Decision tree learning for predicting chronic postpartum depression in the Japan Environment and Children's Study

Area of Science:

• Machine Learning in Healthcare
• Maternal Health Research
• Predictive Modeling

Background:

• Postpartum depression (PPD) prediction models often lack simplicity for community use.
• This study developed a user-friendly decision tree model for chronic PPD prediction.

Purpose of the Study:

• To create a simple, pen-and-paper-compatible prediction model for chronic postpartum depression.
• To identify key predictive variables for PPD in a large maternal cohort.

Main Methods:

• A decision tree model was trained on 84,091 mothers using 84 pregnancy variables.
• Model constraints included a branching depth of 3 and a minimum of 100 participants per branch.
• Chronic PPD was defined as an Edinburgh Postnatal Depression Scale score ≥9 at 1 and 6 months postpartum.

Main Results:

• A 35-branch decision tree achieved an area under the receiver operating characteristic of 0.84.
• "Feeling worthless" was the most effective single predictor among 10 extracted variables.
• Prevalence rates varied significantly across model branches in both training and validation datasets.

Conclusions:

• A simple, high-performing prediction model for chronic postpartum depression was developed.
• The model's ease of use makes it suitable for community maternal health settings.
• This tool can support early identification and intervention for PPD.