Implementation of a Machine Learning Risk Prediction Model for Postpartum Depression in the Electronic Health Records

Area of Science:

• Artificial Intelligence in Healthcare
• Clinical Informatics
• Machine Learning for Medical Applications

Background:

• Postpartum depression (PPD) poses significant risks to maternal and infant well-being.
• Effective prevention, diagnosis, and management strategies are crucial for maternal health.
• Existing clinical workflows may benefit from enhanced decision support tools.

Purpose of the Study:

• To describe the deployment of an AI-driven clinical decision support (CDS) system for PPD.
• To detail the technical architecture and integration of the CDS into clinical practice.
• To facilitate efficient PPD risk management through an automated clinical pathway.

Main Methods:

• Developed an L2-regularized logistic regression model using electronic health record (EHR) data.
• Refined the model with consortium data for generalizability and fairness.
• Utilized Microsoft Azure for scalable deployment and FHIR for data interoperability.
• Implemented CI/CD pipelines for automated deployment and maintenance.
• Integrated CDS risk assessment into the clinical workflow.

Main Results:

• Successfully deployed an AI-driven CDS system for PPD.
• Established a scalable and secure deployment architecture on Microsoft Azure.
• Ensured data interoperability using FHIR standards.
• Automated deployment and maintenance processes via CI/CD pipelines.
• Integrated risk assessment and action options into the clinician schedule.

Conclusions:

• The deployed AI-CDS system offers a scalable and efficient approach to PPD risk management.
• Seamless integration into clinical workflows enhances usability and adoption.
• The system has the potential to significantly improve PPD prevention, diagnosis, and management outcomes.
• Leveraging EHR data and modern deployment practices is key for effective clinical AI solutions.