Machine Learning Takes on Health Care

Machine Learning Takes on Health Care 620 374 IEEE Pulse

Leonard D’Avolio
FIGURE 1 Leonard D’Avolio, Ph.D.

When Leonard D’Avolio (Figure 1: Photo courtesy of Cyft) was working on his Ph.D. degree in biomedical informatics, he saw the power of machine learning in transforming multiple industries; health care, however, was not among them. “The reason that Amazon, Netflix, and Google have transformed their industries is because they have embedded learning throughout every aspect of what they do. If we could prove that is possible in health care too, I thought we would have the potential to have a huge impact,” he says.
That led D’Avolio to spend 13 years employing and developing software as a researcher. Then, in 2015, he founded the Cambridge, Massachusetts, company Cyft as a way to bring machine learning and natural language processing (NLP) to health care.
The first step to tackling health care was to figure out where to start. Preventive care obviously saves money overall, but hospitals are based on a fee-for-service financial model, providing little motivation to promote wellness, D’Avolio notes. Other healthcare organizations, however, do have incentives to focus more on wellness: these include payer-provider systems, health plans that have very heavy care-management components, and third-party organizations that receive a fixed amount of money from health plans to care for patients. Cyft aimed its focus on such value-based organizations, which make up an estimated 5–10% of the healthcare industry.
The approach is working well, D’Avolio reports. Since deciding to apply machine learning commercially in health care, he has worked with 28 healthcare organizations. “I want people to understand that these are not theoretical scientific ideas that aren’t yet helping in the real world,” he says.
To accomplish this, Cyft software compiles patient data in its native format, including clinical notes, health-risk assessments, lab data, behavioral health information, and claims and device data, D’Avolio explains. It then applies machine learning, artificial intelligence, and NLP to analyze all these data to meet priorities set by a particular organization. The software can quickly match an individual to the intervention that will be most beneficial, which cuts healthcare costs. In addition, it can predict an increasing risk for a particular patient. For example, the Cyft software can use data to identify potential physical instability in a geriatric patient that could result in increased risk of a fall, and it can determine whether a patient with a serious mental illness is likely to have an impending in-patient psychiatric event within the next 90 days, which presents opportunities to intervene.
“We formed this company for one reason, and that is to make value-based care wildly successful in terms of both clinical outcomes and financial returns,” D’Avolio says. “We are at a very unique point in time when everyone is saying health care needs to change. If we can focus machine learning and NLP on evaluating all the data we’re collecting for this small percentage of care that is delivered in value-based models, we may be able to demonstrate that learning, quality improvement, and the novel utilization of data are a more attractive path than how health care has been delivered to date.”
D’Avolio adds, “Cyft is our best try at helping to make and keep as many people healthy as possible.”