Predictions on AI in primary care in 2024
My musings on where we are headed this year in the intersection of AI and primary care, and an announcement on a new effort to explore health and humanity in the age of algorithms
Healthcare, AI and hype in 2024
According to Gartner, who publish the now well-recognized “hype cycle” of emerging technologies and their relative maturity and adoption, generative AI and AI-augmentation are approaching their peak in hype, which is unlikely to surprise anyone. Generally speaking, the buzz is exceeding delivery of value from AI at the present moment, particularly when we talk about primary care.
With that said, let’s dive into talking about what I think lies ahead in the near future. Here are some of my thoughts for how AI will impact preventative and primary care in 2024, broken down by the perspective of different players in the care delivery system:
Clinics and administrative staff: LLM-based tools will begin to become the patient’s “front door” for tech savvy care delivery organizations; the key value will be reduction of admin burden, which will translate to improved work experience for admin staff as well as cost savings for clinics (reduced staff). Enhanced patient experience will also be a driving factor, particularly in private markets. Likely there will be overlap with the same groups that were first to adopt online appointment booking. This adoption will be greater for the mid-to-large sized care organizations (which historically have been more likely to adopt technologies to improve efficiency), as compared to independent or small practices.
Patients: People will slowly become more comfortable with LLMs as their first-stop for general medical advice (compared to a Google search). The availability of medical-specific LLMs will additionally mean more trust and credible options for the public to have their questions answered. However, my guess is that by year end, Google search will still remain the primary resource for searching for medical information. Bard (Google’s general purpose LLM), which is currently only available in the US, will eventually become more deeply integrated into Google, though my guess is Google will retain both classic search along with Bard results, given how deeply their legacy search in entrenched in their business model as well in the general public. Additionally, on average, use of an LLM for medical information will not tangibly make patients any more “informed” than if they had used a Google search, and it will not appreciably decant patients away from basic primary and preventative care.
Clinicians: The dominant forms of LLM utilization by primary care clinicians will be ambient scribe and general LLM use for medical reference purposes. Penetration of ambient scribe usage in the market will increase over the year, though I think by year end that number is unlikely to be more than 20-30% in North America (certainly it won’t be the majority). There will also be an increased utilization of stand-alone LLM usage for medical reference / clinical decision support, as the market gets flooded with “medically fine-tuned” GPTs on OpenAI’s store, and there is a proliferation of other LLMs that are either medical foundational models or fine-tuned for usage with medical knowledge. We will see trusted brands bring their own model into the public market, and I anticipate Mayo Clinic and UpToDate to be amongst those throwing their hat in the mix.
Medical schools and academia: The pressure to incorporate and respond to the acceleration of AI into medical education will begin to rise dramatically in this year. Beyond isolated sessions, I doubt significant real progress will be seen this year (nor will AI become a core competency for clinical trainees in 2024, as I advocated for here). The exception to this will be newer programs/schools that may benefit from starting with a clean slate (examples being the Kaiser Permanente School of Medicine and the Toronto Metropolitan University School of Medicine). We will continue to see a rise in continuing education programs / certificates in relation to AI in medicine come onto the market for healthcare providers already in practice that are looking for a primer.
Technology: If 2023 was defined as the year of LLM experimentation and ambient scribe’s entry to the market, 2024 will be the year of maturation and specialization. Ambient scribe penetration in the market will increase (though a significant fraction of the market will be impervious to the change). We will begin to see the second-order benefits of ambient scribe technology, namely value-add automation beyond clinical documentation (e.g. automation tasks that were picked up in the conversation). Ambient scribe tools that benefit from EMR integration will be able to demonstrate their value-add more clearly, with the release of functionality such as automated post-visit tasks (letter generation, billing, coding, etc).
Ecosystem: I foresee more people looking to non-traditional players (i.e. big tech and corporations, private virtual care companies, etc) to access primary care as the year goes on. Major corporate players (who have historically failed to make a meaningful entrance into primary care delivery) will be more likely to succeed in the present climate. It will be interesting to follow of the progress of Walmart Health and Amazon Clinic in particular. I also anticipate there to be consolidation through M&A activity in the smaller specialized companies offering “disease specific” and “virtual care only” models.
Bottomline, my prediction for 2024 is that while we will see an increased usage of AI tools in primary care this year, the gap between what AI can do and it’s real-world usage in clinics will only widen.
