Answers are grounded in this paper and cite their sources.
Systematic review evaluates machine learning using wearable data to detect perinatal mental health issues, suggesting careful validation needed.
Pregnancy and the postpartum period are marked by significant psychosocial transition and elevated mental health vulnerability. When mental health needs go unmet, they can manifest as perinatal mood and anxiety disorders, which affect approximately 1 in 8 women and carry significant consequences for mother and child. While machine learning (ML) applied to wearable data shows promise for mental health detection, its application to the perinatal population remains underexplored and methodologically limited. We conducted a systematic literature review of articles focusing on ML and artificial intelligence applications for detecting perinatal mental health states using wearable-derived data. We searched PubMed, Scopus, ACM, IEEE, and ScienceDirect. We screened a total of 484 records and excluded 472 after abstract review and duplicate removal. We assessed 12 full-text articles, ultimately including 11 studies. ML models trained on wearable-derived physiological and behavioral features demonstrated moderate to strong predictive performance for classifying or predicting perinatal mental health states (e.g., stress, anxiety, postpartum depression, loneliness), with most AUC values exceeding 0.693. Heart rate and heart rate variability-derived features were the most consistently studied and frequently reported important predictors across included studies, and Random Forest and Support Vector Machine were the most frequently applied algorithms. However, most studies relied on internal validation, and substantial methodological limitations were identified: homogeneous samples, non-validated outcome measures, and limited real-world testing. While wearable-based ML shows promise for perinatal mental health detection, the evidence base remains early-stage and rigorous validation and feasibility studies are essential before clinical implementation.
No takes yet. Share an insight, caveat, or question.
Lucca et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: