This story is part of an ongoing series on A.I. in healthcare.
Accolade, a healthcare concierge company, has built what it’s calling an intelligence engine to do things like suggesting when an individual should seek a second opinion about a planned surgical procedure.
The goal is to make interacting with health insurance benefits more like using a modern e-commerce site that takes cues from the shopper’s past queries and offers personalized buying options. That kind of dialog with patients could bring benefits to the large, self-insured employers and insurance companies Accolade serves, such as reduced costs, increased consumer satisfaction, and ultimately, improved health outcomes of individuals. But improving those metrics over the long term is significantly more challenging than applying machine learning to recommend the next product to add to your shopping cart.
The intelligence engine, called Maya—an amalgam of My Accolade—is designed to be aware of the context in which each individual is making a healthcare decision, and guide them toward the best course of action at that point in time. It’s basically what Accolade employees do in phone consultations now, but by applying machine learning technology to the company’s trove of data, recommendations can be automated and improved over time, says Accolade chief product officer Mike Hilton, pictured at top.
Accolade, which has raised about $160 million in venture capital, added many former senior executives from business expense management company Concur to its leadership ranks—Hilton and CEO Rajeev Singh among them—and established a Seattle corporate headquarters and engineering center in 2016. The company has hired data scientists from tech stalwarts such as Microsoft and Amazon, and has focused a significant part of the Seattle operation on technology development, such as the Maya system, Hilton says.
Accolade markets to benefits plans administrators who are also buying other modern healthcare services, such as wellness plans, second-opinion providers, and telemedicine services. The administrators want to ensure that those services are being used to reduce costs as promised, so part of Accolade’s value is in recommending that individuals make use of them at the appropriate time, he says.
Of course, each individual’s healthcare scenario, medical history, and benefits information is different. Accolade’s human healthcare concierges—the company refers to them as health assistants—have at their fingertips all of that data and personalized context, which they use to help individuals make choices about their care. The phone consultations can run eight to 10 minutes long, which presents a challenge for the company as it scales up. In an interview last fall, CEO Singh said Accolade was serving some 700,000 people.
Now, the assistants will also have a machine learning algorithm working with them in tandem to present people with the right choices at the right time, Hilton says.
In addition to augmenting the health assistants’ advice during phone consultations, the Maya engine will also help present the best choices for individuals who use Accolade through its Web and mobile services. Those include a new mobile app the company announced Tuesday that allows direct text-messaging with health assistants.
The Maya engine is designed to help pick the optimal moment to suggest that someone seek a second opinion about a surgery proposed by their doctor, for example. It’s not something that happens very often, Hilton says, but it’s an opportunity for large savings, as surgeries and follow-ups are often very expensive for insurance providers, and don’t always lead to the best health outcomes.
In another scenario, an individual might be seeking a new primary care physician for a child with autism.
Maya’s machine learning algorithms are trained on the recommendations Accolade health assistants have made to individuals in similar situations in the past, as well as data about the individual, Hilton says.
As with any such technology, the recommendations can only be as good as the underlying data.
Accolade taps into what Hilton describes as a “really unique dataset” on people using its service. That includes insurance claims data, benefits plan information, prescriptions, medical history, and behavioral health data, for example.
Accolade adds data from its direct interactions with individuals. Hilton says Accolade has two-way interactions with 70 percent of the individuals covered by its insurer or employer customers at least once a year.
“The more you engage with someone, the more you’re going to learn about them—their habits, the details that go beyond the medical data,” Hilton says.
The Maya engine can also ingest data from third-party service providers, such as corporate wellness plans, telemedicine services, and second-opinion services.
But can Maya predict whether seeking a second opinion today will result in lower costs, increased satisfaction, or improved health outcomes over the long term—the key factors Accolade’s customers are focused on? It’s a work in progress.
Hilton describes that ability to predict as “one of the holy grails of healthcare.”
“The hardest thing that we try to measure is long-term patient health outcomes,” he says.
“You’ve got these short-term, real-time, ongoing learnings, so you keep updating the dataset that is feeding the recommendation engines, and you’re constantly correlating it to these longer-term outcomes that you’re trying to optimize around.”
That’s part of what makes applying this technology in healthcare harder than something like recommending the next Netflix show to watch. “Trying to make someone healthy is more complex,” Hilton says.