As artificial intelligence (AI) gains more momentum in the healthcare sector, CIOs' use of these technologies has expanded. And as a result, the industry is now moving quickly, crafting solutions to meet this growing demand.
To this point, according to the findings of a recent Accenture report based on C-suite executive responses from more than 100 health organizations, AI is poised to become the new user interface (UI) in health IT. The report noted, “The growing role of AI in healthcare is moving beyond a back-end tool to the forefront of the consumer and clinician experience, becoming a new user interface that underpins the ways individuals transact and interact with systems. Emphasizing its growing importance of AI, more than four-fifths (84 percent) of healthcare executives surveyed as part of the research believe that AI will revolutionize the way they gain information from and interact with consumers, and nearly three-quarters (72 percent) of health organizations surveyed are already using virtual assistants to create better customer interactions.”
Peter Borden, managing director of the consulting firm Sapient Healthcare, notes that people have actually been talking about AI for ages across various sectors, and now strong use cases are staring to emerge, especially in the health space where evidence is so important. Borden points to three areas in healthcare that people are paying the most attention to as it relates to leveraging AI: population health insight, or going beyond the core data sets to analyze where populations might need the most attention; augmented intelligence, in which there is not a replacing of a function but rather making one stronger; and precision engagement in which personalization is taken to the next level.
Says Borden, “You might understand population health and what segment is at highest risk for going from pre-diabetes to diabetes, but you need to know how to engage each individual in that sub-segment in the way that makes the most natural sense for them. And there are great emerging stories from that.” He adds, “As people start to understand that it will impact the business, and not just the outcomes, there is now a lot more acceptance.”
When folks think of AI, the first thing that often comes to the minds of many is Siri, which of course is in most people’s pockets via the iPhone. And as Borden notes, “Everyone knows about Watson. Google is also making strong and interesting pushes into the space. And Amazon is making moves into the health space, too. Some of the big cloud players are doing interesting things as well, and because of the nature of the cloud, it allows for analysis of data in certain ways from a processing perspective. So the cloud is a natural partner. But really, every minute it seems like a new and interesting company is popping up,” he says.
One such healthcare AI company that has “popped up” is New York City-based Prognos, which formed just this year and is interested in developing predictive models that use massive datasets (mostly coming from an extensive network of partner labs nationwide) to determine how likely it is that a patient will undergo a specific health event in the future. Prognos builds these models using patient’s anonymized records, and has an exponentially-growing amount of them (over 8 billion by Q1 of 2017), according to company officials.
The company’s co-founder and chief medical officer Jason Bhan, M.D., a family physician who previously worked at Clinovations, a company that helps hospitals implement health IT, recalls a project he was working on for a client in which the organization’s CEO and chief medical officer turned and said to Bhan and his team, “We just spent $150 million on this [EHR] system, what did we do for ourselves?” Bhan said he responded by saying, “Let’s go find out.”
Bhan then dug into the data that was on the client’s IT systems, and the realization was that most of the data coming out of EHRs just is not very useful, he recalls. He adds that the diagnostic information, however— lab data, radiology, and tests that physicians are performing—is a gold mine, so that coupled with years of practice, and also years of making decisions based on looking at lab results, guided him towards thinking more about diagnostics in healthcare. “Prognos was a concept of eradicating disease, but how do we improve health in general by tracking and predicting disease at the earliest? That’s where we got into AI,” Bhan says.
Applying AI to Determine Risk
While the lines are sometimes blurred between machine learning and artificial intelligence, Larry Lefkowitz, Ph.D., chief scientist at SapientRazorfish, a company under Sapient that launched this year to help clients drive digital transformation, explains the difference. “Machine learning is if I have sufficient data representative of the kind of problem I am trying to solve. Then I have a reasonable chance of applying machine learning to be able to ‘understand’ that data enough so it can make the same kinds of predictions and get the same results a person would, given that next set of inputs.” He adds, “When you have tons of data like in radiology or pathology cases, it seems like a really good application of machine learning. So here is my next X-ray, I have told you what X-rays like this might look like and might mean, go ahead and give me an answer.”
Meanwhile, says Bhan, AI is about probability and risk, which he says is essentially a microcosm about what life is about. AI is trying to guess the likelihood of something, he adds, whether it’s a chatbox, or self-driving cars, so the likelihood of turning left or right. In health, it’s the likelihood of getting diagnosed with a certain disease, failing a particular therapy, or the likelihood of the information being presented actually being correct. “In healthcare we have been collecting data for a very long time, without any particular rhyme or reason or downstream use other than a doctor looking at it and trying to interpret it. So when you talk about a machine looking like it and trying to interpret it, you can’t just deploy your typical AI engine from the internet,” Bhan says.
Jason Bhan, M.D.
This is why his company has chosen diagnostics in healthcare AI since it is “known” and since the predictive power of it provides better opportunities than using old entrenched data sets like transactional claim activity or a patient picking up a prescription from a pharmacy, he says. “We are looking at real clinical data that doesn’t tell you whether the patient has diabetes or not, but whether a person is in control or not in control. And looking at those patterns, what’s the next steps for that person?”
Indeed, Prognos specifically takes rich, clean data at a massive scale and processes it using AI “training sets” tailored for specific health events using an AI engine. The company uses this enormous cache of historical data and time-interval data —which continues to grow over time — to build and train its algorithms, and to refine them iteratively in order to develop reliable predictive-analytic capabilities.
Working with Payers
Certainly, the application of predictive analytics to specific patient-related data can help payers—including not only private health insurance plans, but Medicare and Medicaid, as well—to recognize and react to emerging health-related trends more quickly. Unlike claims that can take weeks or months to process, diagnostics test results are available in near real-time.
As such, a core focus for Prognos is the payer market. Its officials attest that an estimated 10 to 15 percent of conditions are never discovered from claims alone, and claims also underreport condition severity by 30 percent. High-quality lab data not only fills that gap but does it quickly, enabling payers to deliver personalized interventions earlier in the disease process, they say.
Via a Prognos cloud solution, the entire management of a payer’s diagnostic data set is taken on, so if the payer has patients all over the country or a region, those patients are going to different labs, and sometimes they get that lab data, while sometimes they don’t, Bhan explains, noting that leveraging the AI in this situation is about cleaning up the data and prepping it for later AI usage. “So the data sits in the cloud. Out from the cloud is what we call care alerts or risk alerts, so using that payer’s actual data set, looking at the patients as they come through and assigning a risk to them,” he says. Using these payer smaller data sets—maybe a couple hundred thousand or a million members—Prognos is applying the same knowledge and algorithms that it has derived from the massive data registry that it has on these smaller data sets. “So everything we learn on the big registry we apply to the payer’s patients and apply risk,” Bhan notes.
Fernando Schwartz, Ph.D., chief data scientist at Prognos, compares this transformation to when computers came around, and how at that time there were inefficient processes with various players and lots of bookkeeping. “But computers revolutionized the way we worked; they made the process much easier and quicker,” says Schwartz, who is new into healthcare, but has been studying data science for years.
He adds, “What we are trying to build here is the next revolution in that sense. All of the files in the computer are replacing papers, but you still need to do something with those files. We are automating workflows for payers—I think of AI of playing that role as well. As much as [it is] playing the part of providing more accurate and personalized healthcare, AI is also helping payers become more streamlined so they aren’t chasing files and records across the system.”
Part 2 of this feature on AI in healthcare will focus on the core challenges that exist today and how physicians are reacting to this oncoming phenomenon