Deploying Predictive Models In A Healthcare Environment - An Open Source Approach.

Area of Science:

• Medical Informatics
• Computational Medicine
• Health Services Research

Background:

• Predictive modeling and data mining show promise in medicine.
• A significant hurdle is integrating these models into clinical practice.
• Existing methods often lack flexibility for widespread adoption.

Purpose of the Study:

• To develop an easily implementable framework for publishing predictive models.
• To enable consumption by diverse clinical applications and reuse across environments.
• To facilitate the integration of machine learning models into routine healthcare.

Main Methods:

• Developed a framework for publishing R or Python predictive models.
• Models are served as web services embedded in containers.
• Utilized exclusively open-source technologies for accessibility and cost-effectiveness.

Main Results:

• The framework allows models to be consumed by virtually any downstream clinical application.
• Models can be reused in various environments without requiring modifications.
• A model for predicting hospital readmissions was successfully deployed using this framework.

Conclusions:

• The proposed framework offers a flexible and simple methodology for model deployment.
• It is readily adoptable by a wide range of healthcare institutions.
• The open-source approach promotes accessibility and broad application of predictive models in medicine.