© 2019 SourceMedia. All rights reserved.

Is Predictive Analytics on the Horizon?

September 20, 2010 - BI reports generated within health care usually are retrospective looks at cumulative data. But what if BI could be used to actually predict patient outcomes and risks, and steer clinicians to steps to provide even better care?

That's the question that Novant Health, a nine-hospital delivery system based in Winston-Salem, N.C., hopes to answer. For the last decade, Novant has been using data mining and analysis tools from MEDai. The health system has compiled benchmark data for 50 distinct disease states, says Jim Lederer, M.D., vice president of clinical improvement.

"Our data is great, and helps us identify opportunities to improve care," he says. "But a better step would be predicting patient experience before it happened, rather than evaluating it after it happened."

Toward that goal, Novant is an alpha site development partner with MEDai on a BI platform called "PinPoint Review." Using the tool, clinicians at Novant will be provided a list of patients ranked on a scale of 1 to 10 for their risk of certain outcomes, including hospital readmission and the development of pressure ulcers during a current stay.

The amount of data needed to create such predictive scorecards is huge, Lederer says. "From tens to hundreds of thousands of HL7 messages every hour," he estimates. The data, which might include lab results, vital signs, and X-ray reports, goes to MEDai in a real-time data feed, run through an artificial intelligence processor, and then returns to Novant in the form of a predictive scorecard.

The system, Lederer says, would not replace clinician judgment, nor would it necessarily be foolproof.

"A prediction is nothing more than a prediction," Lederer advises. "I have to validate it by looking at the patient. If I have a risk of 2 out of 10 for mortality, it doesn't mean I won't die. But if I have someone with a high risk of transfer to the ICU, (I can ask) what can I do to prevent it."

By the end of the year, Lederer expects that the predictor for patient readmission will be complete. The effort will also include predictors for mortality and ICU transfer risk for patients admitted to the general floor, and LOS and mortality risks for all patients in the ICU, he adds. The reports would be delivered real-time thru a portal. "We can parse out the population down to patients who need more attention," he says.

This originally appeared on Health Data Management.

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