Taking machine-learning models in health care from concept to bedside

Machine learning—an application of artificial intelligence in which large amounts of data are processed, patterns are identified, and the information then applied to help answer real-life questions—holds enormous promise for precision health applications. But implementing machine learning in clinical healthcare has been challenging. As the authors of a recent Nature Medicine article point out, “The potential impact of machine learning (ML) in health care warrants genuine enthusiasm, but its limited adoption in clinical care to date indicates that many of the current strategies are far from optimal.”

The authors provide an overview of common challenges to implementing ML in a health-care setting, and describe the necessity of breaking down the silos in ML and of engaging an interdisciplinary team of collaborators “from beginning (problem formulation) to end (widespread deployment)” of the ML process.

Jenna Wiens, PhD, incoming Co-Director of Precision Health at the University of Michigan and Assistant Professor of Computer Science and Engineering, is a first author of the paper, “Do no harm: a roadmap for responsible machine learning for health care.” Mohammed Saeed, MD, PhD, a Clinical Lecturer in Internal Medicine at U-M, is a co-author.

“As the excitement around ML for health grows, it’s important to take a step back and consider what’s needed to have meaningful impact,” Wiens says. “Applying ML to health care is difficult, but the path forward can be a lot clearer with the right team. The paper serves as a roadmap for researchers who are interested in getting involved in this area, and what (and who) it takes to go from problem formulation to implementation.”

The paper grew from a discussion at the Machine Learning for HealthCare (MLHC) meeting in August 2018. A day before the meeting, “an interdisciplinary group of researchers got together to discuss pitfalls, challenges, and roadblocks to deployment of ML in healthcare,” says Wiens. This discussion, she says, culminated in a written piece outlining responsible machine learning for healthcare.

“As a clinician with a background in machine learning,” says co-author Saeed, “I strongly believe that this paper provides useful guidance from a multidisciplinary team, including thought-leaders in machine learning as well as clinicians, with real-world experience in taking machine learning from research concepts to the bedside.”

The authors list critical steps of the ML process as

• choosing the right problems
• developing a useful solution
• considering ethical implications
• rigorously evaluating the model
• thoughtfully reporting results
• deploying responsibly

All steps must be considered for successful ML implementation, by stakeholders from each stage of the process: ML and health IT experts, hospital administrators and regulatory agencies, doctors and nurses, and patients and “framily.” By following these steps and engaging relevant stakeholders early in the process, the authors state, “many issues stemming from the complexity of adopting ML in practice can be successfully avoided.” They conclude, “Doing so will not only decrease the potential for unintended consequences, but also reduce rather than amplify existing social inequalities and ultimately lead to better care.”