The Quiet Power of AI in Healthcare: Closing Care Gaps Before They Become Costly

When considering AI's applications in health care, I don't start from the examples of advanced medical robotics, automated procedures, or from artificially intelligent tools involved in diagnosis and assessment. I see the applicability of health care related AI in performing more routine tasks. Will a hospital system's AI help a doctor find the pieces of important information that have remained stubbornly absent before and will now be in plain sight given the proper context? I do believe that AI's most useful application in the hospital system will be systematic, and even if it is not dramatic or exciting, it will fulfill a crucial purpose. The systematic application of AI to health care will answer questions that have remained unanswered, include oversights in essential lists and prioritize overdue follow up actions. AI will help close care gaps that undermine the public's faith in the health care system, especially if they are persistent and frustrating.

Gaps Stem from Complexity Rather than Negligence

Care teams and operations leaders acknowledge that most gaps occur due to complexity within the system, not from team members failing to fulfill their roles. For example, no one followed up on critical labs because no one person could see that information. Although information related to that failing was in the system, it was scattered across several systems and was not accessible to the right person at the right moment. It is infeasible for a single care navigator to extract that information one system at a time and put it in front of a care team on a continual basis. This is the potential that AI has for transforming healthcare systems. AI has the ability to continuously and seamlessly analyze disconnected information across an entire system, highlight missing information, and present tasks that need to be completed to members of the care team.

Patients may not be able to recognize the complexity, but they feel the broken system. The start of a new appointment involves filling out a blank medical history form and trying to remember what the previous provider documented, while the care team attempts to reconstruct a medical history that should be assembled prior to the appointment. The same medical record is produced and split again and again, and the patient carries the burden of a broken healthcare system.

The Dangerous Gap You Cannot See

This problem has a more difficult version, and I find myself returning to it repeatedly. One of the values in a record can be wrong, but eventually, someone will catch that discrepancy. An absent value, on the other hand, leaves no evidence. The model will think it over and will show no signs of the fact that the one value that would have altered the plan was not recorded. No fairness review can identify the missing fact. The absence of information is the most dangerous form of bias, because nothing will alert to it.

This is exactly why a record that is based on a single event is so unhelpful. Healthcare systems usually record what happened during the encounter and forget everything that happened before, and the history leading up to the encounter. This missing part of the record is most likely the part that would have altered the decision. When a model is given a record that is shaped this way, it learns that single episode and ignores the patient.

The Underrated Work Is Standardizing the Chaos

Anyone who's worked in a health system know they are made up of many silos. In many of the organizations I have worked with, there were five, ten, or even twenty different platforms. Each platform had its own structure and format and its own version of data. By the time this data reached the point of care, it was pretty much useless. The majority of the time spent was devoted to trying to understand and reconcile the data rather than analyze it.

This is the ideal place for the stealthy application of AI to disrupt the status quo. AI, coupled with Natural Language Processing and Entity Recognition, is now more capable of reading data in all of its disparate and scattered forms and digitally piecing together patient histories than ever before. This enables the clinical team to conduct much more extensive and sophisticated patient research.

There are also many other areas within a health system where similar AI technologies are being used, especially imaging. While AI is still a relatively new application in this area, it is already meaningfully enhancing the completeness of the health system records. In my experience of observing rudimentary and ineffective systems, failure is most pronounced where records are incomplete.

How I Advise Health Systems to Start

As you start to build your strategy, operations, or clinical innovations, you do not begin with a tool. You begin with a measurable gap. Look at where care is leaking. What are the overdue screenings? What are the unacted lab results? What are the unanswered referrals? Which gap do you want to make the most impact on? Once you identify the gap, the technology is chosen to support the outcome instead of the reverse.

From this point, integration is the line that is not negotiable. A tool that cannot pull data from all of your systems will never encompass all of your systems. Having a pretty interface on top of partial data is just as bad as providing no data at all. The goal of starting small to learn is missing when designing for integration from the beginning. Start small to learn on one practice, one specialty, or one workflow that matters. The first day you start the design, think of all the challenges you will face when you start to scale, because an unplanned pilot will most likely never be done.

Now for the last point, which is most often missed. Insight is not the goal, action is. A model that produces another dashboard has only created bigger work, not smaller. Value is created when the system provides a specific and timely next step to the right person and that next step is easier to accomplish than to ignore.

The Best Technology Does Not Announce Itself

Many consider AI's application in healthcare as science fiction. In reality, AI is employed in ways that remain quiet to the general public. A concerning screening slips by undetected, an intervention occurs after the fact, or a patient transfers to a different care setting and the paperwork slips smoothly between care providers. This is not newsworthy and rarely even takes a year to deploy. With a little care, AI very much starts to improve the situations described.

This is the kind of improvement that attempts to push the boundaries of AI. It does not look like anything you could create a pretty slide for. Rather, it looks as simple as gathering the necessary information in front of a care provider and giving the care that the patient needed.

Christopher Hutchins Founder and CEO, Hutchins Data Strategy Consultants