Black-white differences in chronic stress exposures to predict preterm birth: interpretable, race/ethnicity-specific...

Area of Science:

• Reproductive Health
• Health Disparities
• Machine Learning in Medicine

Background:

• Racial/ethnic disparities in preterm birth are linked to differential chronic stress exposure.
• Previous research has not fully explored the cumulative, interactive, and population-specific effects of chronic stressors on preterm birth.
• Novel modeling approaches are needed to capture complex, high-dimensional associations that may vary by race/ethnicity.

Purpose of the Study:

• To develop and apply machine learning models to predict preterm birth more accurately.
• To identify specific chronic stressors and risk factors driving preterm birth risk.
• To investigate these associations within non-Hispanic Black and non-Hispanic White pregnant populations.

Main Methods:

• Multivariate Adaptive Regression Splines (MARS) models were utilized for preterm birth prediction.
• Models were developed for non-Hispanic Black, non-Hispanic White, and combined study samples.
• Model performance was evaluated using the Area Under the ROC Curve (AUC) and variable importance analysis.

Main Results:

• Race/ethnicity-specific MARS models demonstrated high accuracy (AUC: 0.754-0.765), performing as well or better than combined models.
• Key predictors across populations included prenatal care visits, premature rupture of membranes, and medical conditions.
• Chronic stressors, such as low maternal education and intimate partner violence, were significant predictors of preterm birth exclusively for non-Hispanic Black women.

Conclusions:

• Findings underscore the importance of targeting mid-to-upstream social determinants like chronic stressors.
• Interventions focused on these factors are crucial for reducing excess preterm birth risk among non-Hispanic Black women.
• Addressing chronic stressors can help narrow the persistent Black-White gap in preterm birth rates in the U.S.