OSHA Recordkeeping (29 CFR 1904) AI Documentation: The Clinical Library Playbook for Occupational Health MDs
Learn how AI documentation distinguishes First Aid from Recordable Treatment under OSHA 29 CFR 1904, generating audit-ready Causality Narratives via HL7 FHIR R4.

OSHA Recordkeeping (29 CFR 1904) AI Documentation: The Clinical Library Playbook for Occupational Injury Clinics
A technical reference for the Clinical Operations Director responsible for OSHA 300 log integrity, employer liability audits, and clean occupational-injury claims.
TL;DR
The core failure remains: Ambient AI scribes generate SOAP notes but do not classify an encounter as First Aid vs. Recordable Treatment under OSHA 29 CFR 1904. That classification decision is where employer liability audits live or die.
The Scribing.io difference: We encode work-related causality directly in HL7 FHIR R4 using the US Occupational Data for Health (ODH) Implementation Guide—a Work-Relatedness Observation linked to an AdverseEvent and the employer Organization, cross-referenced against Procedure (suturing 12001–12007) and MedicationRequest (Rx vs. OTC strength).
The computed output: An algorithmic First Aid/Recordable determination, an auto-populated OSHA 301, an OSHA 300 log update, a modifier-25 NCCI guardrail, and a time-stamped, Provenance-backed Causality Narrative.
Who this serves: Clinical Operations Directors at occupational health, urgent care, and employer-contracted clinics.
On This Page
First Aid vs. Recordable: The Classification Gap No Scribe Solves
Information Gain Pillar: Encoding OSHA 1904 Causality in FHIR R4 + ODH
Scribing.io Clinical Logic: The Dehiscing Palm Laceration Case
The NCCI Trap: Modifier 25, E/M 99213, and CPT 12002
Technical Reference: ICD-10 Documentation Standards
The Provenance-Backed Causality Narrative and Employer Audits
Deployment Playbook for the Clinical Operations Director
First Aid vs. Recordable: The Classification Gap No Scribe Solves
CLINICAL UPDATE 2026: Revised for new CMS CPT G2211 standards, SB 1120 compliance, and FHIR interoperability.
Most AI documentation tools in 2026 optimize for the wrong endpoint. They convert conversation into a structured SOAP, DAP, or BIRP note and export it to the EHR. For continuity of care that is useful, but for a workplace injury clinic the note is only raw material.
The regulatory artifact that matters is the 29 CFR 1904 recordability determination. A narrative note does not make that determination, which is why Scribing.io encodes the classification as computable data rather than prose.
Under OSHA's recordkeeping rule, treatment beyond first aid is a general recording criterion that triggers a recordable event. The regulation enumerates a closed list of first aid measures at 1904.7(b)(5)(ii).
Inside the closed first-aid list: wound cleaning, non-prescription medication at non-prescription strength, simple bandaging, and tetanus immunization.
Outside the list and recordable: sutures, prescription-strength medication, restricted duty, and days away from work.
The competitor pattern in 2026—represented by ambient-note products such as HealOS—markets "audit-ready" notes. Their compliance model checks medical necessity, interventions, response, and plan against payer and state guidelines. That is E/M compliance, not OSHA 1904 recordkeeping compliance.
Ambient scribes vs. OSHA recordkeeping requirements
Documentation Requirement | Generic Ambient AI Scribe | Scribing.io Occupational Logic |
|---|---|---|
Narrative SOAP/DAP note | Yes | Yes |
Work-relatedness assertion (1904.5) | Free text only, unstructured | Structured ODH Work-Relatedness Observation |
First Aid vs. Recordable classification (1904.7) | Not performed | Algorithmic, Procedure + Rx-strength driven |
Restricted-duty / days-away counting (1904.7(b)(4)) | Not performed | Auto-computed restricted-day count |
OSHA 301 Injury & Illness Incident Report | Not generated | Auto-populated |
OSHA 300 log update | Not generated | Line item generated with classification column |
Employer-audit causality artifact | None | Provenance-backed Causality Narrative |
The gap is not transcription accuracy; it is regulatory reasoning. For adjacent specialties documenting ambient encounters, we address the same reasoning gap in our Scribing.io Ai Documentation Alma Providers Telehealth Mdm Reference, where MDM must be encoded, not narrated.
Encoding OSHA 1904 Causality in FHIR R4 + ODH
Here is the architecture that separates Clinical-Grade Scribing from every SOAP-note tool on the market. The determination is not a paragraph a clinician types.
Scribing.io encodes causality directly in HL7 FHIR R4 using the US Occupational Data for Health (ODH) Implementation Guide. The result is a set of interoperable resources that can be queried, versioned, and audited.
Work-Relatedness Observation (ODH): persists whether the injury arose out of and in the course of employment, satisfying the 1904.5 threshold.
AdverseEvent resource linkage: anchors the injury as a discrete recordable event rather than a routine visit.
Organization (employer) reference: attributes the recordable event to the correct OSHA 300 log.
Procedure resource signals: simple wound repair CPT 12001–12007 acts as a deterministic marker that the encounter crossed beyond first aid.
MedicationRequest strength discrimination: OTC at OTC strength stays first-aid; a prescription or Rx-strength OTC pushes the event into recordable territory.
Provenance wrapper on every resource: a time-stamped, attributable chain that makes the Causality Narrative defensible under audit.
Why payer-readiness is not OSHA-readiness
Ambient tools stop at "templates that keep every note payer-ready." Payer-readiness and OSHA recordability are two different data models, and conflating them is the exact exposure a Clinical Operations Director owns.
By modeling causality in FHIR resources, Ambient Clinical Intelligence makes the recordability decision computable. The system independently evaluates "Procedure present? Rx strength present? Restricted duty assigned?" rather than hoping a clinician flagged it.
