Skip to main content
5 min read

The Preliminary Discharge Letter Is Due Same-Day — and It Is an Admissions Problem

Germany's framework agreement on discharge management requires a discharge letter on the day of discharge, and ePA filing has carried sanctions since 2026. Why the deadline routinely fails on data that was never structured at admission.

Dr. Sven Jungmann

Dr. Sven Jungmann

CEO

Ward corridor with a packed patient bag and wheelchair, in the foreground an incomplete discharge letter on a clipboard

Friday, shortly after eleven. The discharge is promised for 2 p.m., the transport service is booked, the daughter has taken the afternoon off. Still open: the preliminary discharge letter, the medication plan, the documents for the nursing handover.

The ward physician is looking for the admission medication. It is in a fax from the general practitioner's office — forty pages, received three days before admission, filed as a single PDF. The cardiology report that the course section refers to twice is a scan nobody ever captured in structured form. Between rounds and an admission interview, the letter gets finished anyway, at 1:50 p.m.

What it is missing, nobody notices at first. The general practitioner notices on Tuesday.

The deadline is in the framework agreement

The Rahmenvertrag Entlassmanagement — Germany's framework agreement on discharge management, concluded between the GKV-Spitzenverband (the German association of statutory health insurers), the National Association of Statutory Health Insurance Physicians and the German Hospital Federation on the basis of § 39(1a) SGB V — obliges hospitals to hand patients a discharge letter (Entlassbrief) on the day of discharge [1, 2]. If the final letter is not ready, and on the day of discharge it rarely is, a preliminary discharge letter is required: with the information continuing care needs, from the diagnoses through the discharge medication to pending findings and initiated follow-up measures.

Since October 1, 2025, the elektronische Patientenakte (ePA — Germany's electronic patient record) has been added: hospitals are obliged to fill it, among other things with the discharge letter; since April 1, 2026, violations carry sanctions [3]. The same-day obligation has thus acquired a second address — and an audit trail that can be evaluated by machine.

In many hospitals, the preliminary letter is still created the way it is in the scene above: under time pressure, from sources that were never captured in structured form, by the person with the shortest tenure on the case.

The problem is older than the day of discharge

The same-day deadline is not a writing problem but a data problem. The discharge letter is the last document of the case; its content largely comes from the beginning. The history and the prior findings arrived with the referral. The admission medication is the reference against which every change in the discharge medication must be explained. The diagnoses at admission carry the case summary (epicrisis). If this information was left sitting as a fax stack at admission, it gets reconstructed on the day of discharge — in the tightest time window of the case. Or it is missing.

That missing input data carries over into missing output data also shows in AI-generated discharge documents. In a study from the University of California, San Francisco, 47 percent of GPT-4-generated discharge documents from the emergency department omitted clinically relevant information [4]. A language model summarizes what it is given; what is hard to find in the sources is especially likely to be missing from the result. With humans under time pressure, it is no different.

And the letter has recipients with deadlines of their own: the practice continuing treatment, the rehabilitation facility, the nursing handover, which cannot work without complete nursing documentation (in German). Every gap travels downstream — as a follow-up question, as a duplicate examination, as a medication error.

Write faster, or structure earlier

Whoever solves the problem on the day of discharge speeds up the writing: better text blocks, dictation, and by now AI drafts. That helps, but changes nothing about the starting position: the fastest draft from a chart full of gaps is a fast draft full of gaps.

The second path starts at admission. Prior findings are captured in structured form when they arrive (in German); the medication is reconciled at admission and maintained as a list; the course is documented continuously between admission and discharge (in German) instead of reconstructed at the end. Then the discharge letter on the last day is essentially compilation plus the physician's review. The same-day deadline remains ambitious. It stops being risky.

How aiomics approaches it

aiomics starts where the data originates. Incoming documents — faxes, scans, prior findings — are processed on receipt, verified and transferred into a structured chart with source links; every field shows which document it comes from. This ingestion with field-level provenance runs in production, as does base generation of document drafts from this chart.

