Analytics Tip: Stop Cleaning Your Data

Adopting a basic set of principles is the key to success with healthcare data analytics, according to John Showalter, M.D., chief health information officer at the University of Mississippi Medical Center.

During a presentation at the Healthcare Analytics Symposium sponsored by Health Data Management, Showalter said the top two principles are: healthcare analytics is all about the individual patient, and don’t waste time and money cleaning your data before analyzing—just start using it.

Also See: HDM Announces 2015 Analytics All-Stars

Analytics enables a provider to identify an individual at risk, and to work with the individual to change risky behaviors and overcome barriers, he said. Focusing on the individual helps to identify barriers, such as no pharmacy on the individual’s bus line. A provider can improve the health of a population by improving the health of one individual. So, Showalter emphasized, stop cleaning your data. It’s dirty. How do you know the codes in progress reports that have been turned into data are accurate and really reflect the patient? You don’t.

Cleaning data is expensive and doesn’t improve analytics, Showalter contended. When cleaning, there is a lot of data thrown out that doesn’t have to be. That’s intelligence you’ll never analyze. Use of neural networks and machine learning technology can help create many associations out of the data. The bottom line is simple for Showalter: “Twitter and Facebook have horribly dirty data. They don’t clean it, they just use it.” It works and healthcare should follow suit.

Current analytics work at UMMC includes sending every clinical note through natural language processing technology to convert SNOMED concepts and sent them out for analysis. This can speed up decisions to take action. Showalter also advocates using geo-coding, as it is cheap and effective. UMMC took two million addresses from its data warehouse and geo-coded them in about three months. Now, physicians and researchers can pull up maps and look at diabetics, healthcare disparities and other metrics by zip code. And the software to do this cost less than $10,000. It’s not predictive analytics, but it shows where the community stands.

Other tips from Showalter, Health Data Management’s 2015 Clinical Visionary All-Star Analytics recipient, include:

Don’t go it alone. Find partners, such as a vendor, hospital or clinic. Don’t reinvent everything. Stop creating silos. Stop trying to shut down competitors and cooperate with them on lowering readmission rates and other initiatives that meet patient needs. “All ships rise with the tide.”

Don’t put analytics results in a portal; put results in the electronic health record. Just like lab tests flow to the EHR, so should analytics. “They are a lab test of your data.”

UMMC contracts with an outside vendor to conduct its predictive analytics and has had some stunning results. For instance, 70 percent of an analyzed high-risk group was predicted to have a heart attack within three to six months. “Stop cleaning data, integrate the results and stop trying to be perfect,” Showalter said.

For reprint and licensing requests for this article, click here.