AI system outperforms docs: researchers

A pair of Indiana University computer scientists set out to devise a computer-based, artificial intelligence system that could “think like a doctor.” But what they created may be even better, producing a system that could outperform physicians in making care decisions over time.

The AI system they developed reviewed and, in effect, second-guessed the decisions of human clinicians in 500 cases involving behavioral health patients with a variety of other chronic illnesses, such as hypertension, diabetes and cardiovascular disease.

For each patient, across eight encounters, they came up with a series of dynamic, alternative, evidence-based care paths that outperformed those of the human decision makers in terms of clinical outcomes and cost, their research shows.

Comparisons between outcomes from actual case histories and the hypothetical histories created by applying the team's computer algorithms were made based on an “outcome rating scale,” or ORS, described as a “validated ultra-brief measure of functioning and symptomology for chronic and acute mental disorders.” The authors said the ORS metric has “over a decade of research supporting its use in similar settings and populations.”

Decisions also were evaluated based on a second metric, the “cost per unit change,” which measures the cost in dollars it takes to obtain one unit of outcome change (for better or worse), using the ORS, over time. The researchers concluded the cost per unit of outcome change was $189 when using their artificial intelligence system, compared with $497 for “treatment as usual.”

Meanwhile, use of artificial intelligence could produce “a 30-35% increase in patient outcomes” with a 50% increase possible with some “tweaking” of the model's parameters.

The findings of Casey Bennett, a doctoral student and instructor at IU's School of Informatics and Computing, and Kris Hauser, an assistant professor of computer science there, were published online Jan. 2 in a report, “Artificial Intelligence Framework for Simulating Clinical Decision-making: A Markov Decision Process Approach,” in the journal Artificial Intelligence in Medicine. Bennett also is a research fellow at Centerstone Research Institute, the research arm of Centerstone, a not-for-profit provider of behavioral healthcare with offices in Columbus, Ind., and Nashville.

“It is likely that the next decade will see the integration of multiple sources of data—genetic, clinical, and socio-demographic—to build a more complete profile of the individual, their inherited risks, and the environmental/behavioral factors associated with disorder and the effective treatment thereof,” Bennett and Hauser wrote in their article. “Combining autonomous AI with human clinicians may serve as the most effective, long-term path. Let humans do what they do well, and let machines do what they do well. In the end, we may maximize the potential of both.”

In an interview, Bennett said the potential improvement in costs and outcomes made possible by using artificial intelligence to guide multiple clinical decisions over time “speaks to the potential, obviously, of utilizing these kinds of technology.”

“We've gotten some angry responses,” in “some random, unsolicited email,” in the wake of publication, Bennett said. “It's going to be a bit of a culture shift. Docs are trained to have a bit of an ego, and they should, if you're making life-and-death decisions, you're going to need one. It's going to be a transition” to using computers to assist with those decisions.

“I think the younger generation is more open to it,” Bennett said. But ready or not, change is coming, maybe within a decade, he said.

“I think there is a new generation of technology coming out in AI, in computational power, that will make a big difference,” Bennett said. “So, I think over the next 10 years, you'll se a big change. With meaningful use and Obamacare, there is going to be a political shift that's going to force change.”

Without it, Bennett believes, soaring healthcare costs will bankrupt the country.

“So, you won't be able to quibble out it,” Bennett said. “It will happen whether people like it or not.”