Current clinical benchmarks indicate under-recording is driven far more by classification error and single-visit export gaps than by transcription error. A note-only tool structurally cannot close this exposure. The same FHIR discipline governs our dental workflows in the Scribing.io Eaglesoft Ai Documentation Dental Ambient Challenges Reference.
Scribing.io Clinical Logic: The Dehiscing Palm Laceration Case
Scenario for review. A 32-year-old warehouse worker sustains a palm laceration from a pallet strap. Track how the classification evolves across two visits.
Visit 1 (Hour 0) — First Aid
Wound irrigation, adhesive bandage, and a tetanus booster are administered. Every measure sits inside the 1904.7 first-aid list.
Scribing.io persists a Work-Relatedness Observation (ODH) and an AdverseEvent, then classifies the encounter as First Aid — Not Recordable. The event is already tracked in FHIR, so it can be re-evaluated later.
Visit 2 (Hour 48) — Dehiscence and Reclassification
The wound dehisces on return. The provider places 3 sutures (CPT 12002), prescribes cephalexin 500 mg, and assigns 3 days of restricted duty.
Medical AI Scribing detects two triggers: a suture Procedure and a prescription MedicationRequest at Rx strength. Either alone crosses the first-aid boundary; together they are unambiguous.
Reclassification to Recordable: the persisted AdverseEvent is upgraded to Medical Treatment Beyond First Aid.
Restricted-day computation: the 3-day duty restriction is counted under 1904.7(b)(4) and written to the 300 log column.
OSHA 301 auto-population: the Injury & Illness Incident Report is filled from the FHIR resources, not retyped.
Causality Narrative generation: a Provenance-backed, time-stamped account links Visit 1 and Visit 2 as one event.
The three failure points Scribing.io eliminates
Risk Without Occupational Logic | Scribing.io Correction |
|---|---|
EMR exports Visit 1 only | Persistent AdverseEvent links both visits |
Employer under-records the injury | Auto-computed Recordable status + 300 log line |
TPA flags 99213 without modifier 25 | NCCI guardrail inserted before submission |
The relevant coding anchors here include S61.411 (ICD-10-CM) for the laceration of the hand and S33.5 (ICD-10-CM) for related lumbosacral strain when lifting mechanics are documented.
The NCCI Trap: Modifier 25, E/M 99213, and CPT 12002
Visit 2 carries a coding hazard that defeats otherwise-clean documentation. When a provider bills E/M 99213 alongside procedure 12002, NCCI edits deny the E/M unless it is distinct.
Modifier 25 signals a significant, separately identifiable E/M service performed on the same day as a procedure. Omitting it triggers a bundling denial from the TPA.
The guardrail fires automatically: Scribing.io detects the 99213 + 12002 pairing and prompts for modifier 25 documentation before export.
It requires substantiation, not a checkbox: the E/M must reflect the reassessment of dehiscence distinct from the suture repair itself.
It prevents downstream rework: a clean first submission protects the occupational claim timeline and the employer relationship.
For the financial case behind this, the denial-avoidance math is modeled in our AI Medical Scribe ROI Calculator.
Technical Reference: ICD-10 Documentation Standards
Occupational injury coding demands laterality, encounter type, and external-cause specificity. Vague codes weaken both the claim and the causality record.
Laceration coding must specify laterality: use S61.411 (ICD-10-CM) with the correct 7th-character encounter extension.
Concurrent strain injuries need capture: S33.5 (ICD-10-CM) when lumbosacral mechanics are involved in the lift.
External-cause codes reinforce causality: the pallet-strap mechanism strengthens the ODH Work-Relatedness Observation.
The clinical detail that competitors miss is that encounter-type extensions (initial vs. subsequent) must remain consistent across both visits, or the two-visit linkage fractures during audit.
The Provenance-Backed Causality Narrative and Employer Audits
Employer liability audits examine causality, not prose quality. The auditor asks whether the injury was work-related, whether treatment exceeded first aid, and whether the 300 log reflects reality.
The Causality Narrative answers all three from structured data. Because it is Provenance-wrapped, every assertion carries a timestamp and an attributable source resource.
Work-relatedness is asserted, not implied: the ODH Observation ties the injury to the employer Organization.
The First Aid to Recordable transition is dated: the narrative shows exactly when and why reclassification occurred.
Restricted days are enumerated: the 3-day count maps directly to the 300 log column.
The artifact survives scrutiny: it passes employer audit because it is reconstructed from resources, not memory.
This is the difference between defending an under-recording finding and never receiving one. The narrative is the durable record a Clinical Operations Director hands to counsel.
Deployment Playbook for the Clinical Operations Director
Deployment succeeds when classification logic is treated as a governance function, not a documentation feature. Sequence the rollout deliberately.
Map your employer contracts first: confirm each employer Organization resource is registered before go-live.
Validate the first-aid closed list: confirm the classification engine reflects the exact 1904.7(b)(5)(ii) enumeration.
Wire the NCCI guardrails: enable the modifier-25 prompt on all procedure + E/M pairings.
Reconcile the 300 log monthly: compare auto-generated line items against manual review during the first quarter.
Audit the Provenance chain quarterly: confirm every Recordable event carries a complete Causality Narrative.
What to measure in the first 90 days
Metric | Target Signal |
|---|---|
Classification override rate | Declining as clinicians trust the engine |
Single-visit export gaps | Zero linked-event breaks |
Modifier-25 denials | Trending toward zero |
300 log reconciliation variance | Under one line item per month |
To scope pricing against clinic volume, review Scribing.io Pricing & Plans and align tiers to your occupational encounter count.
For specialty-specific configurations, the occupational injury workflow lives within the broader Scribing.io specialties library and integrates with your existing EHR export path.


