Exploring the Association Between Behavioral Determinants and Intention to Use a Chatbot-Led Parenting Intervention by Caregivers of Adolescent Girls in South Africa: Cross-Sectional Study.

Abstract

Background: While digital innovation, including chatbots, offers a potentially cost-effective means to scale public health programs in low-income settings, user engagement rates remain low. Barriers to participant engagement (eg, perceived difficulty of use, busyness, low levels of digital literacy) may exacerbate inequality when adopting digital-only interventions as alternatives to in-person programs.

Objective: This cross-sectional study nested within a 2×2 clustered factorial trial that followed the Multiphase Optimization Strategy principles investigated the relationship between behavioral determinants (ie, human and socioeconomic characteristics that facilitate the use of digital health interventions [DHIs]) and caregiver intention to use a digital public health intervention, ParentText, an open-source, rule-based parenting chatbot designed to promote positive parenting, improve adolescent health, and reduce risky behaviors.

Methods: Caregivers of adolescent girls (10-17 years; N=1034 caregivers) were recruited by implementation partners from a community-wide project aimed at HIV prevention in two districts of Mpumalanga, South Africa. A Digital Health Engagement Model was adapted from the technology acceptance model, the PEN-3 model theoretical frameworks, and the Theory of Planned Behavior to investigate the relationship between behavioral determinants and the intentions of caregivers to engage in ParentText. Community facilitators administered baseline surveys to caregivers during intervention onboarding. Regression models tested associations between behavioral determinants (ie, perceived ease of use, perceived usefulness, attitude, hedonic motivation, habit, price value, and social influence) and intentions of caregivers to use the parenting chatbot. Interaction effects were explored to examine whether individual-level sociodemographic and psychosocial characteristics moderate associations between overall behavioral determinants and intentions to use the chatbot.

Results: Caregivers reported a mean of 2.85 (SD 0.79) and 2.90 (SD 0.72) out of a maximum score of 4 regarding their intention to use their mobile data and to continue using ParentText in the future, respectively. Overall behavioral determinants predicted by 76% (odds ratio 1.76, 95% CI 1.72-1.81) the intentions of caregivers to spend mobile data and by 85% (odds ratio 1.85, 95% CI 1.81-1.90) their intentions to use ParentText in the future. Moderator analysis suggested the interaction effects of age, paternal absence, financial efficacy, and stress on the relationship between overall behavioral determinants and intention outcomes.

Conclusions: This is the first known study to investigate the associations between overall behavioral determinants and participant intentions to use a parenting chatbot in a low-income setting. This study identifies behavioral determinants of engagement for improved delivery of DHIs, considering the need to provide low-cost, scalable parenting support through digital platforms that engage parents, especially those in low-income contexts. Future research should explore methods to investigate mechanisms that regulate behavior to enhance the development of DHIs.