Comprehensive Summary
In this study, Zhu et al. examine the effects of social determinants of health on the development of postpartum depression (PPD) using a machine learning (ML) model. Instead of electronic health record (EHR) data, researchers used a national dataset from the Phase 8 Pregnancy Risk Assessment Monitoring System (PRAMS) questionnaire, which was sent to a random sample of women 2 to 6 months postpartum with recent live births. The survey intended to measure population-based information, including maternal health behaviors, healthcare access, stress levels, and income. A final cohort of 92,762 records were obtained and using multivariable logistic regression, researchers were able to determine trends in the data. Women presenting with PPD were more likely to be aged 25-34, White, earn < $20,000 annually, participate in Medicaid, have depression before or during pregnancy, and had received federal assistance during pregnancy. The study also aimed to determine the best ML classification model, which found that Logistic Regression possessed the most accurate class distinction abilities, boasting the largest area under the curve (AUC). The ML methods used in this study were fair predictors of PPD and could be used in the future to streamline the determination of PPD risk factors to improve outcomes in diagnosis and treatment.
Outcomes and Implications
Postpartum depression (PPD) can be a difficult time for a mother and baby and often results in challenges that can last for years. Children of mothers with untreated postpartum depression can grow up with difficulties in emotional regulation, social interaction, and aggression. Through pinpointing and addressing the risk factors and social determinants of health that are contributing to PPD development, there can be a focus on interventions for at-risk women. Due to the widespread nature of the ML databases, data from everywhere in the United States can be used to aggregate survey responses from women across the country, generating a bigger picture of clinical and social factors contributing to the development of PPD. Future research will be altered to incorporate additional lifestyle factors as well as more class inclusion and balancing to ensure equal representation in data collection. Overall, the data is promising in developing clinical support models to automate PPD screenings, making them less time-consuming and allowing the focus to be on prevention and treatment to improve health outcomes.