As someone employed in medical research, I’m a little bothered when my friends self-diagnose using webMD or self-medicate rather than consult physicians. I think those of us trained in health-related fields are especially prone to this; when I get sick, I probably think I’m smart enough to know what’s going on. We can debate whether this is because I’m cocky or because I’m subconsciously protecting myself by pretending to have control over sickness, but clearly running a fever means more computer screens lighting up and fewer doctors’ appointments.
Kris Hauser and Casey Bennett took a different approach: they removed the human component altogether and made a computer program to diagnose people based on symptoms. Their findings published in Artificial Intelligence in Medicine are undeniably thought-provoking. The algorithm so far seems to be 30% more accurate than average physicians in diagnosing diseases. Like the Jeopardy robot, Hauser and Bennett’s program uses Markov decision making processes, a sort of finite-state machine that changes based partly specified criteria such as specific symptom combinations and partly on chance.
This approach combines Markov decision processes and dynamic decision networks to learn from clinical data and develop complex plans via simulation of alternative sequential decision paths while capturing the sometimes conflicting, sometimes synergistic interactions of various components in the healthcare system. It can operate in partially observable environments (in the case of missing observations or data) by maintaining belief states about patient health status and functions as an online agent that plans and re-plans. This framework was evaluated using real patient data from an electronic health record. Such an AI framework easily outperforms the current treatment-as-usual (TAU) case-rate/fee-for-service models of healthcare (Cost per Unit Change: $189 vs. $497) while obtaining a 30-35% increase in patient outcomes.
Turns out the cost of running the program costs about half as much as paying physicians to make diagnoses. Though promising, at this stage the algorithm is too specialized. Ran using a different set of patients might have unpredictable results, and rampant parameter tweaking will be required to maintain the integrity of its results. Additionally, recognizing symptoms accurately is something physicians do effectively that the machine takes for granted. Doctors out there–your jobs are probably safe, for now.