With millions of possible drug combinations for treatment of diabetes, hypertension, hyperlipidemia and heart disease, new AI program offers more thorough analysis and options.
When we had but two insulins and one oral tablet, a sulfonylurea, back in 1950, it was simple to choose the best possible treatment. But 68 years later, we now have new treatments approved just about every month with millions of possible combinations.
If we can develop self-driving cars and fly to the moon — and beyond — we should be able to come up with a computer program that considers all the positives and negatives of the millions of possible combinations of different treatments, and comes up with the best possible treatment for each patient that will fit into their budget.
We don’t need to wait any longer; it is happening today. As research from Indiana University has revealed, a new computer program is doing a better job than doctors when it comes to both diagnosing and treating health conditions — and by a significant margin — using artificial intelligence and robotics.
The system, which uses decision-making processes similar to the Jeopardy-bot, Watson, was recently given the task of analyzing and predicting the health outcomes of 500 real individuals. After putting in the relevant data — which mostly had to do with clinical depression and chronic diseases like high blood pressure and diabetes — the researchers compared the real-life outcomes to the simulated treatment prescriptions.
And what they discovered was amazing in that the computer was 42% better at diagnosing illnesses and prescribing effective treatments than human doctors. To achieve such an impressive outcome, the computer used an artificial intelligence framework that employs sequential decision-making, allowing the computer to simulate a series of alternative treatment paths into the future. The machine can also make assumptions about a patient’s health when data is not available, and re-adjust when new data is introduced. The system is also non-disease-specific and adaptable to virtually any health issue.
The computer deliberates about the future and considers all the different possible sequences of actions and effects in advance — even when it’s unsure of the effects.
So instead of spending money on treatments that will probably be effective only after trial and error, the patient will achieve their treatment goals in a timely fashion with huge cost savings and not have to be hospitalized as often.
No one is suggesting that we hand over the medical care of our patients to the computers now, but to take a serious look at developing a partnership with technology to improve the quality of care and to lower costs.
Kris Hauser and Casey Bennett, the main researchers in this project, stated, “Even with the development of new Artificial Intelligence techniques that can approximate or even surpass human decision-making performance, we believe that the most effective long-term path could be combining artificial intelligence with human clinicians. 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.”
We know that much of what physicians do (checkups, testing, diagnosis, prescription, behavior modification, etc.) can be done better by sensors, passive and active data collection, and analytics. But doctors not only have to measure, they are supposed to consume all that data, consider it in context of the latest medical findings and the patient’s history, and figure out if something’s wrong. Computers can take on much of that diagnosis and treatment and even do these functions better than the average doctor (while considering more options and making fewer errors). Most doctors couldn’t possibly read and digest all the latest 10,000 research articles on diabetes and heart disease. And, most of the average doctor’s medical knowledge is from when they were in medical school, while cognitive limitations prevent them from remembering the 10,000+ diseases humans can get.
Computers are better at organizing and recalling complex information than any smart physician. They’re also better at integrating and balancing considerations of patient symptoms, history, demeanor, environmental factors, and population management guidelines than the average physician. More than 50%-70% [MA1] of the family practitioners — where 90% of type 2 patients go for their care — are below average in knowledge on all the diabetes treatments. Computers also have much lower error rates. Because the cost of medications play a role in patients getting and using their recommended medications, it is also important that we determine which of the best treatments will fit into the patient’s medical budget. So, what are we waiting for?
Please see our January, 2018 special four part series about a new way to find the best treatment for diabetes: a program that is available now at no cost that can not only tell you the best treatment for each individual patient, but will also give you multiple options depending on the patient’s insurance and their medical budget.
Artificial Intelligence Framework for Simulating Clinical Decision-Making: A Markov Decision Process Approach” was published in Artificial Intelligence in Medicine J Community Health. 2015; 40(5): 1002–1007. Published online 2015 Apr 16. doi: 10.1007/s10900-015-0024-2