Loading this page
Sentinel turns FHIR R4 observations into NEWS2 scores and governed alerts, keeping acknowledgements, clinical orders and audit evidence in one live workflow.
Observations, alert state, clinical orders and acknowledgements are often split across systems or informal handoffs. Sentinel gives an entitled care team one explicit workflow: ingest, score, alert, acknowledge, order and acknowledge, with data age and missing NEWS2 parameters kept visible.
A colour change alone does not show whether anyone took ownership. Sentinel records raised and acknowledged states for danger alerts and clinical orders.
An apparently normal score based on old observations is not the same as a current assessment. The live board surfaces when the latest observation was recorded.
Governed actions write PHI-lean audit records into a per-tenant hash chain, so later removal or alteration can be detected by chain verification.
Four movements, one accountable workflow, from a FHIR observation to a governed alert, explicit acknowledgement and verifiable clinical order.
Hospital systems and device gateways can submit FHIR R4 Observation resources or Bundles through a clinic-scoped API key. Sentinel validates supported observations, deduplicates stable source IDs, records accepted readings, and reports anything it skips so partial coverage stays visible.
Sentinel builds the latest encounter snapshot and calculates NEWS2, including the configured SpO2 scale, plus optional clinic hard thresholds. Missing NEWS2 parameters are surfaced rather than silently scored as zero. An optional AI narrative may explain an already-raised alert, but it cannot change the score, severity, or alert decision.
A danger or critical breach raises an idempotent governed alert and updates the acuity-sorted live board. Team members can acknowledge alerts; authorised clinicians can issue clinical orders, and order acknowledgement is recorded separately. Automated calling is a disarmed, fail-safe integration point; it is not a live notification or order-capture channel.
Observation records, alerts, clinical orders and acknowledgements run through governed actions with PHI-lean audit projections. Audit rows are hash-chained per tenant scope: removing or changing a row makes chain verification fail from that point forward.
When a patient deteriorates
Lights up a red tile and waits to be seen.
Raises a governed danger alert, updates the acuity board, and tracks explicit acknowledgement state.
The audit trail
A timestamped log that can be edited after the fact.
Per-tenant audit rows carry sequential hashes, so deletion or alteration is detectable by chain verification.
If no one is at the screen
The alert sits unread until someone looks.
Keeps the alert in a raised state until acknowledged. Automated calling remains disarmed pending a complete dial integration; no call is represented as sent.
Before changing thresholds
Change the policy and learn from the live result.
Managers can replay a bounded historical window against alternate NEWS2 settings and review aggregate counts before a live change.
Interoperability
A bespoke integration project.
A documented FHIR R4 Observation and Bundle boundary, with deployment mapping and validation kept explicit.
What it actually is
A monitoring app bolted onto the ward.
A scoped module on the same governed platform as Concierge, Scribe, Central Command, and Hospital Core.
Sentinel keeps deterministic scoring separate from optional AI narrative. The narrative cannot alter NEWS2, severity or whether an alert is raised. Clinical orders require an authorised clinician, and their privileged audit entry must be committed before the order can be written.
Governed observation, alert, order and acknowledgement actions write PHI-lean rows into a sequential, tamper-evident audit chain.
Orders ride a fail-closed audit: the record is committed before the action. If the audit can't be written, the order doesn't happen.
Sentinel vital values, patient references, alert detail, patient names and order text use AES-256-GCM field encryption at rest. Authorised server routes decrypt only for authenticated, tenant-scoped views.
Managers can run rate-limited what-if replays over a bounded historical window. Stored and returned results contain aggregate counts only, with truncation and baseline-fidelity caveats kept visible.
Sentinel is a scoped module on the same governed platform as Concierge, Scribe, Central Command and Hospital Core. Tenant identity, governed actions and audit infrastructure are shared; product entitlements and rollout scope remain independent. Automated consultant calling is not part of Sentinel’s live workflow.
See the platformSentinel accepts supported HL7 FHIR R4 Observation resources and Bundles through a clinic-scoped public API. A deployment still validates source identifiers, coding, units, encounter mapping and data completeness. The standard boundary reduces coupling without pretending integration work disappears.
FHIR R4
Observation and Bundle ingestion
NEWS2
Deterministic encounter scoring
AES-256-GCM
Sentinel PHI field encryption
Counts only
Stored scenario replay results
Review active encounters by acuity and data age, acknowledge danger alerts, and issue governed clinical orders from one live board. Sentinel supports the response workflow; it does not replace bedside observation or clinical judgement.
Start with a scoped, entitlement-gated FHIR integration. Review tenant-bound access, encrypted clinical fields, the tamper-evident audit chain, and manager-only policy replay before widening rollout.
Evaluate a consistent observation, scoring, alert and acknowledgement model through a bounded pilot. Facility identifiers stay attached to encounters while access and configuration remain scoped to the entitled tenant.