Making predictive analytics part of healthcare requires moving past these challenges

At a health IT conference I attended in Harrisburg last week, I learned that one data stream that payers are interested in is credit scores as a predictor of health. The thinking is the worse the credit score, the worse their health is likely to be. Of course, they could just be behind in their […]

At a health IT conference I attended in Harrisburg last week, I learned that one data stream that payers are interested in is credit scores as a predictor of health. The thinking is the worse the credit score, the worse their health is likely to be. Of course, they could just be behind in their medical bills, but you get the idea. It seems like all consumer data involving transactions is of keen interest for predictive analytical tools, particularly for healthcare.

As UnitedHealthcare CMO Sam Ho points out in a new Rock Health report:

“Current data sets generally revolve around claims data but that’s going to be changing with lots of clinical data and transactional information with lifestyle becoming more readily available.”

The report highlights several approaches to predictive analytics in healthcare and the inherent challenges with each of them, including data aggregation, relationship search, realtime data collection, contextualized data and performance capture. Oncology company Flatiron Health uses data aggregation to develop best practice approaches for specific types of cancer. It involves making unstructured data useful. Another challenge includes integrating that data into the clinical workflow in realtime and all the awhile adhering to HIPAA requirements for keeping patient data private.

In other instances the challenges highlight the reliability of some data. Although patients are intimately familiar with their condition in a lot of ways, the data they convey may inevitably be less than accurate.

Among the biggest market opportunities Rock Health identifies for predictive analytics is “overtreatment” — which costs $192 billion each year, according to the report. It sees predictive analytics as a way to limit treatment to only those patients who will benefit the most from it. That description sounds a bit stark. And it’s the source of some ongoing debate because the conclusion would likely depend very much not only on which data is being used to make this decision, but also who is making that decision in the first place — clinicians or payers.

These are the sorts of clinical decision support tools that the FDA is likely to take a greater role in regulating, but the measures it will use to assess these technologies is still under review.