Machine learning: How can AI improve healthcare?
Artificial intelligence (AI) is “set to revolutionize healthcare,” according to a recent presentation at the Health Information and Management Systems Society (HIMSS) Asia Pacific Conference 2019.
“In fact, it’s set to revolutionize everything that’s going on in our lives at the moment, not just healthcare but all industries,” says Tim Young, general manager for Asia Pacific at DataRobot, one of the exhibitors at HIMSS 2019.
AI is defined as a computer system that can make decisions or perform tasks that normally require human intelligence. And one aspect of AI that, according to Young, can help improve the healthcare industry is through machine learning, which refers to a computer that learns by example.
In a nutshell, machine learning involves automating processes in an organization. However, Young stresses that automation does not mean that humans will be replaced by machines. Rather, “it’s about scaling intelligence.”
AI is expected to solve three major healthcare challenges, namely quality of healthcare, cost and revenue, and operational efficiencies.
According to Young, automated machine learning can help identify patients at risk for sepsis or other hospital-acquired condition, predict medication adherence, predict care gaps, prevent adverse events, as well as detect disease early.
AI can also reduce overall healthcare cost while increasing revenue by predicting readmissions, reducing patient length of stay, forecasting demand, optimizing staff, detecting fraud and stratifying patients by risk. Finally, it increases operational efficiencies by predicting the following: staffing needs, bed occupancy, no-show appointments, service utilization and supply.
But building AI is no easy task. “It’s actually very difficult to implement AI systems,” says Young. “Most of the organizations we talked to are in the skeptic stage, uncertain how they can move forward with this.”
Initially, an organization must identify the “good opportunity” for machine learning.
“Machine learning is a great way of solving problem that it’s a totally different paradigm from any analytic that’s happened before,” says Young. “So the fact that you understand about business intelligence doesn’t necessarily mean that you’re going to understand about how to deploy and leverage machine learning.”
After identifying what problems can be solved with machine learning, the next challenge is to determine what modelling and validation to execute. Thousands of open source algorithms are available free of charge today, but an organization must select which model works best for its needs.
“Once you found the perfect algorithm, the next thing you’ve got to do is to deploy it, and how you deploy the algorithm can potentially have fairly big impact on the organization itself,” says Young.
“Finally, the last thing to do is to make sure that the AI/machine learning algorithm is deployed in the organization in a way that people will actually use and benefit from it,” he adds.
Rexer Analytics, a company that specializes in data science and predictive modelling, claims that only 13 percent of data science projects reach production, and fewer generate real business value.
“The problem that we have is that people don’t understand how to frame up the problem from machine learning,” notes Young.
Another challenge is that the people who are deploying and building machine learning algorithms—the data scientists—are “expensive” and “very much in scarce supply.”
“As you’re embarking on this machine learning journey, the really important thing is to start with something that’s fairly simple... something where you can get results and confidence and build up expertise,” says Young.