This month, The Center for the Future of the Health Professions will be posting our third monthly op-ed column for 2021. Our columns represent strong, informed, and focused opinions on issues that affect the future of the health professions. As mentioned in the past, the Center was developed to provide state, local, and national policymakers and health system stakeholders with accurate, reliable, and comprehensive data and research about the healthcare workforce, so they can effectively plan for a sustainable future and make the best use of available resources.
This month features a discussion of If Artificial Intelligence Analytics Continue to Improve Healthcare, What Substantial Advances Should Clinical Educators Consider in the Redesign of Medical Education? By Valerie Sheridan D.O. DFACOS, FACS, Assistant Dean of Clinical Education, Assessments and Outcomes, GME, Assistant Professor, Surgery at A.T. Still University.
We encourage readers to share their thoughts after reading this important piece.
Just a decade ago, discussions regarding artificial intelligence (AI) may have had many of us imagining a futuristic robot taking over the world. Currently in 2021, robots may not yet be taking over the world but they are performing many tasks within our society, and artificial intelligence continues to creep into our lives on a daily basis. While still a hype to some, AI is well on its way to shaping our futures. The psychological impact of AI in our lives has been significant. AI applications are surfacing everywhere transforming our lives. Smartphone devices, home devices, and media applications are just to name a few. Yet, what exactly is AI, and how does AI fit into our professional lives and a career in medicine?
If one was to ask Siri to define AI, the response may be something like this: “According to the Merriam-Webster dictionary, AI refers to a branch of computer science that deals with the simulation of intelligent behavior in computers.” More specifically, AI involves the art of computer programming developing specific algorithms that allow computers to perform tasks that normally require human intelligence.1 Computers can learn, understand, and make assessments about the world based on the information we provide them. A computer’s strength is that it is able to analyze large volumes of data in an efficient and reliable manner without fatigue.
There are many dimensions of AI. Machine learning consists of algorithms that learn from data through experience without human intervention. The process of machine learning is to observe data and identify patterns that will make predictions based on continued learning.1 Deep learning technology refers to machine learning methods that utilize artificial neural networks. These large mathematical functions include millions of parameters in computer modeling algorithms that loosely mimic the human brain. This process helps improve the accuracy of conventional techniques from older software designs.2 Occasionally, AI and deep learning are used synonymously. Other dimensions of AI include data mining, supervised and unsupervised learning, natural language processing, robotics, speech, voice and image recognition, and precision computing.
It will be important to identify gradual steps that prepare our students for where we want them to be with data and AI in the future during the transformation of medical education. Others3,4,5,6 have already attempted to identify frameworks for medical schools to use as context how AI can be integrated into medical education. Questions remain about what to add, when to add it and what to take away?
Medical knowledge, application of learned skills and professional identity formation remain the core attributes of a competent physician. Machines and robots are gradually acquiring human abilities. Therefore, consideration of determining the role of the physician in the future and the direction medical schools should take to best prepare tomorrow’s physician must occur as the transformation of medical education curricula continues.
Each entering medical school class has at least four years to learn to integrate critical thinking with clinical problem-solving. These crucial years allow students to absorb basic science fundamentals and clinical contexts in preparation for future patient care. It is the hope that these efforts will instill the necessary competence and confidence to manage life and death decisions in their future careers.
So where does the instruction for AI fit in? We have evolved beyond telling machines what to do with data. Computer programs can now learn from patterns and anomalies due to the sheer size and complexities that exist within data. Regardless of how much education a student may have when entering medical school in the area of engineering or informatics, more education may be necessary concerning AI and its many uses in medicine. Convolutional neural networks and deep learning techniques exist in the medical image-focused subspecialties such as radiology, cardiology, pathology, dermatology, and ophthalmology. The ER and ICU are full of data that is also ideal for gathering data information and knowledgewith new AI tools and data strategies.7,8 AI creates better patient outcomes, elevates patient experiences, improves clinician experiences, and reduces costs.7,9,10
At the very least, students with little or no exposure to AI prior to entering into medical school will require knowledge of AI. Through varied learning experiences with AI applications, every medical student should aim to develop the ability to create validated and trustworthy information to provide to patients and the public.
