Captioning at hospital-system scale: HealthStream, Relias, and Epic My Learning caption workflows for HIPAA workforce training, CMS Conditions of Participation, and Joint Commission mandatory education

Why hospital-system L&D is a distinct compliance environment

Corporate L&D compliance consists primarily of the employment law and industry regulation that applies to the employer — OSHA for manufacturing and construction, FINRA for broker-dealers, FDA regulations for pharmaceutical manufacturers. Hospital L&D compliance is different in three structural ways that create a more demanding operating environment than most other sectors.

The CMS participation condition

Every hospital that receives Medicare or Medicaid reimbursement — which includes virtually all hospitals in the United States — operates under CMS Conditions of Participation (42 CFR Part 482 for acute care hospitals). The Conditions of Participation specify requirements for hospital operations, staffing, clinical services, and patient safety — and embedded throughout those conditions are training mandates that L&D teams must satisfy to maintain certification. These are not aspirational guidelines; hospitals that fail CoP compliance risk decertification, which terminates Medicare and Medicaid participation. For most hospitals, that outcome is existential. The training mandates embedded in CoP are therefore not treated the way OSHA training mandates are in manufacturing — as important but occasionally deferred. They are treated as hard deadlines with institutional survival as the backstop.

The Joint Commission accreditation layer

Most hospital-based organisations — acute care hospitals, critical access hospitals, ambulatory surgical centres, behavioural health facilities — seek Joint Commission accreditation, which is a prerequisite for Medicare participation for most facility types. Joint Commission accreditation requires compliance with the Comprehensive Accreditation Manual for Hospitals (CAMH) and is evaluated through triennial on-site surveys that use tracer methodology: surveyors select patients from the census and follow the care documentation through the organisation, examining every system the patient touched and every staff member involved in that patient’s care. Training records for every staff member who touched the patient are reviewed. The Joint Commission’s National Patient Safety Goals (NPSGs) impose additional training mandates in areas including infection prevention, fall reduction, suicide risk, and patient identification. Staff who cannot demonstrate training compliance in these areas during a tracer review create findings that must be resolved before accreditation is restored.

The HIPAA workforce training mandate

HIPAA Security Rule 45 CFR 164.308(a)(5)(ii)(A) requires covered entities to “implement a security awareness and training program for all members of its workforce (including management).” There is no minimum frequency specified in the regulation, but OCR enforcement guidance, state privacy law, and standard-of-care evidence in litigation consistently support annual refresher training as the minimum defensible cadence. Privacy Rule training under 45 CFR 164.530(b) is similarly mandatory. Hospitals that deliver HIPAA training as video content — which is the vast majority of mid-to-large-sized facilities, because classroom delivery at hospital scale requires scheduling across multiple shifts in a 24/7 operation — must deliver that video content with accessible captions to comply with ADA Title I and Section 504 simultaneously with the HIPAA training mandate.

The accessibility mandate applies alongside all of the above

The three compliance layers above create the mandatory training content that hospital L&D must produce and deliver. They say nothing about how that content must be made accessible to employees with disabilities. That obligation comes from ADA Title I (42 U.S.C. § 12112), which prohibits employment discrimination on the basis of disability, and which the EEOC has consistently interpreted to require that training programmes be equally accessible to employees with disabilities. The implementing regulations at 29 CFR Part 1630 require reasonable accommodation in training, which for a deaf or hard-of-hearing employee means accurate captioned training video. Section 504 of the Rehabilitation Act applies directly to every hospital that receives federal financial assistance, which includes Medicare and Medicaid reimbursement — meaning every participating hospital. Section 504 imposes a programme-wide accessibility obligation under 29 U.S.C. § 794 and implementing regulations at 45 CFR Part 84. The WCAG 2.1 AA standard (specifically Success Criterion 1.2.2 — Captions Pre-Recorded, Level AA) provides the technical specification that courts and OCR have recognised as the applicable accuracy benchmark: verbatim, synchronised captions meeting a 99% accuracy threshold as measured by the DCMP Captioning Key formula.

None of the clinical compliance mandates create a waiver or deferral of the accessibility obligation. A hospital that correctly deploys HIPAA training for all 800 employees but delivers that training as uncaptioned video has simultaneously satisfied its HIPAA workforce training obligation and violated its ADA and Section 504 obligations. These two things are not in tension — they are simply independent requirements that both apply.

The hospital captioning vocabulary accuracy problem

The existing analysis of drug name caption failures in medical training video focuses on pharmacology vocabulary — generic drug names, brand names, dosage forms, routes of administration, drug interaction terminology — in the context of clinical education. That vocabulary set is real and the failures are significant, but it represents only one category of the hospital workforce training vocabulary accuracy problem. Hospital mandatory training covers vocabulary across at least six distinct domains that general-purpose ASR engines have not been trained to handle accurately.

Infection control and prevention vocabulary

Infection prevention training is required by both CMS CoP (42 CFR 482.42) and Joint Commission NPSG 07 (infection prevention). The vocabulary in infection prevention training is highly specific and generates consistent ASR failures:

• Pathogen names: MRSA (methicillin-resistant Staphylococcus aureus) → “marsa” or “marshal.” C. diff (Clostridioides difficile) → “see diff” or “city.” CAUTI (catheter-associated urinary tract infection) → “caught-ee” or “cotty.” CLABSI (central line-associated bloodstream infection) → “CLASBI” or “clap-see.” VAP (ventilator-associated pneumonia) → “vap” occasionally survives but “VAP bundle compliance” generates “gap bundle compliance.” VRE (vancomycin-resistant Enterococcus) → “very” or omitted.
• Procedure terminology: PPE donning and doffing → “PPE donning and offing” (substitution of “offing” is a consistent failure). Contact precautions → survives. Droplet precautions → “drop-let precautions” (timing split). Airborne infection isolation → survives but “AII room” → “AI room.” Terminal cleaning → “terminal cleaning” survives. Sterile technique → survives. Hand hygiene → survives.
• Surveillance vocabulary: HAI (hospital-acquired infection) rates, surveillance culture, antibiotic stewardship, carbapenem-resistant organism (CRO), extended-spectrum beta-lactamase (ESBL) → “E-S-B-L” as spoken letters typically survives letter-by-letter but “ESBL-producing Klebsiella” generates “ESBL-producing Kiebsiella” or similar.

