EHR-based models show promise in predicting postpartum depression

EHR-based models show promise in predicting postpartum depression | Image Credit: © Arsenii - © Arsenii - stock.adobe.com.

Strong performance has been reported from clinical prediction models for postpartum depression (PPD) using electronic health record (EHR) data, according to a recent study published in JAMA Network Open.¹

Mental health disorders such as anxiety disorders and major depressive disorders are defined as perinatal mood and anxiety disorders (PMADs). These disorders impact approximately 20% of birthing parents, with PPD being the most common in 13%.²

PMAD screening has been reported in under 1 in 4 women in Los Angeles, California, and almost 50% of PMADs are never diagnosed.¹ This may lead to adverse outcomes such as inhibited childhood development or suicide, highlighting the need for better diagnostic and prognostic models.

Investigators conducted a diagnostic study to evaluate the performance of prognostic models for PPD with psychometric screening as the target variable. Data was collected at Cedars-Sinai Medical Center (CSMC), which has routinely screened patients with a live birth for PPD immediately after delivery since 2017.

Participants included birthing individuals admitted to the postpartum unit or maternal-fetal care unit at CSMC until after delivery. Race and ethnicity data was obtained from these patients, with categories including Black, American Indian or Alaskan Native, Asian, White, and other.

Depression screening was performed using the Patient Health Questionnaire 9 (PHQ-9) and the Edinburgh Postnatal Depression Scale (EPDS). The PHQ-9 measures depressive episodes and symptom severity on scale ranging from 0 to 27, with 0 indicating no depression symptoms and 27 the greatest depression symptoms.

PMADs are assessed using the EPDS. Responses are provided on a 4-point Likert scale, and total scores range from 0 to 30. Higher scores indicate increased PPD symptoms.

Screening results from the PHQ-9 and EPDS were assessed as the target variable. A positive PPD screening was determined by a PHQ-9 score of at least 5, an endorsement of suicidal ideation, or an EPDS score of at least 8.

Logistic regression, random forest, and extreme gradient boosting algorithms were fitted to the data. The initial dataset was initially split into 75% for training, 15% for validation, and 10% for testing.

There were 11,377 PHQ-9 records and 8658 EPDS records encompassing 19,430 patients included in the final analysis. These patients were aged a mean 34.1 years at first delivery, and 7% were Black, 12% Asian, 10% Hispanic White, 56% non-Hispanic White, 3% multiple races, and 11% other race or ethnicity.

The odds of positive screening on the PHQ-9 and EPDS were increased among racial and ethnic minority patients vs non-Hispanic White patients, with odds ratios (ORs) of 1.47 and 1.38, respectively. The current models showed noninferiority, and in some cases superiority, when compared to models from past studies.

When reweighing the samples in the training set, the mean area under the receiver operating curve (AUROC) for the model ranged from 0.610 to 0.635, vs 0.602 to 0.622 when not reweighing. In comparison, prior models had mean AUROCs from 0.602 to 0.635.

The models indicated increased positive rates among racial and ethnic minorities vs non-Hispanic White patients, with a demographic parity of 0.238. Similarly, racial and ethnic minorities had reduced false negatives, with a mean difference of -0.184.

These results indicated modest performance from PPD prediction models using common EHR data to predict psychometric screening results. Investigators recommended additional research to “explore methods to optimize the weights during training that achieve specific performance and fairness goals.”

References

1. Wong EF, Saini AK, Accortt EE, Wong MS, Moore JH, Bright TJ. Evaluating bias-mitigated predictive models of perinatal mood and anxiety disorders. JAMA Netw Open. 2024;7(12):e2438152. doi:10.1001/jamanetworkopen.2024.38152
2. Gavin NI, Gaynes BN, Lohr KN, Meltzer-Brody S, Gartlehner G, Swinson T. Perinatal depression: a systematic review of prevalence and incidence. Obstet Gynecol. 2005;106(5 Pt 1):1071-83. doi:10.1097/01.AOG.0000183597.31630.db