Researchers have developed and externally validated a machine learning model that stratifies risk for postpartum depression at the time of hospital discharge, potentially enabling targeted interventions for prevention and early treatment. The model incorporated prenatal depression screening scores along with routinely collected clinical data and demonstrated good discrimination and calibration in an external validation cohort.
In this retrospective cohort study, researchers from Massachusetts General Hospital analyzed data from 29,168 individuals who delivered between 2017 and 2022 at two academic medical centers and six community hospitals. The study population was limited to patients who had no prior diagnosis of mood or psychotic disorders or antidepressant prescriptions 1 year prior to delivery.
“Postpartum depression [PPD] is common, affecting approximately 15% of recently pregnant individuals, and represents a major contributor to both morbidity and mortality following pregnancy,” the study authors wrote. “It is associated with an increased risk for suicide and self-harm and is estimated to be responsible for 10% or more of all pregnancy-related deaths,” they added.
The model was developed using data from five hospitals and validated in three separate hospitals within the same health system. Predictor variables included information known prior to delivery discharge, such as prenatal Edinburgh Postnatal Depression Scale (EPDS) scores, sociodemographic factors, medical history, medication use, and delivery characteristics. Race and ethnicity weren't included as predictor variables but were used for subgroup performance analyses.
In the external validation cohort, the model achieved an area under the receiver operating characteristic curve (AUROC) of 0.721 (95% confidence interval [CI] = 0.709–0.736) and a Brier score of 0.087 (95% CI = 0.083–0.091). At a specificity threshold of 90%, the model yielded a positive predictive value of 28.8% (95% CI = 26.7–30.8) and a negative predictive value of 92.2% (95% CI = 91.8–92.7).
Overall, 9.2% of individuals in the study met criteria for PPD within 6 months postpartum, defined as having ≥ 1 of the following: a mood disorder diagnosis, an antidepressant prescription, or a positive EPDS screen (score ≥ 13).
The model performed comparably across subgroups stratified by race, ethnicity, age, and hospital type (academic vs community based), suggesting potential for equitable application in diverse populations. The inclusion of prenatal EPDS scores significantly improved model performance: the AUROC of models without EPDS was lower at 0.647 (95% CI = 0.631–0.662). Models using only the maximum or most recent EPDS scores also performed less well (AUROCs of 0.683 and 0.680, respectively).
“These findings demonstrate that a simple machine-learning model can be used to stratify the risk for PPD before delivery hospitalization discharge,” the study authors stated. “This tool could help identify patients within a practice at the highest risk and facilitate individualized postpartum care planning for the prevention of, screening for, and management of PPD at the start of the postpartum period and potentially the onset of symptoms," they concluded.
The researchers noted that more than 98% of births occur in hospital or health care settings, making the delivery hospitalization a key opportunity to implement risk stratification tools. Future studies will assess how the model may be implemented in clinical workflows and how risk information should be presented to clinicians and patients to guide care without bias or unintended consequences.
Disclosures for the study authors are included in the original publication.