The expansion stages are planned and labeled as such:

  • A forms pipeline with validated statutory forms, including the physician letter (Arztbrief) compliant with the statutory health insurance requirements, along with Word export into the facility's own letter template (planned).
  • The derived patient letter in plain language — generated from the signed discharge letter, so that the patient version and the physician version cannot drift apart (planned).
  • An export package for KIM (the German healthcare sector's secure messaging service) and the ePA (planned).

One boundary stands: aiomics transmits nothing to insurers, practices or the ePA. Transmission stays with the hospital information system (KIS) — aiomics prepares, validates and evidences.

What to measure any solution against

  1. Where does the solution start — on the day of discharge or at admission? Ask concretely what happens to the forty-page referral fax when it arrives.
  2. Where does the discharge medication come from? From a list reconciled at admission and maintained since — or from free text that a model gathers together on the day of discharge?
  3. Is the patient letter derived from the signed physician letter, or is it a second, independent generation? In the second case, the two versions can contradict each other, and nobody notices.
  4. What does the system show about information missing from the draft? A draft without an omission display shifts the completeness check entirely onto the physician — on the day with the least room for it.
  5. Who transmits to the ePA? If the vendor wants to transmit itself, clarify the responsibilities first — in case of doubt, it is the hospital that gets sanctioned.

If you want to think through your discharge pathway from admission onward, write to us — no evaluation intent required. Or subscribe to our weekly briefing Visite (German; English edition Grand Rounds is in preparation), covering documentation and AI in German healthcare.

Sources

  1. § 39(1a) SGB V. https://www.gesetze-im-internet.de/sgb_5/__39.html
  2. GKV-Spitzenverband, Kassenärztliche Bundesvereinigung, Deutsche Krankenhausgesellschaft. Rahmenvertrag über ein Entlassmanagement beim Übergang in die Versorgung nach Krankenhausbehandlung (Rahmenvertrag Entlassmanagement), as amended. https://www.gkv-spitzenverband.de
  3. gematik. Elektronische Patientenakte: filling obligations for healthcare providers. https://www.gematik.de
  4. Williams CYK, Bains J, Tang T, et al. Evaluating large language models for drafting emergency department encounter summaries. PLOS Digital Health. 2025;4(6):e0000899. doi:10.1371/journal.pdig.0000899
#AI discharge letter#Discharge management software#Same-day discharge letter#ePA hospital obligations

The forms pipeline, the derived patient letter and the KIM/ePA export package are planned and not yet released; document ingestion with provenance and base draft generation are in production.

Keep reading

Editorial collage: a quarterly report whose figures are connected by threads to queries and control charts, a watermark across the page

Why aiomics for QM reports and quality analytics

A QM report in which every figure resolves to its query and the language model never sees individual data. All of it is planned — and we are laying the architecture open now anyway, so you can measure us against it.

Dr. Sven JungmannCEO
A physician stands at a wall-mounted clinic terminal in late-afternoon corridor light, one hand on the mouse, shoulders slightly dropped at the end of a long shift.
Reflections

The 4 p.m. Hazard: When Bad Software Becomes a Clinical Risk

We regulate how long a doctor may work, but not how hard the day grinds down their judgement. The most dangerous fatigue in a hospital is not in the legs. It is in the part of the mind that makes careful decisions — and clumsy software spends it for free.

Dr. Sven JungmannCEO
A physician stands at a cluttered clinical workstation, several monitors layered with records and alerts, pausing with eyes closed for a moment before the next decision.
Reflections

The Value of AI Isn't Prediction. It's Cognitive Ergonomics.

We keep debating whether AI will replace doctors. The real threat is quieter: a data environment so noisy it stops clinicians from thinking at all. A case for tools that curate the evidence rather than predict the answer.

Dr. Sven JungmannCEO

This analysis comes from the people behind Visite.

Our weekly newsletter on AI in medicine. Every Friday, rigorously checked.

By signing up you agree to receive Grand Rounds by email. Unsubscribe anytime. More in our privacy policy.

Want to see this in your hospital?

30 minutes. Your questions. Our physician-founder shows you the platform personally.

Book a demo

No commitment. No sales pitch. Physician to physician.