Where the figures come from
Each outbreak has one of two kinds of source, shown on its card.
| Read automatically | Curated by hand | |
|---|---|---|
| Source | The outbreak’s Wikipedia infobox, which cites the health ministries and WHO, or, where no infobox exists, the health authority’s own weekly report (West Nile virus in Greece: EODY; measles in the United States: the CDC; Candida auris in the United States: the CDC’s weekly notifiable-disease table). | Official situation reports, each with its source. |
| Covers | Headline totals and the chart. | Milestones, country and province tables, map bubbles and written summaries. |
| Updated | Checked twice a day (08:00 and 20:00 UTC; West Nile virus in Greece twice a day on its own schedule). An official weekly source (EODY’s report, the CDC’s measles page or its notifiable-disease table) adds a reading when a new week is published. | When we enter a report. It does not update on its own. |
| Card shows | When the source was last checked. | The date the figures were last verified. |
| How to trace it | Every reading links to the exact Wikipedia revision or report. The raw infobox or report text is archived with a SHA-256 fingerprint, so any number can be re-checked later. | The official report is linked on each entry. |
Hand-checked milestones are merged with the automatic readings for the headline totals and the chart. On a day with both, the hand-checked figure wins.
Countries that only the news has linked to an outbreak are labelled Unverified until an official source confirms them.
Checks before a reading is kept
| Check | Rule | Exception |
|---|---|---|
| No decreases | Confirmed cases, deaths and recoveries never go down. | None. |
| Largest jump | Confirmed cases or deaths may not rise by more than 25% over the previous reading. | Not applied when the previous reading is more than 7 days old. |
| Deaths and cases | Deaths may not exceed confirmed cases. | None. |
A weekly report is about a week apart from the one before, so the largest-jump check applies only to two reports at most 3 days apart (an extra mid-week report). Instead, a report is kept only when the totals in its text and in its summary table agree, and the table adds up. A week of the CDC’s notifiable-disease table is kept only when its states add up to their regions, the regions to the national row and the parts to the total. Those counts are provisional and the CDC says a cumulative count can go down as states reconcile their data; such a week is not stored, it is logged for a person to review.
A reading that fails a check is rejected and logged, and the previous figures stay on the page.
Historical import. When older history is imported, edits that break these rules (typos, vandalism, edits in progress) are dropped and the longest consistent run of readings is kept.
Reporting lag and the 7-day trend
Sources are updated in batches, so the most recent days often show no new cases simply because the report has not been entered yet. Measured naively, that looks like a sudden collapse. Every 7-day count therefore ends on the latest day the total actually moved, but never more than 3 days before the newest reading, so a real halt in cases still shows up. Days without a reading are filled in a straight line between their neighbours.
Each card leads with a plain-language result for the trend: it compares new confirmed cases in the last 7 days with the 7 days before. There are five possible results:
| Result on the card | When it shows |
|---|---|
| Cases are rising | The whole 95% range of the growth rate is above zero. |
| Cases are falling | The whole 95% range of the growth rate is below zero. |
| No clear rise or fall | The 95% range of the growth rate includes zero. |
| Too few new cases to judge a trend | Too few new cases in the 7 days to judge a trend. |
| Not enough data for a trend yet | Fewer than 5 readings in the last two weeks (2 for a weekly report), or less than 14 days of history. Shown instead of guessing. |
On an outbreak’s dashboard the same five results appear as a badge: Growing, Declining, Plateau, Unclear or Not enough data. The bar chart on each card shows new confirmed cases per 7-day period, newest on the right, and the result and counts beneath it describe the newest bar.
Rates use only the automatic readings (daily, or weekly for a weekly report), never the hand-entered milestones, so a change of source cannot create a false jump.
How the home page orders outbreaks
The home page is titled “Outbreaks we are tracking”, not “the worst outbreaks in the world”: it covers only the outbreaks we follow. They are ordered by a fixed rule. First, by status group, in this order:
| Status group | Meaning |
|---|---|
| 1.Active | Cases are still being reported and the outbreak has not been declared over. |
| 2.Under watch | A signal worth following that has not yet been confirmed as an outbreak. |
| 3.Waning | Transmission has slowed sharply; authorities are counting down to declaring it over. |
| 4.Declared over | Declared over by the health authorities. Kept as a permanent record. |
Then, within each group:
- Most new deaths in the last 7 days first.
- Ties are broken by most new confirmed cases in the same 7 days, then alphabetically.
- An outbreak whose source collects no deaths (Candida auris) counts as zero new deaths here.
- Outbreaks without enough history for a 7-day count come last in their group.
There is no editorial pinning and no combined severity score.
We will revisit the rule once more than about six outbreaks are tracked, and any change will be dated on this page.
Comparing outbreaks
Outbreaks are counted differently: case definitions, testing capacity and reporting delays vary by disease and by country, and some sources report suspected cases while others report only confirmed ones. Two totals side by side are rarely like for like. Each card shows its own source and as-of date; compare the direction of each trend rather than the size of the numbers.
Nursing-home data
Nursing homes is a separate layer: weekly measurements by state and inspection citations by facility, never ranked or counted as an outbreak. Every fetch is checked before anything is stored; a failed check stores nothing and is flagged for review. Each accepted extract is archived, compressed, with the SHA-256 checksum of its text, so every figure can be re-derived.
- NHSN counts: for every week and every count, the states and territories must add up exactly to the national row, and so must the ten HHS regions; vaccinated or hospitalised subsets can never exceed their totals; CDC's percentages must equal numerator ÷ residents; weeks must end on a Sunday. Rates per 100,000 use CDC's resident count for the same homes and week.
- NHSN dashboard rates: CDC's file repeats each rate up to three times, and every copy must agree; every week must have a national row; the chart must still say “per 100,000 residents”. These rates are not checked against the counts, because CDC computes them from a different extract and they differ slightly.
- CMS F880 citations: the rows received must equal the count the API reports; every row must be tag F880 with the release's processing date; codes, dates and flags must be valid; no row may repeat; and the release must not change while it is being read. “Last 12 months” means surveys within the year that ends on the release's processing date. “G or above” means scope/severity G to L: actual harm or immediate jeopardy.
What we do not do
- No forecasts. Every indicator describes the recent past and depends on how complete reporting is.
- No medical or travel advice. For decisions, follow WHO and your national health authority.
- No single “severity” score across diseases.
- The reproduction number (Rt) relies on a serial interval measured for a specific disease, so it is shown only where one is available; the values used are listed on the Data & API page.
Found an error? See corrections.