Friday, April 3, 2015

Uncertain Health in an Insecure World – 32

“Size Matters”

Managing healthcare big data with open architecture computing platforms like Hadoop is possible.

 Most major developed world healthcare systems and jurisdictional health insurance plans are still paying for their capital investments in enterprise level electronic medical records (EMR’s). It remains unclear whether these entities produce big enough data to warrant the cost of doing business with Hadoop level processing power.

Certain industry sectors handle truly massive data, and were the raison d’être for Hadoop’s creation a decade ago as a highly efficient data storage and processing ecosystem. Companies like Google and Yahoo! were there at Hadoop’s inception, and LinkedIn and Facebook have enjoyed legacy benefits since.  These social media behemoths leverage 10,000's of nodes in Hadoop clusters in lieu of using typical enterprise software (Microsoft, Oracle) and disk storage.

Discovery scientists can effectively employ this platform to manage very large gene sequencing data sets. Public health investigators can correlate the U.S. National Oceanic & Atmospheric (NOAA) Administration weather data with children’s asthma admissions. And environmental toxicologists can compare U.S. Environmental Protection Agency (EPA) chemical levels in effluent waste water to long-term cancer rates. But a typical patient generating only 100MB of data per year in a modern healthcare system which must store 200 terabytes of data over 20 years does not demand Hadoop computing power.

By comparison, Facebook adds 500 terabytes of data capacity every day to handle its big data demands: 4.8 billion content changes, 4.5 billion ‘likes’, 10 billion messages and 350 million uploaded photos!

A mid-sized domestic U.S. airline’s fleet of 600+ Boeing 737’s generates >250,000 terabytes every day of unstructured data, most of which is neither stored nor actively used. Yet, in the wake of the Flight 9525 tragedy, these streaming data may eventually become part of a system that can take over the operations of a rogue airplane from the ground.

If healthcare systems used Hadoop’s power, what would doctors and providers actually do with the data to improve the quality or quality of healthcare? 

Hadoop advocates propose using MapR for case-by-case personalized treatment planning, diagnostic predictive modeling, prescription fraud detection, remote vital sign monitoring, legacy medical records searches, and patient self-directed care and prevention management. Some of these Hadoop ecosystem functionalities might even allow for smart programs to learn from trending of streaming data (i.e., machine learning). University of California at Irving Medical Center initiated an small 8-node Hadoop platform in 2012, with some early economies of scale. However, such big data applications are generally beyond the affordability threshold of most healthcare systems or jurisdictions, and most lack the highly qualified personnel to support it.

The Cloud also offers potentially secure healthcare information sharing and data storage capacity, but it is non-analytical. The only way for healthcare systems to predict failures and/or solve problems before they occur is through smart programming or machine learning, whereby the power of analytics can be brought to bear on massive data sets in real time, during the actual patient care moment

People, especially untrained people, are the problem with this promising construct.

The slow adopters for EMR will be the same foot-draggers on big data analytics. Not even severe financial penalties for poor healthcare informatics performance have moved the needle much; in many developed counties, such penalties do not exist. Progressives like the Cleveland Clinic spun off Explorys which offers the allure of combining clinical and claims data to establish population-based health profiles to either measure or mitigate future risk. Wherever you sit in the purchaser-provider split, such information would be powerful as a means to determine insurability and project future costs of care, for individual patients, bundled cohorts and entire countries.

Take a small country, like Denmark – population 6.2 million.

A colleague of mine has recently published research in Nature Communications from the entire Danish population, where standard non-Hadoop computer crunching of medical histories in a 14.9 year medical claims registry led investigators to predict disease progression from previously unseen patterns of health data.  These so-called disease “trajectories” could potentially be a key tool in predicting or even preventing future diseases for selected patients within these risk profiles.
Take a big country, like India – population 1.3 billion.

Big data and analytics has exploded in this, the world’s most dense population center, where the current $1 billion market could more than double to $2.3 billion in 2018. Whether personalizing patient services at scale or trending unstructured data from devices & monitors, India's healthcare sector is now in play.  

For example, Metaome company produced DistilBio, a web-based search or enterprise level computing platform that generates health risk patterns from private & public laboratory data management systems. In a country where 29.5% of the billion plus people are below the poverty line, and where 46% of children are malnourished, there is hope that big data analytics can also improve the country’s rural under-served health system, achieving Svasth Bharath (Healthy India). 

Hewlett Packard and other big data and analytics firms with a firm India footprint in mining, oil & gas, business & retail, video security, social media, engineering, etc. now have a healthcare market presence. And India's software engineers and analysts are relatively abundant, spawned by a growing number of digital health innovator start-ups and accelerators.

Countries and companies considering the use of open architecture computing to analyze their healthcare big data should look critically at the size of their platforms before taking this leap.

The Square is small, but the risks and related rewards should not be underestimated. 

No comments:

Post a Comment