Smartphone-based AI Diet Coach for Postpartum Nutrition Adherence and Outcomes

Thesis Abstract

Postpartum women face unique nutritional demands that, if unmet, can compromise maternal recovery and infant health; despite available guidelines, adherence remains suboptimal due to time constraints, access barriers, and lack of personalized support, highlighting the need for scalable, evidence-based interventions. This study aims to evaluate the effectiveness of a smartphone-based AI diet coach in enhancing postpartum nutrition adherence and maternal-infant health outcomes, by integrating real-time guidance, personalized meal planning, and behavior-change support within daily routines. The specific objectives are to (1) assess whether AI-driven dietary coaching improves adherence to postpartum dietary guidelines over a 12-week period compared with standard care, (2) evaluate changes in maternal biochemical markers (hemoglobin, ferritin, 25-hydroxyvitamin D), anthropometrics (weight, waist circumference), and infant growth indicators (weight-for-age z-scores) at 12 weeks and 24 weeks, (3) determine user engagement and satisfaction with the AI coach and identify determinants of sustained use, and (4) explore perceived mechanisms by which AI feedback influences dietary decision-making through a mixed-methods lens. A mixed-methods sequential explanatory design is employed. The quantitative strand uses a randomized controlled trial with 240 postpartum participants recruited from maternity units and primary care clinics, randomly allocated to an intervention group receiving the AI diet coach plus standard care or a control group receiving standard care alone. Data collection instruments include validated scales for dietary adherence (modified Mediterranean Diet Score), nutrition knowledge questionnaires, and objectively measured intake via 24-hour recalls administered at baseline, week 6, and week 12, alongside biochemical assays (hemoglobin, ferritin, 25(OH)D) and anthropometrics. Engagement metrics (app login frequency, feature usage, and push notification responsiveness) are captured continuously. The qualitative strand conducts semi-structured interviews with a purposive subsample of 40 participants (20 from each arm) and focus groups with 6 healthcare providers, analyzed thematically to elucidate user experiences and hypothesized behavioral pathways. Quantitative analysis employs intention-to-treat principles. Primary outcome analysis tests differences in dietary adherence scores between groups at 12 weeks using ANCOVA, adjusting for baseline adherence, with secondary analyses for adherence at 6 and 24 weeks. Repeated-measures ANOVA evaluates trajectories of biochemical markers and anthropometrics. Multivariate linear and logistic regression models identify predictors of adherence and health outcomes, incorporating covariates such as age, parity, socioeconomic status, and pre-pregnancy BMI. Mediation analyses examine whether changes in nutrition knowledge and self-efficacy mediate the relationship between app use and adherence. The qualitative data are analyzed using inductive thematic analysis to identify patterns related to perceived usefulness, usability, cultural fit, and acceptance of AI-generated recommendations. Triangulation integrates quantitative outcomes with qualitative insights to provide a comprehensive interpretation. Expected findings include higher adherence scores and improved nutritional biomarkers in the intervention group at 12 weeks, with sustained effects at 24 weeks in a subset of users who maintain engagement. Infants in the intervention cohort are anticipated to display comparable or improved growth trajectories, and user engagement is expected to correlate with magnitude of dietary change. The study anticipates that AI personalization, timely feedback, and goal-setting features will emerge as key mechanisms driving behavior change, moderated by self-efficacy and perceived social support. The study contributes to knowledge by providing rigorous evidence on the feasibility, effectiveness, and mechanisms of AI-driven, smartphone-based dietary coaching in the postpartum period, addressing a gap in scalable, personalized nutrition interventions for new mothers. The integration of behavioral theory—specifically Social Cognitive Theory and Self-Determination Theory—with advanced machine learning to tailor recommendations advances understanding of how digital tools can optimize adherence and health outcomes in postpartum populations. Practical implications include informing clinical guidelines, digital health policy, and the design of scalable postpartum nutrition programs. Recommendations emphasize strategies to enhance engagement, equitable access, data privacy, and integration with routine antenatal and postnatal care, along with directions for long-term follow-up studies to assess sustained maternal and child health benefits.