The pandemic has had a number of effects on medical practice, and not all bad. It gave a huge boost to telehealth, accelerating its incorporation into regular outpatient care. Telehealth can be more efficient, provides improved access to care for those with physical or practical limitations (like childcare), and has reduced disparities in access to care based on socioeconomic status and rural location.
Another potential benefit is increased attention given to remote or virtual learning. As with telehealth, this can be problematic when it’s the only option and adoption is forced by circumstances rather than voluntary. But when incorporated in a thoughtful way it can have the benefit of removing barriers to access to high quality training and learning as well as standardize the quality of such education. There is another way to standardize and optimize education – use artificial intelligence (AI) rather than (or to supplement) human instructors.
It is fairly well established that there are differences in practice styles and details that lead to sizable differences in outcomes. These differences can be in subspecialty training or even regional. It is also well established that standardizing practice, staying close to updated best practices, leads to better outcomes. There are ways to accomplish standardized practice within the existing medical education paradigm, through continuing medical education, the publication of standard of care guidelines, and requiring recertification. However, these methods have not achieved the level of standardization that we want. Further, at the highest level of training, such as complex surgical skills, training is still mostly done through individual mentorship. This leads to persistent individual, institutional, and regional differences.
This is where AI can come in – providing a completely standardized yet individualized training experience regardless of other educational variables. The technology now exists for high-fidelity virtual reality simulations of patient care, including procedures, even brain surgery. The latter is the subject of a recent study comparing training outcomes using an AI tutoring system.
The study was performed at The Neurosurgical Simulation and Artificial Intelligence Learning Centre at The Neuro (Montreal Neurological Institute-Hospital) using 70 medical students as subjects. They all learned neurosurgical techniques using the Virtual Operative Assistant (VOA), which is a high-fidelity VR platform that allows for simulation of bimanual surgery. There were three study conditions. In the first students used the VOA and were provided feedback by a deep learning program called the Intelligent Continuous Expertise Monitoring System (ICEMS). In the second group the students were monitored in real time by live neurosurgeons who provided feedback to their technique. And in the third control group the students received no feedback. They found:
The researchers found that students who received VOA instruction and feedback learned surgical skills 2.6 times faster and achieved 36 per cent better performance compared to those who received instruction and feedback from remote instructors. And while researchers expected students instructed by VOA to experience greater stress and negative emotion, they found no significant difference between the two groups.
That is fairly significant. The VOA-ICEMS system also has the advantage of allowing repetition of training almost without limit (based on availability only). It does not require access to animals or actual patients. It also does not require the availability of a neurosurgical expert for training. Students therefore have access to more training and more standardized and thorough feedback.
The authors are not recommending that such systems replace human instructors. They write:
The authors are not advocating for the replacement of the present educational paradigms by automated systems. We recognize that human interaction is vital to learning. Rather, we believe that the integration and the wider availability of intelligent tutoring systems may complement present curricula while minimizing the impact of events such as a pandemic on trainees’ skills development. Surgical education programs will have to adapt in order to smoothly integrate automated and in-person education pedagogy. Intelligent tutoring systems can utilize a variety of simulation platforms to provide almost unlimited opportunities for repetitive practice without constraints imposed by the availability of supervision. In these risk-free environments, numerous adaptable and clinically relevant simulations can be tailored to the needs of learners, consistent with best practices for simulation-based education.
Obviously the application of this technology can go beyond neurosurgical, or even surgical, medical training (and beyond medical training). An AI-controlled virtual learning experience can be a personalized iterative process, optimized for learning and skill development. My first personal experience with something like this was in high school, when I learned chemical notation and equations using a simple program designed like a game. The program provided immediate feedback and I was able to go at my own pace while the tasks increased in difficulty as my performance improved. This is 40-year-old technology.
There are many such systems already incorporated into medical training. At my own institution we have a “virtual patient” neurological simulator where the residents can learn neurological localization and diagnosis. But these systems have not been developed and incorporated to the degree that they should, in my opinion. With the availability of virtual reality and deep learning AI systems, all of medical education can be augmented significantly by such systems. I would argue further that we really have no choice but to leverage this technology. Already medical students have more to learn than four years of medical school can provide, and competition for classroom time is getting fierce. The amount of knowledge to impart is increasing rapidly, challenging the standard curriculum to keep up.
The potential efficiency gains of incorporating automated virtual learning can therefore not be ignored. Further, this technology can provide a new tool to help close the gap between practice in the real world and optimal standard of care. Such systems could be used for continuing medical education, even becoming necessary to maintain certification and licensure.
What is needed now is a massive investment in the development of these virtual systems. That investment would be well worth it, and would likely pay for itself many times over in improvements in practice and outcomes.
