Predictive analytics tool for readmissions boasts record results

82 percent accuracy
By Erin McCann
11:10 AM
Share
empty doctor's office

Researchers at University of Washington Tacoma have developed a machine-learning predictive analytics tool that forecasts with 82 percent accuracy the probability a patient will be readmitted within 30 days.

UW Tacoma's Center for Data Science was charged with the task of taking a closer look at heart failure readmissions data for the five-hospital MultiCare Health System. They needed an algorithm, a tool that would sort through the mountains of clinical EHR data and predict which patients were most at risk for 30-day readmissions, according to UW Tacoma news report.

[See also: A beginners guide to data analytics.]

The team was able to tap Microsoft's Azure cloud platform that helped the team "quickly scale to run multiple experiments simultaneously," explained David Hazel, managing director of UW Tacoma's Center for Data Science, in a news statement.

Now having the data at their fingertips, the team needed to come up with a set of measures that are integral in heart failure diagnosis, which finally led to the development of the Risk-O-Meter predictive analytics tool. The tool, contrary to standard predictive tools, which typically have accuracy rates around 60 percent, has an 82 percent accuracy in forecasting 30-day readmissions.

[See also: Data analytics top concern, but industry stumped about where to start.]

Then, they went even further. So the clinicians now had access to the readmissions tool, but what about the patient. MultiCare and UW Tacoma researchers wanted to put the tool into the hands of the patient, so they could see what risk level they had for readmissions. The tool uses the patient's information and then compares it to the troves of data from other medical records. The patient then gets back a percentage that explains their readmission risk.

Having seen such success with the Risk-O-Meter, UW officials hope to ultimately commercialize the tool.

Behind the development of the tool, there's serious motivation beyond the clinical aspect. In 2011, the federal government estimated that 30-day Medicare readmissions for all causes cost a whopping $26 billion annually, with $17 billion being attributed to avoidable events.

For Washington state, specifically, Medicare readmissions rates ranged from 18 percent to 20 percent. The national average, according to CMS data, currently stands at 15.2 percent.

By the end of 2015, CMS will have levied readmission penalties of $945 million on U.S. hospitals over three years. CMS expects that 2015 alone will result in $428 million in penalties.