Healthcare data mining: OptumLabs collaborative begins to offer insight about what works

Diabetic patients are a popular target population for health systems that hope to improve patients' health with better care coordination and closer management. But new research from an insurer-owned, big-data analytics project suggests diabetes management can be too aggressive.

The findings are some of the first to come out of OptumLabs, a research collaborative owned by UnitedHealth Group's management and analytics subsidiary Optum, of which the Mayo Clinic is a founding partner. Now in its third year, OptumLabs has produced nearly two dozen studies and expanded its membership and ambitions. That growth comes as competition to mine healthcare's ocean of data intensifies. IBM Corp., which in April launched its health analytics arm Watson Health, Geisinger Health System's xG Health Solutions, and Aetna are among the companies vying to crunch big data sets to better understand what works and what doesn't in healthcare. These companies hope to develop strategies, services and products to capitalize on that information.

The Mayo Clinic and the American Medical Group Association, another OptumLabs member, are starting to apply the research consortium's early results. Lab officials say practical application is the goal of all OptumLabs' projects. OptumLabs' members consider which research projects to pursue based on which ones could help inform clinical decisionmaking.

“They think about translation from Day One,” said Dr. John Cuddeback, chief medical informatics officer of the American Medical Group Association.

Using OptumLabs data on more than a million patients with Type 2 diabetes, Mayo researchers studied the potential for overtreatment among those diabetics who receive frequent testing for blood sugar but don't have problems controlling their glucose level. Meanwhile, it's estimated that 8.1 million diabetics in the U.S., or about one-quarter of everyone with the disease, remain undiagnosed.

The AMGA's 40 medical groups will use the Mayo Clinic research to identify variations in diabetes management across physicians. The OptumLabs research results will help identify possible overtreatment of some diabetics. “What we ought to be doing is testing people who haven't been tested or testing those who are out of control,” Cuddeback said.

That study and others like it are possible thanks to the volume and detail of OptumLabs' data, which include a decade of de-identified records for 150 million patients. UnitedHealth Group and Mayo contributed initial data on hospital visits, prescriptions, laboratory results and diagnostic tests. The AMGA contributed additional data, including physician notes from patients' medical records. Other research to date has targeted comparative effectiveness of diabetes drugs and the safety of anticoagulants. The laboratory has 70 projects underway and two established areas of emphasis: heart disease and Alzheimer's disease.

When OptumLabs was launched, Dr. John Noseworthy, Mayo's CEO, said one of the primary research goals was to compare the effectiveness of treatment options. But as the collaborative gains members, research projects have expanded beyond those initial goals.

For example, OptumLabs' huge data set can be used to develop more accurate predictors of illness, experts say. In addition, it could be used to advance cognitive computing, also known as artificial intelligence. Cognitive computing could help develop better tools for mining healthcare data for new discoveries. IBM Watson Health has marketed its services as cognitive computing.

New algorithms developed and tested using OptumLabs data could help sift through numerous correlations between patients' health, treatment and outcomes to find those most relevant, said James Hendler, director of the Rensselaer Institute for Data Exploration and Applications at the Rensselaer Polytechnic Institute, which joined OptumLabs soon after its launch.

OptumLabs is expected to announce a new organizational member that's “very well known in data science,” said OptumLabs CEO Dr. Paul Bleicher. Cognitive computing is a powerful tool “to apply to thorny healthcare problems to develop new insights and tools,” he said.

Together with the National Quality Forum, OptumLabs also may launch an incubator for developing quality measures. The potential partners are currently testing the idea with a simulation.

The incubator would connect funding, data and expertise with researchers and clinicians who hope to develop new quality measures, said Nicole Silverman, the quality forum's chief operating officer. So far, development of good quality measures has lagged behind demand, she added.

OptumLabs' large data set could further be used to speed up testing of specifications for quality measures as they are developed. As it stands, specifications for measures are developed and then tested. Revisions after testing must then be developed and tested again. “It takes two to three years and it's expensive,” Silverman said. Simultaneous development and testing would hasten the process.

OptumLabs collaborative members now include consumer groups, universities and life science companies. AARP joined the consortium a few months after it launched. Pfizer, two health systems and three colleges and universities soon followed. By the end of 2014, two more universities, the not-for-profit Medica Research Institute, drug company Merck & Co. and medical-device maker Boston Scientific were members. Another half-dozen academic and corporate members have joined since January, including the Massachusetts Institute of Technology and Novartis Pharmaceuticals Corp.

Members bring new research agendas, data or funding. “You bring one or the other,” said the AMGA's Cuddeback.

The AMGA brought data. So did the Mayo Clinic, which helped to found the lab but is not an owner. Optum, the consulting arm of insurance giant UnitedHealth Group, owns OptumLabs. Optum ended last year with earnings of $3.3 billion on $47.7 billion in revenue. Optum declined to say whether OptumLabs research has generated proprietary products for Optum.

“Optum's vision in co-founding OptumLabs was to create an environment where health system participants could collaborate and innovate on topics related to improving patient care,” an Optum spokesman said. “The benefit to partners and other stakeholders accrues from having the opportunity to have a place at the table to drive innovation and research translation.”

Optum covers OptumLabs' operating budget, which includes 50 research scientists, project managers and other employees. Optum declined to disclose that budget, but said the goal is for the research laboratory to support itself. Larger research projects may require additional financing.

The research endeavor produced 20 journal articles in its first few years. Those included a closer look at the risk of gastrointestinal bleeding by age for patients who took blood thinner products that had recently entered the market, rather than taking warfarin. That research, published in the British Medical Journal, found a lower bleeding risk for those younger than age 65 who were prescribed the new anticoagulants. But the risk was unsettlingly high for those older than age 75.

The OptumLabs data set was large enough to allow analysis by age subgroups. Earlier research using clinical trial data to explore the risks of gastrointestinal bleeding had raised concerns about medication use among older seniors.

But the clinical trials required for Food and Drug Administration approval are too small to reliably test drug safety among patients separated into groups by age or other traits, said Dr. Jayant Talwalkar, medical director for the OptumLabs collaboration in the Mayo Clinic Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery.

OptumLabs' members expect to continue to add data, including data on patients' experience and satisfaction. Talwalkar said the ability to incorporate more information from patients will strengthen the value of the data set.

One diabetes research project seeks to identify the least costly way to achieve the best performance on quality of care measures. An analysis of OptumLabs data shows largely similar, solid performance on quality measures among the AMGA's member medical groups. But the groups have numerous options for how they treat diabetics, and evidence for the most cost-effective choice is limited.

Researchers can use nuanced clinical data from doctors' notes and information on prescriptions in the OptumLabs data base to study when the most costly diabetes drugs might be more effective than older, cheaper drugs, and vice versa. The difference can amount to $3,000 a year per patient, Cuddeback said.

Correction, Oct. 6, 2015: