What is Missing From Healthcare AI Efforts? User Centricity

whatismissing

At ThetaRho, we strive to provide a non-obvious but informed perspective on the challenges and opportunities for AI in healthcare. We believe that the end-user perspective is underrepresented in the current discourse regarding AI in healthcare. The idea of using AI to transform healthcare is not new. History is littered with well-funded entities placing large bets on AI in healthcare. However, impact, especially macro-measurable impact, remains elusive, especially for users of these technologies. The unfortunate reality is that the lived experience of almost all healthcare professionals is a far cry from the AI-promised utopia.

At this point, there is sufficient evidence to indicate that technologies designed to improve healthcare, especially EHR technology, have increased the workload of users, not reduced it. Indeed, physicians are experiencing record levels of professional discontent and burnout at unprecedented rates partly because of the unfulfilled promises of EHR technology.

In theory, AI can improve the quality and efficiency of healthcare delivery, but we must first consider what it will take to improve user experience. As Thomas Edison famously said, “Vision without execution is just hallucination.” And he was not commenting on GenAI or LLMs.

There are many reasons to believe that the GenAI revolution fundamentally differs from previous innovation waves. Every day, these technologies demonstrate capabilities that were thought impossible only a few years ago. Not to mention the investments made by enterprises large and small. However, the complexity of AI implementations, particularly with GenAI, creates unique challenges for adoption by highly trained physicians.

Current GenAI is akin to the jet engine in that it will eventually fundamentally alter the world. However, jet engines are ineffective when operating alone. You must build planes with controls to tame these jet engines, and airlines must purchase planes to transport people from one location to another. Similarly, these powerful GenAI need to be embedded in larger applications with controls and provide compelling user experiences to be useful.

To be fair, we have tried. However, the story of AI in healthcare is a tale of two cities: Clinical AI, which has received a lot of attention and investment, and Administrative AI, which has received relatively little attention. In both cases, the user experience is frequently an afterthought.

Clinical AI

Hundreds of medical AI algorithms have been developed and approved by the U.S. Food and Drug Administration, some with the goal of “beating physicians” at diagnostic tasks. It is unclear how many of these algorithms have had a significant impact on medical practice. The common narrative surrounding clinical AI is that it is not “AI replacing” humans, but “AI + humans” is superior to simply humans. In other words, the use of AI can reduce the apparent gap between a skilled therapist and a less experienced clinician.

However, research shows that the interaction between AI and humans in clinical settings is complex, and it is impossible to make definitive statements like “AI can enable the inexperienced physician to perform better” without making a lot of assumptions and preconditions that may not hold most of the time. It also emphasizes that AI errors are frequently the driving force behind system performance. Because these are probabilistic systems, they are likely to fail occasionally; nonetheless, without mechanisms to detect and recover from errors when using these systems, they are likely to be less impactful.

Administrative AI

Because diagnosis is such a complex endeavor, one may be tempted to think that Admin AI is potentially easier to develop and deploy than Clinical AI. But this isn't always the case. For example, a patient may visit the same provider numerous times with the same complaint and receive various treatments. It is not always easy to determine what was done previously and build on an existing treatment plan. This frequently has little to do with the physician's expertise in diagnosing and treating patients; rather, it is due to the physician's inability to access all the necessary information at the right time. Daniel Kahneman, the late Nobel Prize-winning economist, refers to this as “noise in the system,”which leads to a lot of unnecessary re-work, avoidable costs to the healthcare system, and negative impact on the quality of patient care.

Making relevant information available for the task at hand available to physicians is critical for improving patient care. We currently operate under the false assumption that the current presentations of EHR data are complete and comprehensive. They are not. Most of the data is available but accessible only through complex menu navigation and, therefore, often missed. Mining the EHR for all the relevant information needed for a patient interaction often requires hundreds of clicks and excessive time that physicians don't have. Accessing data from other organizations, even if available in the EHR, often requires dumpster diving by the physician. In fact, not only are physicians weary from navigating the EHR, they are also concerned about missing critical data, which adds to their stress. Thus, the promise of EHR to alleviate the burdens of physicians and improve patient outcomes falls short.

Admin AI, as envisioned by ThetaRho, can deliver faster, cheaper, and better care.

Faster because physicians will have easy access to relevant contextual information
quickly
Cheaper because physicians will avoid unnecessary costs like repeated laboratory tests when they have access to more complete information
Better because physicians regain the joy of practicing medicine when they are not
wasting time dumpster diving for relevant information or being data-entry clerks

Check out our next blog to learn how we are tackling Admin AI, and join us in the effort.