient engagement? H2a. This study will examine the impact of bundling multiple tests (vs. single test) on patient engagement.

H2b. This study will test whether pairing cancer screening with a chronic disease testing recommendation affects cancer screening engagement. Recommending multiple tests may increase the perceived value of engaging with the healthcare system, boosting engagement with cancer screening; alternatively, patients may find the message more overwhelming or burdensome, reducing engagement with cancer screening.

H2c. This study will test whether pairing chronic disease testing with a cancer screening recommendation affects chronic disease testing engagement. Pairing may increase perceived value and make chronic disease testing seem less emotionally aversive, thus increasing engagement with chronic disease testing; alternatively, combining multiple recommendations may feel burdensome and reduce engagement with chronic disease testing.

The investigators will run ordinary least squares regressions (OLS) with robust standard errors to predict each outcome variable.

To test questions 1a and 1b, the study will focus on participants assigned to the cancer-screening-only condition, and the primary predictors of interest are indicators for whether participants are assigned to the Risk/Importance message condition or the Peace-of-Mind condition (with the Basic message condition as the reference group).

To test question 1c, the investigators will pool participants assigned to the cancer-screening-only and chronic-disease-only conditions and estimate models that include framing-condition indicators, an exam-type indicator, and their interactions.

To test question 2a, the investigators will focus on participants assigned to either cancer-screening-only condition, the chronic-disease-only condition, or the both-exams condition. The key independent variable will be an indicator for assignment to the both-exam condition (vs. the other two conditions combined as reference group). If this indicator is statistically significant, the investigators will compare the both-exam condition separately with cancer-screening-only condition and the chronic-disease-only condition.

To test question 2b, the study will focus on participants assigned to either the cancer-screening-only condition or the both-exams condition. The key independent variable will be an indicator for assignment to the both-exam condition.

To test question 2c, the study will focus on participants who received a message about chronic disease testing, including both those assigned to the chronic-disease-only condition and those assigned to the both-exams condition. The key independent variable will be an indicator for assignment to the both-exams condition.

Questions 1c-2c will only include participants with two exams pending (a cancer screening test and a chronic disease test for diabetes and dyslipidemia).

All regressions will be run with and without control variables, including number of pending exam fixed effects, age (continuous), gender, insurance type (public or private based on the local health insurance plans), and clinic or region indicators if available. For robustness, the investigators will conduct logit models as well.

Additionally, the study will explore whether the peace-of-mind message is more effective than the risk/importance message by comparing the coefficients on indicators for the risk/importance message and the peace-of-mind message in regressions described in the analysis.

Also, the investigators will examine whether the effects of the messaging manipulations on cancer screening engagement change when patients are simultaneously encouraged to complete both a cancer screening and chronic disease testing, relative to when they are only encouraged to complete a cancer screening. The regressions will include indicators for message types (risk/importance and peace of mind, relative to basic), an indicator for whether patients are recommended to take multiple tests (vs. just a cancer screening), and their interaction terms to predict (1) whether the patient indicates interest (by either clicking "Yes, I want help" or providing an equivalent affirmative response in Whatsapp) within 7 days, (2) whether the patient schedules the recommended cancer screening within 7 days, and (3) whether the patient completes the recommended cancer screening within 12 weeks.

Given that the experimental design addresses multiple distinct research questions, the analyses may ultimately be split across two papers focused on different sets of questions, particularly if including all analyses in a single paper would limit clarity or coherence.

The experiment is expected to last for 60 working days and reach 235,000 patients across all arms (including the standard-message arm that is not part of the study of interest). About 145,000 of these patients have two pending exams (cancer screening+chronic disease testing), and 90,000 patients have a pending cancer screening.

As an exploratory analysis, the investigators will examine patients' prior frequency of cancer and chronic disease screening as a potential moderator of intervention effects.