Risk-prediction models for postpartum depression: Assessment of risk of bias using PROBAST+AI.

OBJECTIVE

The postpartum depression (PPD) risk prediction model is an effective risk stratification tool and is expected to play a significant role in the early detection and intervention of PPD. This study aims to summarize the existing evidence on PPD risk prediction models and provide references for their development, validation, and clinical application.

METHODS

We searched PubMed, EMBASE, and Web of Science databases for original articles published in English up to November 9, 2025. Our primary analysis included studies developing or validating risk prediction models for PPD. Additionally, we identified and appraised independent studies that conducted external evaluation of existing models. Studies developing or validating risk prediction models for PPD were included. Two reviewers independently extracted study characteristics, including predictors, model performance, and methodology, and assessed the risk of bias and applicability using PROBAST+AI (Prediction model Risk Of Bias Assessment Tool).

RESULTS

The primary analysis included 13 model development/validation studies, encompassing 64 original prediction models. The most frequently incorporated predictors included maternal age, race/ethnicity, marital status, education level, economic status, history of mental disorders during pregnancy, pre-pregnancy body mass index (BMI), and pregnancy complications. PROBAST+AI assessment revealed a high risk of bias in 10 of 13 model development studies and 12 of 13 model validation studies, primarily due to issues with participant representativeness, predictor handling, and analytical methods. All studies have low applicability concerns.

CONCLUSIONS

While several models have undergone independent evaluation, the majority of existing AI-based PPD prediction models in the development and validation phase suffer from significant methodological limitations, resulting in a high risk of bias. This critically restricts their current clinical utility. Future work should prioritize methodological quality, external validation, and usability to enable real-world implementation.