The proliferation of computers in the daily operation of healthcare providers is generating unprecedented volumes of data. This volume will grow in size and complexity with the expected adoption of low-cost sensors. Some will be prescribe for research purposes, such as pediatricians monitoring asthmatics to find the environmental triggers behind their attacks. Others will be initiated by individuals outfitting themselves with a raft of consumer electronics that measure such signs as heart rate and blood pressure.
Buried in the resulting databases may be answers to how to improve patient outcomes and hold down the cost of healthcare. It’s a great example of how Big Data can be put to good use.
At the end of October, the IEEE (Institute of Electrical and Electronics Engineers) will host its 2015 International Conference on Big Data in Santa Clara, Calif. Embedded in that conference will be two workshops focused on the potential of Big Data in health IT.
One, titled “Deriving Value from Big Data in Healthcare,” will serve as “an interdisciplinary forum for data scientists and clinical researchers to exchange ideas and share information.” The other workshop is called “Mining Big Data to Improve Clinical Effectiveness.” Its stated goal is to “bring together researchers working at the intersection of big data mining and healthcare to share with and learn from each other.”
It’s hard to put a dollar amount on the value of healthcare data but the management consulting firm McKinsey & Company has tried. Based on early successes in the application of big data analyses, McKinsey estimates savings in healthcare costs between 12% and 17%. Extrapolated to the $2.9 trillion spent on healthcare in 2013, this translates to between $348 billion and $493 billion in cost reductions. And that is in 2013 dollars.
Achieving such gains will not be easy. But the potential is there.
In the February 2015 Journal of Digital Imaging, investigators at the University of Maryland in Baltimore described how they leveraged data obtained during the National Lung Screening Trial to develop a clinical decision support tool that could be applied to everyday lung screening.
Patient demographics and lung nodule characteristics of NLST subjects were converted into Structured Query Language tables and uploaded to a web server. Using a web-based application, the researchers queried the database in real-time. The results were nothing short of mind boggling.
By transforming trial results into an easily accessible reference database, data obtained from individual patients could be matched and the patients expertly and objectively assessed. The size and shape of nodules found in these individuals were compared to those of the research subjects. The matched data was then evaluated in the context of patient smoking history, age, and geographic location. The result was a personalized index indicating the likelihood of whether the nodules found in the screened individual nodules were benign or malignant.
Last spring, at the Society for Imaging Informatics in Medicine (SIIM) annual meeting, Dr. Arjun Sharma of the University of Maryland, cited the NLST along with the Prostate, Lung, Colorectal and Ovarian Cancer (PLCO) Screening Trial as examples of the kind of massive data sets from which personalized approaches to cancer screening might be achieved. Bolstering his claim, Sharma described a Web-based app that could match patient information from the PLCO trial to results obtained during individual screening in the context of patient demographics and an assessment of personal risk factors. The matching data led to a detailed analysis that could be used to guide patient-specific treatment and management.
Sharma opined that by using matched data, “an individualized diagnostic decision-support system can personalize imaging, testing, follow-up intervals, intervention, and prognosis.”
Hundreds of thousands of dollars are commonly spent in the medical and radiotherapeutic management of individual cancer patients. In this context, the hundreds of billions of dollars that McKinsey estimates might be saved through the use of Big Data doesn’t seem so outlandish.
Early gains might be achieved by mining genomic and proteomic data bases. Other low-hanging fruit might be found in the data collected by the U.S. Centers for Disease Control and Prevention. Studies of chronic illness and common diseases, supported by government as well as private companies, might be tapped as well. The greatest potential, however, is the largely untapped reserves of data now being built everyday through the use of electronic health records.
The vast majority of these data are not used beyond the confines of the healthcare enterprise that collect them; most are not even used to their potential internally.
Yet, as the raw material of future clinical decision support, they represent the means by which extraordinary improvements in patient outcomes and healthcare savings might be achieved.