New model identifies women at risk for postpartum depression

Researchers at Massachusetts General Hospital have created a machine learning model that identifies women at high risk for postpartum depression (PPD) right after childbirth. This tool allows for immediate intervention, potentially preventing severe symptoms that can arise if treatment is delayed for 6 to 8 weeks post-delivery.

Untreated PPD can lead to serious consequences, including increased maternal morbidity and mortality. It contributes to approximately 20% of maternal deaths by suicide. Mark A. Clapp, MD, the lead investigator and a maternal-fetal medicine specialist, emphasized the importance of early identification in improving patient outcomes. The findings were presented at the American Psychiatric Association’s 2025 Annual Meeting and published online in The American Journal of Psychiatry.

PPD affects up to 15% of women after giving birth and is linked to heightened risks of suicide and self-harm. The condition can severely impact a woman’s mental and physical health, her ability to function, and her relationships with her newborn and family. Traditionally, the American College of Obstetricians and Gynecologists (ACOG) recommended PPD screening only at the postpartum visit, but now suggests screenings at the initial prenatal visit and later during pregnancy.

The study involved 29,168 women with available Edinburgh Postnatal Depression Scale (EPDS) scores who delivered at two major academic hospitals and six community hospitals. Researchers aimed to create a PPD risk stratification model, which utilizes electronic health records (EHRs) to gather data on maternal medical history, medication use, and demographic factors, among others.

The model distinguished between higher- and lower-risk populations, achieving an area under the receiver operating characteristic curve (AUROC) of 0.750 during training, indicating good predictive ability. Out of the participants, 9.2% met at least one criterion for PPD within six months post-delivery. Anxiety disorders, medication use for nausea, headache disorders, and gastrointestinal issues were among the top contributors to PPD risk.

The model’s positive predictive value (PPV) was 24.4% at a specificity of 90%, meaning that nearly one in four women flagged as high risk actually developed PPD. In external validation, the model maintained a PPV of 28.8%, suggesting its robustness across different patient populations.

Clapp noted that the findings show the model could be effectively used across diverse racial, ethnic, and age groups, enhancing the equitable application of care. The researchers aim to integrate this model into existing EHR systems for real-time risk predictions, which could facilitate timely interventions. For example, a simple check-in call from a healthcare provider has proven beneficial for women identified as high-risk for PPD.

Despite its strengths, the study has limitations, including its focus on a specific geographic region and a population predominantly composed of White, college-educated, and privately insured individuals. Researchers acknowledge the potential for misclassification, which is common in studies that rely on diagnostic codes.

Misty Richards, MD, a reproductive psychiatrist at UCLA, emphasized the need for improved diagnostic tools for PPD, as many affected women, particularly those without prior depression histories, often go undiagnosed. Ned Kalin, MD, from the University of Wisconsin, highlighted that the study’s participants who developed PPD were largely undetected prior to this intervention, marking a substantial advancement in identifying at-risk individuals.

The National Institute of Mental Health, the National Institute of Child Health and Human Development, and the Simons Foundation funded the study. Clapp also disclosed his affiliation with Delfina Care as a scientific advisory board member.