AqtaBio Pandemic Risk Agent

Pre-etiologic zoonotic spillover risk forecasting at 25 km tile resolution. Backtest validation on the v0.1.0 historical cohort: AUROC up to 0.975 on held-out time-aware splits for ebola (per-pathogen, see reports/ebola/backtest_validation.json in the closed source). Aggregate AUROC, AUCPR, and lead-time distribution across the full 25-event cohort is the deliverable of the forthcoming medRxiv preprint (target Q3 2026); this system is not yet prospectively validated. Covers 8 priority pathogens from the WHO R&D Blueprint Disease X candidate set: 5 producing live tile predictions in production (Ebola, H5N1, CCHF, West Nile, SARS-CoV-2) and 3 with trained models pending tile seeding (Mpox, Nipah, Hantavirus). Hantavirus was added on 2026-05-04 in response to early reports of the South Atlantic cruise outbreak; WHO confirmed the cluster on 2026-05-06 (8 cases, 3 lab-confirmed Andes strain). Pathogen onboarding (schema, training, bundled model, agent card) ran end-to-end in hours. Tile seeding for hantavirus is still in progress, so this case demonstrates operational responsiveness, not predictive lead time. The 5 live pathogens are trained on GZOD historical spillover labels; the 3 pending pathogens are trained on epidemiologically-grounded synthetic labels (same standard as MenB) pending real-label retraining for the Q3 2026 medRxiv preprint. See `prediction_status` and the training-script citations on each entry. Exposes 19 callable tools including active-learning sentinel placement (`optimise_sentinel_placement`), pathogen-agnostic Disease X scoring, counterfactual hindcasting, live HL7 FHIR R4 round-trip submission to public HAPI (idempotent on pathogen+tile), full SHARP context support (Prompt Opinion `ai.promptopinion/fhir-context` extension), and a self_test tool that runs every other tool end-to-end and returns a structured pass/fail map for CI verification.

Skills

• list_pathogens
  List all monitored pathogens with SNOMED CT codes and geographic scope.
  discovery
• get_risk_score
  Current spillover risk score for a geographic tile with SHAP feature drivers. FHIR RiskAssessment optional.
  queryfhir
• get_hotspots
  Hotspot counts by severity tier (critical ≥0.9, high ≥0.7, moderate ≥0.5) for a pathogen. FHIR DetectedIssue optional.
  queryfhiraggregate
• get_risk_trend
  24-month risk trajectory for a tile. FHIR Observation Bundle optional.
  querytimeseriesfhir
• get_top_risk_tiles
  Highest-risk tiles ranked by score for a pathogen.
  queryranking
• get_top_risk_countries
  Returns the top countries by spillover risk. When a trained CountryClassifierModel sidecar is available for the requested pathogen + month, serves rankings from that model with a `scoring_method: 'trained_classifier'` flag. Otherwise falls back to a heuristic aggregation of tile-level scores (mean / max / upper bound of a bootstrap CI on the country mean), with saturation share flagged so callers can see when a country's rank is driven by clipped scores. Use for country-of-concern guidance (eg pre-positioning by WHO Africa).
  queryrankingaggregatecountrytrained-classifier
• get_system_status
  System health, data freshness, live vs demo mode.
  ops
• generate_outbreak_briefing
  Claude generates a PHO situational brief synthesising hotspots, top tiles, and SHAP drivers into an actionable narrative with 3 recommended actions.
  genainarrative
• explain_risk_drivers
  Claude translates SHAP values into plain-English causal narrative for a tile, with 2-3 specific recommended actions.
  genaiexplainability
• retrospective_validation
  Returns the recorded retrospective attestation for a historical spillover event paired with the publicly verifiable WHO / ECDC / national notification date. Backtest provenance, AUROC and AUCPR are surfaced in the `cross_check` field. Not a live model recomputation.
  validationevidence
• get_multi_pathogen_hotspots
  Detects syndemic convergence: regions and time windows where multiple pathogens simultaneously cross HIGH+ severity, flagging response infrastructure saturation risk.
  syndemicaggregate
• generate_fhir_bundle_for_pho
  Complete HL7 FHIR R4 transaction Bundle (RiskAssessment + DetectedIssue + 12×Observation) with per-entry POST requests, ready to submit to any FHIR server.
  fhirintegration
• get_disease_x_risk
  Pathogen-agnostic pre-spillover risk score addressing the WHO R&D Blueprint's Disease X priority. Aggregates per-pathogen risks into a single 'any zoonotic emergence' signal for the unknown pathogen of the next pandemic.
  disease-xblueprintnovel
• get_hindcast
  Counterfactual timeline analysis. Given a recorded retrospective attestation, returns the actual outbreak timeline alongside an illustrative counterfactual: what intervention window would have been available if a public-health responder had acted N days after the threshold-crossing signal. Honest about caveats; no claim of cases averted.
  counterfactualvalidationnovel
• submit_to_hapi_fhir
  Live HL7 FHIR R4 round-trip: builds a RiskAssessment for a tile/pathogen/month, POSTs to the public HAPI FHIR test server, and returns the assigned resource URL plus HAPI's HTTP status. Makes the 'FHIR round-trip tested' claim a callable proof; anyone can fetch the resource back to verify conformance. Population-level risk only, no PHI.
  fhirinteropverifiable
• get_patient_local_risk
  SHARP-aware patient-local risk. Reads the Prompt Opinion `ai.promptopinion/fhir-context` block (patient_id, fhir_server, access_token), fetches the Patient resource via SMART-on-FHIR, derives the home tile from address.country, and returns AqtaBio's population-level spillover risk for that area. PHI minimisation: only country is retained. Designed to be invoked from a clinician's Prompt Opinion workspace.
  sharpfhirpatient-contextsmart-on-fhir
• emit_riskassessment_to_ehr
  SHARP-aware EHR write-back. Takes the SHARP-propagated bearer token from the clinician's EHR session and POSTs an AqtaBio FHIR RiskAssessment resource to the same FHIR server, attached to the patient reference. Demonstrates round-trip context propagation: the platform's promise of bridging EHR credentials without bespoke token handling is verifiable end-to-end.
  sharpfhirwritebacksmart-on-fhir
• handoff_to_triage
  A2A v1.0 handoff. Takes a FHIR RiskAssessment from the surveillance side of AqtaBio and returns a FHIR Task that hands the matter to a clinical triage specialist agent. The Task carries a deterministic risk-band action (notify / surveil / routine) plus a disclaimer in note that the mapping is not clinical decision support; a public health officer must approve before any operational step. The triage specialist agent card is exposed at /.well-known/triage-agent.json.
  a2ahandofffhirtasktriage
• self_test
  Runs every other tool with sane default arguments and returns a structured pass/fail map. Lets CI / pre-deploy / post-deploy check that all 18 working tools execute without exception. Catches dangling references and missing pathogen branches that would otherwise only surface in a clinician's workspace.
  self-testciops

How to call

https://qjtqgvpd9s.eu-west-1.awsapprunner.com/mcp