Healthcare has entered the era of artificial intelligence, but it faces hurdles to implement the technology, a Henry Schein Inc. executive said Thursday.
Medical institutions have initiated artificial intelligence programs in the last five years, some with notable success, said Daniel Holewienko, executive director, big data and business intelligence at Melville-based Henry Schein, speaking at the annual conference of Stony Brook University's Center of Excellence in Wireless and Information Technology.
He pointed to projects resulting in a 72% reduction in time to detect heart disease at the Mayo Clinic and a 95% increase in accuracy in identifying skin cancer at Stanford University.
Failing to pursue artificial intelligence and its subset, machine learning, which allows machines to learn from data without explicit programming, puts health care organizations "at a competitive disadvantage," he said.
At the same time, Holewienko cautioned that such projects carry risks.
Industry researcher Gartner predicts that through 2022, 85% of artificial intelligence projects in all fields will deliver erroneous outcomes due to faulty data, algorithms or management.
Holewienko said that organizations seeking to harness artificial intelligence face difficulties in: integrating AI projects with existing systems; high costs of adoption; executives' failure to understand the technology; and scarcity of experts to hire.
He advised health care institutions to: avoid trying do-it-yourself and find a partner; begin with smaller projects and build on them; carefully define the problem; use scientific processes; and have sufficient data for the project.
Improving patient services
Another speaker, Emmanuel Egbo, management consultant, health care analytics at Scottsdale, Arizona-based Artha Solutions LLC, described a project where a chain of urgent-care clinics sought to use machine learning to optimize staff scheduling, improve profitability, reduce patient wait time and increase patient satisfaction.
That project harvested data from more than 100 of the walk-in clinics that operate 24 hours a day and sought to identify seasonality, randomness and trends in patient visits.
In the end, Egbo said that patient wait time was reduced by 75% across the chain and resource allocation improved by 85%.
Still, while many health care projects could benefit by applying AI or machine learning, Holewienko warned that not every problem fits that model or provides sufficient data to analyze.
In that case, he said, it's best to walk away.