Part of my attitude relates to an experience I had. And this was within a single HIS. I wanted to…
“Redefining the Boundaries of Medicine” is a Mayo Clinic Press book that was written by Paul Cerrato, MA (senior research analyst and communications specialist, Mayo Clinic) and John Halamka, MD, MS (president, Mayo Clinic Platform). The authors have collaborated previously on three books and several journal articles.
The book is written for readers who are knowledgeable about the “only in the US” healthcare mix of research, medical practice, consumerism, and hardcore capitalism where money has an outsized influence on both individual health and the business of healthcare. Its dense typography and layout is hardly inviting, but it provides an excellent history of how we got to where we are in healthcare (hint: often illogically, stubbornly, and parochially) and how healthcare can be improved.
The book delivers what its title promises. The authors are predictably precise in their citations and conclusions, and they are on the provider front lines rather than ivory tower academics. In addition, Mayo Clinic Platform is working actively to apply data science and technologies to healthcare.
I admit that I wasn’t aware of the previous books that these authors co-wrote and wasn’t exactly sure what Mayo Clinic Platform does or what happened with John Halamka after he left BIDMC three-plus years ago. But I think these authors might be the go-to-experts that the healthcare industry needs as it rushes headlong into artificial intelligence and re-examines itself with an opportunity (or requirement) to change dramatically.
Here are some of the notes I took.
The book leads off with a chapter on artificial intelligence, where the authors observe that the human brain cannot process the amount of new information from journals and conferences, much less apply it at the bedside, and can’t analyze all available information to arrive at an accurate diagnosis. AI is also better than humans in analyzing diagnostic images, although system training must be carefully designed in an environment that has never-ending changes in scanning technology, coding and terminology, EHR configuration, changed institutional practices or order sets, and a changing patient mix that may not be applicable elsewhere.
A fascinating idea is that all broad research, whether powered by AI or not, overgeneralizes to the entire population instead of digging into patient subgroups. For example, a large study on the effect of lifestyle modification on cardiovascular disease was abandoned when no differences were seen between the intervention and control groups, suggesting that lifestyle doesn’t matter. However, applying sophisticated analytical technique found that lifestyle intervention actually worked in two subgroups that were otherwise lost in the large numbers: patients whose diabetes is poorly controlled and in those with well-controlled diabetes who self-report their health as good.
They also note that FDA’s approval of AI devices is inconsistent and often involves retrospective and/or single-site studies.
The authors conclude AI algorithms need to be more equitable and better validated before being placed into clinical use.
Medicine’s history in the US involves paternalistic physicians; diagnosis and treatment protocols that were based on GOBSAT (good old boys sat around the table); and slow acceptance of research findings in favor of personal experience, anecdotes, and opinions lacking evidence.
Randomized controlled trials, especially those that conclude that a therapy was not beneficial, have weaknesses such as too-small sample size and inclusion criteria that may introduce bias or reduce clinical usefulness. RCTs should be supplemented with real-world evidence and cohort studies.
The “heterogeneity of treatment effect” acknowledges that treatment benefit and risk can vary widely among patients. Patients know their conditions and see the effects of treatments firsthand, so N-of-1 trials comparing active treatment with placebo are a good idea.
“Patients like mine” data can help support decisions in the absence of RCT or observational studies now that EHR data is widely available, although it may require experts to turn patient data into actionable evidence.
Rethinking Medical Expertise
The public questions the value of medical expertise. Experienced clinicians use Type 1 thinking, in which pattern recognition can lead to quick conclusions involving common conditions as “disease scripts.” But sometimes it fails dramatically when a patient’s symptoms fall outside the norm. Type 2 reasoning starts with a hypothesis that is refined via logic and critical thinking, which can be more accurate and avoid bias and thinking shortcuts, but takes too long to conduct in high-volume settings.
The authors cite previous studies that found that peer-reviewed journals often rejected research that turned out to be important, questioning whether that publishing process is the best way to gestate new ideas.
Replacing “One Size Fits All” with Personalized Medicine
Full genomic sequencing is increasingly useful. Some experts say it should be performed at birth, whereas now newborns are screened for a small number of genetic disorders.
Large studies on using the antiplatelet drug clopidogrel for blood clots found that the drug outperformed aspirin in just two of each 100 patients, but the real challenge is to identify those two instead of incurring the cost and risks of giving it to everyone.
“Normal” lab ranges are just a statistical convention, and each person’s “normal” may be different and deviation from it may not indicate the presence of disease. Insurance will often pay for only drugs and treatments that appear effective for broad segments of the population.
Researchers search for one or two primary causes of a disease, such as HIV as a cause of AIDS or striving to control the blood sugar of diabetics, and immediately refocus all research on those causes. The outliers are rarely studied, such as the people who are exposed to HIV but don’t develop AIDS and why that might be. Correcting the condition for a given patient doesn’t necessarily deliver the expected benefit.
Too many clinicians still practice the “doctor knows best” model when patients don’t agree with their evidence-based interventions. Policy decisions are rarely made on science alone since beliefs and core values will usually win.
FDA knows that most drugs that it approves offer only slight benefit, but consumers aren’t capable of analyzing studies, especially when faced with direct-to-consumer advertising. The public is easily confused by correlation versus causation and relative value versus absolute risk, such a miracle drug that reduces the risk of some disease by 50% that really means that one person instead of two out of 1,000 patients will get it, which is hardly impressive. Schools do not teach critical thinking skills and the US doesn’t follow the lead of other countries that teach media literacy.
Interdisciplinary Patient Care
Researchers and clinicians need to communicate better. Experts say that NIH-funded research focuses on silos for particular conditions of interest without looking at how they relate to, or are affected by, other factors, which is an outdated understanding of medicine. DARPA might offer a better model.
Clinician fragmentation increased with the growth of specialty medicine, medical group consolidation and insurance programs networks that separated people from their specific doctor.
More than three-fourths of chronic diseases are caused by or exacerbated by lifestyle choices that can’t be easily explained or encouraged in the allotted 15-minute office visit.
Patient-generated data should be fed into EHRs.
You will be stimulated by the ideas the authors express in this book if you are comfortable reading journal abstracts and understand clinical practice, especially if your specialty is informatics. It seems like a slim read at under 200 pages, but is packed with information in being free of self-aggrandizement and pontificating (and again, the typeface is pretty crammed, so it’s got more content than you might think). If you or your organization want to be considered disruptive in healthcare, the authors are giving you great ideas of where you might focus.