Given all the relentless hype about its Artificial Intelligence and its transformative potential for healthcare, it would be understandable if some health systems might be casting about in search of AI or machine learning projects they could try.
But that sort of rushed, ad hoc approach is precisely the wrong one to take, says Tushar Mehrotra, senior vice president of analytics at Optum.
"The only way you are going to get value out of AI is to link the clinical or business problem to the organization’s overall strategy and make sure you have a rich enough data set to train the model so it generates actionable insights," said Mehrotra.
"Making sure you are building and designing your AI effort the right way means putting in the work up front to create a clear understanding of what you are trying to solve so it can be embedded in the decision-making workflow," he said. "Too often, AI projects start with a quest for academic insight."
At HIMSS20, Mehrotra and his colleague, Optum SVP of Artificial Intelligence and Analytics Sanji Fernando will offer their perspectives on how AI can be applied to promote growth and speed strategies for digital transformation.
"The providers that have seen the most success in AI initiatives are organizations that begin planning around what they are trying to solve, rather than open-ended academic experimentation," said Fernando
"From there, consider the data you are using to train your AI models," he suggested. "How rich are the data, how much do you have, and how well do you understand the decisions that will be made off the data?"
Another key question: "With the automation you are creating, what are the outcomes of these decisions aided by the data?" said Fernando. "If the decisions directly impact outcomes in healthcare for patients, there should be a higher hurdle than for decisions around reimbursement, though those are important too."
If those are some of foundational questions health systems should be asking themselves as they ponder potential AI deployments, there are also some common pitfalls to avoid.
"Depending on where they are in the country and in their AI maturity level, some providers need to put more consideration into how they will access certain kinds of talent to accomplish their goals," Mehrotra explained. "While there has been considerable progress in recent years in the distribution of talent beyond the Northeast and West Coast, it can still be tricky. Organizations need to figure out what kind of talent to hire so they don’t, say, bring on 15 data scientists and have them all writing reports."
In addition, "some organizations overlook the level of access they have to the data that will feed the models," said Fernando. "AI models are only as powerful as that data you train them on. You need to know your business and the data your business runs on.
