Outbreak investigation
Turn an episode cohort into candidate exposure links and candidate outbreak clusters from case timing, location and contact evidence, and pathogen genomic relatedness — then adjudicate them. Every result is a hypothesis for investigator review, never proof of transmission or a confirmed outbreak.
Research question
Which episodes are plausibly linked by shared exposure and pathogen relatedness, do they form candidate clusters, and which clusters warrant investigation?
Who this is for
- Hospital epidemiology researchers reconstructing ward transmission hypotheses
- Public-health research teams screening for candidate clusters
- Investigators who need genomic and epidemiological evidence combined, with adjudication
Data requirements
| Data | Required | Purpose |
|---|---|---|
| Infection episodes with index dates | Yes | Timing gate and cluster nodes |
| Exposures (locations + windows, contacts, events) | Yes | Candidate exposure links |
Linked results with sample_id | Recommended | Map episodes to sequenced isolates |
| Microbiology relatedness run | Recommended | True pairwise SNP distances for genomic support |
| Linked Microbiology lineage assignments | Optional | Lineage frequency, replacement, and growth-advantage tracking |
| Exposure locations tagged as ward/ICU/facility/unit/room/bed | Optional | Facility movement and co-location investigation |
Workflow
Model episodes + exposures → Exposure network (+ threshold sensitivity)
→ Adjudicate candidate links
→ (Microbiology relatedness run) → Cluster investigation
→ Adjudicate candidate clusters
→ Lineage dynamics / Facility investigation (optional) → AI interpretation
Step 1 — Exposure network
The Exposure network analysis builds candidate links from recorded contacts, shared locations with overlapping time windows, and shared events, and groups episodes into connected components. Link a Microbiology relatedness run to overlay true pairwise SNP distances (otherwise a coarse shared-lineage proxy is used). An interactive graph shows components and genomically-close links. A network sensitivity sweep re-runs the exposure network across a range of contact-window and genomic-distance thresholds so you can see how candidate links change before committing to one setting.
Step 2 — Adjudicate exposure links
Accept, reject, or mark uncertain each individual candidate exposure link with a recorded rationale, separately from cluster-level adjudication.
Step 3 — Cluster investigation
Cluster investigation combines case timing (max serial days), location/contact evidence, and pathogen relatedness (a linked Microbiology relatedness run) into candidate clusters, persisted for review. An edge requires an epidemiological link and, when relatedness is supplied, genomic distance within threshold.
Step 4 — Adjudicate clusters
Promote, reject, or reset each candidate cluster with a recorded rationale. Adjudicated clusters are preserved across re-runs, and their episode groups are not re-proposed as duplicate candidates.
Step 5 — Lineage dynamics and facility investigation (optional)
Lineage dynamics tracks per-period lineage frequency from linked Microbiology lineage assignments, flags replacement events (a new dominant lineage displacing the prior one), and ranks lineages by a log-frequency growth-advantage slope — a descriptive signal, not a causal fitness estimate. Facility investigation reconstructs per-participant ward/facility movement paths from exposure records and flags episode pairs with overlapping facility windows at the same location as healthcare-associated co-location hypotheses, never proof of transmission.
Expected outputs
- Candidate exposure links with evidence types, SNP distances, and components
- Threshold-sensitivity results across contact-window and genomic-distance settings
- Link-level adjudication decisions (accepted / rejected / uncertain) with rationale
- Candidate outbreak clusters with evidence, genomic support, and status
- Cluster-level adjudication decisions (outbreak / rejected / candidate) with rationale and timestamps
- Lineage frequency-by-period, replacement events, and growth-advantage ranking (when Microbiology lineage results are linked)
- Facility movement paths and co-location pairs (when facility-tagged exposures are recorded)