Cleveland Clinic's use of algorithms for risk stratification results in better population health outcomes

Cleveland Clinic, Getty Images, Douglas Sacha

The Cleveland Clinic's population health team has found a way to deal with all of the data necessary to target at-risk and vulnerable populations.

The team uses algorithms to give them a risk stratification in order to understand the population in need of resources. As importantly, it gives them actionable data at the time of decision-making.

"The clinic has long emphasized the importance of actionable data at the moment of care delivery," said Adam Myers, chief of population health for the Cleveland Clinic. 

Myers spoke with Jonah Comstock, director of content development and editor-in-chief for HIMSS Media, during HIMSS's Machine Learning & AI for Healthcare Digital Summit, Wednesday, December 2, in "Letting Data Connect the Community Health Dots." 

Cleveland Clinic developed the risk algorithm using 120 variables from the electronic health record, claims data, and labs and other results, to determine the likelihood that someone would encounter a problem.

Having the algorithms is the only way to deal with overwhelming amounts of data and a huge influx of COVID-19 patients, he said. The algorithms process the data for meaningful changes so the care teams are not responding to every little fluctuation.

The algorithm automates certain activities, giving well-defined, computer-implementable instructions. While the large Cleveland Clinic uses machine learning and AI in many of its systems, the algorithm's risk-stratification tool is not based in machine learning.

"The risk-stratification tool really helps us target our interventions," Myers said.

Two interventions in particular have shown strong results.

One is within the care at-home division. The team looked at the highest risk patients with comorbidities to use layers of interventions. They have a company that goes to people's homes to help patients with connected devices that feed data into the EMR. 

This program was ramped up during COVID-19, when no one wanted to visit a physician's office or hospital. Pre-COVID-19, care at home had about 2,000 patients. Within two weeks of the pandemic, the program had 200,000 patients. 

The health system saw a 35% reduction in hospitalizations due to in-home monitoring, according to Myers.

All of the information is accessible.

"It's important to be able to look across the entire panel," Myers said, "and determine not just who is coming to see us, but who isn't coming to see us [who] could potentially need our help."

The team takes an hour each week to pause in its regular care for a panel management plan, to see which patients have gaps in care that need to be addressed. 

Through these methods, they've been able to emphasize and increase tighter ambulatory management. In one year, the Cleveland Clinic saw an increase of 9.5% in what Myers called "ambulatory touches." During that same year, in a control population of about 100,000 Medicare patients, it saw a 7.5% reduction in inpatient stays. 

"So, 9.5% investment of additional touches in management on the ambulatory side yielded a substantive reduction in inpatient utilization," Myers said.

For COVID-19 patients, a high-touch COVID-19 monitoring program uses connected devices and frequent outreach for a 7.5% reduction in hospitalizations for COVID-19 patients who are able to manage their care at home.

The Cleveland Clinic geo-mapped their COVID-19 test results to predict the next hot spots. In some cases, the health system was able to inform local health departments of hot spots developing before the health officials had them on their radar.  

Another success due to algorithms has been in diabetes management. To understand blood sugar levels, having a blood sample at the time of the doctor's appointment makes for a more fruitful visit. 

The Cleveland Clinic had established a group of 30 medical assistants who worked to contact patients to get those tests done before an appointment. The system worked, but it was very labor intensive, Myers said. 

The population health team built an algorithm to automate the process into a half-time job for one person. It resulted in better outcomes and a far better use of people, he said, as those 29-and-a-half positions were used for patient care.

Another benefit of the risk-stratification tool for population health is that it checks all the boxes toward value-based care. 

"Technology," Myers said, "is essential for value-based work at scale."

Twitter: @SusanJMorse
Email the writer: susan.morse@himssmedia.com