Skip to main content
Possible SetupHealthcare

Discharge Letter Automation

A possible setup for automated discharge letter creation: HIS integration, AI-assisted draft, mandatory physician approval, and dispatch via the German telematics infrastructure (KIM). Target: roughly 5 instead of 30–45 minutes per letter.

Time Saved
-89%
Industry
Healthcare
Implementation
10–12 weeks

At a glance

It starts with
Discharge Triggered
The machine handles
13 of 15 steps
A person keeps
Doctor Review · QES Signature
Bottom line
-89%Time Saved

The problem

Real talk: nobody likes dictating discharge letters. You spend the day seeing patients, evening rolls in, the stack is still untouched, and the medical review board is asking again.

Twenty minutes per letter on average. A clinic with thirty discharges a day means ten hours of dictation time — every day, spread across doctors who should be seeing patients instead.

The uncomfortable part: letters go out two weeks late, referring colleagues call, the patient waits for rehab — all because there's no pipeline between dictation and dispatch.

Here's the pipeline.

Discharge letters are critical for continuity of care, but the manual process is extremely time-consuming: 30–45 minutes per letter, copy-paste between 5–7 systems (HIS, LIS, RIS, pharmacy, OR documentation). With 100+ discharges daily, that's 50–75 physician hours per day. A significant share of letters miss the typical 24-hour deadline for dispatch to the general practitioner. Quality issues from missing information and inconsistent wording endanger patient safety.

How the process runs — step by step

Scroll through. The diagram stays put and highlights the step you are on.

The workflow01 / 15
  1. 01

    Discharge Triggered

    HIS Trigger

    A discharge is documented in the HIS and automatically triggers the letter creation process.

  2. 02

    Data Collection

    FHIR/HL7

    The system collects all relevant patient data from the HIS via FHIR R4 and HL7 v2 interfaces.

  3. 03

    Parallel

    Data collection is split into parallel paths for optimal processing speed.

  4. 04

    Diagnoses

    ICD-10-GM

    Diagnoses are extracted with ICD-10-GM codes from the system and prepared for the letter.

  5. 05

    Medication

    ATC Codes

    Medication data with ATC codes is compiled from the patient's home medication and inpatient prescriptions.

  6. 06

    Join

    All data collected in parallel is merged into a complete dataset.

  7. 07

    AI Structuring

    LLM (local or EU cloud)

    The AI structures all data according to department-specific templates and generates medically accurate wording.

  8. 08

    Draft Generated

    Template-based

    A complete letter draft is generated, including all relevant sections and recommendations.

  9. 09Human decides

    Doctor Review

    2-3 Min

    The responsible physician reviews the draft and makes adjustments if needed – typically in 2-3 minutes.

  10. 10Branch

    Approved?

    Gateway: Was the letter approved by the physician or are corrections required?

  11. 11

    Corrections

    Adjustments

    If changes are requested, corrections are incorporated and the letter is resubmitted for review.

  12. 12Human decides

    QES Signature

    D-Trust

    The approved letter is signed with a qualified electronic signature (QES) via D-Trust, making it legally valid.

  13. 13

    KIM Dispatch

    To GP

    The signed letter is automatically sent via KIM (gematik) E2E encrypted to the general practitioner.

  14. 14

    Archiving

    Patient Record

    The letter is archived in the electronic patient record and stored for future reference.

  15. 15

    Complete

    Letter Delivered

    The process is complete. The GP has received the signed letter, delivery is confirmed.

Our Solution

Automated data aggregation from all relevant systems via FHIR R4 and HL7 v2 (SAP IS-H, Orbis, iMedOne, Medico). AI-assisted draft using department-specific templates – either with locally hosted open-source models (Llama, Mistral) or via DPA-compliant EU endpoints of commercial LLMs. The physician reviews and approves in roughly 2–3 instead of 30–45 minutes. The signed letter is then dispatched to the general practitioner via the telematics infrastructure (KIM) with end-to-end encryption and delivery confirmation, after a qualified electronic signature (QES, eIDAS-compliant) via D-Trust. Physician approval is always required – the AI does not replace clinical judgment.

