The Patient Is Still Attached
by Serelora
This article was originally published on Medium.
Read full article on MediumOn specialized medicine, portable records, and the peculiar business of reconnecting what never stopped being one person.

Maybe I’m an idiot, but nobody has ever managed to explain to me why my eyes need their own software.
My teeth have theirs. My mind gets a third, assuming anyone bothered to digitize it. The rest of me, the part we call “physical health” as if the retina were made of something else, lives in a fourth. Each has its own login, its own vocabulary, its own billing codes, and its own sales team. All of them describe one person, who is currently in a waiting room filling out the same damn clipboard.
Vision health. Dental health. Mental health. I know what the labels describe. What I can’t work out is when the labels got their own border checkpoints.
What we actually do for a living
Take away the white coats and the mystique, and healthcare is the delivery of science to a patient. We take what we know, acknowledge what we don’t, apply judgment to the person in front of us, and try to do something useful. Then we check whether it worked.
That last part matters. Otherwise, we are delivering recommendations and hoping for the best – somewhat different profession. Which was the art of medicine back in Voltaire’s time, but not really the art of medicine today.
“The art of medicine consists of amusing the patient, while nature cures the disease.”
– Voltaire
The science becomes increasingly specific. We distinguish between conditions that look alike, identify subgroups within diagnoses, and learn why an intervention helps one person and does very little for another. Good. That is medicine getting better at its job.
But what holds true across a population does not tell us everything about the individual sitting in front of us. Clinical judgment has to bridge that distance. Then the judgment has to become something that actually happens.
We have known about this gap for a long time. More than twenty years ago, a large American study found that the adults studied received only about 55 percent of recommended care. Preventive, acute, and chronic care all performed similarly. We knew considerably more than we reliably delivered.
Delivery has a foundation that doesn’t care which organ is involved. Find out what’s going on. Write it down. Decide. Order the thing. Get it paid for, this being ’Murica (although, to be fair, healthcare is paid for everywhere; we just have a particularly convoluted way of arranging it). Tell the patient. Make someone responsible. Check back.
The specialties supply the knowledge, instruments, measurements, and decisions appropriate to the problem. They do not invent the need to obtain a history, communicate a finding, or discover whether an order was carried out.
Even the work inside the electronic record reflects this commonality. Across a large sample of adult nonsurgical outpatient practice, reviewing the chart, documenting, and ordering accounted for roughly three quarters of physicians’ EHR time. The clinical content changed. The underlying verbs remained familiar.
I am not asking a dentist to interpret a retinal scan. I am asking why the infrastructure surrounding both examinations needs to behave as though the patients have never met.
Consider the kitchen
A serious restaurant kitchen understands specialization. One cook does sauces. One works the grill. Another runs the cold station. Pastry lives in its own corner with its own temperature and its own temperament, and nobody sane asks the pastry chef to cover the grill on a Saturday night.
But the stations are working from the same order. Everything meets at the pass, where someone makes sure table six gets its food together.
Now imagine solving the pastry problem by buying pastry a separate ordering system, asking the diner to carry information between stations, and charging the restaurant for an interface so the chocolate tart can discover that a steak exists.
That is roughly the arrangement I am questioning.
There is a related distinction in medicine between managing someone’s care longitudinally and managing that person around an intervention.
A cardiologist following a patient over years is managing the continuing course of that person’s cardiovascular care. An interventional cardiologist, acting in that role, manages the patient in relation to a particular intervention, including its evaluation, execution, and associated care. Both involve management. Their purpose, scope, and time horizon differ.
The intervention belongs within the longer history. What led to it, what happened during it, and what followed all matter to the person continuing the patient’s care.
The software should preserve that relationship without turning it into a referral followed by an administrative scavenger hunt.
The eye doctors had a point
In fairness to the people who bought the eyeball software, the large systems gave them reasons.
One early comparison at an academic eye center found annotated, hand-drawn sketches in all 150 paper notes studied and none of the corresponding electronic notes. The electronic records captured more complete examination findings, but an entire method of representing those findings had disappeared.
The anatomy had not become text-only. The software had.
Policy also distributed the opportunity to digitize unevenly. Psychiatric hospitals and other institutions were excluded from major federal adoption incentives available to acute-care hospitals, and early research documented substantially lower adoption in those settings.
Different parts of care entered the digital era with different resources, under different rules, and with different products available to them. Clinicians needed tools that understood their work. Buying something useful was a reasonable response to being offered something inadequate.
