Experts Say ‘Big Data’ Represents Future of Health Care Research


Physician-researchers debate the inclusion of large electronic health databases in comparative effectiveness research and RCTs 

Mattew RoeIn a didactic symposium at TCT 2015, Matthew T. Roe, MD, MHS, of Duke Clinical Research Institute, Durham, N.C., argued that “big data” are currently transforming cardiovascular research — both comparative effectiveness studies and randomized controlled trials — in ways that lower costs without sacrificing quality. Data are routinely collected by health care systems, and opportunities to leverage this information exist at multiple points for clinical trials, from pre-study feasibility analyses to recruitment of participants to compilation of data during the study, Roe said.

Registry-based trials “combine the advantages of an ongoing clinical registry with the rigor of a randomized trial,” Roe said. Data collected for the registry are used for the randomized trial along with some trial-specific information, but these types of studies have not yet been formally defined, according to Roe.  

One advantage that has been demonstrated is that big data studies greatly reduce costs. Roe contrasted two trials evaluating thrombus aspiration: the TASTE trial, which included 7,244 patients at a cost of 500,000 Euros, and the TOTAL trial, which included 10,732 patients and cost 15 million Euros. The two trials ultimately came to the same conclusion.

“Comparative effectiveness studies and RCTs in cardiovascular disease are being transformed by big data,” Roe said. “The question isn’t should we transform… but how do we transform?”

Reasons for caution 

Robert A. HarringtonRobert A. Harrington, MD, of Stanford University Medical Center, Stanford, Calif., agreed that great opportunities exist in big data, but he raised several areas of concern, some technological and some cultural.

“If we’re going to make big progress in a big data era, we have to have serious societal conversations about issues like informed consent. We also have to have serious conversations about data security,” Harrington said.

Technical challenges include assembling the “tapestry” of relevant data that may come from sources other than an electronic health record and overcoming the lack of interoperability of health record databases from different vendors.

Cultural challenges include consent issues, such as whether the public would be willing to forgo explicit consent for the use of personal records under some circumstances, as well as data security, an issue highlighted by recent widespread data breaches.

But theoretical issues about observational studies and data mining are perhaps the most significant challenges to big data, according to Harrington.

“Is more data better than a little bit of data?” Harrington asked. “Well, that depends on if you want precision or accuracy. You can get precision — you keep increasing your sample size, and big data is good at that. But the question is, does precision move you closer to the truth? Does it give you accuracy?”

Points of agreement 

Roe and Harrington agreed that large health databases provide opportunities for research but that big data methods are better suited to evaluation of therapies currently in routine use than to experimental agents and devices.

Moderator William S. Weintraub, MD, of Christiana Care Health Services, Newark, Del., summed up the discussion: “I think one of the take homes from this [debate] is that questions should always drive the methods, but you have to be very, very careful. … What we really want at the end of the day is accuracy. … Sample size does not overcome bias, so you could have the biggest data system in the world… but if you ask a question that can’t account fully for bias, P-values could be very small but could be absolutely wrong.”   

Disclosures: 

  • Harrington and Roe report relationships with multiple device and pharmaceutical companies. 
  • Weintraub reports no relevant conflicts of interest. 

 

Comments