The Computer Will See You Now

This article published with permission from The Burrill Report.

Indiana University researchers have created an artificial intelligence system able to improve patient health care while at the same time reducing costs.

The expanding costs and complexity of the U.S. health care system hinders doctors’ abilities to make optimal treatment decisions over time, say the researchers. Their non-disease-specific, artificial intelligence framework simulates an environment for exploring various health care policies and payment methodologies, establishing the basis for a computer that can “think” like a doctor.

When the researchers compared actual doctor performance and patient outcomes against their sequential decision-making model, they found that modeling capable of understanding and predicting outcomes of treatment could improve patient outcomes by an average of almost 50 percent while reducing healthcare costs on the average by the same amount.

“We’re using modern computational approaches to learn from clinical data and develop complex plans through the simulation of numerous, alternative sequential decision paths,” says graduate student Casey Bennett. “The framework here easily out-performs the current treatment-as-usual, case-rate/fee-for-service models of health care.”

Bennett, together with Kris Hauser, assistant professor of computer science, used 500 randomly selected patients from a dataset containing more than 5,800 patients with major clinical depression diagnoses, the majority of whom had co-occurring chronic diseases like diabetes, hypertension, and chronic pain, complicating their diagnoses and treatment plans. This made them perfect subjects to test the computer model’s ability to learn from iterative input, provide a diagnosis, and recommend treatment options.

The current model builds on previous work demonstrating the ability of an artificial intelligence to correctly make a single diagnosis and treatment decision. The simulation approach combines Markov decision processes — to provide a mathematical framework for modeling decision making in situations where outcomes are partly random and partly under the control of a decision maker — with dynamic decision networks, a tool for supporting a decision-maker faced with multiple uncertain variables and many opportunities to collect information.

The model simulates the outcome of numerous alternative treatment paths; maintains beliefs about patient health status over time even when measurements are unavailable or uncertain; and continually plans and re-plans as new information becomes available.

The expanding use of electronic health records and the growing number and complexity of genetic and biological datasets lay the foundation for clinicians to be more efficient in their work, but also more overwhelmed by available information.

“Simulation modeling can help improve decision-making and the fundamental understanding of the healthcare system and clinical process,” say the researchers. “With careful design and problem formulation, we hypothesize that such an AI simulation framework can approximate optimal decisions even in complex and uncertain environments, and approach — and perhaps surpass — human decision-making performance for certain tasks.”

Copyright 2013 Burrill & Company. For more life sciences news and information, visit The Burrill Report.