Early Warning
Outbreak-grade detection on top of routine surveillance: proportion-drop tests with FDR control, weekly-count CUSUM, 14-day cluster scan and MDRO-phenotype CUSUM. Every warning is a statistical hypothesis for investigation — never a confirmed outbreak.
Alert budget
Loading the alert-budget baseline…
Warning feed
Rolling 4-week window · ranked by evidence score
Running the detection stack — the first pass can take ~30 s while the config backtest warms up.
Transmission map
Schematic hospital layout — node tint = current MDRO prevalence · ring = warnings live in the selected week · edge width = patients transferred. Click a ward to filter the page.
Layout is schematic (role-positioned), not geographic — the dataset carries no coordinates. Flows are reconstructed from repeat cultures (same patient, different ward, ≤30 days); a warning ring means the detectors fired that week — a hypothesis for investigation, not a confirmed outbreak.
MDRO weekly prevalence
Loading MDRO prevalence…
MDRO by ward
Loading per-ward MDRO burden…
Recent MDRO isolates
Loading recent MDRO isolates…
Patient pathways
Reconstructing patient transfers…
Does the detector work?
Deterministic backtest of the detection stack against injected outbreak scenarios and an outbreak-free baseline.
The full harness replays labelled outbreaks through every detector and measures detection latency, false alarms and PPV. It takes ~20 s on the first run — run it on demand.