October 10th, 2008

Making the giant health database work for you

Posted by Dana Blankenhorn @ 10:27 am

The big health database is at the center of both Presidential campaigns.

Whether you’re Republican or Democrat, your guy (or gal) thinks that automating health records is a “silver bullet” from which billions in savings will flow.

But for that to happen some really massive databases need to be integrated and made useful, in hospital emergency rooms and doctors’ offices. You need your brain scan to show up on the screen, not your second cousin’s.

This is not easy. (Picture from MRI Segmentation paper filed at Tel Aviv University in Israel.)

Netrics, which is also aiming at helping build national security databases, is one of the leaders in this new integration. CEO Stefanos Damianakis was kind enough to map out the challenges for ZDNet Healthcare yesterday.

“First you need to connect the systems. Then you have to really integrate the data. You have to deal with natural variations in the databases. You can’t assume the data is perfect.”

Companies like IBM and Microsoft, which have recently entered the business through acquisitions, try to do this with data matching, he said. To find my name, for instance, you search through the database for common variants, like Blankenship.

It’s a “rules-plus” approach. By contrast, Netrics is using machine learning and mathematical modeling, techniques that emerged 20 years ago in what was then called Artificial Intelligence or AI.

“The way we deal with it is not to make the data 100% perfect, complete, and right. Keep it in reasonable shape and create smarter software that can interact with the data and find records, and link records, across databases.”

Netrics’ mathematical models start by trying to figure out what a human being might do in a medical records room, then it lets the operator teach it the rest of their techniques for getting things right.

In this way, the Netrics system learns, and its results improve, while the compute power required to reach these results gradually declines.

It’s these medical records experts whose function has been lost in the translation from paper to magnetic ink, Damianakis concluded, and unless that is recovered digital systems will never work as well as paper.

Google’s algorithms try to use metadata in order to perform this function using unstructured data. Find what others have clicked on in response to a query and you can order 67 million pages on electricity with some accuracy.

But medical databases are structured. ”The link structure analysis won’t work. You have to throw a different technology at it,” he concluded.

This sounds rather deep, analytical and complex but it’s really important. The one, big database we rely on to make health care reform work won’t really be one database, but many databases, whose contents will have to be matched to be useful.

Applying human intelligence, on the fly, to the problem of data matching is key to making the one big database more useful than the old paper records room.

That in turn is the key to taking records online.

And everyone agrees that taking records online is the key to health reform.