Incorporating Machine Learning and Statistical Methods to Address Maternal Healthcare Disparities in the United States: A Systematic Review

Background: Maternal mortality in the United States (U.S) has more than doubled over the past 30 years, with certain communities facing rates up to three times higher, irrespective of income or education (WHO, 2024). Key drivers include unequal access to care, medical bias, and systemic inequities such as racism and poverty. Addressing these disparities requires an integrated strategy spanning healthcare reform, community engagement, and targeted research. Innovations, particularly in machine learning (ML) and advanced statistical analysis, play a critical role in advancing this effort. This systematic review aims to evaluate the use of machine learning (ML) and statistical methods such as predictive modeling and data-driven decision-making to identify and address maternal health disparities. It seeks to understand how sociodemographic factors, healthcare practices, and technology impact maternal health outcomes, guiding future research and informing equitable care policies.

Methods: This review is registered with PROSPERO (CRD42024578664) through the National Institute for Health Research. Following the PRISMA guidelines, we systematically reviewed peer-reviewed studies published between January 1, 2012, and February 2024. Inclusion criteria included studies done in the U.S with focus on maternal health disparities within the U.S., utilizing ML and statistical methods and involving maternal and infant health data up to one year post-delivery. A thorough search was conducted through PubMed, Web of Science, and ScienceDirect, using a rigorous screening process to ensure that only relevant and peer-reviewed studies were included. Data extraction emphasized study objectives, design, sample size, and analytical tools, while the risk of bias was assessed using the Cochrane risk-of-bias tool for randomized studies and the ROBINS-I for non-randomized studies.

Results: Out of 4,581 initially identified articles, 147 met the inclusion criteria. The findings reveal that 88% of the studies employed statistical methods to examine correlations, treatment effects, and public health initiatives, providing essential insights for clinical and policy decision-making. Meanwhile, 12% of the studies applied ML techniques to explore complex, nonlinear relationships in data. Most studies highlighted significant racial and socioeconomic disparities, focusing less on geographic inequality. The reviewed studies also emphasized the need for culturally sensitive and inclusive health practices to address these disparities effectively.

Conclusions and Implications: The findings demonstrate the transformative potential of ML and statistical methods in addressing maternal health disparities. However, the findings also highlight several limitations, including methodological constraints, data quality issues, and the need for broader dataset utilization. Future research should focus on improving methodological precision, enhancing dataset diversity, and integrating social determinants of health into research models. These steps will help develop targeted interventions and informed policies to reduce maternal health disparities and promote equity in healthcare outcomes. The findings underscore the need for interdisciplinary collaboration to equitably integrate technological advances, like machine learning and statistical analysis, into maternal care, focusing on improving outcomes for minority populations and underserved remote areas.