"Electronic health records contain a treasure trove of data, but they are simply too large for humans to process," said
Hauser is one of two computer scientists from
Hauser is the lead investigator, and he's working with fellow assistant professor of computer science Sriraam Natarajan and Regenstrief investigator Dr.
"Patients, doctors, hospitals, insurers and policymakers coexist in a complex and often dysfunctional web of hidden information, costs and objectives, but they also face the problem of information overload," Hauser said. "AI systems can digest relevant information and put it all on the table, ultimately making health care more transparent and cost-effective."
The team will use statistical relational learning techniques to find patterns in large electronic health care databases, then put those patterns into a mathematical framework. That framework will use documented observations along with observation probabilities to maintain a probability distribution over a set of possible treatment sequences. Through exhaustive searches, the decision support system will then identify the optimal treatments.
"The system will present information to doctors about what might be the best treatment plan, along with the plan's expected outcome and costs," Hauser said. "It will provide doctors with several options they can weigh in combination with their own expertise. We are not trying to eliminate the human element. Computers can easily integrate a patient's data from a lot of different visits to hospitals and doctors' offices, but doctors have their own intuition and insight that a computer can't see."
Hauser said the system will be used in hospital emergency rooms by doctors seeing re-admitted patients because some of them are frequent ER users.
"They represent a huge strain on our health care system," he said. "They are very costly to treat in the ER, and often their problems either don't need to be treated, or could be addressed by seeing a primary care physician or making lifestyle adjustments."
Grannis, who's also an associate professor of family medicine, said recent studies have demonstrated that leveraging health care data can enhance physicians' ability to offer more efficient and effective care.
"This grant will allow us to combine two proven AI-based clinical decision support approaches to create a more powerful, more accurate tool," he said.
Hauser said past research has shown statistical relational learning techniques can help predict cardiac arrest in some patients based on demographic and lifestyle data, and can improve health outcomes by 40 percent.
"What we're doing is using AI to automate the job that statisticians have been doing for hundreds of years," Hauser said. "It's not magic. Today statisticians can already figure out statistical models for predicting things like heart attacks and cancer rates. Our systems use computers to do the same thing, but do it much faster."
Hauser said the systems can also help physicians make the best diagnoses and order the most appropriate and effective tests.
"Some doctors over test, maybe ordering X-rays and MRIs for a sore ankle, which can really run up medical bills," he said. "If the treatment will be the same regardless of the test outcome, then our system will recommend not getting the test."
Clinical partners in the prototype testing will include Centerstone Research Institute, the research arm of Centerstone, a behavioral health service provider in
Additional members of the research team include Regenstrief's Dr.
(c)2014 the Herald-Times (Bloomington, Ind.)
Visit the Herald-Times (Bloomington, Ind.) at www.heraldtimesonline.com
Distributed by MCT Information Services