What is harder to defend is treating the resulting fragmentation as a permanent requirement of medicine.
The useful drawing interface can remain. The private informational kingdom surrounding it is negotiable.
And then there is billing, which has an impressive ability to become the organizing principle of things that were supposedly built for other purposes.
When office-visit coding and documentation rules changed in 2021, an early evaluation found immediate shifts in billing patterns without meaningful changes in note length or time spent in the record.
The revenue machinery adapted. The paperwork retained its position.
A practice needs to be paid. Nobody is suggesting that clinicians live on gratitude and conference tote bags. But supporting an invoice and preserving the clinical meaning of an encounter are different jobs. A good system should do both without making the patient’s history subordinate to the bill.
What we accumulated was a mixture of legitimate clinical needs, policy boundaries, purchasing decisions, and commercial opportunities. Once those boundaries existed, work appeared around them. Somebody had to request the record, export it, receive it, identify the patient, find the relevant passage, and enter the information again. Somebody else had to explain why the systems still disagreed.
These are real people doing necessary work under the circumstances. The question is whether we need to keep reproducing the circumstances.
What should actually travel
Interoperability ought to mean that another authorized system can receive information and use it meaningfully. Even the federal definition includes use without special effort on the user’s part.
That is a more demanding proposition than successfully sending a file.
A PDF arrives. The transfer succeeds. Somewhere, a dashboard turns green. The receiving clinician still has to read seventy pages to discover that the diagnosis on page four was ruled out on page sixty-two.
The file moved. Much of the work did not.
This is why I care about building a clinical identity. A patient identifier helps establish whose information we are looking at. A clinical identity should help us understand what is known about that person, how it became known, what remains uncertain, and how it has changed over time.
It is a maintained representation of the person’s clinical history and circumstances. It will always be incomplete. Its usefulness depends partly on making that incompleteness visible.
It is also not a diagnosis list with a photograph attached.
The mechanism has to be concrete. Incoming records must first belong to the right individual. An uncertain match should remain uncertain until it can be reviewed. A beautifully organized history attached to the wrong person is still the wrong history.
Then the information has to become usable without losing its origins. A laboratory result appearing in a laboratory feed, a specialist’s report, and a primary care note may be one observation represented three times. The system should preserve those sources without inventing three separate measurements.
Time needs similar care. When something happened, when it was documented, and when another clinician learned about it may be different dates. Those differences matter when reconstructing what someone knew when they made a decision.
So does the status of a finding. Suspected, confirmed, resolved, and ruled out mean different things. A patient’s report, a clinician’s interpretation, a billing entry, and a model’s inference carry different kinds of evidence.
Reconciliation should make disagreement understandable. It should not quietly award victory to whichever document arrived last.
The relationships matter just as much as the individual facts. A finding informed a decision. A decision produced an order. Someone accepted responsibility. The action happened, was delayed, was declined, or never occurred. Something followed.
The record should preserve documented reasoning without allowing an algorithm to invent a causal explanation merely because two events happened near each other.
Consider a prescription that was never started. A rejected claim and a patient’s message about cost might explain what happened. Those events belong in the same account when the evidence supports the connection. A claim alone does not establish that someone took a medication, and a missing claim does not establish that they did not.
Without those distinctions, an administrative obstacle can acquire a clinical label. The patient becomes “noncompliant” with a treatment they could not obtain.
This is the history another authorized clinician should be able to pick up. The patient should not have to become the sole surviving witness to their own care.
Different tools, same person
On top of that foundation, let the specialties be as specialized as they need to be.
Give the ophthalmologist the drawing and imaging tools. Give the dentist tooth-level charting. Give the interventional team the measurements, device connections, and procedural documentation its work requires. Each should extend a common account of the person’s history and ongoing care.
For years, much of what made a clinical application useful had to be explicitly designed, programmed, and maintained for a particular workflow. Different requirements produced different products, and those products grew into separate systems.
AI gives us more ways to retrieve relevant evidence, assemble a history, interpret documentation, and adapt how information is presented for a particular task. That creates an opportunity to make specialty workflows more flexible while keeping the underlying record, permissions, and validated clinical requirements consistent.
We do not need every application independently reinventing patient intake, documentation, ordering, billing, messaging, and follow-up. We need a common foundation capable of supporting those activities in the forms the clinical work requires.
Different evidence requirements become rules the system applies. Different response times become constraints it respects. Different clinical questions bring forward different information.
Those differences belong inside the architecture.
