Conversations regarding uses of AI in healthcare continue to accelerate to what seems to be a daily topic. Frankly, this is the most interesting technology I have witnessed in my 40 years in business. In conversations with leaders in business, education, and medicine, it is hard not to hear the talk about AI many times a day. I am fortunate to be in these industries. I have witnessed first hand the emergence of other technologies from the internet, cloud computing, blockchain, etc., nothing has taken the imagination and interest of so many industries as the “AI Craze.”
Medical Schools are not only seeking the best ways to serve their students and faculty stay ahead of emerging trends, but also relevant in the day-to-day. Universities, under threat from dropping national enrollment[1], are racing to produce specialized certifications in AI that one has to wonder how one gets “certified” in something so new – anyone that can spell “AI” is a Subject Matter Expert. AI is one of those in a generation movement that you can’t ignore. Many tried to ignore the Personal Computer (PC), including the behemoth IBM that put its chips in the mini-computer and mainframe market in the early 1980s. That paid a high penalty to play catch-up, get out of business, or get swallowed up as did Tandem, DEC and Harris.
The lesson to be learned from this snippet of history is that ignoring the “newcomer” because it may not fit your business model can bring change. Change has a way of shaking the trees and causing the over-ripe fruit to fall and make way for the “new growth.” The old adage applied then and is being reapplied now: “Lead, Follow, or Get Out Of The Way!”
Today, we have hundreds of companies exploring AI, in particular ChatGPT4 (and soon to be 5 and 6), to help people diagnose their own issues or interpret medical results. This is may be just a little more than what was called in the industry “Dr. Google.” Everyone that is reading has gone to Google® or WebMD® many times to investigate some symptoms they or a family member may have, or to find something about some disease someone has commented on. Well, using AI for triage, is a bit like people I knew in the 1980s that use computers to store cooking recipes. Yes, it can do that, but you are missing about 99.99% of the computing power of even the 8086 16-bit Intel microprocessor.
We are convinced that AI is changing and will continue to change healthcare in ways we are imagining, but most likely in ways we have not yet conceived. AI may be the analog (pun intended) of ARPANET (1969) that lead to HTTP (Hypertext Transfer Protocol) and became the Internet. How many people then could see all we are doing today with the Internet? We think the same will happen with the use of AI in healthcare.
Back to healthcare. Decision support is a great way for doctors to be assisted with rare conditions, as well as with daily patient load. Recently, I heard an oncologist talk about how AI could help him, and he noted that 90% of the time, there was no need for AI. The diagnostics and doctors training could do the job, but the last 10% could be broken into two parts. One where AI can help get to the diagnoses faster, and then the 5% that are really complicated. It may be that generative-AI will help us explore the “known unknowns” of healthcare. We know there must be something however, we don’t know where or what it is. The ‘key’ to unlocking the strength of AI is in phrasing the question as we brought out in Part 5 of this series – the “new” field of “Prompt Engineering.” Certainly, AI can help the Primary Care Physicians by quickly reviewing each patient to be seen categorizing and synthesizing their ‘complaints’ in context of their medical history, and providing a quick ‘jumpstart’ for the physician before meeting with the patient. A bit like the patient reviews being conducted during residency with the class teaching doctor but in this case synopsized by “Dr. AI.”
The challenges we have in the development of meaningful AI healthcare platforms is bridging the gap, chasm as it were, between technologists and data scientist, to how a provider would use it or needs it to work. The most significant issues we faced in the development of Electronic Medical Records (EMR) in the U.S. was due to developers that were creating advanced technology that was NOT easy, practical, or time efficient for doctors to use. The issues of “work flow” were ignored by many developers because they did not understand, nor did they investigate, how the physician practice functioned. This led to a great deal of extra effort by doctors creating burnout, dissatisfaction, and ultimately their discarding the EMR and reverting back to their ‘tried and true’ paper system – even though it meant that they would see a reduction in their Medicare reimbursements. Afterall, the computer system should be serving man and not the other way around, and if not, then what is the computer systems’ value?
If we adapt the experience from the EMR-era we should be conscious of ensuring that we are not only creating an AI platform for the sake of AI, but also a platform that truly empowers the physician. One inherent value in an AI-healthcare program is the machine learning (ML) capability. Every interaction becomes a part of the ‘data lake’ for the AI to access. Over time, it will truly advance the science of medicine. However, we have to ‘design’ this into the platform so that we have ‘structured’ data or at least restructured data from which to draw.
We know from experience, having built an EMR that collaborated between technologists, practice administrators, financial managers, nurses, and doctors. The platform answers all the needs, but it could not be sold for a loss forever, because that system was expensive to build, and even more expensive to maintain. It was also complicated every year, when the government changed the mandates and requirements. Today, many physicians are upset with the unfulfilled promises of EHRs.
From the AMA[2]:
A study published prior to the pandemic, “Association of Electronic Health Record Design and Use Factors With Clinician Stress and Burnout,” identified an association between key elements of the EHR design and physicians’ well-being.
These factors are:
Information overload.
Slow system response times.
Excessive data entry.
Inability to navigate the system quickly.
Note bloat.
Fear of missing something.
Notes geared toward billing, not patient care.
“Physician burnout as a whole is a multifaceted problem, but people are retiring early or going part-time because they’re spending so much time on the computer, they don’t feel they get enough time to take care of their patients,” said Philip J. Kroth, MD, MS, lead author of the study. He chairs the biomedical informatics department at Western Michigan University, Homer Stryker M.D. School of Medicine.
We think AI can and should be different from the experience of the EHR – we’re smarter now by having gone through that process. Do all those cancelled, defunct, or even those EHRs still in use, individually contain the data we need to use in our AI platform? We think NO! We are confident that the difference will be that this time around it will be consumer-driven, the consumers will demand their data to be augmented with AI, and have AI help show them their future self, and what changes will feel like. Lead them to a higher wellness state – a healthier self with greater quality of life and longevity.
Imagine if after aggregating your OWN medical records, an AI platform could tell you what your life will feel like at 50, 60, 70, 80? What if it tells you that based upon a comparison of your medical data with others in your demographic, health-state that your life expectancy is actually 69.2 years? What If you were only 63. Will you ask AI what you can do to live longer?
[1] https://nscresearchcenter.org/stay-informed/
[2] https://www.ama-assn.org/practice-management/digital/7-things-about-ehrs-stress-out-doctors