Numerous industries are being disrupted by growth in new technologies, especially information technologies, and healthcare is no exception. Advances in robotics, wireless sensor networks, 5D printing, and cloud technologies are reshaping countless industries. I am intrigued by the increasing importance of automation, machine learning, and artificial intelligence (AI) in healthcare. Let us explore three questions together:
- Where are common applications of AI and automation in healthcare?
- What implications for physician assistants (PAs) arise from increased automation and AI in caring for patients?
- Did AI bring back the 's that causes any self-respecting PA to cringe?
I nearly panicked recently when I caught sight of the following two headlines from online articles about new healthcare technologies, which might lead a person to think the PAs of the future are not people at all. At the very least, I was ready to e-mail the AAPA communications team to combat those pesky apostrophes. The articles actually detailed advances in automation and AI within healthcare.
Bright.MD raises another $8M for “virtual physician's assistant” SmartExam (www.mobihealthnews.com/content/brightmd-raises-another-8m-virtual-physicians-assistant-smartexam)
Healthcare Chatbots: The Physician's Assistant of the Future? (http://blog.kantarhealth.com/blog/brian-mondry/2016/11/28/healthcare-chatbots-the-physician's-assistant-of-the-future)
Next, let us sort out AI and automation. According to Merriam Webster, artificial intelligence is the capability of a machine to imitate intelligent human behavior. Automation, on the other hand, is the automatically controlled operation of an apparatus, process, or system by mechanical or electronic devices that take the place of human labor.
A widely adopted automation in healthcare is appointment reminder software that automatically reminds patients of their upcoming scheduled appointments, with options to customize the message and/or time it is delivered for patient preference. Similarly, missed appointment notification systems can alert a PA to a potentially worrisome pattern of missed appointments for a patient identified as high-risk. Robotics, commonly deployed in areas such as pharmacy and surgery, are automations proven to increase efficiency and safety.
According to CB Insights, about 86% of healthcare provider organizations, life science companies, and healthcare technology vendors are using AI technology. The most common applications seem to fall into one of ten categories: managing medical records and other data; doing repetitive jobs such as analyzing tests, interpreting radiologic studies, and data entry; helping design treatment plans; digital consultation (such as the Babylon app); virtual nurses (such as the Molly app), medication management (such as the AiCure app); drug development; precision medicine; health monitoring; and healthcare system analysis.1 Numerous tech giants are investing heavily in AI applications for healthcare as well, such as Microsoft's Healthcare NExT initiative and Google's Deepmind Health.
IMPLICATIONS FOR PAS
The Guardian's Richard Vize anticipates two key implications of AI for medicine—making affordable healthcare more accessible to everyone and augmenting (or replacing) the current pool of medical professionals.2 I do not believe these technologies will widely replace providers such as PAs or physicians in the foreseeable future. The practice of medicine is too intimate. Geoffrey Hinton, a computer scientist at the University of Toronto, was quoted in TheNew Yorker on how learning machines will affect clinical medicine.3 He suggested radiologist interpretations may become obsolete as deep learning systems in radiology outperform them. He qualified his analysis with, “A deep-learning system doesn't have any explanatory power ....The algorithm can solve a case. It cannot build a case.”3
It seems inevitable that AI and automation will profoundly change the way PAs predict illness and practice preventive medicine. A recent analysis of healthcare AI startup companies found that one in four was focused on exactly that.4 The episodic and inconsistent way we currently assess and manage risk for future illness or disease complications is fraught with inaccuracies and bias. If you think about it, healthcare to date has relied on data that are limited by the number of cases and are challenged by sampling requirements.
Millions of Americans now wear a FitBit or Apple Watch, and it is logical to consider continuously collected biometric or lifestyle-related data interfacing with AI systems to evaluate health risks. Could PAs evolve to become experts to harness that data and facilitate risk mitigation strategies for affected patients? As care paradigms continue to move toward value and emphasize population health, greater reliance on automation and AI (including its machine learning applications) seems necessary. The Institute for Health Technology Transformation has reported that automation makes population health management feasible, scalable, and sustainable. I would add that meaningful population health is an impossible goal without significant adoption of both automation and AI. Linking clinical research to large population data pools could likewise accelerate studies with more generalizable results, increase sensitivity to recognize uncommon benefits or adverse risks earlier, and uncover new connections between lifestyle and health (or disease) at the individual and population levels.