Patient safety and care delivery vocabulary

Joint Commission NPSGs and hospital patient safety programme training generate vocabulary failures that directly undermine the instructional intent of the training:

• SBAR communication: “SBAR” (Situation, Background, Assessment, Recommendation) → “S-bar” or “esbar” or “as bar.” In a patient handoff training video, a sentence like “use the SBAR framework when transferring a patient to the ICU” becomes “use the esbar framework when transferring a patient to the ICU” — the training term for a mandated communication protocol is unrecognisable.
• Code status and escalation: Rapid response team (RRT) → “R-R-T” survives letter-by-letter but “RRT activation criteria” → “our-our-tee activation criteria.” Code blue, code stroke, code STEMI → the “code X” construction generally survives. ACLS (advanced cardiac life support) → “ACLS” survives but “ACLS provider” → “ACLS provider” survives. DNR (do not resuscitate) → survives but “DNR/DNI” (do not intubate) → “D-N-R/D-N-I” with variable handling of the slash.
• Assessment scales: Braden scale (pressure injury risk) → “Bradden scale” or “BRAEDEN scale.” Morse Fall Scale → “Morse fall scale” generally survives. Glasgow Coma Scale → survives. NEWS2 (National Early Warning Score) → “news two” survives. PHQ-9 (depression screening) → “P-H-Q-9” or “phq nine.”
• Sepsis bundle vocabulary: qSOFA (quick Sequential Organ Failure Assessment) → “cue-sofa” or “Q-sofa.” SEP-1 (CMS sepsis care measure) → “sep one” or “step one.” Lactate clearance → survives. Broad-spectrum antibiotics → survives. Blood culture before antibiotics → survives.

Clinical documentation and EHR vocabulary

Most hospital workforce training programmes include substantial EHR navigation training, particularly when a new Epic, Cerner, or Meditech version deploys. EHR training vocabulary fails in ASR because it combines proprietary system names with clinical workflow terms:

• Epic-specific vocabulary: Epic Hyperdrive → “epic hyperdrive” survives as common words but “Hyperspace” (Epic’s prior client name) → “hyper space” or “hyperspace” (spacing variable). SmartPhrase (Epic documentation tool) → “smart phrase” survives. Dot phrase → “dot phrase” survives. In Basket → “inbox” (substitution to the familiar concept). MyChart → “my chart” survives. Storyboard (Epic patient summary sidebar) → “story board” or “storyboard”. BestPractice Advisory (Epic alert) → “best practice advisory” generally survives. LDA (Line, Drain, and Airway in Epic) → “L-D-A” or “LDAP.”
• Cerner-specific vocabulary: PowerChart → “power chart” survives. FirstNet (Cerner ED module) → “first net” survives. Dynamic Documentation → survives. Dragon Medical One integration → “Dragon Medical One” survives.
• Documentation process vocabulary: Meaningful Use → survives. CPOE (Computerised Physician Order Entry) → “C-P-O-E” or “see poe.” MAR (Medication Administration Record) → “mar” survives as a word but “MAR reconciliation” → inconsistent. eMAR → “E-MAR” or “E mar.”

Credentialing and certification vocabulary

Hospital clinical staff are required to maintain current certifications — BLS, ACLS, PALS, TNCC, NRP, and specialty-specific certifications. Training for these certifications, and onboarding content that walks new staff through credentialing requirements, generates vocabulary failures:

• BLS (Basic Life Support) → “B-L-S” survives. ACLS (Advanced Cardiac Life Support) → “A-C-L-S” survives. PALS (Paediatric Advanced Life Support) → “P-A-L-S” or “pals.”
• TNCC (Trauma Nursing Core Course) → “T-N-C-C” or “T-N-double-C.” NRP (Neonatal Resuscitation Programme) → “N-R-P” survives. ENPC (Emergency Nursing Paediatric Course) → “E-N-P-C” or “ENP-C.”
• Skills validation → survives. Competency assessment → survives. Annual skills fair → survives. Peer review → survives.

Accuracy benchmarks for hospital training content

Measured against a reference transcript prepared before submission — the methodology required by DCMP Captioning Key and validated in the QA methodology post — hospital workforce training content shows the following baseline ASR accuracy ranges without a custom clinical vocabulary glossary:

The accuracy gap for technical hospital content (infection control, patient safety, EHR) is wider than for HIPAA security training because infection control and patient safety vocabulary is more concentrated in acronyms and multi-word clinical terms. A 74% baseline on infection control training content means an average of 22 errors per 10-minute module — substantially above the 13-error limit that WCAG 2.1 AA permits at 99% accuracy on a 1,380-word 10-minute module. The LMS native auto-caption accuracy comparison shows that no major LMS reaches the 99% threshold for technical content without external glossary support.

HealthStream caption workflow

HealthStream is the dominant clinical education LMS by installed base in US hospitals, with approximately 4,800 health system clients. Its captioning architecture distinguishes between three content types that require different caption approaches: HealthStream-produced content from its managed content library, hospital-uploaded custom content, and integrated content from third-party publishers.

HealthStream content library (HLC content)

Content produced by HealthStream and delivered through the HealthStream Learning Centre (HLC) content library already includes captions. Most HealthStream-produced content has been captioned as part of its production workflow, but caption quality varies significantly by content age and content type:

• Post-2020 content: Generally captioned to 99% WCAG 2.1 AA standard. HealthStream updated its content production standards following ADA Title II enforcement guidance. Spot-check a sample by downloading the SRT file if available via the content management portal or by reviewing the caption track in the learner view.
• 2015–2019 content: Mixed. Some content is captioned; some has auto-generated captions that have not been quality-reviewed. For clinical vocabulary content (infection control, medication safety, patient safety), auto-generated captions from this era frequently fall below 85% accuracy on drug names and procedure terminology.
• Pre-2015 content: Frequently uncaptioned, particularly for video-based modules that predate HealthStream’s captioning production workflow. If your hospital is using pre-2015 HealthStream content for mandatory annual training — which is common for foundational modules that have been maintained without replacement — audit for caption status before the next annual training cycle.
• Third-party publisher content delivered through HealthStream (Relias content on HealthStream, AMN Healthcare content, NetSol content) may or may not be captioned. Third-party publisher content retains its original caption status; HealthStream does not add captions to third-party content as a matter of course. Check each content publisher’s captioning policy separately.

