The future of electronic medical records: Why resistance to change is rational

If interoperability is so good, why isn’t it everywhere? That question has plagued Canadian digital health for more than two decades. We have standards. We have HL7 FHIR, an international standard for sharing health data. We have APIs, tools for connecting disparate databases. We have data repositories and strategies.

Yet patients still repeat their histories. Clinicians still search for data across systems. Information moves, but mostly over fax and often without coherence.

And patients notice.

They want access to their information. They deserve access to their own information. And most physicians want to give it to them. But too often, what sits inside the electronic medical record (EMR) is fragmented, inconsistently coded, buried in narrative text, or difficult to extract in meaningful form.

Dirty data does not just slow systems and understanding. It blocks transparency and sharing.
Interoperability in Canada has been treated primarily as a connectivity problem. In reality, interoperability is an alignment problem.

Healthcare is cautious for good reason. But caution is not the core barrier. The deeper issue is that the burdens and benefits of interoperability do not fall on the same actors.

The physician entering structured data does not necessarily experience the downstream system value. The clinic assuming cybersecurity exposure does not necessarily capture the analytics or research benefit.

When risk and reward are misaligned, resistance is rational.

There is also a harder truth. Interoperability, implemented poorly, can increase costs, increase cognitive load, and increase perceived privacy and cybersecurity risk. When physicians resist sharing data, they are not ‘resistant to change’. They are acting rationally, by avoiding risk where there is no benefit for them.

More connections mean more complexity. More inbound data means more information to interpret.

More inbound data means exposure to more medico-legal risk. More endpoints mean more cybersecurity risk. Without careful architecture, interoperability redistributes effort and risk without redistributing the benefits that drive adoption.

This is most visible in primary care, where the longitudinal patient story lives. Here, prevention, chronic disease management, prescriptions, lab results, diagnostic imaging, referrals, specialist notes, emergency department and hospital discharge summaries and follow-up converge. It is also where data is often messy. Cleaning and standardizing it requires time and expertise that clinics do not have.

When we ask why data does not flow, the answers are practical. The data is too dirty to share reliably. Cleaning it takes effort. Sharing increases perceived risk. Integration increases costs.

If we begin from this reality rather than from aspiration, the path forward becomes clearer.

The next stage of electronic medical records is not more exchange. Surprisingly, it already exists to a great extent. What is needed instead is clean, validated, standardized data inside the clinic, created in a way that reduces human effort rather than adds to it.

If information can be coded automatically to international standards such as SNOMED, ATC, and LOINC, the burden shifts from clinicians to systems. If data is captured once and reused many times, duplication falls and clinician effort falls.

If structured summaries can be generated instantly from validated data, cognitive load decreases. At that point, interoperability stops being about compliance and coercion and starts being about capability and clinical value.

We have evidence that this model can work. Through CPCSSN, Canada’s Primary Care chronic disease surveillance system, primary care data from multiple EMRs is extracted, transformed into a common structure, coded, cleaned using natural language processing, and de-identified for secondary use.

In return for participating, clinicians receive meaningful information that assists them during the patient encounter, reinforcing participation.

The lesson is simple. Reciprocity drives sustainability.

This architectural shift is no longer optional. Bill S-5, through the new Artificial Intelligence and Data Act, signals Canada’s intention to regulate high-impact AI systems with transparency, accountability, and risk management at the forefront. Health AI tools, from online appointment booking to predictive analytics, depend on structured, reliable data.

However, regulation without infrastructure creates burden. Infrastructure without regulation creates risk.

If messy, inconsistently structured data flows into AI systems, exposure will increase and increase exponentially. Garbage in, super garbage out.

If, on the contrary, validated, standards-based data flows within Canadian governance frameworks before it is shared, responsible AI becomes possible. Bill S-5 raises the bar for accountability. Clean data makes that bar achievable.

Data sovereignty will be the driver for change: Increasingly, data sovereignty is entering our collective vocabularies. Data sovereignty is not about isolation. It is about agency. Who controls how clinical data is structured? Who benefits from the intelligence derived from it? When cleaning, normalization, and modeling occur outside Canadian governance frameworks, strategic control shifts. Even if raw data remains within our borders, value can and will migrate outward.

Dirty data creates dependency and sovereignty risk. Clean data creates choice.

When identifiers remain local and only structured, purpose-specific information flows outward under clear governance, Canadians retain control. Researchers gain analyzable datasets. Vendors and startups can compete on quality and innovation rather than proprietary access and deep pockets. AI systems operate on validated inputs that meet regulatory expectations.

But sovereignty and architecture are not ends in themselves.

Patients are the reason this matters.

For patients, interoperability is not about APIs. Rather, it is about not having to retell their story in moments of vulnerability. It is about medications reconciled accurately across settings. It is about a specialist seeing a coherent summary rather than disconnected fragments. It is about logging in and actually understanding their own health record.

Doctors want to provide that transparency. They want to share information confidently. They want to support patient access. They are hampered not by unwillingness, but by the condition of the data itself and rational reasons for not sharing.

The boring design principle: Interoperability is meant to be infrastructure. It should be boring. That is not an insult. It is a design principle. When it is working, no one notices. No one should notice. Information flows safely. Transitions are smooth. Patients experience continuity without thinking about standards.

If we want interoperability to succeed in Canada, the sequencing must change. First, invest in automated cleansing and validation within primary care so that data becomes useful at the point of care. If physicians don’t benefit from clean data, their incentives to share are considerably less.

Second, design architectures that reduce risk by keeping identifiers local and limiting external access to what is necessary. Third, ensure that every interoperability initiative delivers visible value to clinicians and patients before broader system gains are pursued.

Interoperability is not a privilege granted to those who can afford integration. It is a promise to patients that their information will follow them safely, intelligently, and meaningfully.

That promise does not begin with another API.

It begins with clean data, inside Canadian primary care.

Karim Keshavjee, MSc, MD, MBA, is program director, Master of Health Informatics, Institute of Health Policy, Management and Evaluation at the Dalla Lana School of Public Health, University of Toronto.