3 lessons the healthcare industry could learn from the intelligence community

When it comes to handling the growing hoards of data healthcare has at its disposal, the intelligence community may offer healthcare leaders a better model than the titans of tech.

And the industry could use the help, according to an op-ed in the New England Journal of Medicine.

Kevin Vigilante, M.D., chief medical officer of Booz Allen Hamilton, said well-meaning efforts to adopt practices of large tech companies like Amazon or Google to unlock the benefits of data are missing the mark.

“It’s not so much that they do anything wrong—it’s a more narrowly defined problem space where the decision frame is much more constrained to whether the consumer is more likely to buy this or that.”

RELATED: NIH gives first look at All of Us precision medicine research health database

Vigilante and co-author Steve Escaravage, a senior vice president in Booz Allen’s digital, analytics and strategy practice, see more potential in the way intelligence agencies handle information when it comes to organizing and parsing all that information—in part because the challenges intelligence agencies have had to overcome parallel those with which the healthcare industry is currently struggling.

Here are three key lessons healthcare leaders could learn from the way intelligence agencies handle big data:

1. How to deal with unstructured data

Unlike the narrow types of data consumer companies parse to get insight on buying decisions, the healthcare industry gets data from a much wider variety of sources and in a much wider variety of formats. To get the full benefit of all that information, it somehow needs to be integrated and processed so it can undergo analysis.

Intelligence agencies developed the concept of a “data lake” after the 9/11 Commission cited them for data gaps they could have closed by sharing information they had on hand. According to Vigilante, federal healthcare agencies face an analogous situation now. Without the ability to integrate and share information, the analysis that could trigger new connections, hypotheses and discoveries could end up sitting in agency databases indefinitely, just waiting to be uncovered.

2. How to share data without sacrificing security

Healthcare data are also subject to a huge amount of regulation—and for good reason. The intelligence agencies have deep experience handling sensitive information, however. The authors’ key recommendation here is something intelligence data scientists call “cell-level security.” That means systems control access to pieces of information at the most granular level possible. Vigilante cites systems such as Accumulo, which allows fine-grained control.

“If you imagine a data lake as a spreadsheet of infinite size, each cell has an item of data in it, and each item has a metatag that can be regulated to say who can access that item of data in that cell by their role, or even limit the time of day they can look at it—it’s an extraordinary level of security,” he said.

RELATED: HIMSS19: ONC unveils long-awaited information blocking rule

3. How to measure and ensure data quality

One of the other issues with large, varied data sets is the wide spectrum of quality in the underlying data. “In big tech, with enough reinforcement and alignment it’s okay to have low-quality data streams because they’re drowned out," Escaravage said.

Healthcare and intelligence have more urgent needs for precision. “One observation could indicate a completely different set of circumstances—and it’s not just quality, but methods: you have to understand the provenance of the data, the platform that did the collection, and the end-to-end lineage.”

Intelligence agencies deal with that issue through strict tagging of data and go a step further by requiring the deep involvement of data scientists within the overarching data pipeline. That’s an area healthcare will need to address, because this type of analysis requires a more dynamic approach to dealing with the source data than medical researchers typically take, according to Escaravage.

“Having the skillset to query into a data lake versus a static report is completely different,” he said.