Nor does a shared foundation require every practice to buy one enormous product. That would be an extraordinarily convenient interpretation for the company selling the foundation. Standards-based application platforms already demonstrate how specialized tools can work across electronic record environments. A practice should be able to replace an inadequate component without undertaking an archaeological expedition through its patients’ histories.
The same principle applies to AI models. A general model can support common work across the record. A particular task can bring in the relevant structured data, clinical evidence, specialty instructions, and tools. Where a specialized model is appropriate and adequately evaluated, use it.
Specialization can reside in the model, the supplied knowledge, the tools, the task definition, or a combination of them.
We might be reviewing a longitudinal course, evaluating an intervention, reconciling medications, or determining whether the documented evidence meets an authorization requirement. Each task directs attention to different relationships within the same person’s history.
Research on clinical guideline interpretation has already shown how much model performance can improve when relevant guidance is supplied and prepared carefully. That gives us a useful direction for building these systems. It does not make every polished answer correct.
The output should remain identifiable as an interpretation or proposal, with supporting evidence and unresolved assumptions available for review. Required approvals should be recorded. An action should update the history with what actually happened.
“The model suggested it” and “the patient received it” must never become the same entry.
Models, interfaces, and specialty applications should be replaceable. The clinical history is what needs continuity.
The record should be able to leave
This is where the architecture becomes economic.
When documentation, billing, authorization, referral, and research each demand another reconstruction of the same patient, we pay repeatedly to recover meaning from information we already have. Then we buy integrations to keep the reconstructions approximately aligned.
Some of that work is necessary. Secure infrastructure, maintenance, terminology mapping, reliable identity matching, and error handling cost money. People doing that work deserve to be paid.
But a necessary engineering service and a commercially useful obstacle are different things.
Organizations that exchange health information continue to describe fees and business practices as barriers, even as their reports of potential information blocking have declined. Some connections are difficult because the engineering is difficult. Others remain difficult because somebody benefits from the arrangement.
A larger language model does not renegotiate the contract.
Meaningful portability changes that equation. Interoperability allows systems to exchange and use information. Portability allows a patient or practice to move the usable record into another authorized environment while preserving the clinical meaning accumulated along the way.
A download button is useful. A download button followed by six months of reconstruction is a rather theatrical form of freedom.
The receiving system should be able to preserve dates, sources, corrections, clinical states, relationships, and unresolved work. Supporting documents and images should remain available through permitted copies or dependable access. An unfinished task should remain recognizable as unfinished, with responsibility explicitly accepted by the receiving person or team.
A source link that stops working when the contract ends is a rather conditional form of medical history.
This standard has to apply to Serelora too. We do not get to criticize captivity and then offer a nicer cage.
Every year of accumulated history should increase the usefulness of the record without making the vendor harder to replace merely because departure would be painful. When a customer can leave with a usable history, the company has to offer a better reason to stay.
A coherent record also need not place every original document under one company’s control. Different organizations can retain custody while authorized systems maintain a usable account across them. That requires shared definitions, dependable access, traceable updates, and permissions governing which information can be used for which purpose.
Shared does not mean visible to everyone. Unity of meaning does not require centralization of control.
For an independent practice, that distinction is practical. A physician should be able to participate in a larger clinical network without selling the practice or surrendering the ability to choose tools.
We should earn money because we help useful work happen. “Your history is here and leaving is unpleasant” is a much less inspiring business model.
Who gets to ask the next question
A usable longitudinal record also changes what an independent practice can investigate.
A physician notices that patients with apparently similar measurements are following different courses. Perhaps the difference is treatment. Perhaps it is access, documentation, a missing variable, or chance. The practice should be able to examine its own population and determine whether the observation deserves further investigation.
Other practices should be able to test it under appropriate governance. Some questions will require approved data pooling. Others may permit local analyses with appropriately protected results exchanged between sites. The scientific question, permissions, and safeguards should determine the arrangement.
An interesting pattern is a reason to do the next piece of scientific work. What changes is who can begin.
The independent clinician can originate the question, help characterize the population, contribute to validation, and participate in what follows. The ability to notice something worth studying should not be reserved for whoever owns the largest warehouse.
There is economic value in that capacity. It comes from preserving clinical distinctions, the circumstances of decisions, the work that followed, and the outcomes that can actually be observed.
Patients and clinical teams contribute to that value. Their participation, permissions, and any compensation arrangements deserve deliberate treatment. They should not become an afterthought once someone discovers a market for the resulting information.
