How artificial intelligence is used in the development of patient treatment regimens: the US experience
An AI-assisted pneumonia care scheme has already saved a U.S. clinic $1,356 per patient, reduced hospital stays by two days, and significantly reduced hospital return visits.
Flagler Hospital in St. Augustine, Florida is using artificial intelligence to improve the treatment of pneumonia, sepsis and a dozen other costly, high-mortality diseases.
Typically, AI is used by large academic medical centers rather than public hospitals. But Flagler Hospital also decided to use it.
AI has identified a new, improved pathway for pneumonia and sepsis care after analyzing several thousand hospital patient records and identifying common denominators for patients with the best treatment outcomes. The hospital quickly implemented a new regimen for the treatment of pneumonia, changing the order established in the EHR Allscripts (Electronic Medical Record – an analogue of MIS in Ukraine).
She expects to save $1,356.35 (35%) in the cost of treating one patient with pneumonia over existing costs. At the same time, the length of the patient’s stay in the hospital is reduced by two days. Also, with the help of AI, a new treatment regimen for sepsis was developed.
The process is going so well that the hospital has expanded its plans to use artificial intelligence from IT developer Ayasdi. In particular, if, according to the original plan, new patient care schemes were to be developed for 12 diseases within three years, now I plan to develop schemes at a rate of one per month.
There is a growing number of providers in the medical services market that use artificial intelligence in the process of providing medical care. These are IBM, Jvion, Medial EarlySign, Pillo Health and Splunk.
Ayasdi uses a branch of mathematics called topological data analysis to group patients who are treated in the same way and identify associations between those groups, explained Michael Sanders, MD, chief medical information officer at Flagler Hospital.
“Once we have loaded the data, we use an AI learning algorithm to create treatment groups,” he continued. – In the case of our pneumonia patient data, Ayasdi created nine treatment groups. Each group of patients was treated the same way and we were provided with statistical data from which we could understand this group and its difference from others.”
Analysis of data groups
Flagler Hospital doctors analyze treatment groups and select, for example, the Goldilocks group. This is the group with the lowest treatment costs, the shortest length of stay in the hospital and the least likelihood of repeat visits, making it the “right” group.
“Then we use AI to generate a treatment plan for this group, showing all the events that should happen in the emergency department, on admission and during the entire stay of the patient in the hospital,” explained M. Sanders. “These events include all medications, diagnostic tests, vital signs, intravenous fluids, procedures, and meals, and the ideal time for each to replicate that particular group.”
M. Sanders is using AI to implement AI-derived Flagler hospital treatment procedures in both the emergency department and the hospital. The hospital has a PIT (IT doctors) team that uses AI input to recommend this procedure and get departmental approval to implement it. The hospital administration is considering changes as part of the oversight process. Statistical differences between groups are carefully studied by M. Sanders, his computer scientists and IT physicians before the Goldilocks group is selected.
Extracting data from the IIS and the financial system
Flagler Hospital extracts AI-enabled data from its HIS, CPM (Corporate Performance Management) analytics platform, corporate data warehouse and financial system using 2,300 lines of SQL code.
“Once we have selected the Goldilocks group, we press a button in the user interface that generates a treatment plan in about 30 seconds,” Sanders explained. – This occurs when the action and the time to act (order of medication, prescription, laboratory order, laboratory results, etc.) occur in at least 50% of patients in the treatment group.
The hospital can then set the time, set the equivalent event to occur in the treatment plan, obtain consent, and finally publish it. This includes what is needed in the emergency room, on day 0, day 1, day 2, etc.
“While this process can be repeated manually or semi-manually, it would take years of work to even come close to unlocking some of this knowledge.