Quickstart: Your HIPAA-Compliant Clinical OS
Enable HIPAA mode, configure AI triage with deterministic safety fence, pair BLE medical devices, set NEWS2 alerts, enable AI medical scribe, wire telehealth — the full setup in 6 steps, in under 30 minutes.
Enable the HIPAA Compliance Profile
Navigate to Admin → Health → Compliance and enable the hipaa profile. This activates a tenant-wide policy enforcement layer:
What HIPAA mode enforces:
- AI provider allowlist — only BAA-signed providers can process PHI
- Pre-configured BAA providers — AWS Bedrock, Azure OpenAI BAA, AWS Transcribe Medical, AWS Polly
- Audit logging — every PHI access logged with user + reason + IP + timestamp
- Encryption — at-rest (AES-256-GCM per-tenant key) + in-transit (TLS 1.3)
- Retention policies — minimum 7-year retention for medical records (configurable per record type)
- Access controls — minimum-necessary principle enforced; users see only patients in their care assignment
- Disposable evidence — all PHI exports watermarked with viewer identity for legal trace
- Sub-processor gating — non-BAA-signed providers automatically blocked from PHI processing
Business Associate Agreement (BAA):
- Your Aeion Health contract includes Aeion's BAA covering ePHI processing
- BAA specifically lists covered sub-processors
- You provide your covered-entity name + signing officer; Aeion countersigns
- Available under Admin → Health → Compliance → BAA Documents
Mode switch is one-way for production safety — once HIPAA is enabled, you can't accidentally turn it off (requires support assistance + audit log entry).
Configure the AI Triage Engine
Navigate to Health → Triage Settings. The triage engine uses a two-tier architecture:
Tier 1 — Deterministic ESI Safety Fence:
- Implements AHRQ ESI Implementation Handbook algorithm exactly
- Evaluates vital sign danger zones (HR / BP / RR / O2 sat / temp)
- Pain scores (0-10 with red-flag pain pattern detection)
- Trauma keywords (configurable per ED protocol)
- Altered mental status indicators
- Returns deterministic ESI level (1-5)
Tier 2 — AI Adapter (configurable):
- Default: an integrated AI adapter using a BAA-signed LLM
- Custom: plug in your own clinical decision-support model through our triage adapter framework
- Returns suggested acuity + reasoning
The hard rule: when AI and ESI disagree, the more conservative (higher-acuity) level always wins. This is not a confidence override — it's a deterministic rule. An AI hallucination cannot down-triage a critical patient.
typescript
// Configure the AI triage adapter for your tenant
await aeion.api.patch(/v1/health_triage_settings/${settingsId}, {
llmProvider: "azure-openai-baa",
llmModel: "gpt-4o",
requireClinicianRole: true,
disclaimerText:
"AI suggestion subject to clinician override per ESI safety fence",
// ESI fence always-on, not configurable
});
Audit trail: every analysis call writes a permanent record regardless of outcome — success, adapter no-op, or error. The encounter is never left without a record. If AI fails, clinician sees "AI Analysis Unavailable" and manually sets acuity (logged with full attribution).
Connect BLE Medical Devices
| Device class | Examples |
|---|---|
| **Pulse oximeters** | Masimo Radius PPG, Nonin 3231, iHealth Air Pro |
| **Blood pressure cuffs** | Welch Allyn Connex, Omron HEM-907XL, iHealth Track |
| **Glucose meters** | Roche Accu-Chek Guide Me, OneTouch Verio Reflect |
| **Thermometers** | Withings Thermo, iHealth No-Touch |
| **Weight scales** | Withings Body+, BodiMetrics Performance Monitor |
| **ECG** | Withings ScanWatch (ECG approved devices only) |
| **Pulse + SpO2 spot-check** | Nonin Connect-2, BCI Apex Pro |
Configure NEWS2 Deterioration Monitoring
| Score range | Severity | Recommended action |
|---|---|---|
| 0 | Low | 12-hour observation interval |
| 1-4 | Low-medium | 4-6 hour observation |
| 5-6 | **Medium** | Urgent review by competent ward-based clinician |
| 7+ | **High** | Immediate clinical review by critical-care team |
| Single param ≥3 | Variable | Targeted intervention for that specific abnormality |
Enable AI Medical Scribe
In Telehealth → Settings, enable AI Medical Scribe. The scribe operates during telehealth sessions:
During the session:
- Audio captured + transcribed in real-time via Whisper (BAA-signed deployment)
- Patient + clinician speech distinguished via diarization
- Background music + medical-equipment noise filtered
After the session:
- AI generates a structured SOAP note:
- Suggested ICD-10 codes (with confidence + source quotes)
- Suggested CPT codes for billing
- Prescription pre-fill (clinician must explicitly approve each)
- Patient instructions auto-drafted (medication, lifestyle, follow-up)
Clinician workflow:
- Review the AI-generated note (typically takes 30-60 seconds)
- One-click approve OR edit specific sections
- Approval signs the note (electronic signature)
- Note posts to the patient's chart + ICD/CPT codes route to billing
Audit trail: every AI suggestion logged with confidence, source quotes from transcript, and clinician action (accept / edit / reject). Trains per-tenant fine-tuning over time without sending PHI to model training.
