Automating Medical Record Review in Workers’ Comp Claims

Medical records are the heaviest part of a comp file and the most time-consuming to review. They arrive as one enormous scanned PDF — hundreds of visits, multiple providers, out of order, often duplicated. Automating this review is less about "reading" and more about turning that dump back into structured, addressable information.

Step one: split the bundle

Before anything can be extracted, a scanned bundle has to be segmented into its actual documents. A reliable pipeline detects document boundaries — a new letterhead, a form header, a caption block — and reassembles page ranges into discrete records instead of one 3,000-page blob.

Step two: extract the facts that drive the claim

From each record the system pulls the fields a defense reviewer would flag by hand:

• Provider and specialty, and the date of each visit.
• Body parts treated and the mechanism of injury described.
• Work status — full duty, modified, off work — and any restrictions.
• Diagnoses, treatment, and any impairment ratings.

Step three: build a chronology you can verify

The output that matters is a timeline: every visit, in order, each entry linked to the page it came from. That deep link is what separates a usable tool from a liability — the reviewer confirms a fact by clicking straight to its source, not by re-reading the file.

Accuracy and the human gate

OCR on decades-old faxed records is imperfect, and no extraction is 100% correct. The discipline that makes automation safe is grounding: every extracted value carries its citation, edits are audited, and a person signs off before the chronology becomes part of the case. Automation compresses the review; it does not remove the reviewer.