BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
A Florida community hospital is using AI tools to identify new care paths for improving treatment of several high-cost, high-mortality conditions such as pneumonia and sepsis.
"Hospitals for, I would say years or even centuries, have been looking for some way to tackle the variations that exist in clinical management," said Michael Sanders, M.D., chief medical information officer at Flagler Hospital in St. Augustine, Fla. "But even more important now that we have reached a point where healthcare is really not very affordable in this country, you need to reduce the cost. We also need to, as best we can, improve the care and reduce readmissions, which of course is a cost to the hospital itself. Every hospital has been working on this process for a long time. But we have not had the tools to do so on a large scale until Ayasdi."
Ayasdi features a general-purpose AI platform upon which applications are built, said Jonathan Symonds, Ayasdi chief marketing officer. The application adopted by Flagler Hospital, clinical variation management, looks into a hospital's data and analyzes events to create a consensus care path for surgical and nonsurgical procedures.
The clinical variation management application uses a "framework for machine learning that combines unsupervised learning techniques in a way that allows you to understand all the patterns that exist in that data," Symonds said.
Using AI tools to generate care paths
Sanders said he began investigating a partnership with Ayasdi, a machine intelligence software company, after reading about another hospital that used Ayasdi's AI platform to evaluate clinical variation.
Jonathan Symondschief marketing officer, Ayasdi
Sanders felt AI was the "right tool" that would allow the hospital to develop new care paths by taking large amounts of data from the hospital EHR, analytics and financial platforms, and enterprise data warehouse to "begin to reduce clinical variation and improve quality and reduce cost."
One of the main benefits of using AI, Sanders said, was being able to use the hospital's own data. "Our population is quite varied and we wanted to do our analysis based on our own data. To do the kind of work we did without a system like Ayasdi would've taken us years to accomplish."
The first condition the hospital tackled was pneumonia and after running a pilot program using the AI tools for about nine weeks, Sanders said the hospital developed a new care path for the condition. The AI tools divided patients into groups and found where they differed in terms of cost, terms of stay and readmission rates. From the care paths suggested by the AI analysis, the hospital selected the group with the lowest cost, readmission rate and length of stay. The hospital implemented the new pneumonia care path by changing the order set in its Allscripts EHR and expects to save about $1,350 per patient, reduce the length of stay by two days and experience a sevenfold reduction in readmission.
"Our readmission rate for pneumonia, 2.9%, is down to 0.4% following the care path," Sanders said.
The hospital rolled out its second care path, for sepsis, shortly after the care path for pneumonia, and has now laid out 18 care paths it hopes to establish within the next 18 months.
Continuing development of new care paths
As the hospital continues to use AI tools to generate new and improved care paths for other conditions, ranging from strokes and seizures to COPD and knee and hip replacement, Sanders said he expects to see lower costs for patients and employers, lower readmission rates and a shorter length of stay "throughout all areas of the hospital."
"If you reduce the cost, reduce the readmission, reduce the length of stay, by and large you have improved the care of patients," Sanders said.
Ayasdi's AI tools are not only used by Flagler, but other, larger healthcare organizations as well, including Intermountain Healthcare, Mount Sinai and Mercy. Symonds said what stands out about Flagler's use of AI tools to develop care paths is the success the hospital had despite its small size. The community hospital only has 335 beds.
"We didn't see a smaller community hospital without any data science resources as being somebody who could execute this, but they have excellent skills, they know where their data is and they have the intellectual curiosity to make it work," Symonds said. "Their roadmap for future care paths is ambitious but achievable. It's really opened our eyes to the possibility of looking at other community hospitals in addition to larger hospitals and systems we generally talk to."