Healthcare Data Interoperability and the Trust It Carries

In a conversation over Signal Room, Dr. Barry Chaiken, former board chair of HIMSS, made the request to "fix interoperability" rather than presenting the issue in the abstract and making a general case for the need to fix interoperability. He made the point that you can use ATMs to withdraw cash virtually anywhere in the world. However, one hospital in Boston may not have access to an MRI taken by a hospital that is 1 mile down the road. This is because the two hospitals use competing software systems. The issue in this case is not with the MRI image. MRIs use a standard file format. The issue, and the answer to Dr. Chaiken's request, is that the software systems refuse to share the image.

While a systems interoperability request may sound like a plumbing complaint, at Hutchins Data Strategy Consultants, we understand that everything he mentioned about AI, trust, and the future of the healthcare system is dependent on solving that one problem. This is often the most significant gap we see when dealing with clients. Clients may want to develop advanced systems, but the data needed to support those systems is trapped in software that refuses to share across systems.

The Value of Records That Will Not Integrate

Chaiken showed that the cost of fragmentation is real, not theoretical. When doctors are unable to access records of previous tests, they order repeat tests, which is a waste of resources and can be detrimental to the patient. The inability to access existing records can result in a patient's condition being misdiagnosed. There is also a cumulative, long-term cost of having incomplete records. Without integrated records, there is no unified data to be the basis of a system to teach AI.

Chaiken said there are more achievable goals to integrate systems than the solutions proposed by the healthcare industry. He said healthcare is more complex than the banking system, and there is no excuse for the healthcare industry to be in the state that it is. The companies that design and implement the electronic medical systems should not be finding the current state to be acceptable.

Data Belongs to the Patient

Chaiken said that the data is owned by the patient and not the hospital or the company that implements the system. Patients are willing to share their data to improve the systems and to serve a better purpose.

He deliberately avoided making an anti-commercial argument. Clearly, there is nothing wrong with these businesses being for-profit and making profits. What patient ownership entails is an obligation to "give back," and one "gives back" by providing genuinely valuable tools, as well as addressing the interoperability dilemma, as opposed to using the dilemma as a competitive moat. That is a significant difference. When the patient is the owner, hoarding data behind systems with little to no compatibility starts to look less like a competitive strategy and starts to look like a failure to fulfill the data sharing agreement.

Trust Is the Foundation, and AI Has the Potential to Compromise That

Chaiken's worries regarding AI extended beyond the healthcare vertical, and this was the reason behind his emphasis on the importance of the data layer. He described trust as the foundation of society and provided a simple example: at some recent point in time, you left a bag or a cart unattended in a shop and asked a stranger to keep an eye on it, and you never really considered the possibility that they would take something. He argued that this innate trust is part of our humanity.

He fears that AI could erode that instinct to trust. He referenced the recent elections occurring in countries around the world that utilized deepfakes and asked if we have the ability to demarcate real from fake anymore. If we get to the point where we can no longer trust one another, there is a breaking point in the social contract. He framed this situation as a fear, not a forecast. He has clarified this is not a guide to the advanced AI threat, but rather the U.S. and the EU have taken very different approaches to dealing with advanced AI. The U.S. recommends more of a 'get there quicker' approach to advanced AI, whereas the EU prefers more of a 'take it slower control and regulations.'

He says that the same principle that helps manage a health care system applies to trust. He connected trustworthy AI to employees feeling acknowledged and included. Building the type of trust that an AI-based system is genuinely reliant on, begins with leadership. Trust is not a marketing word; it is the condition that has to already exist for the deployment to hold.

A Model Trained Elsewhere Is Not Ready Here

Chaiken's primary point is that interoperability is the supply line of AI, and he was clear about the consequences of a data model that is indifferent to the kind of people it is designed to serve. Training a model on a population in Oklahoma or Indiana, for example, would not work in New York City, which is more diverse and pulls staff from every corner of the globe — and the opposite would be just as bad. Training is specific to a region.

This is where his interoperability argument and his bias concern intersect. A shared, standardized data set is not just a convenience for a clinician retrieving a scan. It is the substance that will help determine whether a data model is designed to serve the population in front of it or an entirely different population. The risk of bias that he identified was that the danger is rarely intentional. It is the unintentional gap — the assumption, the data that remains uncollected. From the clinician's side, as he put it, you do not know what you do not know about a patient, and you do not know whether the thing you are missing is the thing that will make a difference. AI can help reduce the chances of that gap, but will not eliminate it.

The 2050 Upside Depends on the Foundation

Regarding the optimistic scenarios on the advancements of AI by 2050, Chaiken does not reserve his comments. He foresees AI discovering new avenues for patient treatments. He even foresees drug development to be less time consuming and more efficient. One example for his prediction is the use of AI programs to assist with protein folding and complex interactions. While this sounds like an obscure task to someone with minimal knowledge to the field of biology, Chaiken claims it is the foundation of the subject as a whole.

Chaiken even sees the possibility of running AI modeled clinical trials to be better than real trials with the capability of digitally modeling a patient with a specific genetic makeup to see how the drug would interact with the patient, as well as inclusion of patients that would typically never participate in clinical trials beyond an academic medical center. Chaiken believes AI will be able to customize and better a treatment regimen in a way staffing cannot. For example, a used regimen that a working single mother can adhere to, would be better than an optimal regimen that a working single mother cannot adhere to.

Chaiken believes there is AI that is promising, but hospital administration sees these tools as a means of making revenue and increasing automation without concern for the patients. Chaiken believes that these possible outcomes would be of benefit to patients if the data used was trustworthy and consolidated.

He finished with a recontextualization targeted at top managers. According to him, we tend to consider AI systems smart. In truth, beyond the scope of human influence, AI is dumb -- a purely statistical entity lacking knowledge, emotions, worldview, values, and opinions. The intelligent actor is the person using the AI. It is the user's experience and moral compass that dictates the appropriate use and the constraining of the system. AI, unregulated, is a dangerous entity with the potential to cause hallucinatory events, yet, with human judgment and discretion, it becomes a powerful and useful tool.

How Hutchins Approaches Healthcare Data Interoperability

We treat interoperability as the precondition for everything a health system wants AI to do, not a downstream IT detail. Before a model is chosen, we work on whether the data it would consume actually joins into one trustworthy picture of the patient — across the systems, formats, and standards that have to interlock for that to happen. That work is inseparable from data readiness for AI and from the data governance that decides how information is shared, by whom, and under what obligation to the patient who owns it. These themes run throughout The Signal Room podcast, where practitioners describe what coordinated, trustworthy data actually takes in practice.

The discipline is to keep the patient at the center of both questions at once — the technical one, can these records connect, and the human one, does this preserve the trust under which the data was shared. Get the foundation right and the optimistic version of 2050 has something solid to stand on. Skip it, and even a capable model is reasoning over a record that was never whole.

Authoritative sources

• HL7 FHIR interoperability standard
• ONC — HealthIT.gov
• U.S. Department of Health & Human Services — HIPAA
• NIST AI Risk Management Framework
• WHO — Ethics & governance of AI for health