Predicting postpartum psychiatric admission using a machine learning approach.

Area of Science:

• Perinatal mental health
• Machine learning in healthcare
• Public health surveillance

Background:

• Accurate identification of mothers at risk for postpartum psychiatric admission is crucial for timely intervention.
• Existing methods may not fully capture the complexity of risk factors.
• Developing predictive models can improve postpartum mental healthcare access.

Purpose of the Study:

• To develop and validate a prediction model for identifying women at risk of postpartum psychiatric admission.
• To utilize administrative health data for early risk detection.
• To enable preventive interventions or facilitate timely psychiatric admissions.

Main Methods:

• Utilized administrative health data from live births in Queensland, Australia (2009-2014).
• Included mothers with pre-existing mental health indicators during pregnancy (n=75,054).
• Employed multiple machine learning algorithms, including boosted trees, to predict admissions for psychotic, bipolar, or depressive disorders within 12 months postpartum.

Main Results:

• The boosted trees model demonstrated strong predictive performance with an AUC of 0.80 (95% CI: 0.76-0.83).
• This model outperformed benchmark logistic regression and elastic net models.
• Key predictors included maternal mental health history, anthropometric measures, and social/lifestyle factors.

Conclusions:

• A big data approach shows significant potential for identifying mothers at risk of postpartum psychiatric admission.
• Early identification allows for targeted follow-up and support post-discharge.
• This can potentially reduce the need for admission or expedite care when necessary.