Who Actually Controls Your Medical Record?

There is a rule in US healthcare that I think deserves more attention than it gets, because it inverted a power relationship that had held for a very long time.

You have a right to electronic access to your health information, through an application of your choosing, without special effort or fees.

Read that again with the emphasis on of your choosing. Not an app the hospital approved. Not a portal they built. Any application you authorise.

Why this needed a law

For most of the digital era, your medical records were technically yours and practically theirs.

You could request them. You would receive a PDF, or a printout, or a fax — formats that are readable by humans and useless to software. Which meant you could not easily take your history to a new provider, feed it into a tool that checks for drug interactions, or aggregate care from five different organisations into one picture.

None of that was necessarily malicious. Some was ordinary institutional inertia. But some of it was a business model. If moving your records is difficult, moving providers is difficult, and health systems that make it difficult keep patients.

The 21st Century Cures Act information blocking provisions addressed this directly. It is now unlawful for regulated actors to unreasonably interfere with the access, exchange, or use of electronic health information.

The interesting part is not the penalty. It is that health systems must now serve third-party applications they have no commercial relationship with, on a patient's authority. For organisations used to controlling every integration through a contract, that is a genuine reversal — you are required to serve clients you did not approve and cannot vet.

Why the technical part is hard

You might reasonably assume this is straightforward. Build an API, let apps request records, done.

Three things make it harder than that, and they are the same three things that make hospitals unable to share records with each other.

Everyone codes things differently. Your diagnosis is stored using one system's internal codes. Another system uses different ones. Unless both translate into shared international standards, the data arrives technically delivered and practically unreadable — like receiving a document in a language you do not speak.

Nobody is sure you are you. There is no national patient identifier in the US. When your record arrives somewhere new, that system has to guess whether you match an existing patient, using your name, date of birth, address and phone number. All of which get entered inconsistently, change over time, and are sometimes wrong.

Guess too eagerly and two people's records merge — someone's allergy list now contains a stranger's data, and the merged record looks perfectly consistent, so nobody notices. Guess too cautiously and you get duplicates, and a doctor views one of three partial versions of your history.

Records disagree and nobody knows who is right. One says you take 10mg, another says 20mg. Neither is obviously stale. And you cannot simply trust the newer one, because medical timestamps record when something was written down, not when it was true — a note typed today might describe a change from last year.

The two kinds of system

This is where two acronyms that sound interchangeable actually differ.

An EMR is the digital version of one clinic's paper chart. Complete within those walls, never designed to leave them.

An EHR assumes your care happens in many places and the record has to assemble a coherent picture from all of them — which means it must publish data others can read and consume data others produce.

On screen they look identical. Both have notes, prescriptions, results. The difference is entirely in whether the hard problems above were solved from the beginning, and it stays invisible until the day it matters.

Where AI changes the stakes

This has become more consequential recently, and not for the reason people expect.

AI tools in healthcare need organised information covering your whole history. A record that only knows what one clinic did, storing most of the detail as typed prose, cannot support them.

The failure mode is what makes this serious. A drug interaction checker that cannot see prescriptions written elsewhere will report that everything is fine — because it genuinely cannot see the interaction. That is arguably more dangerous than having no checker at all, because you would trust its answer. A risk-prediction tool trained on one hospital's data learns that hospital's habits rather than actual medicine, then degrades quietly when used elsewhere while still producing confident-looking numbers.

There is also a newer wrinkle. The AI tool clinicians actually want is the one that listens to the appointment and drafts the notes, so the doctor can look at you instead of a screen.

For that to work safely, the record has to know that software wrote an entry, and whether a human has checked it yet. Most older systems cannot express that at all — they were built assuming every entry was typed by a person, so there is nowhere to record "a machine suggested this and nobody has verified it." Adding that capability turns out to be one of the harder rebuilds in health software, because it touches everything.

What it means practically

If you are a patient: you can request your records in a machine-readable form through an app, and obstruction is not permitted. That right is newer and stronger than most people realise.

If you are building or buying healthcare software: this is not terminology. It decides your legal obligations, roughly doubles your engineering cost if you do it properly, and determines whether you can use AI tools later without a rebuild.

And the costs are lopsided. Building the sharing capability you did not strictly need costs money. Discovering you needed it after launch costs a rewrite, plus everything lost in the market while carrying it out.

Full technical version: EHR vs EMR: What the Difference Means for Your Build

Frequently Asked Questions

What right do patients have to their health records?

Patients have a right to electronic access to their health information through an application of their own choosing, without special effort or fees. Regulated actors may not unreasonably interfere with that access, exchange or use.

Why was a law needed for this?

Because records were routinely provided in formats readable by humans but useless to software, such as PDFs and faxes. Some of that was institutional inertia, but difficulty moving records also made it difficult to move providers, which suited the organisations holding them.

Why is building this technically difficult?

Because systems code the same clinical concepts differently, there is no national identifier to confirm two records describe the same person, and there is no reliable rule for which version wins when records disagree with each other.

Why can the newest record not simply be trusted?

Because medical timestamps generally record when information was documented rather than when it was true. A note written today may describe a medication change from a year earlier, so preferring the newest entry can discard current information.

How does this affect AI tools in healthcare?

Tools such as interaction checkers need a complete history. One that cannot see prescriptions written elsewhere reports no problem because it cannot see one, which is more dangerous than having no tool, since the reassuring answer invites trust.

Why can older systems not handle AI-drafted notes?

They were designed assuming a person typed every entry, so there is no way to record that software produced something and that no clinician has verified it. Adding that distinction requires changes across the entire system.