AI clinical tools
BridgeERP HMS bundles a set of AI clinical tools that reduce documentation load and surface the right action at the right moment: an ambient scribe that drafts notes from the spoken encounter, AI-first draft cards for the note, orders and prescription, smart text auto-expansions, clinical decision support, and a predictive engine that proposes next actions. Every AI output is a draft — the clinician reviews, edits and signs; nothing is committed automatically.
Where to find it
- HMS Core → Clinical → Ambient AI — the ambient scribe: Recordings, AI Tasks → Scribe Tasks and Edit Log, plus Configuration (Providers, Redaction Rules, Scribe Templates, Settings) and Usage → Usage & Quota.
- HMS Core → Clinical → AI-First Clinical — the draft cards: Drafts, Decision Log and Settings.
- HMS Core → Clinical → Clinical Decision Support — CDS Rules, Alerts, Order Sets, Preventive Care, Appropriateness Criteria and Configuration.
- HMS Core → Clinical → Predictive AI — Dashboard, Rules, Predicted Actions and Rejection Feedback.
- HMS Core → Clinical → Doctor Inbox — the unified clinician inbox that aggregates results, messages and AI items.
- Smart text is configured through its phrase library (smart-text phrases, templates and tags) and used inline in note fields.
Before you start
- AI provider. Configure at least one Provider under Ambient AI Configuration (a model such as a transcription or language model is named per provider).
- Redaction rules. Define Redaction Rules so identifiers are stripped before audio/text is sent for processing.
- Quotas. Review Usage & Quota so the practice stays within its AI budget; cost estimates are tracked per task.
- Consent. Patient consent to record should be captured before ambient capture (see Telehealth for the consent record).
Ambient scribe: from speech to SOAP
The ambient scribe listens to the consultation, transcribes it, and synthesises a structured note you can apply to the encounter.
- Start a scribe session (or capture an encounter recording) for the consultation. It begins in Draft.
- Run Transcribe — the session moves through Transcribing to Transcribed, producing transcript segments.
- Run Synthesise — through Synthesising to Ready, the model drafts a structured note from the transcript.
- Review and edit, then Apply to Consultation — the session becomes Applied to Consultation and the note populates the SOAP fields. You can Cancel at any point.
Every clinician edit to an AI draft is captured in the Edit Log, so the practice can measure how much the model gets right and govern its use.
Scribe session states
| State | Meaning |
|---|---|
| Draft | Created, not yet processed. |
| Transcribing / Transcribed | Audio being converted to text / text ready. |
| Synthesising / Ready | Note being drafted / draft ready for review. |
| Applied to Consultation | Draft accepted into the SOAP note. |
| Cancelled | Session abandoned. |
AI-first draft cards
The AI-First Clinical tool presents the encounter’s AI output as draft cards, each a discrete proposal the clinician accepts or rejects. Card types are SOAP Note, Orders (Lab/Imaging), Prescription, Follow-up and Billing Codes. Each card carries a confidence band and holds both the raw model response and the doctor-edited text.
- Open the consultation’s drafts (or the Generate Note (SOAP) action on the Chart Desktop).
- Review each card. Edit the text where needed.
- Press Accept to commit the card — it becomes Accepted or, if you changed it, Edited & Accepted, and the real records (note, orders, prescription) are created. Press Reject with a reason to discard it.
| Card state | Meaning |
|---|---|
| Draft | Proposed, awaiting review. |
| Accepted | Committed as generated. |
| Edited & Accepted | Committed after clinician edits. |
| Rejected | Discarded with a reason. |
| Regenerating / Failed | Being re-drafted / generation failed. |
The Decision Log records each accept/reject decision — the audit trail for AI-assisted documentation.
Card types, confidence & decisions
Each draft card declares what it proposes to create, and the model attaches a confidence band so you can triage your review — trust the high-confidence cards, scrutinise the low ones.
| Card type | Creates on acceptance |
|---|---|
| SOAP Note | The structured consultation note. |
| Orders (Lab/Imaging) | Lab and imaging orders for the encounter. |
| Prescription | A prescription with medication lines. |
| Follow-up | A follow-up plan / appointment prompt. |
| Billing Codes | Suggested billing codes for the encounter. |
Confidence is banded as High (≥70%), Medium (50–70%) and Low (<50%). Every decision a clinician records on a card — Accept, Reject, Edit or Regenerate — is written to the Decision Log with the user and timestamp.
Scribe task status
Behind a scribe session, each call to a transcription or language model runs as a scribe task with its own status, tracked under Ambient AI → AI Tasks → Scribe Tasks with token counts, cost and latency.
| Status | Meaning |
|---|---|
| Queued | Accepted, waiting for a worker. |
| Running | The provider job is executing. |
| Succeeded | Completed; output captured. |
| Failed | The job errored; see the error message. |
| Retrying | Being re-attempted after a failure. |
| Cancelled | Stopped before completion. |
Smart text
Smart text expands a short trigger into a longer block of structured text inside a note field — a normal-exam phrase, a counselling paragraph, a discharge instruction. Phrases are categorised, can be specialty-specific, and can be personal or shared. Usage is logged so popular phrases surface and the library stays curated.
Clinical decision support
CDS is the rule engine that fires alerts on clinical events (order placed, result received, diagnosis recorded, vital recorded, encounter opened, discharge, medication administered). Rules carry categories such as allergy, drug-drug interaction, dose check, duplicate therapy and preventive-care reminders, and a severity from Info to Hard Stop. The prescribing-side behaviour of CDS — how alerts present and how overrides are governed — is documented on the Prescriptions page.
Predictive / at-risk actions
The Predictive AI engine evaluates rules against patient data and proposes predicted actions — a lab test, prescription, referral, monitoring, patient education, imaging, vaccination or a procedure bundle. Each predicted action is Suggested until a clinician acts on it:
- Open Predictive AI → Predicted Actions or the Dashboard.
- For each suggestion, press Approve (it becomes Approved, then Executed when the underlying lab request, prescription or referral is created), Reject with a reason, or Snooze it for later.
- Rejections feed Rejection Feedback, which is used to tune the rules over time.
Doctor Inbox
The Doctor Inbox aggregates the clinician’s actionable items — results to acknowledge, messages, and AI suggestions — into one prioritised list driven by inbox rules, so nothing waiting on the doctor is missed between patients.
AI clinical menus at a glance
The AI tooling spreads across several clinical menus. This is where each piece lives:
| Menu path | What it manages |
|---|---|
| HMS Core → Clinical → Ambient AI → Recordings | Encounter audio recordings for the scribe. |
| HMS Core → Clinical → Ambient AI → AI Tasks → Scribe Tasks | Individual transcription / synthesis jobs. |
| HMS Core → Clinical → Ambient AI → AI Tasks → Edit Log | Clinician edits to AI drafts. |
| HMS Core → Clinical → Ambient AI → Configuration → Providers | The transcription / language model providers. |
| HMS Core → Clinical → Ambient AI → Configuration → Redaction Rules | Identifier-stripping rules. |
| HMS Core → Clinical → Ambient AI → Configuration → Scribe Templates | Note templates the scribe writes into. |
| HMS Core → Clinical → Ambient AI → Usage → Usage & Quota | Consumption and budget tracking. |
| HMS Core → Clinical → AI-First Clinical → Drafts | The draft cards for an encounter. |
| HMS Core → Clinical → AI-First Clinical → Decision Log | Accept / reject / edit decisions. |
Roles & access
AI drafting and acceptance are clinician actions — only the doctor commits a draft to the record. Provider configuration, redaction rules, quotas, CDS rule authoring and predictive-rule tuning are administrative/managerial tasks.