Turn incoming paperwork into usable context.
Prepare reviewable information from reports, referrals and discharge documents while keeping the source close.
Explore the story
Medrella roadmap. This experience is being planned; scope and readiness will be demonstrated before implementation.
What this workflow connects.
The scope below guides discovery and demonstration. This capability is on the Medrella roadmap.
- 01
Referral, report and discharge-document intake
- 02
Draft field extraction linked to the source
- 03
Uncertainty flags and patient-matching review
- 04
Approval before information enters the clinical record
Useful information arrives in uneven formats
Patients bring scanned reports, photographs, referral letters and discharge summaries produced by different organisations. The information may be relevant, but locating it during a consultation can take time.
Medrella's planned document intelligence aims to organise that material and prepare structured drafts for staff review. The purpose is to make source information easier to use, not to treat an extracted field as independently verified clinical truth. A document can be incomplete, outdated or associated with the wrong person. Those possibilities remain important even when the model produces a clean summary.
Preserve provenance alongside extracted fields
Each document should retain its origin, date and relationship to the patient record.
A reviewer needs to know whether a value was entered by staff, received through an interface or extracted from an uploaded file. The source should remain accessible from the proposed structured information. Units, dates and reference context matter, especially when reports from different organisations are compared. The assistant should flag unreadable or uncertain content rather than silently filling the gap. Patient matching and document ownership require explicit verification; a name detected on a page is not sufficient identity evidence.
Organise for the next clinical question
A useful summary is shaped by the encounter and the person's task. A clinician preparing for a follow-up may need a concise timeline, while a records team may need document type and date for filing.
Those views should be generated from the same retained evidence without erasing detail. The system should avoid presenting an AI interpretation as the issuing laboratory's conclusion. It should also distinguish a historical observation from a current finding. The intended experience helps people find relevant material and inspect it, rather than replacing the clinical review with a generated paragraph.
Keep the write-back boundary explicit
Extracting a medicine name or diagnosis from a document should not automatically change the current medication list or problem record. Staff need to decide whether the information is relevant, current and appropriately attributed.
The proposed workflow should show the changes before they are applied and retain a record of approval. Duplicate uploads and corrected documents need clear handling. Access permissions and retention should follow the purpose of the source material. These requirements are part of the integration design, because a useful standalone extractor can still create confusion when connected directly to authoritative clinical records.
Test with the documents your team actually receives
A pilot should use de-identified or synthetic examples representing low-quality scans, multiple pages, handwritten annotations and conflicting dates. Review extraction errors and the effort needed to correct them.
Check whether users can reliably return to the source and distinguish an approved field from a draft. Measure time spent locating and organising information alongside accuracy and correction burden. Medrella's document intelligence remains planned. The scope should identify supported document types and the human review process before any claim is made about automated capture of the hospital's incoming paperwork.
Connected thinking.
What would more time
for care make possible?
Let's explore it together