The Data Scientist - the WPC Healthcare Blog

Crowdsourcing an Answer

How Data Science Solves Healthcare Problems

Facebooktwittergoogle_plusredditpinterestlinkedinmail

Question: What do data scientists do for fun—and to sharpen their keen analytical skills? They solve incredible problems (not your daily newspaper’s Sudoku puzzle).

As healthcare begins to dream about predictive modeling (the holy grail of population health management), data scientists will become part of a comprehensive data and analytics strategy for health systems and payers as we move to value-based care. At WPC we’re just a bit ahead of the curve, and we have the chops to prove it.

Like many industries, healthcare (especially advanced research organizations) is getting into the crowdsourcing game to solve complex problems at an affordable price point. Recently, for example, the Mayo Clinic threw out a problem to a group of data scientists on a global basis—think 99Designs for logo design.

This particular needle-in-a-haystack project focused on the more than 2 million people with epilepsy and the half million who are subject to drug-resistant seizures. Physicians looking to apply DBS (deep brain stimulation) as a treatment for epilepsy need highly specific information on the data surrounding pre-seizure brain activity. So EEG data from both human and canine subjects were provided to approved teams.

Stay with us here. The goal was to find a variable to identify an algorithm to predict seizure activity. In other words, who will have a seizure and who will not. Two hundred teams took up the 60-day challenge, and we are pleased to report that Team WPC came in second by .0006 of a percentage point. I’d say that’s a photon finish (and a lame attempt to spin “photo” finish). The benchmark for this project was 65% accuracy, and the top two teams (us and the other guys) hit 95%—that’s an A +++ in the data science world.

So while projects like this are not directly related to his fulltime work at WPC, lead data scientist Damian Mingle uses this type of data puzzle to hone his skills, applying what he knows and testing what he doesn’t. Mingle and his team have created an artificial neural network that acts as a starting point for analysis. (Kids don’t try this at home) From there they apply creativity and curiosity to generate a particular solution. It’s like a proprietary secret sauce that must blend with new input in order to not “blow up.”

Although healthcare is the business of WPC, Mingle focuses on a wide variety of projects because data are, after all, data. To generate their “answer” to the epilepsy question the team utilized what they’d learned from trying to locate black holes based on the halo effect of starlight by separating signal from noise. They won that one, by the way.

So back to WPC and the analytics solution set offered to clients on a national basis. The goal of the Taproot Insight offering (part of the company’s Taproot Solution Suite) is nothing short of predictive analysis. Knowing what happened in the past is important to addressing issues within the organization, but getting proactive requires a bit of a crystal ball.

As part of the WPC process, foundational steps are taken to do descriptive and exploratory data analysis as a baseline with the goal of normalizing the data and identifying the outliers. From there, inferential analysis is used to create incomplete data sets. Not surprisingly, healthcare data are messy compared to other industries due to its status as a comparative juvenile requiring a level of sophistication to drive actionable insights. Addressing the “missingness” of the data is critical in moving forward. Bridging gaps in understanding is key to finding answers.

We then take it one step further (yes, there’s another step). In order to create meaningful change and improvement, predictive analytics must move into something we call the “mechanistic phase.” That’s the successful implementation of what’s been discovered so that when action is taking place, the outcomes are expected. In fact, Dr. Benjamin Brinkmann of the Mayo Systems Electrophysiology Lab asked our data scientists to document the algorithm so that their researchers could use it as a basis for innovation.
That’s the top of the pyramid. We don’t just solve problems; we put the answers into repeatable practice. So if you have a health problem, you want an expert, right? We think you deserve the same thing in regard to predictive analytics, the kind that transforms data into understanding.