Think (learn, teach) Outside of the Book Medical Education is Changing due to AI –As Technology Advances, Don’t Forget About the Patient
Justin L. Lockman, MD, MSEd, Aditee P. Ambardekar, MD, MSEd, Stephanie A. Black, MD, EdM, Alan Jay Schwartz, MD, MSEd
Original Articles
Chang BS. Transformation of Undergraduate Medical Education in 2023. JAMA. 2023 Oct 24;330(16):1521-1522. doi: 10.1001/jama.2023.16943. PMID: 37698855.
Hswen Y, Abbasi J. AI Will-and Should-Change Medical School, Says Harvard's Dean for Medical Education. JAMA. 2023 Nov 21;330(19):1820-1823. doi: 10.1001/jama.2023.19295. PMID: 37878288.
When educators grasp how students understand and utilize medical knowledge, they undoubtedly are better able to facilitate learning during the student-teacher interaction. Students, including those of anesthesiology, and including both trainees and practitioners in CME settings, have always been products of their time.
Until recently, the medical education milieu has been Flexnerian. In 1910, Abraham Flexner, a high-school teacher, delivered a report commissioned by the Carnegie Foundation on the state of medical education in 155 medical schools in the US and Canada.1 His report defined a new, required medical education standard, including:
a) a university education prior to entering medical school
b) a scientific biomedical focus
c) initial basic science laboratory learning
d) subsequent clinical training in university hospitals (as already developed, but not universally adopted, by Osler and Halsted).
The Flexnerian approach influenced how medical students, postgraduate trainees, and practitioners learned. They became managers of memorized biomedical information gathered in classroom lectures and classic textbooks.
Over the more than 100 years since the publication of the Flexner report, medical education has been based upon scientific discovery and students’ ability to grasp the biomedical basis of disease.2. There is no question that many of the advances in patient care and medical innovation over the past century have stemmed directly from Flexner’s insistence on standardizing a curriculum (or having standards at all!).
Yet there are downsides to classroom lectures and classic textbooks. Such lectures and textbooks can often be woefully out of date in rapidly changing fields. And students trained this way risk becoming data reporters instead of data interpreters/integrators who can think holistically about patient health and wellbeing. As we have previously discussed, there are good reasons why engaging content and experiential learning will always beat reading slides. One could even argue that cultivating a transformation to being data investigators is just as germane to contemporary medical practice as interpreting or integrating. Given the varied inputs our patients seek (including the uncurated opinions of the interwebs), it behooves the modern medical learner to become well-versed in the skill of critically apprising information from all sources.
Fast forward to 2023. There is a similar revolution happening in medical education, although it’s more of a “silent” one. Chang, in his 2023 JAMA “Viewpoint”3 and during an interview by JAMA’s Editor-in-Chief, 4 notes that artificial intelligence (AI) is upon us (like it or not!). Knowledge acquisition, content pattern recognition, selecting clinical diagnoses, predicting the likelihood of a patient’s clinical course, and generating clinical notes about patients are but a few of the powers AI enables for medical education and clinical care. A variety of applications of AI for anesthesiology are promising.5,6
Does this apply to pediatric anesthesiology? The answer is strongly in the affirmative!
“Themes of applications of AI in anesthesiology:
(1) depth of anesthesia monitoring,
(2) control of anesthesia,
(3) event and risk prediction,
(4) ultrasound guidance,
(5) pain management, and
(6) operating room logistics.”5
We must be cautious here, as the effectiveness of AI in clinical practice has not been validated. In fact, there are legitimate reasons to have concern about AI being unsupervised in medical decision-making and treatment planning. But is there anyone among us who wouldn’t love to have an AI system make the OR schedule more efficient with less input from us?
Chang’s “Viewpoint” aims to consider the strengths and weaknesses of AI as they are incorporated into medical education.
AI strengths:
“Aided by…AI-based tools, students should be expected sooner than before not just to regurgitate gathered data or identify relative likelihoods, but also to grapple with the intricate contextual nuances relevant to their patient’s situation and to make balanced decisions that account for a messy set of individual circumstances. These are skills that the most senior physicians have traditionally had in good measure, of course, based on the accumulation of years of experience. The brute force capabilities of AI are counterbalanced and complemented by the inherent human advantages of recognizing and understanding complex patterns of behavior and relying on psychologically informed strategies to improve patient comprehension, adherence, and satisfaction”3
Put another way: if data can now be gathered faster and more accurately by an AI system, students (who no longer must spend time doing the gathering) should be expected to interpret and integrate the data and formulate plans far more effectively than a generation ago. This means that all of us, as educators, will need a complete structural overhaul of our education models to keep up!
