Predictive analytics lowers readmissions
September 15, 2014 in Medical Technology
The challenge for the Carolinas Healthcare System was to reduce the readmission rate for patients with chronic obstructive pulmonary disease. The solution: predictive analytics.
One of the biggest problems, according to Jean Wright, chief innovation officer at CHS, was in identifying patients at risk of readmitting before they leave the hospital and enable care providers to intervene.
[See also: Predicting what's next: Isn't that the work of the physician?]
Faced with CMS readmission penalties the healthcare system had to take swift action.
[See also: Predictive modeling gains adherents]
“We knew at that point in time that COPD was not one of the conditions that was currently being measured and penalized under the CMS system but we knew it was on the agenda for 2013 and 2014,” said Wright.
CHS selected San Juan Capistrano, Calif.-based Predixion Software. Predixion utilizes Machine Learning Semantic Model that enables predictive applications to be quickly implemented into production environments, making the predictive process, according to the company, easily adaptable and repeatable.
Nish Hartman, director of healthcare at Predixion Software said that preventable hospital readmission costs $25 billion in wasteful spending.
“What makes prediction impactful, said Hartman, “is what we call the last mile. This is serving up the risk scores in interventions in a very clear and simple-to-use interface. This is where the true value of your data is realized at the point of care with actionable insights delivered in real time that can be acted upon immediately.”
An action plan was implemented and CHS rolled out a model in 2013.
Wright told participants at a webinar titled: Bringing Predictive Analytics to the Point of Care, on Sept. 10, that initial implementation was completed in six months and included seven hospitals. The solution is now fully deployed at 13 of CHS’s metro-based hospitals where 100,000 plus patients have been managed with clinicians knowing their 30-day risk for readmission.
Wright said that in a year and a half to a two-year period CHS was able to drop the readmission rate from 21 percent to 14 percent.?
CHS took a multidisciplinary approach in attacking the problem of readmissions and a key part of the project team included nurses. “We knew nursing was a critical element. I know how important nursing is particularly when it comes to transition of care, discharge planning and case management,” said Wright.
Wright explained that as they built the model they decided to segment patients into four groups: very high, high, medium and low risk.
The Predixion tool allows case managers to review patient lists that contain increased risk indicators and decreased risk indicators that allows them to see specific variables that caused someone’s risk to either up or down.
For example, what caused a patient to have left hemispheric stroke? “If one of these variables had to do with her anemia or her electrolytes if I kept her another day or so and improved that clinical condition, I could in that moment of time, measure her risk of readmission and see if it would impact change. So this is real decision making using live clinical information at the point of care,” said Wright.
Wright said that by utilizing predictive analytics tools, you could make a decision and use an intervention that the team chose based on their knowledge of the evidence and then create a permanent trail as to which intervention you have applied to a specific patient.
“About 80 percent of the time when we work off the model, we are right smack on target. If we say that someone has a high risk of readmission the patient is probably going to be readmitted and if we say it’s very unlikely, then they’re probably not going to be readmitted,” said Wright.