The Algorithm Is In: Is Adoption of Healthcare AI Outpacing Understanding?
Our Nation’s strategy for better healthcare hinges on putting digital technologies to work.
Today’s powerful tools make it easier to capture and share patient information, coordinate care, and streamline clinical workflows. However, these technologies also have created some unintended consequences. Health systems have massive data sets to manage, and providers face burdensome documentation requirements, voluminous charts to review, and overflowing inboxes.
As the healthcare industry looks for relief, artificial intelligence (AI) may offer solutions. AI-enabled systems can help humans perform key functions—diagnostics, decision support, administrative tasks—faster and, in some cases, with enhanced accuracy. At the same time, AI’s risks warrant serious consideration.
As healthcare AI moves from promise to practice, we must mitigate its potential to exacerbate biases, privacy violations, access divides, and health inequities. We must also ensure that AI-enabled health technologies are safe for patients.
To steer the healthcare industry toward a safe, effective rollout of AI technologies, we need rigorous research and evaluation.
Interest, Adoption, and a Window for Gathering Evidence
Across the digital healthcare technology community, it’s clear that AI has reached an inflection point. Scan a list of recent conferences, research journals, and trade publications, and you’ll see a spiking interest in the promise and pitfalls.
For a variety of reasons, this interest has not yet translated into broad-based adoption. According to the Brookings Institute, healthcare lags far behind other industries in implementation of AI-enabled technologies. While some may be eager to accelerate the pace, this delay also presents a window of opportunity—one that DHR and AHRQ as a whole, as well as other agencies and coalitions, can use to begin gathering evidence to help guard against AI’s risks while maximizing benefits for providers and patients.
This is no small feat. The healthcare AI landscape is a blur of challenges: rapidly accelerating development, nascent but evolving policy guidance, the understandable clamor among providers for burden relief, and the numerous familiar hurdles associated with adopting new technology into clinical practice.
A Running Start on Healthcare AI Research
Already, DHR has studies underway that are exploring healthcare AI, amounting to 17% of our new grants funded since 2020 (AHRQ-Funded Projects).
We also have an immense body of research on digital healthcare technologies to help frame our thinking about AI.
For example, we can evaluate the lessons learned in the early years of health IT implementation, when adoption of electronic health records (EHRs) was lagging behind promises for improving care quality. We can also develop tools to assess patient safety outcomes associated with healthcare AI, as we did for computerized physician order entry systems. Similarly, we can apply the results of ongoing efforts to develop a framework and guide to counter persistent digital divides and healthcare inequities, which may be exacerbated by AI-enabled solutions.
These and other findings from decades of work to date can help guide the research needed to support implementation of healthcare AI.
The Need for Agile Research Mechanisms
Even as we seize the advantages of existing models and frameworks, an AI-enabled world demands an agile approach to research. While traditional grant mechanisms that fund 5-year studies are important, they will not produce the timely evidence we need to keep pace with the development of or the burgeoning need for healthcare AI. Rather, studies must take an iterative approach—learning and refining as they go—to produce actionable insights and recommendations before adoption outpaces understanding.
AHRQ is well positioned to play a role in applying current and emerging research methods to answer pressing questions about AI-enabled healthcare solutions.
To make this happen, priorities may need to shift to fund more research that is conceptual or exploratory. Existing grant mechanisms such as R21 and R33 grants are ideal for pilot studies that can be scaled. However, given the blazing pace of enabling technologies like ChatGPT, even these mechanisms may be too slow, driving funders to consider yet more agile study approaches.
Balancing Research Priorities
Even as the excitement around AI dominates conversations, research priorities must strike a balance. Vital questions about EHRs, clinical decision support systems, and other foundational technologies remain—particularly their effects on health outcomes and health equity. These systems represent the backbone of America’s healthcare transformation, and many of the insights gained from their continued study will be directly applicable to healthcare AI.
It’s imperative that DHR receive adequate funding to conduct the rigorous research this field needs.
While the pace of healthcare AI may necessitate new and more nimble research methods, AHRQ’s DHR program remains steadfast in its purpose and commitment. Across our endeavors, we will continue to work toward a digital health ecosystem that ensures patients and their families have access to safe, effective, equitable, and high-quality care.