In a fascinating and thought-provoking report in Nature, a team of scientists trained a class of artificial neural networks, using images, to classify skin lesions and identify skin cancer.5 Their system was able to classify skin cancer with a level of competence comparable to dermatologists. Researchers in the United Kingdom compared use of the American College of Cardiology/American Heart Association (ACC/AHA) guidelines with four machine-learning algorithms to predict a first cardiovascular event in a patient population over a 10-year period. All four techniques, which analyzed data from the records of about 380,000 patients, performed significantly better than the ACC/AHA guidelines. The authors concluded that machine-learning significantly improves accuracy of cardiovascular risk prediction, increases the number of patients identified who could benefit from preventive treatment, and could help avoid unnecessary treatment for others.6
AI seems to have the capacity to make healthcare more effective and individualized for patients. Arguably, the most irritating effect of the electronic health record is how it has subjugated medical providers to data entry technicians versus optimizing our time as clinical professionals. In a twist of fate, maybe AI offers a reliable means to make medicine more personal again. If machines can generate and record the information we need, providers could have more time to interact with patients to focus on health maintenance and managing treatment for illness. These technologies seem on track to eventually reinforce our diagnostic skills and fill gaps in our memory, amidst an increasingly complex and expansive biomedical knowledge we must harness to practice competently.
The integration of automation and AI with telemedicine seems another probable strategy to improve access to affordable care. Some 400 million people worldwide, according to the World Health Organization, lack access to even basic medical services. Many millions more cannot afford it. Can we digitally extend the reach of our current healthcare workforce to improve access, if technologic systems create more efficiencies, assist in triage, or supplant technical or clinical activities that do not require a trained professional to complete?
Let us reconsider the implications suggested by Vize (making affordable care more accessible and augmenting our workforce). Vize added, “This has profound implications for medical training and what defines a leading clinician. It will be those who can harness AI to their own medical knowledge and their human skills of context and empathy who will be the leaders of their profession.”2 In any regard, the potential consequences of automation and AI for medicine and the PA profession are many, including legal, ethical, regulatory, and educational issues, not to mention an impetus to reconsider the very nature of collaborative practice. Recently, the Canadian government recognized the importance of emerging technologies to the future of healthcare in that country, and began a dialogue and formal stakeholder engagement process to embrace that future.7 Maybe we should consider following its lead.
1. Novatio Solutions. 10 common applications of artificial intelligence in healthcare. http://novatiosolutions.com/10-common-applications-artificial-intelligence-healthcare. Accessed January 8, 2018.
2. Vize R. Technology could redefine the doctor-patient relationship. The Guardian
. March 11, 2017. http://www.theguardian.com
/healthcare-network/2017/mar/11/artificial-intelligence-nhs-doctor-patient-relationship. Accessed January 8, 2018.
3. Mukherjee S. AI versus MD: what happens when diagnosis is automated. The New Yorker
. April 3, 2017. http://www.newyorker.com
/magazine/2017/04/03/ai-versus-md. Accessed January 8, 2018.
4. Shriftman J. Four things I learned building chatbots for major brands in 2017. https://venturebeat.com
/2018/01/05/4-things-i-learned-building-chatbots-for-major-brands-n-2017. Accessed January 8, 2018./
5. Esteva A, Kuprel B, Novoa RA, et al Dermatologist-level classification of skin cancer with deep neural networks. Nature
. 2017;542(7639):115–118. www.nature.com/articles/nature21056. Accessed January 9, 2018.
6. Weng SF, Reps J, Kai J, et al Can machine-learning improve cardiovascular risk prediction using routine clinical data. PLOS One
. http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0174944. Accessed January 8, 2018.
7. Canada Senate. Standing Senate Committee on Social Affairs, Science, and Technology. Challenge ahead: integrating robotics, artificial intelligence, and 3D printing technologies into Canada's healthcare systems. https://sencanada.ca
/content/sen/committee/421/SOCI/reports/RoboticsAI3DFinal_Web_e.pdf. Accessed January 8, 2018.