A Florida-based community hospital is using AI tools to examine patient data and formulate new treatment protocols for deadly illnesses such as sepsis.
Artificial intelligence (AI) is sometimes seen more as hope and hype than as a solution for healthcare that can improve patient outcomes and reduce costs.
Now, a community hospital in Florida is not only proving healthcare AI can generate observable results in the clinical arena, but also showing that initiatives can be launched and sustained without the deep pockets of a large health system.
Flagler has already realized financial gains from new care pathways for pneumonia and sepsis. The pneumonia and sepsis care pathways have generated nearly $850,000 in costs savings in less than a year. Once care pathways are established for about a dozen other high-risk conditions, cost savings could be as high as $20 million over the next three years, says Michael Sanders, MD, CMIO at Flagler.
For health systems and hospitals seeking to capitalize on AI technology, Flagler is a case study on how AI can be used to develop new care pathways that simultaneously cut costs and improve clinical outcomes.
In 2017, Flagler decided to use AI to mine the hospital's clinical information and focus on deadly and costly conditions. The goal was creating new care pathways for conditions such as sepsis to boost clinical outcomes and drive down cost of care.
Early this year, Flagler signed a three-year AI technology contract with Menlo Park, California–based Ayasdi to help with this initiative.
The AI-powered process reviews thousands of patient records from the hospital, then identifies the patient cohort with the best outcomes, such as the lowest direct variable cost, readmission rate, length of stay, and mortality. The commonalities of this patient cohort drive development of new care pathways such as revision of the emergency department order sets with new treatment protocols.
The hospital also has internally published a new care pathway for COPD, and is set to rollout diabetes, total knee replacement, heart attack, and coronary artery bypass graft pathways next.
"Any hospital can do this," Sanders says.
Here are three ways Flagler implemented AI into its organization to improve patient outcomes and reduce total cost of care.
1. Create a Staffing Strategy
To limit costs, Flagler decided to staff its AI capabilities internally, Sanders says.
"We do not have a data scientist at our hospital. All of this work has been done internally, which I think any community-based hospital can do. We have a couple of folks, including myself and a SQL query guy who has great SQL skills," he says.
Most of the work done by humans in a new care pathway effort involve setting up the parameters and queries for data extraction. Once the data has been verified, the AI tools sift through the hospital's data to combine patients into cohorts.
2. Start With a Simple Pilot
In the pilot phase of Flagler's AI initiative, Sanders wanted to pick a condition with relatively few variables to compute.
Picking the right place to start—pneumonia—was essential to the effort, Sanders says. "I wanted something that would be relatively easy and straightforward, so we could get our feet wet with something that wasn't too complicated."
It took nine weeks to complete the pneumonia project—from data extraction to AI analysis, to refinement of new treatment protocols, to internal publication of the new care pathway.
The next care pathway project—sepsis—took two weeks, with learning gains speeding data extraction and the treatment protocol refinement process.
3. Collect and Crunch Patient Data
Flagler took multiple steps to extract and organize the patient data used to form new care pathways:
- Patient data is extracted from the hospital's Allscripts EHR, surgical system, enterprise data warehouse, and financial system
- Data is loaded to a cloud technology, where it can be stored and manipulated
- Three rounds of semantic and syntactic validation are conducted to make sure the data is accurate
- AI tools are used to carve out patient cohorts
Once the analytical work has been completed, the AI team members engage the hospital's physicians to craft the new care pathways.
"We sit down with one physician from every physician group in the hospital. We have been having meetings every first and third Wednesday of the month, and we go through the care paths and make changes that are deemed appropriate based on our evidence and experience," Sanders says.
The new care pathways are used to set treatment protocols in the emergency department and inpatient wards, he says. "We publish the care paths and make changes to the ED order sets and admission order sets based on the care paths, [then] we begin monitoring them."
Operationalizing Care Pathways
Care pathways are road maps for treatment, Sanders says.
"The pathway includes everything, from the moment patients enter the ER to when they are admitted and discharged. There are two sets of orders involved—the care that is given in the emergency department and the care that is given after admission," he says.
Positive patient outcome data has fueled physician engagement in the AI-driven care pathways, Sanders says.
"We had 557 patients with septic shock. For all but 19 patients, the sepsis order set was used. When we looked at the data, the mortality in our hospital is below the national average at 9%. If a patient had the sepsis order set used, they had a readmission rate of 6%. If we look at the group where the order set was not used, the readmission rate was 21%."
Christopher Cheney is the senior clinical care editor at HealthLeaders.
Community hospitals can launch and maintain sophisticated AI initiatives.
AI initiatives can focus in on patient data at health systems and hospitals to target patient cohorts for data analysis.
Using AI to develop new care pathways can improve patient outcomes and reduce total cost of care.