The association of early life experiences with prenatal depression: A machine learning-based predictive model.

BACKGROUND

Prenatal depression is highly prevalent and associated with early-life adversities like adverse childhood experiences (ACEs) and school bullying. Machine learning (ML) can effectively capture complex, non-linear predictive patterns for depression.

OBJECTIVE

We aimed to identify early-life factors predicting prenatal depression and develop a practical, interpretable ML predictive model.

PARTICIPANTS AND

SETTING

A cross-sectional study recruited 1193 pregnant women via convenience sampling at two hospitals in Hubei, China.

METHODS

Feature selection used a 10-fold cross-validation framework strictly within the training set. Nine ML algorithms were benchmarked using AUROC, AUPRC, F1-score, calibration plots, and Decision Curve Analysis (DCA). SHapley Additive exPlanations (SHAP) evaluated feature dependencies on the independent test set.

RESULTS

CatBoost emerged as the optimal model. SHAP identified early family economic status, peer victimization, and maltreatment as primary predictive drivers. Crucially, dependence plots revealed prominent non-linear threshold effects: the mere occurrence of relational traumas drove risk more profoundly than their cumulative frequency. Furthermore, complex interactions showed that early material deprivation moderated the associations between peer bullying, abuse, and depression.

CONCLUSION

The CatBoost-SHAP framework provides a robust, transparent tool for risk stratification, highlighting non-linear associations between early adversity and prenatal depression. Integrating early-life adversity assessments into first-trimester screening and electronic health records can facilitate precision prenatal interventions.