Predicting common maternal postpartum complications: leveraging health administrative data and machine learning.

Area of Science:

• Maternal Health
• Health Informatics
• Machine Learning in Healthcare

Background:

• Postpartum complications pose significant risks to maternal health, necessitating accurate prediction methods.
• Timely identification of high-risk pregnancies can improve postpartum care and reduce hospital readmissions.

Purpose of the Study:

• To predict the risk of common maternal postpartum complications requiring inpatient care.
• To evaluate the utility of routinely collected administrative health data for risk prediction.

Main Methods:

• Utilized administrative health data from 422,509 inpatient live births in Queensland (2009-2015).
• Employed gradient boosted trees with five-fold cross-validation for model development.
• Assessed model performance using the area under the receiver operating curve (AUC-ROC) in independent validation data.

Main Results:

• Models demonstrated good discrimination for postpartum hypertensive disorders (AUC=0.879) and obstetric surgical wound infection (AUC=0.856).
• Prediction of postpartum sepsis and haemorrhage showed poor discrimination with the current data and methods.

Conclusions:

• Routinely collected health data can effectively predict certain postpartum complications, aiding clinical decision-making.
• The findings support the integration of predictive analytics into postpartum care pathways.
• Enhancements in routine data collection may improve the prediction accuracy for other postpartum complications.