MoCab: A framework for the deployment of machine learning models across health information systems

Area of Science:

• Healthcare Informatics
• Machine Learning in Medicine
• Health Information Systems

Background:

• Machine learning models are crucial for advancing healthcare services.
• Integrating machine learning into health information systems (HISs) presents interoperability and data format challenges.
• Existing platforms like EPOCH®, ePRISM®, and KETOS highlight the need for standardized solutions.

Purpose of the Study:

• To introduce the Model Cabinet Architecture (MoCab) framework.
• To leverage Fast Healthcare Interoperability Resources (FHIR) for standardized data handling in HISs.
• To address challenges in deploying machine learning models across diverse healthcare systems.

Main Methods:

• MoCab utilizes a structured framework with a Data Service Center for FHIR data retrieval.
• The Knowledge Model Center formats data for predictive models, and the Model Retraining Center ensures continuous model updates.
• Clinical Decision Support (CDS) Hooks and Substitutable Medical Apps Reusable Technologies (SMART) on FHIR are integrated for alerts and application development.

Main Results:

• MoCab was demonstrated with scoring, machine learning, and deep learning models using synthetic EHR data.
• The framework successfully integrated predictive models with health data for clinical decision support.
• Implementations confirmed MoCab's practical utility and seamless HIS integration via CDS Hooks and SMART on FHIR.

Conclusions:

• MoCab demonstrates significant potential for improving machine learning model interoperability across various electronic health records (EHRs).
• The framework effectively addresses key challenges in adapting machine learning models within healthcare settings.
• MoCab paves the way for enhanced machine learning utility and broader adoption in healthcare, despite FHIR adoption hurdles.