In the United States, a pregnant individual will attend approximately 15 prenatal visits with a medical provider for the monitoring of an uncomplicated pregnancy. During these visits, a vast amount of demographic and clinical information is collected and entered into the electronic health record (EHR). Much of the information is related to monitoring the pregnancy, such as measurements of weight and blood pressure; however, there is information in the medical record that could be used to predict risk for perinatal depression.
In a recent study from Mass General Hospital, researchers examined whether information included in the medical record at the time of discharge after delivery could help us to identify individuals at increased risk for postpartum depression.
Study Design
For this analysis, the research team conducted a retrospective cohort study of all individuals who delivered babies between 2017 and 2022 at two large academic medical centers and six community hospitals in the Boston area.
The analysis excluded individuals considered to be at high risk for PPD based on their clinical history and/or individuals who were already receiving mental health care: patients with a diagnostic code reflecting a current mood or psychotic disorder and those who had received an antidepressant prescription in the 12 months preceding delivery.
Using the EHR, the researchers identified women with postpartum depression within the first six months after delivery, where PPD was defined as (1) a diagnosis of a mood disorder, (2) an antidepressant prescription, or (3) a positive screen (13 or higher) on the Edinburgh Postnatal Depression Scale administered after delivery.
Predictors used in the modeling included sociodemographic factors, medical history, and prenatal depression screening information, all of which were documented in the EHR before discharge from the delivery hospitalization.
Results
For the analysis, The cohort included a total of 29,168 individuals; 2,696 (9.2%) met at least one of the criteria for postpartum depression during the six months following delivery. The model was trained and optimized using a sample of 15,018 patients. When the researchers tested their risk model on a second group of 14,150 patients (external validation), the model performed well:
- Discrimination (AUC 0.721): The model could reasonably distinguish between patients who would and would not develop postpartum depression. (An area under the curve or AUC of 0.721 indicates that the model is much better than chance.)
- Calibration (Brier score 0.087): The model’s predicted risks matched the observed outcomes closely, meaning its predictions were reliable.
- Positive Predictive Value (PPV, 28.8%): Among those the model identified as high risk, about 29% actually developed postpartum depression.
- Negative Predictive Value (NPV, 92.2%): Among those the model identified as low risk, about 92% did not develop postpartum depression.
Can We Use the Electronic Medical Record to Predict Risk for PPD?
The current study indicates that machine learning can be used to construct a model using data collected from the EHR and that such a model can be used to predict risk for postpartum depression within the first six months after childbirth. While not every patient identified as high risk using this model will not go on to develop postpartum depression, the tool is very good at identifying those who are unlikely to develop PPD, which can conserve valuable resources by targeting and following those at highest risk.
The findings are consistent with several other studies which have used the EHR to predict risk for postpartum depression. In all of the studies, using data from the EPDS screening improved the performance of the model, highlighting the importance of screening for depressive symptoms during pregnancy. The main difference in this study is that it excluded women who were already identified as being at high risk based on recent psychiatric history; these were women who were diagnosed with a mood disorder or had presented with depressive symptoms during pregnancy or who were being treated with an antidepressant medication.
On one hand, this may seem like a shortcoming of the study; however, this decision makes sense. The patients used to train the predictive model developed in this study would not normally be considered at increased risk for PPD; as recommended, they would receive some type of screening for depressive symptoms during the third trimester of pregnancy or after delivery, but they would not receive any special type of monitoring and would be considered “low” risk. The model developed in the current study would help us, as the title of the article implies, to stratify risk. In this population of pregnant women who would clinically be considered to be at low risk, there are women who are at increased risk for PPD, although they may be harder to identify.
While previous studies have documented that the women at highest risk for PPD are those with depressive symptoms during pregnancy — about 10% to 15% of women — we have not been so good at defining risk in the remaining 85% of the women. This type of model would help us to better define risk in populations that would normally be considered to be at low risk. This is especially important in women with no history of psychiatric illness.
Ideally we would like to be able to identify women at risk for postpartum depression before it occurs. This would not only allow us to increase monitoring when needed and to treat early if PPD emerges, but it may also provide an opportunity to initiate preventative interventions. Currently our strongest predictors of risk include a history of depression prior to pregnancy and depressive symptoms during pregnancy. These models build on these robust risk factors, and include other risk factors (i.e., age, BMI) to improve our ability to predict and quantify risk.
Ruta Nonacs, MD PhD
References
Clapp MA, Castro VM, Verhaak P, McCoy TH, Shook LL, Edlow AG, Perlis RH. Stratifying Risk for Postpartum Depression at Time of Hospital Discharge. Am J Psychiatry. 2025 Jun 1; 182(6):551-559.