Hospital-uploaded custom content in HealthStream

Content developed by the hospital and uploaded to HealthStream — orientation videos, department-specific procedure training, leadership communications, facility-specific policy updates — requires the hospital to supply captions. HealthStream supports SRT and VTT caption sidecar files for uploaded video content. The upload workflow:

1. Content Manager access: In the HealthStream Content Manager (CM) portal, navigate to the video asset you are captioning. You need Content Manager access role or the equivalent administrative permissions for your HealthStream instance.
2. Caption track attachment: In the asset properties panel, look for the “Captions” or “Accessibility” section depending on your HLC version. Upload the SRT file. HealthStream’s SRT parser requires standard SRT formatting: sequential numeric cue identifiers, HH:MM:SS,mmm–HH:MM:SS,mmm timestamp format, and blank-line separation between cues. BOM characters in the SRT file cause parse failures; strip BOM before upload.
3. Caption track labelling: Label the track with the language (English) and track type (Captions). HealthStream requires the language designation for the caption track to display in the learner player. An unlabelled track may not appear in the player caption menu.
4. Learner-side verification: After upload, verify caption display in a learner account in the target browser environment. HealthStream’s video player behaviour varies between the HTML5 player used in current HLC versions and the Flash-based player used in legacy deployments. If your facility is still on an HLC version that uses Flash-based video delivery for some content types, the caption track activation mechanism is different and the learner-side display may not work identically to the modern HTML5 player.
5. Mobile verification: Verify caption display in the HealthStream mobile application (iOS and Android) if your workforce uses mobile for training completion. Caption track activation in the HealthStream mobile app requires a separate check; SRT files that display correctly in the browser player may require additional configuration to display in the mobile app depending on HLC version.

HealthStream native ASR captions

HealthStream introduced an automatic speech recognition captioning option in its platform for hospital-uploaded content. This option generates captions automatically during video processing. Do not use this as your primary captioning solution for mandatory clinical training content. The native ASR engine performs well on general speech but falls to the 74–88% accuracy range on clinical vocabulary content without a custom vocabulary configuration — which HealthStream’s native ASR does not support for hospital-uploaded content. The auto-generated captions are useful as a starting draft for human review, but delivering auto-generated captions for HIPAA or CoP-mandated training without quality review does not meet WCAG 2.1 AA and creates ADA exposure. Use an external captioning service with clinical vocabulary support and import the corrected SRT file.

HealthStream caption delivery log

HealthStream does not maintain a caption delivery log that documents when the SRT file was attached, what accuracy standard the file was produced to, and whether the caption track was verified in the learner player. This log must be maintained by the hospital L&D team separately. A minimal caption delivery log entry for HealthStream-hosted content should include: asset ID, asset title, caption file version, date attached, attachment method (Content Manager direct or API), caption accuracy percentage as measured against reference transcript, measurement date and method, and learner-side player verification date and environment. This log becomes the documentary evidence in a Joint Commission tracer review or OCR investigation. For how to structure the complete audit trail, see the enterprise LMS caption audit methodology.

Relias Learning caption workflow

Relias Learning (now Relias, a Bertelsmann Education Group company) is the dominant LMS for behavioural health organisations, post-acute and long-term care facilities, and community health settings. Its content library is the largest in the behavioural health and post-acute care segment, and its captioning architecture is entirely SRT-sidecar-based: Relias does not provide native ASR captioning, and its content library caption coverage is historically inconsistent.

Relias content library caption status

Relias has captioned its core compliance content library — annual mandatory content (abuse prevention, patient rights, fire safety, infection control, HIPAA) — but content library caption coverage thins significantly in specialty clinical areas, particularly for pre-2019 content. If your organisation uses Relias for any of the following content categories, audit caption status before the next training cycle:

• Behavioural health medication education (psychotropic medication vocabulary has high failure rates on clinical ASR)
• Post-acute care assessment tools (MDS 3.0, OASIS-E, functional assessment vocabulary)
• Wound care and skin integrity (wound staging vocabulary, wound care product names)
• Dementia care protocols (person-centred care vocabulary, behavioural support terminology)
• Documentation training for Point Click Care or MatrixCare (both have proprietary navigation vocabulary)

Relias custom content caption import

For content developed by your organisation and uploaded to Relias, the caption import workflow operates through the Relias Content Editor:

1. Content Editor access: In the Relias administrator portal, navigate to Content → My Content. Select the course and enter the content editor. You need the Content Manager or Administrator role.
2. SRT file upload: In the course content properties, locate the video asset. The caption attachment option appears in the video asset properties panel. Upload the SRT file. Relias accepts SRT format; VTT files require conversion before upload depending on the Relias platform version your organisation is running.
3. Caption track language designation: Select the language for the caption track. If your organisation serves workers whose primary language is not English, and if you have caption files in additional languages, upload each language track separately with the correct language designation.
4. Player verification: Relias uses a proprietary video player embedded in its course shell. Verify that the caption toggle is visible in the learner view, that caption display activates correctly, and that captions remain synchronised through the full duration of the video. Timing drift that develops after the first 10 minutes is a common failure mode in Relias when the source SRT file has slight timestamp accumulation errors.
5. Mobile app verification: Relias has iOS and Android mobile applications used by nursing home staff, home health aides, and post-acute care workers who complete training on personal devices between client visits or during shift transitions. Caption display in the Relias mobile app should be verified separately from browser-based display.

