Document Intelligence for Healthcare: Structured Data from Lab Reports and Faxes

It's a persistent surprise to people outside public health and clinical operations just how much reporting still arrives as a fax or a scanned PDF. Reportable lab results, especially from smaller labs and clinics without electronic lab reporting (ELR) set up, routinely land as unstructured documents that a human then has to read and manually key into a case management or surveillance system. That manual step is slow, error-prone, and doesn't scale when volume spikes — exactly when you need it to.

What "document intelligence" actually means here

This isn't classic OCR that just converts an image to a blob of text. Modern document intelligence — tools like Azure Document Intelligence — combines OCR with layout understanding and field extraction: it recognizes that a document is a lab report, finds the patient name field, the test name, the result value, the reference range, and the ordering provider, and returns them as structured key-value data rather than a wall of unstructured text you'd still have to parse.

The realistic pipeline architecture

1. Intake — documents arrive via fax gateway, secure email, or upload, and land in a monitored storage location.
2. Extraction — the document intelligence service processes the file and returns structured fields with confidence scores per field.
3. Validation / trust logic — high-confidence extractions can auto-populate downstream systems; low-confidence fields route to a human review queue rather than getting silently accepted or silently dropped.
4. Export — validated data maps into HL7v2 ORU messages or FHIR DiagnosticReport resources for downstream systems, closing the loop into whatever case management or EHR platform is waiting for it.
5. Audit trail — every extraction, correction, and approval logged, which matters for both quality improvement and compliance review.

The real cost comparison

Commercial document-processing SaaS platforms typically price per document processed, often with a platform or per-seat fee layered on top of that. For an agency processing thousands of documents a month, that adds up fast and scales with volume rather than staying flat.

Building on Azure Document Intelligence directly — at published consumption rates, roughly $1.50 per 1,000 pages for prebuilt OCR up to around $30 per 1,000 pages for custom extraction models — means you're paying cloud consumption pricing for the actual processing, with no seat licenses and no platform markup. The tradeoff is that someone has to build and maintain the pipeline logic (validation rules, routing, export mapping) instead of getting it bundled into a SaaS product. For most public health agencies, that build is a matter of weeks, and it's a system you own outright afterward rather than renting indefinitely.

Where this fits into a broader modernization effort

Document intelligence isn't a standalone project — it's most valuable when it feeds directly into the same EMPI and FHIR interoperability layer handling your electronic data. A lab result extracted from a fax should resolve to the same patient identity, and land in the same downstream systems, as a result that arrived electronically. Built in isolation, a document intelligence pipeline is a nice demo. Built as part of the broader data infrastructure, it closes a real operational gap.