Will artificial intelligence replace your primary physician in the future?
Look at the current healthcare crises. Due in part to changes enacted by the Affordable Care Act (ACA), there will be a need for an additional 52,000 primary care physicians doctors (nearly a quarter) by 2025. The mandate of Electronic Health Records (HER) makes your health history accessible digitally, so personalizing diagnosis and treatment should be possible.
Enter “Modernizing Medicine” a so-called web-based repository of medical information.
It is basically data mining from patients and doctors mined and if you enter what you are looking for it spits out recommendations. Think “Trip Advisor” for medical treatment.
Also, revealed at ASCO this week was an MD Anderson program, based on the WATSON platform that suggested standard-of-care diagnosis for cancer patient cases and was correct about 80% of the time. This system was named "Oncology Expert Advisor".
Although I am excited by the prospects of combining the information on EHR with large-scale data mining to get personalized care for patients (let’s face it, it sounds uber-cool on paper), my 20 years experience in Pharma and regulatory is having some questions about the use of this program.
Classically, recommendations are made by consensus groups of well-respected, highly experienced “key-opinion leaders” in a specialty field. They come together for about a week, present, debate, look at all the evidence for use of products in a disease state, and then publish a “consensus paper”, which is then, in turn, reviewed by a panel of peers, critiqued and sent back and forth until it is finally published. This process is renewed every few years to keep recommendations fresh.
Evidence for this is generally that available in peer-review literature, graded A-D, based on large-scale clinical trials to case reports.
The recommendations are produced as algorithms or consensus guidelines, which the doctor has an opportunity to follow and choose based on their patient’s individual profile. The choice is always, and should always, be left up to the licensed medical professional.
Comparing the classical method to this new database: one is based on evidence-based medicine, thoughtful debate by knowledgeable men, peer-review and a tried-and-true process; the other, is a handful of single case reports gathered together and run through a program on the computer to pick the “winner.” I will say that for those diseases that don’t currently have consensus papers for treatment, however, this may be useful for helping a physician find a possible treatment, such as their example.
This new method is not currently being regulated by the FDA. This also confuses me, as when I work with my Pharma clients to develop tools (websites, apps, sliders, booklets) for physicians, I am very careful NOT to tread on the diagnosis and treatment line, as I know the regulatory is very strict in that regard. I see this program as being very blurry on this treatment recommendation line. I am very interested to see how far these type of programs are taken before the FDA steps in and begins to regulate them and demand references and consensus guidelines to be used as part of the programming.
http://am.asco.org/ibm%E2%80%99s-watson-based-oncology-computing-system-recommends-treatment-high-accuracy
http://www.wired.com/2014/06/ai-healthcare/