Generative AI is an impending sea change for healthcare delivery. This technology has the potential for transformative impact across every part of the care continuum, from helping providers deliver remote care more efficiently, to presenting a more personalized and intuitive patient experience, to easing the enormous workload of clinical documentation. But while the potential of AI is already well-publicized, far less attention has been given to the processes and strategies that must be put in place if providers and provider practices are to realize that potential.
Generative AI is far more than individual capabilities that can be grafted onto existing models of care delivery. Instead, providers must begin viewing it as a strategic discipline unto itself, one which will require early investment and thoughtful planning if they are to take full advantage of its potential and build the digitally enabled healthcare delivery models of tomorrow — efficiently, safely, and ethically.
Healthcare’s capacity crisis
The emerging potential or generative AI comes at a pivotal time. The combination of burnout, healthcare worker shortages, and increase in the demand for care are creating a crisis of capacity that spans all levels of the healthcare system — and it’s a crisis that literally cannot be addressed under existing models of care delivery.
A Mercer analysis of the labor market estimates there will be a shortage of more than 3 million healthcare workers by 2026. Raw demographics also indicate that the labor pool in general will shrink by almost 30% by 2033. At the same time, the United States’ aging patient population is driving up healthcare demand and dramatically changing patient preferences about how and where they receive care. The only ways to address the simultaneous phenomena of rising demand and a shrinking workforce are to change how care is delivered — who does it and where it’s done — and to adopt technology that can augment the capabilities of what humans can do and automate the tasks that they don’t need to. Synergy and alignment between these two approaches is essential for an effective solution.
On top of demographic shifts, the “how” and “where” of healthcare delivery is already changing dramatically. We are moving away from a bricks-and-mortar model of healthcare toward a more personalized and flexible paradigm where changing patient preferences and the rise of concepts like aging in place are putting a greater emphasis on digital and home-based care. These shifts are also placing greater importance on the widespread adoption of remote monitoring, digital therapeutics, and other capabilities.
These new channels of care delivery do not fully make up for healthcare’s capacity gap on their own, but they open the door for AI-augmented capabilities by driving more flexible, modular, and digitally enabled platforms of care. This is essential to scale the healthcare workforce with more speed and impact among practices, providers, and patients.
The changing face of AI
Generative AI is transforming awareness of artificial intelligence at all levels of society and will support our day to day in ways we can and can’t even fully anticipate. The outputs driven by these models can support use cases that range from chat bots to guide patients through basic medical questions or help them access the right services at the right time, to intelligent summary tools that surface critical insights from encounter notes for busy clinicians.
What is different about Generative AI “going retail” is the scale and breadth of its applicability. Where past automation and AI technologies were limited to specific use cases or processes, (e.g.: AI tools for detecting likely cancer in radiography images), generative AI is highly flexible and able to mimic many of the capabilities of human communication, sometimes even creating new insights from existing data or predicting future outcomes from clinical and demographic insights. However, keeping humans “in the loop” will be essential to safe benefit. Experience with a decade of AI use in healthcare indicates that its real benefit is not in replacing physicians or other human healthcare workers, but in increasing the efficiency and capabilities of humans by elevating decision-making and automating low-value processes such as EHR documentation.
Through AI’s power in prioritization and summarization, (which it does better than many humans, once trained), provider workload can be dramatically reduced by using AI tools to flag and summarize contextually critical information, increasing capacity without adding staff, while allowing doctors and nurses to focus more on top-of-license capabilities and reduce signal to noise ratio in an overwhelming tsunami of information from multiple sources. Smart chatbots can also interactively augment patient experience in areas such as scheduling, transportation, and clinical triage, potentially reducing the staffing requirements necessary to meet the 24/7 needs of patients while improving response time, access, and patient satisfaction.
By identifying areas where AI can replace and augment human effort, it will not just make human workers more efficient but also free up their time to focus more on high-value activities such as face-to-face patient interaction and the actual work of delivering care, rather than paperwork or administrative tasks (a leading cause of healthcare worker burnout).
Blazing the way to healthcare’s AI-enabled future
Envisioning tomorrow’s clinical care requires us not only to understand AI as a technology, but how it fits into the larger picture of process automation and identifying, evaluating, and prioritizing those use cases is where its real value lies.
One of the critical elements in understanding generative AI is awareness of its dependency on the underlying data used to train it. The tendency toward “hallucination” stems from the model-driven bias to produce “something” from the data and can result in quite “real”-looking output that is not only factually inaccurate, but plausibly referenced. This is far more likely when the data underlying the model is hugely broad (i.e., the publicly available internet, as in ChatGPT). This becomes less of a risk as the data driving the model is better aligned to the intended domain of inquiry, (e.g.: GPT-4 with a large clinical data set and access to referenceable academic publications) but can still be subject to the bias of population selection. All of this demands that both the model architecture and the data to train it will require new skills and capabilities, as well governance and oversight to ensure its accuracy and proper usage at multiple levels. Without accurate data or the necessary oversight, generative AI cannot deliver on its promise of safely reducing provider burden.
The healthcare industry is only now beginning to unlock the potential of both automation in general and generative AI in particular. As physicians and leaders, we must begin laying the groundwork now to create the necessary conditions for success. The critical first step is self-education in what these technologies can and cannot do, and where they will add the most value to our work.
Make no mistake: healthcare AI is here, and it’s advancing at a rapid pace. Current implementations may be focused primarily on non-clinical areas such as revenue cycle, prior authorization, and patient communications, but broader clinical adoption is not far off. And what’s more, simply adopting these technologies is not enough. As providers, we must be ready to completely rethink our models of patient engagement, care delivery, and business fundamentals if we are to gain the benefit of the technology. By preparing now, we position ourselves to be at the forefront of this revolution — rather than left struggling to catch up.