Electronic Health Record

Anyone who has visited a doctor leaves behind a file there. It records what the doctor found, which medications were prescribed, and what blood values or X-rays looked like. An Electronic Health Record is exactly this file, only as a data record on a computer instead of paper in a cabinet. The difference is bigger than it first sounds. Paper exists in exactly one place and must be read by humans. A digital record can be copied, searched, sent to a clinic in another city, and analyzed by software.

Why the record decides over medical errors

The practical benefit shows up in emergencies. Someone is brought unconscious into an unfamiliar clinic and cannot say anything about themselves. If their record exists digitally, doctors can see within seconds their allergies, their pre-existing conditions, and their current medications. Without this information, every treatment begins with guesswork.

A second point concerns interactions between medications. Older people often take eight or more preparations simultaneously, prescribed by different doctors. None of these doctors necessarily knows about the others' prescriptions. A shared digital record can automatically warn when two active substances interact dangerously.

For the AI industry, such records are also an enormous treasure trove of data. Millions of documented disease progressions are the material from which models for diagnosis or risk prediction can be trained. But this is exactly where the conflict arises: health data is among the most sensitive information there is, and no one wants it to end up with an employer or an insurance company.

From free text to uniform codes

An Electronic Health Record is at its core a database. Every entry has fixed fields: patient identifier, date, type of entry, content. Much of it is not stored as free text but as a code from a fixed catalog. For diagnoses, for example, there is the globally used ICD catalog, in which every disease has its own abbreviation.

This standardization is the actual trick. If one doctor freely writes “sugar” and another “type 2 diabetes mellitus,” no software can reliably recognize both as the same thing. If, however, the same code appears in both records, the matter is unambiguous. So that systems from different manufacturers can exchange data, there are additionally transmission standards, the currently most important one being called FHIR.

In reality, however, a lot of unstructured text remains: doctor’s letters, surgical reports, handwritten notes. This is where language models come in, i.e. AI systems that process text. They summarize long reports or extract diagnoses from continuous text. This is not yet reliable everywhere, because an error in a medical record can cost someone their life in an emergency.

The electronic patient record in Germany

In Germany, the implementation is called the elektronische Patientenakte, or ePA for short. Since 2025, every person with statutory health insurance receives it automatically, unless they explicitly object. This principle is called opt-out: you are included as long as you do not actively say no. The record is managed via an app provided by the health insurance fund, in which individual documents can also be blocked for certain doctors.

In the USA, the market has been established for years and is dominated by a few large providers, above all Epic Systems and Oracle Health. These companies regularly appear in business news, usually when they build AI functions into their systems. A common use case is automatic documentation: software listens in during the doctor’s consultation and writes the entry into the record from it.

A common misconception is that an Electronic Health Record is simply a folder of PDF files. Scanned findings are indeed often part of it, but the point lies in the structured, analyzable data behind it. Equally wrong is the idea that all doctors can view everything at any time. Access is strictly regulated by law and generally tied to authorization by the patient.