Subscribe to our Feed

Thought Leaders in Big Data: Interview with Mike Byers, CEO of HighRoads (Part 4)

Posted on Tuesday, Dec 10th 2013

Sramana Mitra: Let’s say one of your clients – IBM, for example – wants to use this big data infrastructure to understand how to respond to their responsibilities [as laid out] in the Affordable Care Act. Is that correct?

Mike Byers: No. I was referring to insurance companies. But let’s look how an employer will look at that data. They are going to look at their plans from a couple of perspectives. Are they providing what the government is referring to as a qualified health plan? The extreme from that position is if the plan is too rich, they are going to end up having to pay something referred to as the “Cadillac tax.” So they need an efficient way to understand that the benefit plan information comes above the floor the government says that they have to, but is below the ceiling to prevent excess taxes. Having that data in an efficient model is critical, and that is what they rely on us for.

SM: The Affordable Care Act is going to create a whole bunch of heuristics against which you are going to do big data analytics on behalf of your major corporation clients and then give them additional actionable items based on that work.

MB: That is an accurate statement.

SM: How does your technology work in helping these insurance companies better process and handle large volumes of data?

MB: Insurance companies need to process very large volumes of claims data for both transactional and analytical purposes.  Historically, due to the unstructured forms of the plan data, transactional claims processing was often performed without the benefit of being able to match a claim against the specifics of the corresponding plan. This resulted in inaccurate, or incorrect, claims payments. HighRoads structured format can serve as the “master data” of the claims transaction process, providing a rigorous basis for processing claims.

From an analytic perspective, the plan data provides a backdrop of dimensionality to the claims data. For example, an emerging challenge for payers is to encourage the “young invincibles” to purchase healthcare plans in order to deepen the risk pool and lower costs. The payers have the problem of figuring out what would entice this young, typically healthy demographic to buy plans. Both sales and claims data can be used to perform the relevant analysis, but the “space” in which the analysis is performed is established by plan data. A payer may consider adding services to plans that the invincibles may find attractive, such as extended health club benefits, or dietary services or other wellness services that invincibles find compelling. This can be done in a kind of A-B testing – providing different combinations of services to see which invincibles buy by analyzing sales data. Here HighRoads structured form of plan data with the history of versions can provide the foundation of the sales data analysis.

Beyond looking at sales, payers can explore further evidence of what draws invincibles by looking at claims data to see which services they use or don’t use. Here again, the plan data provides the dimensional backdrop. Claims data represents consumption events of health services. You can tell when a service has been consumed just by looking at claims data. But non-consumption is tougher; it can be detected only when a service was available, but when there were no claim against that service. The plan data informs as to the availability of a service for the invincible, the claims data informs whether a service was consumed or not. The plan data defines the space of analysis, which when populated with claims data, may have sparse sub-spaces indicating disinterest of services on the part of invincibles. Consumption or non-consumption analysis of health services can be further decomposed to evaluate invincible behavior regionally, by gender, by age, etc.

More generally, when large scale claims analytics needs to be decomposed across clusters of servers, plan “master” data can guide how to break the job into separate streams, providing the natural fissures in the data by which the analysis can be parallelized such as enrollment period, plan, service, etc.

This segment is part 4 in the series : Thought Leaders in Big Data: Interview with Mike Byers, CEO of HighRoads
1 2 3 4 5

Hacker News
() Comments

Featured Videos