Like many other hospitals and health systems, Flagler Hospital, a 335-bed community hospital in St. Augustine, Fla., had a board-level mandate to address its unwarranted clinical variation with the goal of improving outcomes and lowering costs, says Michael Sanders, M.D., Flagler Hospital’s chief medical information officer (CMIO).
He noted that traditional methods of addressing clinical variation management have been inefficient, as developing care pathways often takes up to six months or even years to develop and implement. “By the time you finish, it’s out of date,” Sanders says. “There wasn’t a good way of doing this, other than picking your spots periodically, doing analysis and trying to make sense of the data.”
Available analytics software is incapable of correlating all the variables within the clinical, billing, analytics and electronic health record (EHR) databases, he notes.
Sanders says he was intrigued by advances in machine learning tools and artificial intelligence (AI) platforms capable of applying advanced analytics to identify hidden patterns in data. Working with Palo Alto, Calif.-based machine intelligence software company Ayasdi, Flagler Hospital initiated a pilot project to use Ayasdi’s clinical variation management application to develop care pathways for both acute and non-acute conditions and then measure adherence to those pathways.
Flagler targeted their treatment protocols for pneumonia as an initial care process model. The AI tools from Ayasdi revealed new, improved care pathways for pneumonia after analyzing thousands of patient records from the hospital and identifying the commonalities between those with the best outcomes. The application uses unsupervised machine learning and supervised prediction to optimally align the sequence and timing of care with the goal of optimizing for patient outcomes, cost, readmissions, mortality rate, provider adherence, and other variables.
The hospital quickly implemented the new pneumonia pathway by changing the order set in its Allscripts EHR system.
The AI application uncovered relationships and patterns that physicians either would not have identified or would have taken much longer to identify, Sanders says. For instance, the analysis revealed that for patients with pneumonia and COPD, beginning nebulizer treatments early in their hospital stays improved outcomes tremendously, hospital leaders report.
As a result, for the pneumonia care path, Flagler Hospital saved $1,350 per patient and reduced the length of stay (LOS) for these patients by two days, on average. The hospital also reduced readmission by 7 times—the readmission rate dropped from 2.9 percent to 0.4 percent, hospital officials report. The initial work saved nearly $850,000 in unnecessary costs — the costs were trimmed by eliminating labs, X-rays and other processes that did not add value or resulted in a reduction in the lengths of stay or readmissions.
“Those results are pretty amazing,” Sanders says. “It’s taking our data and showing us what we need to pursue. That’s powerful.”
With the success of the pneumonia care pathway, Flagler Hospital leaders also deployed a new sepsis pathway. The hospital has expanded its plans for using Ayasdi to develop new care pathways, from the original plan of tackling 12 conditions over three years, to now tackling one condition per month. Future plans are to tackle heart failure, total hip replacement, chronic obstructive pulmonary disease (COPD), coronary artery bypass grafting (CABG), hysterectomy and diabetes, among other conditions. Flagler Hospital expects to save at least $20 million from this program in the next three years, according to officials.
Sanders and his team plan to continue to use the AI tool to analyze new data and adjust the care paths according to new discoveries. As the algorithms find more effective and efficient ways to deliver care that result in better outcomes, Flagler will continue to improve its care paths and measure the adherence of its providers.
