The July 2014 issue of Health Affairs is on "Using Big Data To Transform Care." Here are some papers that caught my attention.
Big Data in Health Care: Using Analytics To Identify and Manage High-Risk and High-Cost Patients, by D. Bates, S. Saria, L. Ohno-Machado, A. Shah and G. Escobar. (Paper) This article presents, at a high level, six use cases "where some of the clearest opportunities exist to reduce costs through the use of big data: high-cost patients, readmissions, triage, decompensation (when a patient's condition worsens), adverse events, and treatment optimization for diseases affecting multiple organ systems."
Optum Labs: Building a Novel Node in the Learning Health Care System, by P. Wallace, N. Shah, T. Dennen, P. Bleicher and W. Crown. (Paper) This paper is about Optum Labs, a joint entity created in early 2013 between the Mayo Clinic and Optum, an organization part of UnitedHealth Group. Partners now include AARP, AMGA (the American Medical Group Association), Boston Scientific, Boston University School of Public Health, Lehigh Valley Health Network, Mayo Clinic, Optum, Pfizer, Rensselaer, Tufts Medical Center and the University of Minnesota School of Nursing.
Here are examples of research findings:
- There are a number of seemingly equivalent choices for second-line medication treatment for uncomplicated type 2 diabetes. (Second-line medication is medication you give if the patient doesn't respond as expected to first-choice or first-line medication.) An analysis of the Optum data, however, showed that a specific drug led to less cost and a longer interval until insulin was required. This has the potential to change guidelines used by care providers.
- Another Optum study showed the number of total knee replacement surgeries has increased steadily for all age groups, as has the prevalence of diabetes and obesity among patients undergoing these surgeries. Hence, it seems important in future studies to differentiate patients with multiple comorbidities.
- Another Optum study showed that "the statistical linkage of clinical information from laboratory test results to claims data reduced bias in treatment effects estimates by 24 percent for all ambulatory visits and by 79 percent for visits related to Hepatitis C, relative to the use of claims data alone."
A key overall challenge (and priority) remains to "enable clinicians to connect findings from big-data analyses directly to the care of an individual patient".
Leveraging the Big-Data Revolution: CMS is Expanding Capabilities to Spur Health System Transformation, by N. Brennan, A. Oelschlaeger, C. Cox and Marilyn Tavenner. (Paper) CMS has collected claims, assessments and surveys for many years, and has long made them available to researchers. However, data was previously shipped to researchers in the form of encrypted hard drives, at a rather significant cost for the researchers. In late 2013, the agency launched the CMS Virtual Research Data Center. Researchers can now access the data from their own computers but they "may download only aggregated data sets and analytic results." This is obviously expected to decrease the cost to the researchers (no more encrypted hard drives arrivign through the mail!) Innovations have also been made in the qualified entity program to improve provider quality reporting and in sharing data with providers and states. Further, the Blue Button program "allows patients to easily access their own health information in electronic form."
Early Experiences with Big Data at an Academic Medical Center, by J. Halamka. (Paper) This describes the experience of Beth Israel Deaconess Medical Center (BIDMC) in the Boston area in adopting electronic applications. While data storage now amounts to 3 petabytes (1 PB = 10^(15) bytes) and grows at a rate of 25% a year, the main issue has shifted to the transformation of this raw data into usable information.
In order to gain insights, BIDMC created "Clinical Query, a web-based tool that allows for analysis and visualization of the clinical data collected by BIDMC." The author explains: "The principle behind Clinical Query is that investigators will want to ask questions before conducting research that will help them understand the potential statistical power of a clinical trial or the availability of data for clinical research." The resulting web-based query tool uses Boolean (and/or) expressions and open-source code from the Harvard's Integrating Information from Bench to Bedside project. The author identifies the potential issues related to Big Data: (1) data quality, (2) variations in data collection over time, (3) inconsistent use of medical terminology, (4) patient privacy concerns, (5) need for experts to meaningfully analyze data.
This issue of Health Affairs was very interesting. I do wish that an InnoCentive-type competition where de-identified claims and diagnostics data would be posted online (perhaps using the CMS virtual system with its built-in protections) for all interested parties to study and analyze. When the data is only accessible to some groups of researchers, opportunities for insights are lost because other teams of researchers may have approached the problem a different way or tested a different hypothesis. Further, I also think that the field is ill-served by the current lack of mathematical models on the topic. Health Affairs never presents mathematical models to begin with, but when it comes to analytics, there comes a point where the buzzwords have to stop and the analytical insights have to begin - and those analytical insights have to extend beyond the data-mining insights that are already being derived. (Data-mining is a good first step, but operations research can do so much more.) Given the tremendous progress that has been made in a few short years, though, the state of analytics applied to healthcare treatments is sure to become quite impressive indeed.