Artificial intelligence-based multimodal prediction of peripartum depression: time for a paradigm shift

Every year, 16,000 women in Sweden suffer from peripartum depression (PPD)– depression during pregnancy or the first year after childbirth.

PPD carries a sizable societal and economic burden (5 billion SEK per year), impacting negatively on women, partners and children.

Prevention should be prioritized and effective preventive interventions do exist; but are only cost-effective when targeting high-risk women. However, accurate prediction of risk is currently lacking. Previous depression, resilience and socioeconomy are strong individual correlates.

In parallel, biomarkers and digital phenotyping (e.g., physical activity patterns, internet use) have also shown promising potential in predicting mental illness. Thus far, no study has tested the combined predictive power of all the above variables.

We propose a novel, interdisciplinary project to maximize the predictive accuracy in identifying women at risk for developing PPD using a participatory approach.

We aim to: (i) analyze the most predictive biomarkers for PPD in a large ongoing cohort (mom2b.se) (ii) integrate these results to develop highly accurate predictive algorithms for PPD risk stratification in a multi-modal dataset, using state-of-the-art machine learning techniques; (iii) assess women’s and healthcare providers´ attitudes and preferences relating to an early perinatal risk assessment for depression.

Results of this unique project can inform precision medicine approaches within maternity care and psychiatry.