Suggested for consideration are four substantial advances that aim to empower our medical graduates with the capability to thrive in future health systems utilizing AI. These include modifications to the current medical educational framework in content, skills, leadership and health systems science.
Design engineers ensure that all design criteria yield a functioning model. Educators need to become like design engineers. The creation and application of the curricular design will need to find innovative and regenerative ways to keep the students learning and engaged with a valid and reliable assessment process. AI generated projects such as learning apps, coding exercises and programing comprehension will assist in achieving desired skill outcomes. Students will need to know the mechanics and processes of AI systems that they will be expected to use in the clinical environment. This continued transformation of learning must get away from the memorization-based curriculum to teaching competence in integration and utilization of information and data. It remains extremely important for the student to understand how the biomedical sciences, clinical knowledge and analytical experience are connected.
Technology is not foreign to students entering medical school. Therefore, they can be inspired with the use of technology to ponder, imagine, reflect, analyze, memorize, recite and create.
Innovative methods regarding instructional efforts should attempt to avoid any unintended consequences to the medical profession. For example, deep reinforcement learning is a method of successful analytics. The game Go is Asia’s most popular game of martial strategy. A successful deep reinforcement learning program, “Alpha Go”, defeated the human GO champion years ago.11 In this reinforcement type of deep learning, there was a goal-oriented algorithm designed for real life, real time complex decisions. Alpha Go’s victory publicly demonstrated the learning capacity that AI-based technologies possess. Humans did not teach the computer. It taught itself how to master the game by playing it millions of times with another computer and independently responded to its opponent’s moves.11 Risk parameters are important and need to be set keeping humans involved in this endeavor.
Many students are familiar with gamification. If a student interacts with a knowledge-transfer system designed on current educational principles such as cased-based, and complex scenario-based situations, these systems using Natural Language Processing (NLP) can present flexible and realistic virtual patients in situations that require critical thinking and clinical reasoning. The systems are also able to adjust the level of complexity that suits the student’s performance level.3, 4, 12, 13, 14 The impact of AI is not only in the delivery of the cases, but also in the analysis and solving of the educational problems.15 These learning techniques can be adapted to improve effectiveness for achieving competency, thereby leading to better quality of care for future patients.
Machine learning may become a core competency in the future. It should not come as a surprise to any that we must learn from our data. Adopting tools that can help in these endeavors to analyze, educate and operate more efficiently remain an ultimate goal.
Every clinician considers the bedside of a patient a place of honor. This is where a fellow human being allows the provider the privilege of looking at, touching and listening to their bodies. Our skills and discernment must be worthy of such trust.16
Due to a variety of circumstances, bedside skills of physicians have deteriorated as available technology has evolved.16,17,18
However, the education regarding the application of these skills continues to accommodate these advancing technologies. For public safety measures, this remains important, as tradition has yet to change. Osteopathic physician board certification still consists of multiple-choice questions and a clinical skills exam.
Some tasks are very simple for humans and incredibly difficult for AI. Therefore, in an attempt to understand AI, it is important to understand humans and the process of how we make decisions. Computers are getting better at predicting; however, the decision-making process remains extremely valuable to physicians as it is one process that is difficult to explain to a computer. Human judgement will not easily be replaced by analytics, which is why humans need to remain involved in the process of machine learning techniques.19, 20, 21 The continued importance of sound judgement and ethical standards will be of utmost importance in the education of medical students against analytic capabilities.
In the world of economics, judgement is a complement to prediction.22 Human prediction skills will decrease as machine learning intelligence improves. However, this allows the opportunity for human judgement skills to increase. As physicians continue to make decisions about medical treatments and ethical considerations for patients, emotional support will be necessary. These are judgement skills best provided by humans.
