A roadmap to implementing machine learning in healthcare: from concept to practice

Area of Science:

• Clinical informatics
• Machine learning applications in healthcare
• Pediatric patient outcomes

Background:

• Machine learning (ML) adoption in healthcare settings has been historically slow.
• The Pediatric Real-world Evaluative Data sciences for Clinical Transformation (PREDICT) initiative was established at a pediatric hospital.
• PREDICT aims to develop, deploy, evaluate, and maintain clinical ML models using electronic health records (EHR) data to improve pediatric patient outcomes.

Purpose of the Study:

• To illustrate common challenges encountered during clinical ML deployment in a real-world healthcare setting.
• To provide practical examples and solutions derived from the PREDICT experience.
• To offer insights into overcoming barriers to ML integration in pediatric care.

Main Methods:

• Identification of common challenges in developing and deploying clinical ML models.
• Addressing issues related to clinical scenario identification and data infrastructure.
• Establishing machine learning operations (MLOps) and integrating models into clinical workflows.

Main Results:

• Demonstration of successful strategies for overcoming identified challenges in clinical ML deployment.
• Presentation of pragmatic solutions for implementing ML in healthcare while adhering to best practices.
• Evidence of effective integration of ML models into existing clinical workflows.

Conclusions:

• The PREDICT experience offers valuable lessons for the broader healthcare community regarding ML implementation.
• Continuous refinement of ML deployment strategies is necessary as experience and deployment numbers grow.
• Successful adoption of ML in healthcare requires addressing infrastructure, operational, and workflow integration challenges.