Relias behavioural health vocabulary accuracy

Behavioural health training content — the primary content type for many Relias clients — has vocabulary accuracy challenges distinct from acute care hospital content. Diagnostic vocabulary (DSM-5 diagnoses, psychiatric medication names, evidence-based treatment modality names) generates specific ASR failure patterns:

• SSRI (selective serotonin reuptake inhibitor) → “S-S-R-I” survives letter-by-letter. Specific SSRI names: sertraline → “sertraline” survives. Escitalopram → “escalopram” or “escitalopram” (variable). Venlafaxine → “venlafaxine” survives. Aripiprazole → “aripiprazole” occasionally, “aripriprazole” or “ariprazole” more commonly.
• DBT (Dialectical Behaviour Therapy) → “D-B-T” survives. CBT (Cognitive Behavioural Therapy) → “C-B-T” survives. EMDR (Eye Movement Desensitisation and Reprocessing) → “E-M-D-R” or “Emder.” ACT (Acceptance and Commitment Therapy) → “act” survives as a word.
• MDS 3.0 (Minimum Data Set) → “MDS 3.0” or “M-D-S 3.0”; “MDS assessment” survives. OASIS-E (home health assessment tool) → “OASIS-E” or “oases E.” ADL (Activities of Daily Living) → “A-D-L” survives. IADL → “I-A-D-L” survives.

A Relias client in post-acute or behavioural health should build a separate vocabulary glossary for the content categories they train on — different from the acute care hospital glossary, which has different priority term sets. The architecture of a multi-vertical glossary and the case for domain-specific versus unified glossary design is covered in the glossary architecture post.

Epic My Learning caption workflow

Epic Systems distributes its EHR with an integrated learning management system called MyLearning (marketed variously as Epic MyLearning, Epic Learning Home, or Epic Training depending on version and deployment). Epic MyLearning is used by Epic client organisations for EHR training, go-live preparation, and on-the-job reference. It is not a general-purpose hospital LMS — most Epic clients use HealthStream or a separate LMS for non-Epic training, and use MyLearning specifically for Epic system training — but it is a significant captioning surface because EHR training is often the highest-volume training content at Epic sites during upgrade cycles.

Epic-managed content (Application Curriculum)

Epic provides an Application Curriculum of EHR training content to client organisations as part of the Epic implementation and maintenance relationship. This curriculum content is produced by Epic and delivered through MyLearning. Epic has made accessibility improvements to its Application Curriculum in recent release cycles, and most Curriculum content developed after 2022 includes captions. However:

• Content produced for earlier Epic versions (Interconnect, earlier Hyperspace releases) may be uncaptioned or have auto-generated captions without quality review.
• Epic supports caption localisation for non-English-speaking workforces, but localised captions for languages other than English require separate procurement through Epic’s implementation services; they are not included in the standard curriculum.
• When Epic releases an upgrade (annual in current release cadence), new Application Curriculum content accompanies the release. New content caption status should be reviewed at each upgrade cycle before deploying mandatory upgrade training.

Hospital-built Epic training content

Epic client organisations build substantial volumes of custom training content — site-specific workflow training, local configuration videos, go-live preparation modules, superuser training — using Epic’s Kaleidoscope authoring tool or by creating video content externally and importing it to MyLearning. Custom hospital-built Epic content follows the same SRT import pattern as other LMS platforms:

1. Epic MyLearning content management: Access the content library through the Epic MyLearning administrator interface. Navigate to the video or simulation asset for which you are adding captions.
2. Caption file format: Epic MyLearning accepts SRT format for caption sidecar import. Prepare the SRT file against the reference transcript, verified to 99% WCAG accuracy before upload. Epic-specific vocabulary must be pre-loaded in the captioning service glossary before processing: SmartPhrases, dot phrases, Epic module names (Ambulatory, Inpatient, ED, Surgery, Beacon oncology, Stork obstetrics, OpTime, Anesthesia, Willow pharmacy, Cupid cardiology, Beaker laboratory, Radiant radiology, Phlox pathology) — all generate ASR failures when encountered without glossary seeding.
3. Player caption display: Epic MyLearning uses its own video player for content delivered within the MyLearning shell. Caption display should be verified in the Epic MyLearning player specifically, as the caption track rendering behaviour may differ from a standalone HTML5 player.
4. Simulations and click-through training: Epic’s Kaleidoscope simulations are click-through training, not video, and do not use caption tracks. Screen reader accessibility for Kaleidoscope simulations is a separate accessibility consideration outside the scope of video captioning. This guide covers only video caption delivery in Epic MyLearning.

EHR upgrade training captioning at go-live scale

An Epic go-live or major upgrade requires training hundreds or thousands of staff across all clinical roles in a short window. At a 500-bed academic medical centre, an Epic upgrade may require training 3,000 staff in 4–6 weeks before the go-live date. If the training content includes video components (recorded walk-throughs, simulation overviews, role-specific workflow videos), and those videos are being produced in the weeks before go-live, the caption turnaround SLA from the captioning vendor becomes a critical path item. A captioning vendor that cannot deliver corrected SRT files within 24–48 hours for go-live training content creates a gap between the video delivery date and the training deployment date. During go-live preparation, this gap is not acceptable. Establish the go-live captioning SLA with your vendor before the training content production sprint begins, not after the content is in the editing queue. The elements of a go-live captioning SLA are covered in the vendor SLA and contract review checklist.

MedBridge captioning

MedBridge is used primarily by outpatient physical therapy, occupational therapy, and speech-language pathology organisations, as well as skilled nursing facilities that train rehabilitation staff. It is not a hospital LMS in the acute care sense, but it is a significant captioning surface in the post-acute and outpatient rehabilitation segment.

MedBridge content library caption status: MedBridge has substantially captioned its continuing education and clinical skills library. The rehabilitation science vocabulary in MedBridge content (anatomical terminology, exercise physiology, manual therapy techniques, assessment scale names) has generally been handled at reasonable accuracy in their captioned content, but organisation-uploaded custom content (site-specific orientation, protocol training, facility-specific procedure videos) follows the same SRT import pattern as other platforms. MedBridge supports SRT caption import for custom content in its platform administration tools.