Wire Telehealth + Integration
Navigate to Health → Telehealth → Settings. Configure:
WebRTC video session settings:
- Resolution: 1080p (default), 4K available for dermatology / ophthalmology
- Codec: VP9 / AV1 for bandwidth efficiency
- Recording: HIPAA-compliant, per-tenant encryption
- Quality monitoring: real-time RTT + packet-loss feedback to clinician
Pre-visit workflow:
- Patient receives appointment link (Appointments module integration)
- Self-service: pre-visit questionnaire, photo upload (skin / wound), insurance verification
- Auto-load patient chart for clinician before session starts
During-visit features:
- Co-pilot transcript on clinician screen (real-time)
- One-click ICD lookup
- Prescriptions recorded directly on the patient's chart, with refill and fulfillment tracking
- Lab results recorded and trended on the patient's chart
- Referral generation with auto-fill
Post-visit:
- SOAP note generation (Step 5)
- Visit summary auto-sent to patient
- Follow-up appointment auto-suggested (calendar slots offered)
- Billing claim auto-prepared (clinician reviews + submits)
Patient portal:
- Patient login (HIPAA-compliant authentication)
- View past visits, scheduled appointments, prescription refills
- Secure messaging with care team
- Family-member proxy access (configurable per patient)
Integration Recipes
Appointments → Telehealth. Standard telehealth appointment booked via Appointments → auto-creates a Realtime room with the patient's pre-visit context loaded. Clinician opens and starts.
Bridge → BLE devices auto-streaming. Continuous monitoring devices (ECG in OR, pulse-ox on monitored patient) auto-stream through Bridge into the encounter record. NEWS2 recalculates with every new data point.
Helpdesk → patient support tickets. Patient questions / billing disputes flow through Helpdesk with PHI-tagged routing (only HIPAA-cleared support reps can see PHI in tickets).
Realtime → telehealth video. All telehealth sessions run on the Realtime module's WebRTC infrastructure. Same compliance posture, same audit trail.
Marketing → patient outreach (HIPAA-aware). Marketing campaigns can target patient segments (e.g., "diabetic patients due for HbA1c"), but PHI-aware: campaign content uses tokenized references that resolve to PHI only at delivery time, never stored in marketing logs.
Forms → patient intake. Pre-visit forms (medical history, consent, HIPAA notice acknowledgment) flow through Forms module with HIPAA-compliant storage. Auto-populates the encounter record.
Troubleshooting
BLE device not pairing. Verify Bridge is running + Bluetooth is enabled on the host. Some devices require specific firmware versions — check the device's manufacturer page. For Linux Bridge: ensure user has dialout group access.
Triage AI returning errors. Check Admin → Health → Triage → Audit Log. Common causes: (1) BAA provider experiencing outage (auto-fall-back to ESI safety fence), (2) cost cap reached (raise budget), (3) PHI sanitization rejected the input (check for unusual characters / format).
NEWS2 alert firing repeatedly. Tune the trajectory window — too short produces noise from minor fluctuations. Default 4 hours. Also check the patient's baseline — for chronic conditions, the scoring may need patient-specific overrides.
Telehealth video quality poor. Check the network quality indicators in the session UI. Suggest the patient + clinician switch to wired Ethernet if possible. Switch codec to H.264 if VP9 / AV1 isn't supported on either end.
AI scribe missing details. Common cause: audio quality. Verify the clinician's microphone (high-quality headset recommended). Background music + alarms can confuse transcription — record in quieter rooms when possible.
HIPAA mode blocking a needed feature. Check the BAA provider list (Admin → Health → Compliance → BAA Providers). If the feature needs an LLM that isn't BAA-covered, you'll need to request that vendor add BAA support, or use a covered alternative.
Frequently Asked Questions
Pure deterministic logic based on the AHRQ ESI Implementation Handbook. Evaluates vital-sign danger zones, pain scores, trauma keywords, altered mental status independently of any AI suggestion. When AI and ESI disagree, the more conservative (higher-acuity) level always wins. Rule, not heuristic.
Yes. Bring your own clinical model and register it through our custom AI adapter framework, alongside the built-in LLM-backed adapter. Your model's output is always subject to the deterministic ESI safety fence before reaching clinicians, so a misbehaving model can never escalate or de-escalate a patient on its own.
Bridge reads BLE devices locally. Vital readings flow into the encounter record via the local Bridge filesystem, encrypted at rest. Data never transits the public cloud unencrypted. DICOM rendering on local GPU when Bridge has 16GB+ VRAM; cloud fallback only with active BAA.
Every analysis call writes an immutable audit row regardless of outcome — success, adapter no-op, or error. The encounter is never left without a record. Clinician sees "AI Analysis Unavailable" + manually sets acuity, logged with full attribution.
ESI triages incoming patients by resource need. NEWS2 monitors inpatients for deterioration. Complementary — ESI gets patients to the right area; NEWS2 catches them if they decompensate.
Yes. Every Aeion Health contract includes a Business Associate Agreement covering ePHI processing. BAA specifically lists covered sub-processors (AWS Bedrock, Azure OpenAI BAA, AWS Transcribe Medical, etc.). PHI never sent to providers outside this list.
Scribe suggests but never auto-prescribes. Every prescription requires explicit clinician approval (1-click) before it's recorded on the patient's chart. Full audit trail.
Yes for migration — Aeion Singularity ships FHIR R4-based connectors for Epic and Cerner that import patients, practitioners, and related records into Aeion Health. This is a one-way migration pull, not a live bidirectional sync, so plan the cutover accordingly rather than assuming continuous two-way sync with your existing EHR.
Not currently a built-in integration. Vitals reach the patient chart via BLE medical devices (Step 3) or manual/lab entry; consumer wearables aren't wired in today.
Yes via the patient portal — 21st Century Cures Act compliant (US). Patients see their notes with clear "AI-drafted, clinician-approved" labeling. Open Notes standard format. Patients can request amendments through the portal; clinician reviews + approves.