There are many reasons for this, for which the baseline inertia to change in healthcare being just one point. A more nuanced factor is that the pace of change with which AI-based technologies (LLMs in particular) are advancing is putting virtually every industry on edge. Even though the state of primary care is dire in both the US and Canada, healthcare is simply not designed in current configuration to be able to handle such a velocity of change.
Remember that fax machines are still pervasive in healthcare, despite the fact that more compelling technology has been available for decades.
The suppressing effect of rapid change
Alvin Toffler described the idea of “future shock” in his 1970 book of the same title as being a state of disorientation due to an excessively fast pace of technological and social change, leading to irrational decision-making.
Building on Toffler’s concept, I propose the idea of “velocity incapacitation” taking place in healthcare, whereby there will be an inverse relationship between the speed of advancement of AI and the ability of the healthcare system to be able to leverage the advances.
The healthcare system takes time both in evaluating and adopting new technologies, inherently for both intentional (“do not harm”) and unintentional (Kafkaesque processes) reasons. While technology has been moving fast for quite some time, the current AI revolution is unique a few reasons beyond the incredulous rate of change we’re witnessing: 1) the blackbox nature of much of it’s function which makes it more abstract to understand and the lack of necessary explainability; 2) the significant centralized control with big technology companies (in many cases), leading to gaps around version control and predictability and 3) a notable lack of consensus amongst experts as to where things are headed and their exact impacts.
Hypothetical scenario: imagine for a moment that a group of clinics was looking to make a decision on utilizing a generative AI chatbot with patients. The group is composed largely of enthusiastic clinicians and administrators, with a few hesitant team members. They do their due diligence to ensure this new tool adheres to requires privacy and safety standards, and establish a set of expectations for both what it can do and how they intend to interact with it based on current state. However as the decision-making process proceeds, the technology in question gains impressive new functionality. Public announcements take place that it can now make clinical decisions deemed superior to the average primary care clinician, that it outperforms human clinicians on empathy and compassion in voice conversations, and that it has a realistic avatar that accompanies it. Hesitation grows within the group, as they try to reconcile the implications of these net new capabilities. Meanwhile, a virtual clinic company launches that is powered entirely by this new generative AI chatbot.
While this phenomenon will certainly not be isolated to the healthcare industry, it will be uniquely pronounced here and this will juxtaposition tremendously with just how badly the ecosystem could benefit from efficiency gains and reforms.
“Science fiction is the sovereign prophylactic against future shock.” - Alvin Toffler
I remain grounded in my cautiously optimistic position that this wave of AI advancement will bring about positive change, albeit not as much as we might think or hope for. Deeper and more systematic change is necessary, and technology (however capable it may be) will be unable to bring this about in isolation. We will have to entertain approaches and collaborations we have not seen in the past. We will in fact need to use our imagination. I intend to dive deeper into these ideas in the coming weeks.
Health and humanity in the age of algorithms
I’ve never been one to give much significance to the transition of one year into another, but this year it’s felt different. A combination of personal run-ins with the healthcare system, a dizzying acceleration in the evolution of AI and an increasingly complicated global geopolitical climate have given a new raison d'être for me to write. This year I will both narrow my focus to the intersection of AI and preventative/primary healthcare, while also broadening the reach to capture what is happening around this world in this field. I also intend on bringing a different perspective through my words. As opposed to purely looking at the future as a physician-technologist (as I have in my writing in the past), I hope to write more authentically as a father of two young children, as a resource for navigating healthcare that many in my family and community of friends rely on, and as a global citizen who still believes there is a shared humanity worth preserving.
If you’re sensing a philosophical tone in my narrative, you’re reading this right. Not only do I feel like our times call for more soul-searching around what health really means to us (and who “deserves” it), but also that there may be benefit in what I hope will be a messier but warmer contrast to a rising tide of utilitarian content that solely focuses on AI’s impact in healthcare. Each approach serves its purpose. Given where I live, I will, of course, continue to have a strong focus on what’s happening in and around the US and the Canada, but I truly wish to learn, connect and reflect on what is and isn’t working in preventative health across the globe.
Stay tuned as I hope to unveil more details about this new effort in the coming weeks.
Until then, I remind and urge those practicing on the frontlines (or training to do so) to begin making themselves AI ready. It will be an eventful year. Change is the air, and it is up to us to ensure it is the kind we have yearned for.