Estimating baseline cutoffs for DHA dosage in preterm birth prevention: a Bayesian personalized change-point analysis

Preterm birth (PTB, <37 weeks gestation) is the leading cause of infant mortality and significant health and socioeconomic burdens that affect millions of newborns and families. While docosahexaenoic acid (DHA) supplementation has shown promise in reducing PTB risk, its effectiveness in reducing the most consequential early PTB (ePTB, <34 weeks gestation) depends on baseline DHA levels, with lower DHA levels and intake linked to a higher risk of PTB and ePTB that can be reduced by high-dose DHA supplementation. Given the higher costs of high-dose DHA, personalized treatment strategies based on baseline DHA levels are needed. We proposed a novel Bayesian personalized change-point model to optimize DHA supplementation strategies based on individual baseline DHA intake. By incorporating Bayesian change-point, dynamic linear, and normal mixture models, our approach estimates optimal DHA baseline thresholds and distribution. We applied this model to real-world data and simulated trials to demonstrate its ability to improve secondary analysis and trial design by adjusting for baseline DHA heterogeneity. This personalized approach can help clinicians identify optimal DHA supplementation doses for individual patients, and it can be applied to other trial studies where the heterogenous characteristics of patients can be quantified.