Establishing a UNC perinatal wearable data lake forpassive detection of postpartum depression
2026 Award: $58,820
Every year, millions of new mothers silently struggle with postpartum depression, not because help isn’t available, but because the condition is never detected. Most women go undiagnosed. Our research aims to change that by turning the wearable device a woman already owns into an early warning system. Using AI to analyze patterns in heart rate, sleep, and activity data, we are developing a tool that can passively flag women who may be at risk, no clinic visit, no questionnaire, no extra burden required. This project will validate and expand that tool at UNC and build the data infrastructure needed to bring passive, technology-driven mental health screening into clinical practice.
Need/Problem: Postpartum depression affects about 1 in 6 new mothers, yet fewer than 1 in 3 women with the condition ever receive a diagnosis. Current screening relies on questionnaires at clinic visits, which depend on women recognizing their own symptoms, feeling comfortable disclosing them, and having regular access to care. In North Carolina, mental health conditions are the leading cause of pregnancy-related death, responsible for roughly one-third of maternal deaths. Most women with postpartum depression go unidentified and untreated, not because effective treatments don’t exist, but because we fail to detect the condition in the first place.
Grant Summary: Millions of Americans already own consumer wearables and fitness trackers that continuously collect data on heart rate, sleep, physical activity, and energy levels. Our research has shown that these signals change in measurable ways among women who develop postpartum depression. This project will use that insight to develop and validate an AI-based tool that passively flags women who may be at risk, using data their devices are already collecting, so they can receive timely follow-up and care.
Goals & Projected Outcomes: This project has two goals: 1) to validate our existing postpartum depression detection algorithm in a larger, real-world clinical population at UNC; and 2), to improve the algorithm so it works across all major consumer wearable devices, not just Fitbit. Together, these aims will produce a validated, device-agnostic detection tool and establish the UNC Perinatal Wearable Data Lake, a first-of-its-kind research dataset linking wearable signals with clinical health records for 600 perinatal patients. The data and infrastructure created here will support a future large-scale NIH grant application and serve as a shared resource for researchers across UNC studying women’s mood and perinatal mental health conditions.
Eric Hurwitz, PhD
Danielle Lowe, MD
Grant Details: Our team previously demonstrated, using data from the NIH’s All of Us Research Program, that machine learning models can accurately distinguish women with postpartum depression from those without, using only wearable device data from heart rate, sleep, physical activity, and energy expenditure. These are promising results, but they were developed in a relatively small sample using only Fitbit devices. To translate this work into clinical practice, the algorithm must be validated in a larger, more diverse population and extended to work across all types of consumer wearables. Through UNC’s Secure Health Informatics Research Environment (SHIRE), we will recruit 600 perinatal patients and invite them to retrospectively share wearable data collected during pregnancy and the postpartum period. Linked with clinical records, this dataset (the UNC Perinatal Wearable Data Lake) will allow us to rigorously test and refine the algorithm in a real-world setting and build infrastructure designed to support future psychiatric research broadly.