For organisations that use MedBridge alongside HealthStream or Relias, the captioning programme architecture should treat MedBridge as a separate caption surface requiring its own audit, not assume that caption coverage carries across platforms. A rehabilitation department that captions all its clinical training in HealthStream and uploads procedure videos to MedBridge without captions has partial coverage that fails the WCAG programme-wide standard.

HIPAA workforce training: caption compliance architecture

HIPAA workforce training captioning is a subset of the broader hospital caption programme, but it deserves separate treatment because of the specific documentation trail that HIPAA enforcement creates. An OCR investigation of a HIPAA complaint — or a proactive OCR review of a hospital’s HIPAA programme — may request documentation that workforce training was conducted and was accessible to the full workforce, including employees with disabilities.

Training content categories under HIPAA

HIPAA workforce training obligations under 45 CFR Part 164 cover several distinct content areas, each of which should be verified for caption compliance independently:

• Security awareness training (164.308(a)(5)(ii)(A)): Annual security awareness training for all workforce members. If delivered as video, must be captioned. Topics include phishing recognition, password hygiene, physical security, mobile device use, PHI handling in digital systems. Vocabulary: phishing → survives, malware → survives, ransomware → survives, PHI (Protected Health Information) → “P-H-I” survives, EHR → “E-H-R” survives.
• Password management (164.308(a)(5)(ii)(D)): Password change procedures, password reset workflows. Low vocabulary failure risk; general IT vocabulary survives ASR reliably.
• Malicious software protection (164.308(a)(5)(ii)(B)): Malware recognition and response. Standard IT security vocabulary.
• Log-in monitoring (164.308(a)(5)(ii)(C)): Suspicious log-in detection, monitoring procedures. Standard IT security vocabulary.
• Privacy Rule training (164.530(b)): Privacy practices, minimum necessary standard, patient access rights, authorisation requirements. Patient rights vocabulary: minimum necessary standard → survives, Notice of Privacy Practices → survives, NPP → “N-P-P” survives, authorisation vs consent distinction → survives.
• Breach notification training (164.400–164.414): Breach identification, internal reporting, notification timeline (60-day rule). Breach risk assessment vocabulary: low probability of compromise → survives, 60-day notification requirement → survives.

HIPAA training documentation for OCR review

An OCR investigation or proactive compliance review will request documentation that HIPAA workforce training was conducted, who received it, when, and that it was accessible. The documentation package for a HIPAA training module should include:

• Training completion record (from LMS: learner name, module name, date, completion status)
• Caption delivery record (caption file version, attachment date, accuracy measurement, player verification)
• Caption accuracy measurement result (DCMP-protocol measurement against reference transcript, percentage, measurement date)
• Accommodation request log (any accommodation requests related to HIPAA training accessibility, and resolution documentation)

The caption delivery record is what most hospitals do not have. Completing HIPAA training does not automatically generate a record that the training was captioned. The L&D team must maintain a separate caption delivery log for HIPAA training modules and update it each annual training cycle (because HIPAA training is refreshed annually, and the caption file associated with the new version must be separately documented).

CMS Conditions of Participation: caption compliance integration

CMS Conditions of Participation establish mandatory training requirements in multiple clinical and operational domains. The documentation required for CoP compliance creates the training record that Joint Commission reviewers examine during accreditation surveys. Caption compliance must be woven into the CoP training record architecture, not maintained as a separate accessibility compliance record that no one looks at until an OCR complaint arrives.

CoP training mandates that generate high-volume video content

The following CoP training areas are most commonly delivered as video in hospital L&D programmes and therefore generate the highest-volume captioning obligation:

• Infection control (42 CFR 482.42): Annual infection prevention training for all direct care staff. High vocabulary failure risk (pathogen names, isolation procedures, PPE protocol).
• Patient rights (42 CFR 482.13): Patient rights training for all clinical staff, including advance directive procedures, grievance processes, and restraint and seclusion protocols. Moderate vocabulary failure risk.
• Emergency preparedness (42 CFR 482.15): Annual emergency preparedness training, including fire safety, disaster response, and evacuation procedures. Moderate vocabulary failure risk (HICS - Hospital Incident Command System → “HICS” survives; HEICS → “HEICS” or “H-E-I-C-S”).
• Quality assessment and performance improvement (42 CFR 482.21): Staff training on QAPI programme participation, data collection methods, and performance improvement processes. Moderate vocabulary failure risk (QAPI → “Q-A-P-I” or “quapee”; PDSA cycle → “P-D-S-A cycle” survives; run chart → survives; statistical process control → survives).
• Medical staff credentialing (42 CFR 482.22): Physician training on hospital policies, credentialing requirements, clinical protocols. Mixed vocabulary.
• Nursing services (42 CFR 482.23): Annual nursing competency training and skills validation. High vocabulary failure risk (clinical skills vocabulary, assessment scale names, clinical procedure terminology).

CoP survey readiness and caption documentation

CMS CoP surveys — conducted by State Survey Agencies acting under CMS authority — review training records as part of the survey process. Unlike Joint Commission tracer methodology, CMS surveys typically request broader population-level data: all staff training records for a specified time period, training curriculum documentation, and evidence that training was provided to new staff within specified timeframes. Caption compliance documentation should be integrated into the training record system in a way that allows it to be produced alongside the training completion data in response to a CMS information request. Maintaining caption compliance records in a separate spreadsheet from the LMS training record creates a retrieval problem when documentation must be assembled on short notice.

Joint Commission tracer methodology: caption audit trail requirements

The Joint Commission’s tracer methodology creates a specific and demanding documentation environment. A surveyor conducting a tracer will select a patient from the current census, review the patient’s record, identify every clinician involved in the patient’s care, and then request training records for each of those clinicians. The training records must demonstrate that each clinician has completed the required training for their role in the specified timeframe. The surveyor may ask to see the actual training content — or at minimum the training record that documents what was trained, when, and with what tools.

