Machine learning prediction models for postpartum depression: A multicenter study in Japan

Area of Science:

• Perinatal mental health
• Reproductive psychiatry
• Clinical informatics

Background:

• Postpartum depression (PPD) is a significant global concern.
• Effective perinatal mental health care is crucial.
• Early identification of at-risk mothers is essential for timely intervention.

Purpose of the Study:

• To develop and validate machine learning models for predicting PPD.
• To assess the utility of the 2-week postpartum checkup in identifying high-risk women for PPD.

Main Methods:

• A multicenter retrospective study of 10,013 women in Japan.
• PPD defined as Edinburgh Postnatal Depression Scale score ≥9 at 4 weeks postpartum.
• Models developed using logistic regression and machine learning with routinely collected clinical data.

Main Results:

• Machine learning models showed similar predictive performance to logistic regression using pre-discharge data (AUROC 0.569-0.630 vs. 0.626).
• Including 2-week postpartum checkup data significantly enhanced PPD prediction in Ridge and Elastic net models (AUROC 0.702 vs. 0.630 and 0.701 vs. 0.628).

Conclusions:

• Machine learning models did not surpass conventional logistic regression for PPD prediction.
• The 2-week postpartum checkup proved valuable for identifying women at high risk of PPD.