Luis Louro - Fotolia
In 2014, Cleveland Clinic wanted to get a better handle on controlling the cost of knee replacement surgeries done at its surgery centers. The health system performs about 2,600 of the procedures annually, so even a small cost reduction on each one could add up to huge savings for the clinic and patients. To find what it was looking for, the provider turned to predictive analytics techniques and tools -- but its analytics team also had to spend time convincing surgeons to base treatment plans on what the data was telling them.
Steven Spalding, Cleveland Clinic's medical director for enterprise information management and analytics, said his team developed assessments built into the organization's electronic health record system that score knee-replacement patients on their likely length of stay and their probability of being discharged to someplace other than home, an indication of surgical complications. The analysts, in partnership with surgical specialists, also developed recommended care strategies for surgeons to follow based on individual patients' risk factors.
When the predictive models were implemented, they had some impact on hospital stays and the rate of complications from the surgeries. But outcomes really improved when the health system finally got all of the surgeons on board with following the prescribed care paths, something that didn't happen initially.
"We made some progress, but we realized it's not the tool or the analytics," Spalding said, speaking at the 2015 Big Data in Healthcare Summit in Boston. "It's the people and the process -- and unless you have those things lined up, who cares if you've integrated big data and produced an awesome model? You've got to have the buy-in."
In the end, it took continued pressure from senior leadership to get everyone on board. The chair of the clinic's Department of Orthopaedics sent a letter to all surgeons explaining that compliance with the care paths was mandatory, which eventually led to better participation.
High stakes pave way for more analytics
Steven Spaldingmedical director for enterprise information management and analytics, Cleveland Clinic
There are increasing business reasons for healthcare providers to implement predictive analytics software and applications. Knowing which patients are likely to experience adverse events like infections or injury, return to the hospital after being discharged or have surgical complications could point the way toward preventive treatments that could potentially improve patient safety and reduce readmissions. Having a clear vision of how patients will respond to care is becoming more necessary than ever as the healthcare industry moves to replace the traditional model of paying hospitals and physicians for the number of procedures they do or patients they see with pay-for-performance plans.
The federal Affordable Care Act, signed into law in 2010, introduced changes to Medicare that incentivize participation in such plans and penalize hospitals that have high readmission rates. In addition, a growing number of private insurers are moving providers to quality-based reimbursement programs.
Spalding said Cleveland Clinic has increasingly been shifting to value-based payment contracts with insurers, which is a primary reason why the health system has been interested in pushing greater use of analytics to help its medical teams manage the health of patients more proactively. "We were making some bets financially, and it became clear to us that if we were going to do this at scale, we had to think about how to standardize care," he said.
Make the business case for analyzing data
But healthcare organizations need to be mindful of the value of what they're doing with predictive analytics. Robert Kritzler, deputy chief medical officer at Johns Hopkins HealthCare LLC in Baltimore, said he has seen doctors ask for new predictive models to be implemented without knowing exactly what they plan to do with the information the models will generate. Before any time and money is spent developing healthcare predictive analytics capabilities, he advised, providers should ask themselves why they need the functionality.
"The fact that other health systems are doing it is not an answer," Kritzler said. "That's lemmings all running off the hill. We've been pushing real hard to get the real answers."
At Johns Hopkins, data analysts have developed models to predict patients' length of stay and who is likely to be readmitted after discharge, two factors that are tied very closely to the health system's profitability. They also track quality of care metrics to ensure that the organization isn't spending too much on unnecessary or even non-recommended care.
Know your analytics audience
Meeting the needs of doctors is also a must. Organizations might be able to predict a lot about their patients through the use of predictive analytics techniques, but physicians generally don't have time to sift through lengthy reports or play around with data in self-service software. As a result, analyses need to be targeted and streamlined, according to conference speakers.
Ari Robicsek, vice president of clinical analytics at NorthShore University HealthSystem in Evanston, Ill., said he has built models to investigate potentially interesting questions about how patients respond to specific treatments, only to realize in the end that there was no practical way to implement the results into the clinical workflow. "We've got to know the user before we build the model," Robicsek said.
It's a problem Kritzler has dealt with as well. He said that any data analysis he puts in front of his physicians is intended to be immediately useful, like the estimate of patients' risk of readmission.
"When you get down to it, they don't want data," Kritzler said. "They want knowledge. They want actionable information. If you flood them with data, you get absolutely nothing out the other end."
What kind of ROI should you expect from predictive analytics techniques?
Consider whether you need a data scientist for predictive analytics
Test your knowledge on creating predictive analytics programs