AI Documentation in Rehab: Between Order, Extension, and Discharge Report

Thursday afternoon in a neurological rehab clinic. On the ward physician's desk lie three extension requests, two discharge reports with deadlines drawing closer, and a findings report for a reduced-earning-capacity pension procedure. All six documents concern patients whose story has long been documented — in the referring colleague's order form, in the application to the payer, in the admission findings, in the progress notes, in the reports from physical, occupational, and speech therapy.

Yet each of these documents begins the same way: open the record, search, retype. The social history is in the admission findings, the medication in the chart, the documented Barthel index in the nursing report. None of it is lost; it is simply not stored anywhere in a way that a new document could take it over. The physician will finish two of the six documents today, the rest on Friday after rounds — or at home over the weekend.

And meanwhile, at the front of the building, tomorrow's pile is growing. A single rehab site whose figures we know processes 23,472 inquiries per year — roughly 94 per working day, of which about one in nine becomes an admission. Each of these inquiries is a document package, often a fax bundle of 5 to 40 pages, frequently incomplete.

A chain of documents that do not know each other

Rehabilitation has a document world of its own, and it is longer than that of the acute hospital. At the start stands the order — in the statutory health insurance sector, form Muster 61 [1] —, then the application to the payer, admission, after a few weeks the extension request, at the end the discharge report. For DRV cases (DRV: German statutory pension insurance, which funds rehab), the discharge report has a prescribed structure; since September 2024 it has been transmitted via a uniform electronic channel, and it feeds into the pension insurer's quality assurance [2].

Every link in this chain asks for the same things: diagnoses, course, functional findings, social history, rehab goals, medication. And every link has an addressee who reviews. A gap nobody notices at admission becomes a query on the extension request four weeks later — and, in the end, a weakness of the discharge report.

The tools offered to clinics for this mostly come from the acute-care world. Hospital information systems and most AI documentation tools are built around the acute Arztbrief (physician letter); extension requests, DRV form numbers, and reduced-earning-capacity assessments rarely appear in them. Rehab clinics therefore have a wide choice of tools that solve the wrong problem.

The lever: a record that knows its origins

Vendor-neutrally, the usefulness of AI documentation in rehab is decided by one question: does the record know where each entry comes from?

When every field — every diagnosis, every date, every functional finding — is linked to its source passage, the nature of the downstream documents changes. The extension request becomes an evidenced excerpt from the record; the discharge report collects what was documented over weeks instead of reconstructing it on the last day. The side effect is defensible documentation: if a payer asks, the record shows with a click which document, of which date, an entry comes from.

Three conditions belong to this. Incoming documents — faxes, physician letters, prior findings — must be captured in structured form and verified against the existing record, otherwise the pipeline silently adopts errors. The rehab document world must exist as a class of its own, with its forms, outlines, and reviewing bodies. And the start of the chain, the inquiry inbox, needs a system that makes incompleteness visible before the patient is in the building — how that works in detail is described in our article on the mechanics of the rehab bottleneck (in German).

What runs today — and what is planned

aiomics is in production in German rehab clinics. The admission board runs in production operation at one site of a large German hospital group; at a second site, the Falldialog board (Falldialog: the pre-review case dialogue with payers) is in use for payer procedures.

The admission board captures inquiries in structured form — including the 40-page fax bundle — and shows for each card which documents are present and which are missing. "Complete" always means: the document is present. Whether its content suffices remains a physician's judgment. Cases with requested documents outstanding disappear from the board and reappear on their own as soon as documents arrive, with a note on what is new; accepted cases come back into view three working days before the confirmed admission date, public holidays included. That sounds unspectacular. It replaces the follow-up folder that is the real management software in many admission offices.

Sorting follows administrative criteria — receipt date, deadline, document presence. There is deliberately no field for clinical urgency in the data model; the medical assessment stays with the physicians who read the documents. And so that a configurable board does not grow wild, adaptability is tiered: the workflow stages themselves are locked and change only through the aiomics release process, labels and lists can be adjusted by the facility's administrators, and mandatory flags only with countersignature by the QM officer. Every configuration change is logged with a rationale and can be reverted with one click.

Beneath it lies the verified record, also in production: incoming documents are extracted and checked against the existing record; every field carries a four-level confidence label — from "verified" through "single source" and "patient-reported" to "AI-extracted" — and shows its source passage on click.

On this record works the draft generation, in production as a first draft stage: a competent first draft that physicians review and take responsibility for. Behind it stands a curated library of around 50 document types and 80 regulatory overlays; for rehab, this includes DRV forms such as S0051, G0260, and G2220 as well as a neurology specialty package that also covers the findings report in reduced-earning-capacity procedures. Every template names its sources and its validation date — the underlying catalogs change annually, and the library is maintained accordingly.

Planned, and expressly labeled as planned here: the validated form pipeline, which writes drafts directly into the facility's Word letter template, including Muster 61 and the extension request; and the transmission preparation, which deterministically checks data sets before dispatch — including the uniform DRV channel (in development). Transmission itself remains in every case with the hospital information system or the facility's established transmission route; aiomics does not send to payers.

That the pension insurance has operated a uniform electronic route since 2024 makes rehab rewarding terrain: one standard, one format — and considerably less vendor attention than in the acute sector. The rehab patient survey also belongs in this picture: the central DRV survey delivers facilities their results late, aggregated, and only for DRV cases — the in-house view remains a task of its own. The aiomics survey suite is in pilot operation at rehab clinics for this; the expansion is planned.

Two further pieces (both in German): rehab application management: the documentation gaps between DRV and statutory health insurance and indication review in rehab with structured input.

What to measure any vendor against

1. Does the system run in production in a German rehab clinic today? Ask for the production workflow rather than the demo — and since when it has been running.
2. Does it know the rehab document world? Have them show you how an extension request comes into being and which DRV forms are on file. You can recognize acute-care tools by the fact that at this point they start talking about the Arztbrief.
3. Does every adopted field show its source? In the meeting, click on a diagnosis in the draft and check whether the source passage appears — and how fast. A provenance display that first triggers a search will not be used in ward routine.
4. What happens with incomplete inquiries? Ask how the system recognizes subsequently submitted documents and resubmits cases on its own.
5. Who transmits to the pension insurance and the health insurers? The sound answer is: your hospital information system or your established transmission route.

If you want to see what this looks like in the daily routine of a rehab clinic, write to us — we will show you the production deployment. Or subscribe to Visite, our weekly briefing on documentation and data quality in German healthcare — an English edition, Grand Rounds, has a waitlist.

Sources

1. Kassenärztliche Bundesvereinigung. Vordruckvereinbarung (Anlage 2 zum Bundesmantelvertrag-Ärzte), Muster 61: Verordnung von medizinischer Rehabilitation. https://www.kbv.de/
2. Deutsche Rentenversicherung. Der ärztliche Reha-Entlassungsbericht: Anforderungen und elektronische Übermittlung. https://www.deutsche-rentenversicherung.de/