What tracer methodology reviewers look for in training records

Based on Joint Commission tracer activity and accreditation findings in the public record, training record gaps that surface in tracer reviews include:

• Staff not trained in areas required for their role (e.g., ICU nurse without documented CLABSI prevention training)
• Training completed outside the required timeframe (annual mandate completed 14 months ago rather than within 12 months)
• Training completed by the staff member but not documented in the LMS in a way that produces a retrievable record
• Training content that does not match the staff member’s role (completing general fire safety training but not the role-specific emergency response training required for their department)

Caption compliance is not a Joint Commission tracer primary target — surveyors are not specifically looking for caption delivery records in the same way they look for training completion dates. However, if a surveyor asks about accessibility accommodations for training, or if a tracer review includes a staff member who has a documented disability, the ability to demonstrate that training content was accessible is part of the training record integrity picture. A hospital that cannot demonstrate accessible training delivery is in a weaker position in a discrimination complaint investigation that may follow a Joint Commission survey.

Building caption delivery into the Joint Commission audit trail

The practical approach is to integrate caption delivery documentation into the same record-keeping system used for training completion records. This does not require a new system; it requires adding caption delivery status as a field in the training record database (or an attached log) so that when training records are retrieved for tracer purposes, caption status is retrievable alongside completion status. The minimum fields to add:

• Caption status (yes/no/accommodation pending)
• Caption file version
• Caption accuracy percentage
• Caption accuracy measurement date
• LMS player verification date

For the broader framework of how to structure an enterprise LMS audit trail that satisfies multiple simultaneous compliance requirements, see the enterprise LMS caption audit methodology post. For the specific governance policy structure that supports audit readiness, see the caption programme governance policy template.

NPSG-related training and caption compliance priority

The Joint Commission’s National Patient Safety Goals represent the highest-priority training areas from an accreditation risk perspective. NPSG-related training content should be the first priority for caption remediation in any hospital back-catalogue project:

• NPSG 01 (patient identification) → two-identifier protocol training
• NPSG 02 (effective communication) → SBAR training, handoff communication training, critical value communication
• NPSG 07 (infection prevention) → hand hygiene training, central line care, CAUTI prevention
• NPSG 09 (fall prevention) → fall risk assessment, environmental modification training
• NPSG 15 (suicide risk) → suicide risk screening training (Columbia Protocol), environmental risk assessment

If your back-catalogue remediation budget is constrained and you must prioritise, caption NPSG-related training first. Joint Commission accreditation findings in NPSG areas create immediate remediation obligations; accessibility gaps in NPSG training create both accreditation risk and ADA liability simultaneously.

Back-catalogue remediation at hospital scale

Hospital video training libraries accumulate over time without a centralised caption programme. A 500-bed academic medical centre that has operated HealthStream for 8 years and uploaded custom content throughout that period may have 1,500–3,000 custom video assets in its LMS, a significant fraction of which are uncaptioned. Adding Relias content, Epic training content, and any other video training surfaces, the total uncaptioned content volume may be substantially larger.

Inventory and risk-weighted prioritisation

The first step of back-catalogue remediation is inventory — not remediation. Before committing to a remediation schedule, produce an inventory of all video training assets across all LMS platforms used by the organisation, including:

• Asset title, platform, upload date
• Current caption status (captioned / uncaptioned / auto-generated unreviewed)
• Mandatory / elective designation
• Assigned learner population (all staff / role-specific / department-specific)
• Annual training cycle (is this content delivered annually as part of mandatory training?)
• NPSG relevance (is this content related to any current NPSG?)
• CoP mandate relevance (does this content satisfy a specific CoP training requirement?)
• Content age (pre-2020 / 2020-2022 / post-2022)

From this inventory, apply risk-weighted prioritisation using the framework developed for large-scale caption backlog remediation, adapted for the hospital environment:

• Tier 1 — immediate (90-day remediation window): Mandatory annual training currently in deployment cycle; NPSG-related content; CoP-mandated content for clinical staff; HIPAA training content for all workforce; orientation and onboarding content for new hires currently in process. These videos are being watched right now by employees. Every day they remain uncaptioned is a day of non-compliance.
• Tier 2 — next annual cycle (180-day remediation window): Mandatory annual training not yet in the current deployment window; role-specific clinical competency content; Joint Commission standards-related training.
• Tier 3 — planned remediation (12-month window): Department-specific procedure training; non-mandatory but regularly assigned content; clinical skills reference videos.
• Tier 4 — archive decision (18-month window): Content over 5 years old not in active rotation; content flagged for replacement in the next content refresh cycle. Make a retain-and-caption vs. retire decision before spending on remediation.

The LMS ingestion bottleneck at hospital scale

The binding constraint on back-catalogue remediation in hospitals is not vendor processing capacity — it is LMS ingestion throughput. Attaching SRT files to existing HealthStream courses, verifying caption track activation, checking learner-side display across browsers and the mobile app, and documenting the completion in the caption delivery log takes approximately 15–20 minutes per video in HealthStream. For Relias, the workflow is slightly more streamlined, approximately 10–15 minutes per video. For Epic MyLearning, 10–15 minutes per video for hospital-built content.

At 15 minutes per video, 800 uncaptioned videos = 200 hours of LMS administrative work. That is 5 weeks of full-time equivalent effort for a single administrator — before any content processing begins at the vendor side. Staffing the LMS ingestion work separately from the caption production work is essential for maintaining remediation throughput. An instructional designer who is managing the caption vendor relationship, reviewing incoming SRT files for accuracy, and doing LMS ingestion work simultaneously will complete the back-catalogue remediation in 12–18 months instead of 6 months. The captioning at scale post discusses how to staff and sequence large-volume caption operations.