AI weaknesses:
AI employs technology, yet the “real” patient may be left out of the equation. Chang correctly advises us to:
“…remember that one of the most technologically irreplaceable elements of the curricula is the guided development of skillful interpersonal communication and the expert physical examination.”3
Recognize that, “AI may be able to engineer generically empathic prose, but the much more complex verbal and nonverbal patient-physician communication that characterizes the best clinical visits will likely elude it for some time.”3
Understand that “…right now they [AI tools] are still prone to hallucinations, or basically making up facts that aren’t really true and yet saying them with confidence.”4 Like some trainees, perhaps!
Appreciate and guard against the fact that AI “…may inherently be structurally biased…ChatGPT and these other large language models [LLMs] are trained on the world’s internet...a noncopyrighted corpus of material. To the extent that that corpus of material was generated by human beings who in their postings and their writings exhibit bias…whether intentionally or not, that’s the corpus on which these LLMs are trained. So, it only makes sense that when we use these tools, these tools are going to potentially exhibit evidence of bias.”4 It is critical to appreciate that AI only works with what we teach it through our human selective exposure—so the inputs matter! In other words, garbage in = garbage out.
The pre- and post-anesthesia visits, pain management consultation, and communication with critically ill children and their families are examples of settings where anesthesiologists will need to combine the benefits of AI and the doctor-patient interaction with caution regarding AI weaknesses.
For an anesthesiology example of such an AI weakness, consider the situation when your anesthesiology trainee is singularly focused on sophisticated monitoring technology in the OR and ICU and “forgets” to examine the patient? How many times have you queried a trainee about a patient’s ventilation and their response is to look at the capnograph rather than the patient, rarely using a stethoscope to assess ventilation?
“…training in what is often already lamented as the lost art of the physical examination, and the development of sufficient experience to confidently appreciate and interpret the resulting findings, will remain distinctively important for future clinicians.”3
Chang’s viewpoint suggests that, in fact, the technological advances possible with AI will give “the physician more opportunity and leeway to spend time talking with [the] patient the way it really ought to have been all along.” It is a “back to basics” of sorts—thorough history, the art of the physical exam, empathetic communication, and a reuniting of the humanistic side of medicine. These are the aspects of patient care for which many of us went into medicine and AI simply cannot replace. Fortunately for us, if used wisely, we can leverage the benefits of AI to manage the noise and logistics and bureaucracy, freeing us to engage in the sacred acts of physicianship that add the most value for the patient, physician, and society.
AI is here to stay. We may think we are still far from a time when students will learn from computers how to be silly with children. And perhaps even farther from a time when patients would rather interact with AI than with their physician. But these ideas are important, and we think they can’t be ignored: we need to keep up with this revolution or be left behind by it. Fortunately, or not, we are in the driver’s seat of AI and its applications in medical education. How we navigate that may determine the outcome of the journey.
What do you think? Email Myron at Myasterster@gmail.com and he’ll post in a Friday Reader Response.
References
1. Flexner A. Medical Education in the United States and Canada: A Report to the Carnegie Foundation for the Advancement of Teaching. Bulletin No. 4. Boston, Mass: Updyke; 1910.
2. Duffy TP. The Flexner Report-100 Years Later. Yale Journal of Biology and Medicine. 2011; 84; 269-276.
3. Chang BS. Transformation of Undergraduate Medical Education in 2023. JAMA: 330 (16); 1521-1522
4. Hswen Y, Abbasi. AI Will-and Should-Change Medical School, Says Harvard’s Dean of Medical Education. JAMA: 330 (19); 1820-1823
5. Alekberli T, Alkeberli F, Miouni C, Muhammed AA. Smart-anesthesia Application with Artificial Intelligence for a Preoperative Evaluation of Patients. Research and Innovation in Anesthesia 2000: 5; 21-22
6. Hashimoto DA, Witkowski E, Gao L, Meireles, O, Rosman G. Artificial Intelligence in Anesthesiology Current Techniques, Clinical Applications, and Limitations. ANESTHESIOLOGY 2020: 132; 379-394