Innovating Care in Machine Learning
Phoenix Children’s Hospital Executive Vice President and Chief Innovation Officer David Higginson offered his insights on the practical applications of machine learning at his organization.
“We don’t use vendors to do this. We build these products ourselves,” Higginson said. ML models are bound to local data, so taking a vendor’s model that worked in California won’t necessarily work in Arizona. Plus, trying and failing with a vendor’s solution would be costly.
So far, Higginson said, his team has built systems that help diagnose malnutrition in children, predict when a patient could be a no-show, offer support for staff retention, and identify people who might donate money to the hospital.
“Not every cool idea can be implemented,” he said. “If you don’t have the end in mind, it’s all for nothing.”
The automated ML tools he and his team use come with a price tag, so they need to show concrete benefits for the organization. “You really have to think about the problems — how am I going to extract value, either clinically or financially? — or don’t do it at all,” he said.
LEARN MORE: Catch up on the opening HIMSS23 keynote discussion on AI and health equity.
The key is finding the right problem that can be addressed with an ML model. “You have to reframe problems as a prediction problem for this to work. Many, many problems don’t lend themselves to being predicted,” Higginson said.
Get used to failure and stay away from issues that require high accuracy, he said, but added that many problems exist that don’t require that.
“I cannot tell you the number of times I’ve stopped wasted work by asking people, ‘If you had a prediction that was 85-plus percent accurate, what the heck would you do with it?’ Most people can’t figure that out,” Higginson said.
Keep this page bookmarked for our ongoing coverage of HIMSS23. Follow us on Twitter at @HealthTechMag and join the conversation at #HIMSS23.