New skills will be necessary for future physicians. As data quality and quantity is a challenge, machine learning-based training will be critical to provide future physicians with the conceptual skills to interpret data output. Further skills will be necessary to interpret good data from biased data along with the skill of managing Big Data for the benefit of patient care. Learning and predictions from the data will only be as good as the quality of the data. Identifying the strengths and weaknesses of computers and AI will assist in achieving future success in physician education and patient care. These skills will assist students in gaining insight into diagnostics, care processes, treatment variability, and patient outcomes.23 Using these skills appropriately will help students as they become physicians make better and safer decisions.
Repeatedly stated, medicine is an art as well as a science. It is the element of being an art that artificial intelligence will have difficulty replacing. The nonanalytic, humanistic aspects of medicine, most importantly the art of caring, should be further prioritized in the medical education curricula. Communication, empathy, shared decision-making, leadership, team building, and creativity are all skills that will continue to gain importance for being an effective physician in the future.24 Enhancing soft skills in the area of humanities will be of prime importance. Well-developed soft skills will promote successful soft leadership.
Anticipation of incorporating new elements into any medical education curriculum will require organized and influential leadership. If student engagement is expected, it is clear that student participation in the creation of AI integration will be beneficial. This instruction regarding ‘soft leadership’ will be important for future influence, persuasion and motivation of these new concepts.
Education throughout generations has evolved. Current learners appreciate environments that allow them to multitask, use digital technology, share ideas, work in teams and have flexibility and choice.25 They excel in influencing people through persuasion. While sharing their contributions, they have the ability to align their energies and efforts to accomplish desired goals. Mentoring and coaching efforts are encouraged to channel these abilities to the right mindset, and skill set.
Participation in this learning and leading process will guide students in becoming competent and well-prepared physicians. The resetting of educational and learning environments will progressively occur. It has become increasingly important to identify and utilize one’s emotions in a positive way overcoming challenges, diffusing conflict and communicating effectively with others.
Analytical information continues to promise that ethical, legal and social implications will emerge, and medical students will need to consider these and the questions they will raise. Quality and patient safety issues will more than likely rely on AI algorithms designed, at least initially, by humans. These algorithms will carry the ethical biases of the designers.26 Not only will medical students need to be cognizant of potential biases in data, it will be important for them to understand how best to respond to these biases or lead in efforts to adjust them. They will need to be aware of data usage and security laws.27
Health Systems Science
Health Systems Science (HSS) is a framework used to represent a set of methods and tools that focus on systems—rather than parts—as the context for defining and solving complex problems, and for fostering more effective learning and design.29 Most computer-based algorithms in medicine are “expert systems”. They encode knowledge on a given topic and are able to draw conclusions about specific clinical scenarios, such as detecting drug interactions or judging the appropriateness of obtaining imaging studies.33 Expert systems work the way an ideal medical student would—they take general principles about medicine and apply them to new patients.
AI is ‘prepared to be the engine’ or system that drives improvements across the healthcare continuum.34 Engines are made of many moving parts that individually would not be able to function in moving a vehicle. Yet, together, they work as a system enabling the vehicle to move great distances. This concept of team building creating a functional system, with various contributing parts, needs to continue as a priority within the future of the medical education curricula. It takes all parts of the system to function successfully. Physicians, patients, and technology must work together. Without this knowledge and practice, tracking disease prevalence, treatment methods, and patient responses using data collection, analysis, and dissemination may prove difficult to precisely, and individually improve treatment protocols. Based on evidence of what is working and what is not across various disease states and populations, this is evident.28
When properly deployed, machine-learning systems can help resolve disparities in healthcare delivery if algorithms are designed to compensate for known biases or other areas of needed research.30 It will continue to be important for medical students and physicians to learn and identify meaningful data on the social determinants of health to improve health outcomes. Predictive analytics exist in health-related areas such as chronic disease management, disease outbreaks, radiation illnesses and sepsis.7, 8, 31
As mentioned, importance remains within the curricula on good interpersonal relationships and reasoning. Collaborating people with machine learning techniques can achieve better outcomes. Learning is continuous and a journey for all. Together, focusing on analytical strategies, tools and practices will be instrumental for better patient care and will take these modified curricula to a new age and beyond.
