A recent machine learning study analyzed nearly 9,000 U.S. mothers to uncover distinct, context-specific patterns of perinatal depression risk. The findings reveal how socioeconomic conditions, health circumstances, and lived experiences interact with clinical variables to shape that risk.
Phillip Sherlock, Ph.D., research scientist in the Anita Zucker Center for Excellence in Early Childhood Studies, led a group of researchers, in collaboration with the NIH Environmental Influences on Child Health Outcomes (ECHO) consortium, to explore the risk factors of prenatal and postpartum depression. This is an overwhelmingly overlooked and understudied phenomenon, with rates continuing to grow.
Prenatal depression (PND) and postpartum depression (PPD) are forms of depression emerging during the perinatal period, spanning pregnancy through the months following birth. Globally, PND affects an estimated 21% to 29% of mothers and PPD affects an estimated 27.6%, with variation depending on the population and measurement approach.
Phillip Sherlock, Ph.D.
PPD and PND can adversely affect the mother and child. Key risk outcomes include maternal mortality and morbidity, increased risk for infant death, poor maternal-infant attachment and impaired parenting behaviors. Because of these factors, existing research has been conducted to identify early risk indicators for women at risk for PPD and PND. However, current research often assesses risk based on a single variable.
Sherlock and his team analyzed clustered data from 8,936 mothers, employing a conditional inference tree (ctree) algorithm, an AI-driven approach that recursively partitions data to reveal how multiple variables interact with each other.
Sherlock’s team examined a broad range of predictors spanning four domains: sociodemographic factors such as education, employment, marital status, and insurance; reproductive factors such as gestational age and pregnancy-related health conditions; psychosocial factors such as childhood trauma; and cohort membership across the 16 ECHO studies.
Sherlock and his team found that the same variables did not consistently present as risk factors across groups. Socioeconomic factors and lived circumstances heavily influence whether a predictor becomes a risk at all. As Sherlock puts it, “Risk does not operate in a vacuum. No single factor determines a mother’s path to depression. The same circumstances that challenge one woman may find another thriving because of the protective factors and strengths in her life. Resilience matters just as much as risk, and recognizing both is what makes truly personalized maternal care possible.”
Published in the Journal of Multivariate Behavioral Research, this study has shown a need for precision public health strategies using machine learning within pregnancy care. Based on Sherlock’s findings, researchers found that the more important question is not simply “what raises the risk for depression?” but “for whom, and under what circumstances?”