Scanned PDF Audiograms vs. Structured Data: Why a Filing Cabinet of PDFs Still Fails STS Detection

Somewhere along the way, most hearing conservation programs had a scanning project: the paper audiograms went through a copier, a folder of PDFs landed on a shared drive, and the program was declared “digital.” It wasn’t. A folder of scanned audiograms is a filing cabinet with a search bar — the documents are preserved, but the thresholds inside them are pixels, not data. Standard Threshold Shift detection under OSHA 29 CFR 1910.95(g)(10) is arithmetic performed on threshold values against a baseline, and arithmetic cannot run on a picture. This post explains precisely where the PDF archive fails, what structured audiometric data looks like, and how to convert one into the other without corrupting three decades of clinical history.

What STS Detection Actually Needs

An STS determination is a defined computation: for each ear, take the worker’s current thresholds at 2000, 3000, and 4000 Hz, average the change from the baseline thresholds at those frequencies, optionally apply the age corrections in Appendix F, and flag an average shift of 10 dB or more. Three inputs are non-negotiable:

• The baseline thresholds as numbers — for that specific worker, at those specific frequencies (why the baseline and not last year’s test, see Baseline vs. Annual Audiogram).
• The current thresholds as numbers — same worker, same frequencies.
• The worker’s identity and age — to pair the right records and apply the right correction row.

Every one of those inputs is on a scanned audiogram and available from none of them. That is the entire problem in one sentence.

Where the PDF Archive Fails, Mechanically

• No computation: thresholds locked in images cannot feed the STS formula, so shift detection for legacy history is manual — open, read, key, compare — or skipped.
• No trend visibility: a worker drifting 3 dB a year never crosses a single-year alarm but is 12 dB down over four years. Catching that requires the series as data; a folder of scans hides it perfectly.
• No validation: transcription errors, impossible values, and duplicate scans sit undetected because nothing machine-readable exists to check.
• Fragile retrieval: finding one worker’s complete series depends on file-naming discipline applied consistently for 20 years — which no organization has. Access requests under 1910.95(m)(4) become folder archaeology.
• Fragile custody: shared drives get migrated, reorganized, and permission-locked; PDFs on unmanaged storage fail the 30-year horizon the same way paper does, just more quietly.

None of this means the scanning project was wasted — preservation matters, especially for fading thermal-print originals that might otherwise be illegible by now. Scanning is step one. It is just not the last step.

Not All PDFs Are Equal: Text-Based vs. Image-Based

Before planning a conversion, sort the archive into two piles:

• Text-based PDFs — generated digitally, typically by a vendor’s reporting system. The text is selectable, and thresholds can sometimes be parsed programmatically. Layouts vary enough between vendors and report versions that parsing still needs verification, but the yield is high. If a former vendor produced these, it also means structured data existed upstream — and requesting the actual export beats parsing its printout every time (see How to Recover Audiometric Records From a Previous Testing Vendor).
• Image-based PDFs — scans or photos of paper. No text layer at all: extraction means OCR with human verification, or straight transcription. Most legacy archives are mostly this pile.

What Structured Audiometric Data Looks Like

The target format mirrors the record content 1910.95(m)(2) already requires — one record per test, with discrete fields:

Converting a PDF Archive: The Three Realistic Paths

• Trained transcription: humans key thresholds from each scan into the structured format, with double-entry or second-person review on every record. Slow but dependable; appropriate for small archives.
• Assisted extraction: OCR proposes values, a human verifies every threshold against the image before acceptance. Faster on legible scans; verification is not optional — a 25 misread as 75 (or a right ear swapped with a left) silently corrupts every downstream STS calculation for that worker.
• Provider conversion: a service ingests the archive in whatever mix of formats exists — image PDFs, text PDFs, paper, spreadsheets — and returns validated, per-employee structured records. This is typically the right answer at multi-hundred-worker scale, and it is exactly what Soundtrace’s historical records audit does at no cost.

Whichever path you take, run the validation and per-employee indexing discipline described in the complete historical digitization guide: range and date checks, duplicate flagging, baseline identification, and assembly into one continuous series per worker.

Keep Both: Source Documents and Working Data

Conversion does not retire the PDFs. The scan (or paper original) remains the evidentiary source document — the signed artifact you produce in a claim or inspection — while the structured record is the working data. Link every structured test record to its source image so the chain from computation back to original is one click, and land both in managed custody: backed up, retention-controlled, and indexed by employee. That pairing — not either half alone — is what survives the 30-year horizon and the 15-working-day access clock (on the latter, see Meeting OSHA’s 15-Working-Day Access Requirement).

Frequently Asked Questions