Machine intelligence + clinical intelligence = medical intelligence. 31
The education of the future physician continues even during these challenging times. Indeed, important and necessary elements already exist in the medical education curricula. However, it is apparent that changes will need to occur as the evolution of digital technology and artificial intelligence applications persist in healthcare systems. Medical student success will continue with preparation including applicable information and skills. Providing enhancements within the curricula in the mentioned areas of content, skills, leadership and health systems science will best prepare future physicians. Employers need physicians who will work at the ‘top of their license’ with other members of the healthcare team. They need physicians having the knowledge to appropriately use data platforms such as smart phones, social media, and other devices that focus on analyzing patient outcomes and improving performance in patient care.
For many, the education and use of AI remains a mystery. For AI to be trustworthy, algorithm design requires the moral and ethical responsibilities of the human. Sure, fear exists that AI opens the door for cybercriminals or will replace physicians based on evidence where AI is proving superior to humans in several aspects of medicine.35, 36 It is true that administrators, educators and students will require new skills and expertise for closer alignment of machines and humans in healthcare. However, this is an obligation we have to the communities where those students, as physicians, will serve.
Potentially, lives can be saved on a scale that is unimaginable because of the impact of these technologies. Ready or not, artificial intelligence is here and here to stay. Those of us preparing the future physician will need to learn to use it to be here to stay as well.
*This is a personal viewpoint based on literature.
1. White, Michael. Machine Learning (2nd edition) 2018.
2. Walczak S, Velanovich V. An Evaluation of Artificial Neural Networks in Predicting Pancreatic Cancer Survival. JGastrointestinSurg. 2017; 21 (10):1066-1612. doi :10.1007/s11605-017-3518-7.
3. Masters, K. (2017) ‘Preparing medical students for the e-patient’, Medical Teacher, 39(7), pp. 681–685. https://doi.org/10.1080/0142159X.2017.1324142.
4. Masters, K. (2019) ‘Artificial Intelligence in Medical Education’, Medical Teacher, 41(9), pp. 976–980. https://doi.org/10.1080/0142159X.2019.1595557.
5. Chen, J. (2017) Playing to our Human Strengths to Prepare Medical Students for the Future”. Korean Journal of Medical Education,22(3), pp193-197.
6. Srivastava TK, Waghmare L. Implications of Artificial Intelligence (AI) on Dynamics of Medical education and Care: A Perspective. Journal of Clinical and Diagnostic Research. 2020:14(3):JI01-JI02.
7. Chang AC, Hunt J: Toward unreasonable effectiveness of cardiac ICU data: Artificial Intelligence in pediatric cardiac intensive care. Pediatr Crit Care Med 2014; 15:565-567
8. Rush B, stone DJ, Celi LA: From big data to artificial intelligence: Harnessing data routinely collected in the process of care. Crit Care Med 2018; 46:345-346
9. PwC Health Research Institute analysis of PwC Bot.Me: A revolutionary partnership, PwC Consumer Intelligence Series survey, 2017, and PwC Health Research Institute Provider Survey, 2017
10. 2018 HIMSS U>S> Leadership and Work-force Survey. (2018). Healthcare Information and Management systems Society. Retrieved from https://www.himss.org/sites/himssorg/files/u132196/2018_HIMSS_US_LEADERSHIP_WORKFORCE-SURVY-Final_Report.pdf
11. Silver D, Huang A, Maddison CJ, etal: Mastering the game of Go with deep neural networks and tree search. Nature 2016: 529: 484-489.
12. Chary, M., Parikh, S., Manini, A. F., Boyer, E. W., et al. (2019) ‘A review of natural language processing in medical education’, Western Journal of Emergency Medicine, 20(1), pp. 78–86. https://doi.org/10.5811/westjem.2018.11.39725.
13. Wijayarathna, G. K. and Zary, N. (2019) ‘Feasibility in using de-identified patient data to enrich artificial applications in medical education’, EDULEARN19 Proceedings, 1(July), pp. 7598–7604. https://doi.org/10.21125/edulearn.2019.1837.
