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

Background: The adoption of machine learning (ML) has been slow within the healthcare setting. We launched Pediatric Real-world Evaluative Data sciences for Clinical Transformation (PREDICT) at a pediatric hospital. Its goal was to develop, deploy, evaluate and maintain clinical ML models to improve pediatric patient outcomes using electronic health records data.

Objective: To provide examples from the PREDICT experience illustrating how common challenges with clinical ML deployment were addressed.

Materials and methods: We present common challenges in developing and deploying models in healthcare related to the following: identify clinical scenarios, establish data infrastructure and utilization, create machine learning operations and integrate into clinical workflows.

Results: We show examples of how these challenges were overcome and provide suggestions for pragmatic solutions while maintaining best practices.

Discussion: These approaches will require refinement over time as the number of deployments and experience increase.