Deep HIS Integration
FHIR R4/HL7 v2 connection to all common German HIS: SAP IS-H, Orbis, iMedOne, Medico. Automatic aggregation of ICD-10-GM, OPS, medication, and lab data.
AI-Assisted Letter Draft
Modern LLMs (locally hosted open-source models such as Llama/Mistral, or commercial models such as Claude or GPT) with department-specific templates. Consistent structure and clinically appropriate wording – final approval is always done by the responsible physician.
QES-Compliant Signature
D-Trust integration for qualified electronic signatures. Legally valid under eIDAS, fully integrated into the workflow.
Automatic KIM Dispatch
Direct dispatch via gematik KIM to the general practitioner. E2E encrypted, with delivery confirmation, fully automatic address resolution.

What comes out of it

Possible setup, not a packaged product

The figures shown are target values and expected magnitudes for a possible setup – based on industry benchmarks, public studies of comparable setups, and our own tests on a real stack. They are not measured outcomes from a specific customer project; actual results depend on company size, process maturity, and integration depth. We do not offer this setup as a packaged product. We help teams design, automate, and run such processes themselves – through architecture consulting, workshops, and implementation support with n8n. For regulated third-party systems with certification or license requirements (e.g. HIS, gematik, DATEV-certified), we partner with specialized providers.

~5 Min
Time per Letter (target)
~-89%
Time Saved (target)
150+
Letters/Day (with approval)
≥ 95%
On-time Rate (target)

Expected magnitude: roughly 89% less time per letter, materially improved on-time rate, 150+ letters per day with physician approval

Before vs. After

Time per Letter
Before
30–45 minutes
After
~5 minutes (with physician approval)
Data Sources
Before
Manual copy-paste from 5–7 systems
After
Automatic aggregation via FHIR/HL7
24h On-Time Rate
Before
Often below 70%
After
Target ≥ 95%
Signature
Before
Handwritten or without qualified signature
After
QES (legally valid under eIDAS)
Dispatch to GP
Before
Fax, mail, or unencrypted email
After
KIM (E2E encrypted, with delivery confirmation)

Technical facts

Technology Stack

n8nFHIR R4HL7 v2SAP IS-H APIClaude/GPT-4D-Trust QESKIM (gematik)PostgreSQL

Integrations

Seamless connection to your existing infrastructure

SAP IS-H

HIS

Complete integration via SAP interfaces for patient data, diagnoses, and medication

Orbis (Dedalus)

HIS

Native FHIR R4 connection for real-time data retrieval

D-Trust

Signature

Qualified electronic signature for legally valid documents

KIM (gematik)

Communication

Secure dispatch via telematics infrastructure to general practitioners

Security & Compliance

Enterprise-ready with highest security standards

GDPR Compliant

Full compliance with EU General Data Protection Regulation. Data processing in Germany.

ISO 27001

Data processing in ISO 27001 certified data centers.

KIM Encryption

End-to-end encryption via gematik's telematics infrastructure.

On-Premise Option

Optional operation in your own infrastructure without external data transfer.

Frequently Asked Questions

We support all major German HIS: SAP IS-H, Orbis (Dedalus), iMedOne (Telekom Healthcare), Medico (CompuGroup), and more. Connection is via standardized FHIR R4 and HL7 v2 interfaces.
Yes. The AI creates a draft that is reviewed and approved by the responsible physician. The legally valid QES signature (D-Trust) complies with eIDAS standards. The physician remains responsible for the content.
Implementation typically takes 10-12 weeks, depending on the complexity of your IT landscape. This includes: HIS connection, template configuration, testing phase, and training.
We recommend two operating models: (1) On-prem with locally hosted open-source models (e.g., Llama, Mistral) – patient data never leaves your infrastructure. (2) Cloud LLM (Claude, GPT) via DPA-compliant EU endpoints, with pseudonymisation prior to transmission. The right choice depends on a DPIA conducted with your data protection officer.
The solution is designed as a decision-support system – physician approval of every letter is mandatory. The AI does not make autonomous medical decisions. A concrete classification under the EU AI Act (transparency and high-risk requirements) and, where applicable, the MDR is established during the pilot together with your compliance team.
This page describes a possible setup, not a completed customer project. The figures shown are target values and expected magnitudes drawn from industry benchmarks and comparable setups – not measured outcomes from a specific hospital.
No. We do not build a packaged connector between your software and Orbis/SAP IS-H/iMedOne that we run and support long-term. What we deliver: architecture consulting, team workshops, and implementation support with n8n – so you (or a specialized HL7/HIS integrator) can build it cleanly. For regulated interfaces requiring a Dedalus connector license, gematik certification, or MDR-relevant components, we recommend specialized partners (e.g. ID Berlin, ICW, Health-Comm, NEXUS).