Method

How the forecast works

Sithi publishes the latest public data on flu, RSV, COVID-19 and norovirus in England and a forecast for the next eight weeks. This page explains where the data comes from, how the forecast is made, how well the method has done on past winters and where it falls short.

Data sources

• UKHSA data dashboard (ukhsa-dashboard.data.gov.uk): flu and RSV test positivity for England, flu hospital admission rates for England and each region, RSV admission rates, COVID-19 test positivity and weekly norovirus laboratory reports with UKHSA's five-season average.
• UKHSA weekly national flu and COVID-19 surveillance report data files (GOV.UK): flu and COVID-19 test positivity in hospital (Pillar 1) testing for the nine regions, RSV positivity for four larger areas and COVID-19 admission rates by region.
• Department for Education, Explore Education Statistics, "Pupil attendance in schools": pupil absence through illness by council area.

UKHSA updates its data on Thursday afternoons. Our server collects it automatically on Thursday evening and again on Friday morning. If a collection fails or the data does not pass our checks, the site keeps the last good data and shows its date. Contains public sector information licensed under the Open Government Licence v3.0. The data is reused with attribution to UKHSA and the Department for Education; no endorsement by either is implied.

The forecast in plain words

Each winter, flu and RSV tend to rise, peak and fall in a broadly similar shape. The forecast fits that shape to this season's data so far, using previous seasons as a starting guide for the height, width and timing of the peak. As more weeks of data arrive, the current season's figures count for more and the starting guide for less.

• One to three weeks ahead, the forecast is the average of two things: the latest figure held steady, and the fitted seasonal curve. This keeps short-range forecasts close to what is happening now.
• Four to eight weeks ahead, the forecast follows the fitted curve alone. This is why the forecast line can step up or down between the third and fourth week when a season is moving quickly.
• Hospital admissions are reported late and are usually revised upwards in later weeks. The forecast scales up the most recent weeks a little to allow for this, using the size of past revisions. Test positivity is barely revised, so it is not adjusted.
• Gaps and glitches. Missing weeks, zero placeholders and sudden unexplained drops are held back and logged rather than treated as real changes. The number held back is shown with each forecast.
• Ranges. Each forecast comes with an 80% range, worked out from the size of past forecasting errors. In roughly eight weeks out of ten the true figure should fall inside it. The ranges are wide, particularly early in the season and four or more weeks ahead.
• Norovirus is forecast for England only, using UKHSA's five-season average as the guide. Its data history is short and the forecast has not been tested on past seasons, so treat it as experimental.
• School illness absence. For each council area we compare pupil absence through illness with its usual level and project it two to four weeks ahead using the regional outlook.
• COVID-19 is shown as the latest data only. We do not forecast it.

How accurate has it been?

We tested the method on the three winters from 2023/24 to 2025/26. Each week's forecast used only the data exactly as UKHSA published it at the time, taken from 123 weekly data files, so the test reflects what could have been known on the day.

• Compared with simply assuming next week will be the same as this week, the seasonal curve method reduced the average error by between 29% and 41% at two to eight weeks ahead.
• The harder benchmark is the seasonal average: what usually happens in the same week. Against it, the current version's flu positivity forecasts four to six weeks ahead were about 10% more accurate overall. That average hides a wide spread: about 40% better in the unusually early 2025/26 season, but 25% and 6% worse in the two seasons before. For RSV the gain was about 8% and for flu admissions about 7%.
• The method called the week of the peak to within one week about 60% of the time, no better than the seasonal average.
• UKHSA's weekly report appears about ten days after the end of the week it covers. A forecast "two weeks ahead" of the latest data is therefore only a few days ahead of the date you read it, and an eight-week forecast gives about six and a half weeks of real notice.

From October 2026 every published forecast is recorded, with a time-stamp and a file fingerprint, before the outcome is known. Each is graded once its target week has two further weeks of data, against both the seasonal average and the "same as this week" benchmark. We will publish the results, including the weeks in which the forecast was wrong.

Limits

• The forecast is a statistical model of published surveillance data. It does not measure illness directly and cannot know about events that have not yet appeared in the data.
• Regional hospital admissions for 2026/27 come from a new, larger UKHSA collection, so comparisons with earlier seasons are indicative only.
• Regional Pillar 1 positivity has been published only since December 2024, so the guide from past seasons rests on little more than one season.
• Early in the season, the timing of the peak leans heavily on past seasons and is likely to move from week to week.
• School data covers term time only, is about two weeks old when published and has gaps at half-term and Christmas.
• Sithi's colour bands (low to very high) are our own and are not the same as UKHSA's activity levels.

Sithi is independent analysis of public data. It is not produced by, affiliated with or endorsed by the NHS, UKHSA or any government body, and it is not medical advice. For health concerns use NHS 111 online or call 111.

Open data

The figures behind the page are available as live.json (UKHSA data as collected) and forecast.json (the current forecast). You are welcome to reuse them with attribution to Sithi and to UKHSA under the Open Government Licence.