Access to care or one’s record should never depend on accepting a commercial data arrangement.
I want the ability to generate knowledge to move outward alongside the ability to use it. Otherwise, decentralization amounts to asking more people to supply the same warehouse.
Healthcare begins with delivering science to a patient. A properly structured record can also help that patient’s experience contribute to the next scientific question.
That is a more interesting ambition than generating the same report faster.
Somebody still has to stand at the pass
There is a trap in this argument, and the technology industry has fallen into it often enough that we should probably put up a sign.
Connecting records does not, by itself, make anyone healthier.
Automated notifications have been shown to shorten the time to action on important test results without significantly improving the overall proportion receiving documented follow-up. The information arrived sooner. That was useful, and insufficient.
A heart-failure program that combined close follow-up with active treatment optimization did improve clinical outcomes. People received different care, and their outcomes changed.
That should tell us something about what we need to build.
A handoff needs an accepting person or team, a clear next step, an appropriate time frame, and a way to determine whether the action occurred. An unanswered assignment should remain visibly unresolved.
Putting someone’s name in a field does not constitute an agreement with that person.
The organization also has to provide the capacity to do the work. A referral does not create an available appointment. A reminder does not provide transportation. A portal does not make treatment affordable. Those barriers should remain visible instead of disappearing behind a completed administrative task.
And the patient still needs a conversation. A clinician reviewing an outside encounter, reconciling what happened, and agreeing with the patient on what comes next is doing substantive clinical work. Removing the effort of hunting down the record gives that conversation more room.
Some friction is waste. Some friction is the point.
A kitchen does not work because the ticket printer works. Someone has to stand at the pass, notice what has not arrived, and do something about it.
We should measure the system accordingly. Saving staff time, reducing cost, increasing appropriate follow-up, improving clinical outcomes, and widening access are distinct achievements. One does not entitle us to claim the others.
A perfect audit trail can document a terrible decision. More completed actions can include more unnecessary care. A better-organized record can be valuable without entitling its vendor to claim that it has made patients healthier.
That claim has to be earned.
One patient, one ticket
This is what we are building Serelora around. A shared clinical foundation that supports specialized work while keeping it connected to the person’s continuing history. A record that preserves what was known, what remained uncertain, what was decided, who accepted responsibility, and what followed.
Something useful across specialties, reusable for legitimate purposes, and portable enough that its value does not depend on staying with us.
Then we have to test whether it delivers.
I want the person operating on my retina to know an extraordinary amount about retinas. I just do not see why that expertise requires a private filing cabinet.
Specialize the knowledge all you like.
The patient should not need a software migration to remain the same person.
It’s just health.
References
- McGlynn EA and colleagues. The Quality of Health Care Delivered to Adults in the United States. New England Journal of Medicine, 2003.
- Overhage JM, McCallie D Jr. Physician Time Spent Using the Electronic Health Record During Outpatient Encounters. Annals of Internal Medicine, 2020.
- Sanders DS and colleagues. Electronic Health Record Systems in Ophthalmology, Impact on Clinical Documentation. Ophthalmology, 2013.
- Wolf L, Harvell J, Jha AK. Hospitals Ineligible for Federal Meaningful-Use Incentives Have Dismally Low Rates of Adoption of Electronic Health Records. Health Affairs, 2012.
- Apathy NC, Hare AJ, Fendrich S, Cross DA. Early Changes in Billing and Notes After Evaluation and Management Guideline Change. Annals of Internal Medicine, 2022.
- United States Congress. 21st Century Cures Act, Section 4003, Interoperability. 2016.
- Mandel JC and colleagues. SMART on FHIR. Journal of the American Medical Informatics Association, 2016.
- Kresevic S and colleagues. Optimization of Hepatological Clinical Guidelines Interpretation by Large Language Models. npj Digital Medicine, 2024.
- Office of the National Coordinator for Health Information Technology. Trends in State, Local, and Regional Health Information Organizations’ Experiences of Perceived Information Blocking, 2019–2025. Data Brief 85, June 2026.
- Dalal AK and colleagues. The Impact of Automated Notification on Follow-up of Actionable Tests Pending at Discharge. Journal of General Internal Medicine, 2018.
- Mebazaa A and colleagues. Safety, Tolerability and Efficacy of Up-titration of Guideline-directed Medical Therapies for Acute Heart Failure. STRONG-HF trial. The Lancet, 2022.
RELATED ARTICLES
Explore more insights and perspectives from our team.