Vendor throughput for hospital back-catalogue work

Hospital back-catalogue work has content characteristics that affect vendor throughput and pricing. Clinical audio with high background noise (nurse call system alerts, overhead paging, ventilator sounds, telephone rings) requires audio pre-processing before ASR can achieve acceptable accuracy. Budget for audio remediation cost — typically 10–15% of the content batch will require active audio clean-up before captioning, not just noise reduction filters. Clinical content also has a higher review burden than soft-skills content because the error density on technical vocabulary is higher; reviewer time per video is longer than the generic estimate. Request a pilot batch from the vendor that includes a representative sample of your highest-difficulty content before agreeing to a volume contract for hospital back-catalogue work. The vendor pilot programme design post covers how to structure a pilot that generates reliable throughput and accuracy data for hospital content.

Hospital vendor contract: caption SLA requirements

The vendor contract for a hospital captioning programme must contain terms that address the specific compliance context hospitals operate in. The standard SLA terms in a captioning vendor agreement — turnaround time, file format, price per minute — are necessary but not sufficient for a hospital environment. The following clauses are not standard in most vendor agreements and must be negotiated:

Business Associate Agreement

Audio submitted to a captioning vendor from hospital training videos will frequently contain PHI. Training videos that include patient cases, clinical scenario demonstrations, actual patient voices (in patient communication training), or identifiable patient images constitute PHI under HIPAA. Even training videos that do not include patient content may reference specific patient situations or include audio from care environments. A Business Associate Agreement (BAA) under 45 CFR 164.314 is required before audio containing PHI is transmitted to a captioning vendor. The BAA must cover:

• Permitted uses and disclosures of PHI (processing only, no secondary use)
• Safeguard obligations (encryption in transit, encryption at rest)
• Breach notification to hospital within 60 days
• Subcontractor BAA obligations (if the vendor uses subcontractors in the ASR or human review workflow)
• Return or destruction of PHI on contract termination
• Audit rights for HIPAA Security Rule compliance review

The vendor contract review checklist covers the BAA provisions in detail, including the subcontractor chain clause that most generic vendor agreements omit. If the captioning vendor uses a cloud ASR provider (AWS Transcribe, Google Speech-to-Text, Azure Cognitive Services) as part of its processing pipeline, that ASR provider may also need to be covered by a BAA. Request the vendor’s subcontractor list before executing the BAA.

Clinical vocabulary support obligation

The vendor contract must specify the clinical vocabulary glossary support arrangement: who owns the glossary terms, who maintains the glossary as clinical vocabulary changes (new drug formulary, new procedure names, new EHR module names), how quickly glossary updates are reflected in processing, and what happens to accuracy performance if the glossary is not maintained. A hospital that builds a 200-term clinical vocabulary glossary in year one and does not update it for three years will see accuracy regression as new clinical vocabulary enters training content (new medication names, new EHR terminology from system upgrades, new procedure vocabulary from clinical programme expansions). The contract should specify:

• Glossary term limit (minimum 500 terms for hospital content; 1,000+ for academic medical centres with specialty-specific training across multiple service lines)
• Glossary update SLA (new terms reflected in processing within 48 hours of submission)
• Glossary portability at termination (hospital owns the glossary as structured data in a portable format, not as a vendor-proprietary configuration)
• Annual glossary review obligation (both parties review term coverage annually against new content types)

Annual training cycle turnaround SLA

Hospital mandatory annual training deploys in concentrated windows — often January through March for the calendar-year training cycle, or aligned to fiscal year. All-staff HIPAA training, infection control updates, and CoP-mandated content may deploy simultaneously, creating a demand spike on the captioning vendor. The contract must specify:

• Standard turnaround SLA (24–48 hours for up to X minutes of content per day)
• Rush SLA (same-day or 12-hour turnaround for go-live and annual training cycle peaks)
• Annual training cycle capacity reservation (number of minutes per week the vendor guarantees capacity to process during the annual training deployment window)
• SLA breach remedy (credit or fee reduction; not just “best efforts” language)

Accuracy guarantee with DCMP measurement protocol

The accuracy guarantee must specify the measurement methodology, not just the target. A vendor accuracy guarantee of “99% accuracy” without specifying the measurement protocol is not a binding commitment to WCAG 2.1 AA compliance. The guarantee must state:

• Accuracy measured by DCMP Captioning Key formula: WER = (S + D + I + T) / N × 100, where S = substitutions, D = deletions, I = insertions, T = timing errors over 2 seconds, N = total caption words
• Reference transcript prepared independently before submission (not post-correction vendor self-measurement)
• Measurement on a representative sample of each content category, including highest-difficulty clinical vocabulary content
• Remediation obligation when measured accuracy falls below 97% on any batch (not just below 99% average)

For the complete accuracy guarantee clause structure and the other 41 contract review points, see the caption vendor SLA and contract review checklist.

Records retention for regulatory review

Caption delivery records must be retained for the same period as the training records they accompany, which is determined by the most restrictive applicable requirement. For hospital training:

• HIPAA requires retention of training documentation for 6 years from creation date or last effective date (45 CFR 164.530(j)(2))
• CMS CoP surveys may request training records for the preceding 3 years
• State employment law may impose additional retention requirements for training records in employment discrimination contexts
• Joint Commission requests records from the current accreditation cycle (3 years)

The vendor contract should require the vendor to retain processing records (submission logs, processing metadata, accuracy measurement records, glossary version history) for a minimum of 6 years and to deliver those records to the hospital on request within 10 business days. This is the audit-rights provision that the caption vendor audit rights and examination evidence post identifies as the clause most hospitals fail to include at contract signature.

Building the hospital caption programme: a 90-day launch plan

A hospital L&D team launching a caption programme for the first time — or formalising an ad hoc approach into a governed programme — can achieve the critical milestones in 90 days. The sequence is designed to address the highest-risk content first while building the infrastructure for ongoing compliance. The fuller programme build framework is in the L&D caption compliance programme guide.

