Flagler Hospital, a 335-bed community hospital based in St. Augustine, Florida, is projected to save more than $20 million after AI technology helped it reduce costs, average length of stay and readmissions for pneumonia patients, Michael Sanders, MD, Flagler’s chief medical and informatics officer, said during the HIMSS conference in Orlando.
“Other hospitals need to understand that you don’t need the data scientists. It’s nice if you have one, but you (probably) can’t afford one if you’re a hospital like us or smaller,” Sanders told AI in Healthcare. “Many (hospitals) are much smaller than us, and they can still do this.”
In early 2018, Flagler began a case study aimed at addressing the challenges of clinical variation, which is described as the overuse, underuse, different use and waste of healthcare practices and services with varying outcomes. The hospital turned to an AI-powered clinical variation management tool—offered through AI software company Ayasdi—to develop and measure adherence to care pathways for acute and non-acute conditions.
Using the hospital’s patient data, the tool uses unsupervised machine learning and supervised prediction to find the best care pathway in hopes of producing better patient outcomes, reduced costs, readmissions, mortality rate and increase provider adherence.
“You need to be able to practice and develop care process models that reflect those patient groups and patient types,” Jonathan Symonds, Ayasdi chief marketing officer, said. “If you do that, it drives the adherence way up because the doctors know these are the results of my best doctors, doing their best work against our patient set, and that cannot be understated.”
The hospital first tested the tool on its treatment protocols for pneumonia patients, with the process lasting a total of nine weeks. The study found the best care path for pneumonia patients saved $1,350 per person and reduced average length of stay by two days. Readmission rates also dropped from 2.9 percent to 0.4 percent. According to the case study, the AI tool saved nearly $850,000 in unnecessary costs by eliminating labs, X-rays and other processes for pneumonia.
“To do what we did with the pneumonia sample would have probably taken us a couple of years,” Sanders said. “The beauty of Ayasdi is you’re looking at (all of the data), everything that happened, and let the algorithm tell you what you should be looking at.”
The process can be extrapolated out for major savings.
“Ultimately, what (Flagler) has been able to do from a production level speaks volumes to what this can do for healthcare on a macro basis. It’s an $812 billion problem—clinical variation management in the U.S. alone,” Symonds said. “They’ve shown, and they’ve proven, that you can productize this inside of a hospital in a way that can really change the game.”
Following its study on pneumonia, the hospital also applied the AI tool to sepsis and chronic obstructive pulmonary disease (COPD). Overall, it plans to apply the AI tool to 18 conditions over the next 18 months and is expected to save at least $20 million from the program over the next three years.
Sanders said he can’t predict what type of impact AI will have on the healthcare industry in the future. But from his experience with AI and Ayasdi, he said the technology has allowed the hospital system to better find problems and solutions in medicine.
“From my personal experience, (AI) has had a big impact,” Sanders said. “It’s probably going to change the way we operate.”