Artificial intelligence is permeating the healthcare industry in a number of ways, but not all AI is created equally. The technology now in place in hospitals and health systems across the U.S. tends to be on the simpler end of the AI spectrum, and while this has resulted in operational efficiencies, there's an opportunity to invest in next-level technologies that could bring about more transformative change.
Specifically, there's an opportunity in healthcare to improve precision medicine and reduce administrative complexity, thereby cutting the cost of doing business. In this regard, the AI landscape is still evolving.
Venky Ananth, senior vice president and head of healthcare at Infosys, spends a lot of time thinking about the spectrum of AI. On the low end of the spectrum is software-led and physical robotics automation, which typically still require human judgement to be effective. Toward the middle of the spectrum is AI that has predictive capabilities, and can suss patterns out of historical data to prognosticate what's likely to happen next.
At the highest end of the spectrum is cognitive computing, which uses more sophisticated techniques to solve specific business problems. It's not as widely used in healthcare as AI technologies elsewhere on the spectrum, but this could be changing as more organizations understand its value.
"The value is highest on the cognitive side, but it's also the most difficult to achieve because it needs significant data capability and technical capability to implement," said Ananth.
While robotic process automation has been a part of the way healthcare does business for some time now, it's really been in the past two years that a large amount of operational processes have been automated, and they've been associated with reduced administrative costs and more accurate prescribing of medicines.
"People seem to realize the value in this technology," said Ananth. "AI is in the right stage where, now, organizations are getting really serious about it. Adoption is happening on the ground. They're asking, 'What problems can it solve for me?'"
PRECISION MEDICINE, REDUCED COSTS
Cognitive AI technologies hold promise in the realm of precision medicine by reducing the trial and error associated with matching a patient to a drug. The process can be tricky depending on the specific condition in question.
For a condition like diabetes, there are certain biomarkers present, so it's easier for a clinician to recommend a specific treatment process. But in the case of a disease such as mental illness, there are no biomarkers involved. The patient answers questions from a psychiatrist and then may take a particular drug for about six weeks or so, a process that Ananth said is linked to about a 50% success rate. Same odds as a coin toss.
This can go on for months. And if finding the right medication lengthens into a six-month process, the patient's quality of life continues to decline because not only are they having, say, depressive symptoms, but they could also be dealing with negative side effects from ineffective drugs.
"What we're working on is, without changing the drug itself, is (finding) a way we can use techniques to pinpoint biomarkers, and through a combination of genomics and metabolism, we match that with a drug, and we figure out the response rate for the drug," said Ananth. "There are some preliminary clinical trials. And then the endgame is, how do we scale this out and roll it out at a larger healthcare industry level?"
When it comes to administrative costs, Ananth said about a quarter of the total healthcare spend in the U.S. is attributable to waste, a figure backed up by recent research published in October in JAMA. In that study, the authors reported the total estimated annual cost of waste was $265.6 billion for administrative complexity alone, including billing and coding waste, and physician time spent reporting on quality measures.
Ananth said administrative complexity is often the byproduct of care fragmentation -- a number of players are actively engaged in the delivery of care, from payers and providers to specialists and pharmacy benefits managers.
Where AI shows promise in this regard is in the pursuit of the somewhat elusive holy grail of interoperability, or the ability of disparate electronic health records and information systems to communicate with one another. Driving interoperability will likely be the best path forward for reducing the amount of repetitive administrative tasks that occur at different stages of care.
Despite the promise of AI, there are some obstacles to overcome.
AI is actually a fairly old technology by Ananth's estimation, and for a long time was prevented from becoming more mainstream by the massive computing costs involved, which made it commercially unviable.
One of the challenges to greater adoption of AI is the cost of implementation. Investments can eat into short-term profits, which many providers want to avoid.
"Legacy systems are not designed for that," said Ananth. "Data exists in vast reserves in silos. This is what we mean by interoperability. Interoperability means, 'How can my claims data and benefit management data and my pharmacy data all come together to make it useful, and make a difference at the point of care?' Bringing them together is a huge task. You've got to align incentives. Why would I, as a provider, spend time trying to work with the payer and pharmacy to work together? Nobody has any incentive to do that.
"It doesn't come cheap," he said. The skills to engineer it are huge. It's a tough nut to crack."
It's critical, said Ananth, for healthcare professionals to understand the power of AI and its potential to make them more efficient no matter what role they're in, be it a nurse practitioner, cardiologist or anyone else on the care continuum.
Adoption won't be simple and straightforward, and will need a focused effort, but the potential gains are massive. Change will happen through education, said Ananth -- talking about AI in the industry and spreading the word.
"What works is, think of one or two problems you want to solve -- problems that would have impact and would matter to your organization," he said. "Then ask, 'What would it take to solve this problem?' It could be a process you think is quite flawed. Then you could ask, 'What is my process today, and what should it be tomorrow?'"