Days 1–14: inventory and prioritisation

• Complete video asset inventory across all LMS platforms (HealthStream, Relias, Epic MyLearning, any others)
• Apply four-tier risk prioritisation: Tier 1 = mandatory annual training currently in deployment + NPSG content + HIPAA training + orientation videos for current new hires
• Identify total Tier 1 content volume in minutes
• Establish which content was produced by the hospital and requires external captioning vs. content provided by HealthStream or Relias that may already have vendor-supplied captions (even if unreviewed)
• Document current caption status for all Tier 1 assets

Days 15–30: vendor selection and pilot

• Issue a brief RFP or conduct a scored vendor evaluation for healthcare captioning services. Priority criteria: BAA availability, clinical vocabulary glossary support, DCMP accuracy measurement, turnaround SLA for annual training cycle peaks. The captioning RFP playbook covers the full evaluation process.
• Run a pilot batch of 30–50 minutes of Tier 1 hospital content across all content categories (infection control, HIPAA, patient safety, EHR training, credentialing). Measure DCMP accuracy against reference transcripts you prepare.
• Evaluate the pilot results per the vendor pilot programme design framework.
• Execute the vendor agreement including BAA, clinical vocabulary glossary terms, annual cycle SLA, accuracy guarantee with DCMP measurement, and 6-year records retention.

Days 31–60: Tier 1 remediation

• Submit all Tier 1 content to the vendor with the clinical vocabulary glossary. Stagger submissions to manage LMS ingestion workload alongside vendor throughput.
• Review returning SRT files for accuracy on a DCMP spot-check basis before LMS ingestion.
• Attach SRT files to HealthStream, Relias, and Epic MyLearning assets. Verify learner-side display in browser and mobile environments for each platform.
• Document each completed caption delivery in the caption delivery log.
• For any Tier 1 content that HealthStream or Relias has provided with vendor-supplied captions, review the caption accuracy on a sample of the highest-vocabulary-difficulty content. If below 97% on DCMP measurement, flag for replacement with corrected SRT files.

Days 61–90: programme governance and Tier 2 pipeline

• Establish the new content gate: no new training video deploys to any LMS without a captioned SRT file attached and verified in the learner player. This prevents accumulation of new uncaptioned content.
• Create the caption programme governance policy. Minimum: accuracy standard (99% WCAG DCMP), measurement method, remediation trigger (below 97% on DCMP spot-check), exception procedure (live training without CART availability), annual review cadence.
• Integrate caption delivery log into the LMS training record management workflow.
• Begin Tier 2 remediation pipeline.
• Schedule the annual review for 12 months from programme launch: vendor accuracy vs. year-1 benchmark, SLA adherence, glossary term audit (new clinical vocabulary added in the past 12 months), regulatory update review (WCAG 2.2 criteria applicable to caption player UI, new OCR enforcement guidance, annual CoP update).

The budget planning process for a hospital caption programme should include: vendor cost per minute for the Tier 1 volume (typically 2,000–10,000 minutes depending on hospital size), internal LMS ingestion time cost (15–20 minutes per video at L&D coordinator hourly rate), glossary build time (one-time: 30–40 hours for a 500-term clinical glossary), and ongoing annual cadence cost for new content. The caption programme budget planning guide has the unit economics model and the ROI calculation for the Finance presentation.

Eight failure modes in hospital caption programmes

Based on the pattern of documentation gaps and compliance findings in hospital L&D settings, these are the eight failure modes that most commonly cause problems in regulatory review:

1. HealthStream content library not audited for caption status before annual training deployment. The assumption that all HealthStream library content is captioned is incorrect. Pre-2020 content, third-party publisher content, and content from some specialty clinical areas may be uncaptioned or carry auto-generated captions below the 99% threshold. Annual training cycle deployment without a caption status check on each module in the curriculum is the most common source of large-scale non-compliance at hospital organisations that believe they have a caption programme.
2. Hospital-uploaded custom content treated as outside the caption programme. L&D teams that have contracted with a captioning vendor for mandatory annual training content frequently omit hospital-specific orientation videos, department-level procedure training, and leadership communications from the caption programme because “those are internal.” ADA Title I applies to all employee training. There is no exemption for informally produced or narrowly distributed content.
3. SRT file attached but caption track not verified in the learner player. A caption file that is attached to a HealthStream or Relias video asset but fails to display in the learner player due to a format error, a timing offset accumulated from the SRT source, or a platform-specific rendering issue provides zero accessibility benefit. Verification in the actual learner environment — in the target browser, in the mobile app, at the playback point where timing drift is most likely to occur — is not optional.
4. No BAA with captioning vendor before PHI-containing audio is submitted. Training content that includes patient scenarios, clinical demonstrations, patient voice, or identifiable care-setting audio constitutes PHI. Submitting this content to a captioning vendor without an executed BAA is a HIPAA Security Rule violation (45 CFR 164.314 requires BAAs with all business associates who access PHI). This failure mode is common because L&D teams do not always know which of their training videos contain PHI and assume the captioning vendor is not a business associate.
5. Clinical vocabulary glossary not maintained after initial build. A clinical vocabulary glossary that was accurate when built will become less accurate as the hospital’s training content evolves — new EHR module names, new clinical programme vocabulary, drug formulary additions, new infection control protocol terminology. A glossary that has not been reviewed in 18 months is not protecting accuracy on new content. The glossary maintenance workflow should run on the same annual cadence as the caption programme annual review.
6. Caption delivery log not maintained. The caption delivery log is the instrument that demonstrates, at the document level, that training content was accessible to employees with disabilities at the time the training was conducted. Without it, the hospital can demonstrate training happened but cannot demonstrate it was accessible — the same documentation gap that the examination evidence post identifies as the most common finding in regulatory reviews.
7. Back-catalogue remediation scoped without accounting for LMS ingestion time. Remediation plans that are built around vendor processing throughput and do not account for LMS ingestion administrative time consistently miss their deadlines. At 15–20 minutes per video for HealthStream attachment and verification, 500 videos requires 125–167 hours of internal L&D time on LMS tasks alone. This work cannot be compressed by increasing vendor throughput.
8. No new content gate established after back-catalogue remediation begins. A hospital that spends 6 months remediating 800 uncaptioned historical videos while continuing to deploy new uncaptioned content is running a leaking bucket. The back-catalogue is shrinking while the uncaptioned inventory is growing. The new content gate — no video deploys without a verified SRT file attached — must be established at programme launch, not after the back-catalogue is clear.

FAQ