14. Afzal, S., Dhamecha, T., Gagnon, P., Nayak, A., et al. (2020) ‘AI Medical School Tutor: Modelling and Implementation’, in Artificial Intelligence in Medicine. 18th International Conference on Artificial Intelligence in Medicine, AIME 2020, August 25–28, 2020, Minneapolis: Springer, pp. 133–159. https://doi.org/10.1007/978-3-030-59137-3_13.
15. Jović, J., Milić, M., Cvetanović, S. and Chandra, K. (2019) ‘Implementation of machine learning based methods in elearning systems’, in The 10th International Conference on eLearning (eLearning-2019), 26-27 September. Belgrade, Serbia, pp. 39–44.
16. Verghese, A, MD; Shah N, MBBS, PhD; Harrington R, MD; what this computer needs is a physician Humanism and Artificial intelligence JAMA, 2018; 319 (1): 19-20.
17. Cabitza F, PhD; Rasoini R, MD; Gensini G, MD. Unintended Consequences of Machine Learning in Medicine. Viewpoint-JAMA August, 2017. 318(6): 517-18.
18. Alrassi, J. Technology Can Augment, but Not Replace, Critical Human Skills Needed for Patient Care. Academic Medicine, 2018; 93(8): 1105-1106.
19. Circulation. 2015;132:1920-1930. DOI: 10.1161/CIRCULATIONAHA.115.001593.)
20. Karches, KE. Against the iDoctor: Why Artificial Intelligence Should Not Replace Physician Judgement, Theoretical Medicine and Bioethics, 2018; 39:91-110.
21. Arnol J, Davis A, Fischhoff B, Yecies E, Grace J, Klobuka A, Mohan D, Hanmer J. Comparing the predictive ability of a commercial artificial intelligence warning system with physician judgement for clinical deterioration in hospitalized general internal medicine patient: a prospective observational study. BMJ Open 2019:9:e032187.
22. Agrawal A, Gans J, Goldfarb A. The Simple Economics of Machine Intelligence. Economics, November 2016.
23. Barton AJ. Big Data. JNursEduc. 2016;55(3):123-124. doi:10.3928/01484834- 20160216-01.
24. Johnston SC. Anticipating and Training the Physician of the Future: The Importance of Caring in an Age of Artificial Intelligence. Academic Medicine, 2018: 93(8):1105-1106.
25. Roberts DH, Newman LR, Schwartzstein RM, Twelve tips for facilitating Millennials’ learning. Med Teach. 2012: 34(4): 274-278.
26. Straw, I. The automation of bias in medical Artificial Intelligence (AI): Decoding the past to create a better future. Artificial Intelligence in Medicine. 2020:p. 101965. https://doi.org/10.1016/j.artmed.2020.101965
27. PwC Health Research Institute analysis of PwC Bot.Me: A revolutionary partnership, PwC Consumer Intelligence Series survey, 2017, and PwC Health Research Institute Provider Survey, 2017
28. Paruk F, MD. Think 4 keys to success with AI and machine learning. HIT December 2018.
29. Skochelak S, etal. Health Systems Science. Elsevier, 2019.
30. Char DS, Shah NH, Magnus D. Implementing Machine Learning in health Care—Addressing Ethical Challenges. New England Journal of Medicine 378; 11.
31. Leveso NG, Turner CS, an investigation of he Therac-25 accidents. Computer. 1993; July: 18-41.
32. Neves J (1984) A logic interpreter to handle time and negation inlogic databases. In: Muller R, Pottmyer J (eds) Proceedings of the1984 annual conference of the ACM on the 5th generation challenge.Association for ComputingMachinery, New York, pp 50–51.
33. Brynjolfsson and McAfee, 2011; MGI, 2013.
34. Hamet P, Tremblay J. Artificial intelligence in medicine. Metabolism. 2017;69:S36-S40. doi:10.1016/j.metabol.2017.01.011.
35. Gulshan V, Peng L, Coram M, et al. Developmet and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in retinal Fundus Photographs. JAMA 2016; 316 (22):2402—10.
36. Esteva A, Kuprel B, Novoa RA, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature 2017; 542 (7639):115—8.