Design and Pilot Implementation of a Chatbot to Support Breast Cancer Screening Among Black Women

Our study purpose is to evaluate the effectiveness of a culturally tailored chatbot intervention to improve access to mammography scheduling and breast cancer screening (BCS) completion among average risk Black women who receive primary care at the University of Washington (UW) and are due for breast cancer screening. This is a prospective, pilot randomized controlled trial that will evaluate the effectiveness of an outreach strategy in improving BCS completion via mammography. Individuals eligible for enrollment into the trial will be randomized to the two following interventions: 1. Arm 1: Chatbot intervention: SMS/text message with breast cancer screening educational and scheduling information; reminder message at 2-weeks. 2. Arm 2: Control: usual care. Random assignments will be based on a randomization scheme called Big Stick Design with a maximum tolerated imbalance between groups of 4. This design allows random assignments to be generated in a way that limits the imbalance between arms within strata, while ensuring that the sequence is not predictable based on previous assignments. Randomization will be stratified within arms by 1) age (40-55 years vs. 56-74 years - to account for potential differences in recommendations for mammogram frequency and comfort with technology/chatbots), 2) history of any prior breast cancer screening (prior screener vs. never screener), 3) campaign messaging status (received UW Medicine Population Health breast cancer screening campaign outreach message in August 2026 vs. did not receive outreach message in August 2026). We plan to stagger the timing of chatbot outreach based on mammogram access in coordination with the UW radiology department. For example, if there are 20 slots available in the next 2 weeks across sites, we may send chatbot outreach to 100 eligible participants (with expected \~20% response rate to successful outreach interventions among people overdue for screening). If there is a delay between randomization and intervention group chatbot outreach (i.e. due to staggering), we will confirm if participants have scheduled mammography in interim and are therefore no longer eligible for study participation before intervention outreach messaging is sent. Our primary outcome of interest is proportion of participants who completed BCS at 90 days following initial chatbot outreach. Our secondary outcome of interest is proportion of participants who scheduled or completed BCS at 90 days following initial chatbot outreach (i.e. measuring implementation effectiveness). We will also perform a sensitivity analysis measuring both primary and secondary outcomes at 180 days following initial chatbot outreach (i.e. to account for delays in scheduling/outreach). We will perform intention-to-treat (ITT) analyses. First, we will use the Pearson's chi-square test of homogeneity to assess for statistically significant differences between randomization group and control group mammogram completion rate at 90 days post chatbot outreach (primary outcome), or mammogram scheduling/completion rate at 90 days post chatbot outreach (secondary outcome). We will then conduct a multivariable logistic regression analysis to test the association between randomization group and mammogram completion at 90 days post chatbot outreach (primary outcome), or mammogram scheduling/completion at 90 days post chatbot outreach (secondary outcome), adjusting for individual- (e.g., age, insurance status, patient portal use, primary care visits in last year, previous BCS), clinic- (e.g., clinic group), and system-level characteristics (e.g. campaigns messaging), including stratification factors, to address covariate imbalance between groups and improve precision in the analysis. We will estimate adjusted risk differences from these models using predictive margins. Finally, we will conduct unadjusted exploratory subgroup analyses. We will explore intervention effect consistency between: 1) participants who use the patient portal and those who do not use the patient portal; 2) participants who receive primary care through a university hospital-based clinic, county hospital-based clinic, or community-based clinic; 3) among intervention group participants who responded to text messages and/or reached scheduling message (i.e. interacted with chatbot) vs. those who did not.