Sithi Illness Index

Methods

This page sets out how the index and the household survey work, in the spirit of the British Polling Council's disclosure rules: who runs it, what is asked, how answers are weighted, how many people took part and how precise the figures are. Sithi is not a member of the British Polling Council, whose membership covers political and social polling, but we follow the same rules on disclosure voluntarily.

Who runs it

Sithi is a trading name of Neurithmetic Ltd, registered in England & Wales no. 15177502. Registered office: The Courtyard, Shoreham Road, Upper Beeding, Steyning, West Sussex, BN44 3TN. The index is designed and run by Sithi. It is not funded by any health body, pharmaceutical company or test supplier. 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.

What the index measures

Two weekly measures for the UK, each scaled so that 100 is a typical winter week: the average week from October to March over the three winters 2023/24, 2024/25 and 2025/26. A reading of 150 means half as much illness again as a typical winter week; 50 means half as much.

• Coughs, colds and flu: respiratory illness, including flu, RSV and colds.
• Sickness bugs: vomiting and diarrhoea, mostly norovirus.

Readings are described in words as low (below 50), typical (50 to 139), high (140 to 199) and very high (200 or more). In the three reference winters about a quarter of weeks were low and about one in ten were very high. These bands are Sithi's own and are not UKHSA's activity levels.

Release

Every Tuesday at 10:00 UK time, covering the Monday to Sunday just ended. The server builds the figures earlier that morning and publishes them at 10:00. If any check fails, nothing is published and the previous week stays on the page with its date. Figures, once published, are not changed except through the corrections log.

How each figure is worked out

Coughs, colds and flu

From method version 2.0 this figure is an estimate of what UKHSA's laboratory and hospital measure for the week just ended will show when UKHSA publishes it on the Thursday. It is built in four steps.

1. Start from last week's UKHSA figure: the average of flu test positivity, RSV test positivity and the flu hospital admission rate for England, each relative to its average in a typical winter week.
2. Carry on part of last week's change. If the UKHSA figure rose by 20% last week, this step adds a rise of about 10%.
3. Move with this week's searches. We use Google searches in England for "chest infection", "chesty cough" and "cough" in the week just ended. If they rose by 10% on the week before, this step adds a rise of about 7%.
4. A small pull towards the level searches suggest. If searches are high for the time of year compared with the UKHSA figure, the estimate is nudged up a little, and down if they are low.

The three factors (0.55, 0.74 and 0.026, applied on a log scale) were fitted to the October to March weeks of the four winters from 2022/23 to 2025/26 and are fixed until the next review. The estimate may not move by more than a factor of two in a week. If searches cannot be collected, steps 3 and 4 are left out, step 2 uses a factor of 0.71 and the page says so. Once the household survey is in the index it makes up 40% of this figure and the four steps the other 60%. The index page shows how many points each step added in the latest week.

Version 1.0 also used UKHSA's syndromic bulletins (NHS 111, GP and emergency department) for this figure. In testing they did not improve the estimate once last week's UKHSA figure and searches were included, so they are no longer used for it.

Sickness bugs

Unchanged from version 1.0. The inputs are combined with fixed weights, shared out over the inputs available that week.

Weights and factors are fixed in advance and are not changed quietly: any change is announced in the corrections log. NHS England's winter situation reports (beds closed by diarrhoea and vomiting) will be considered as an extra sickness bug input once they resume, expected in early December. COVID-19 is not part of the index because it does not follow a regular winter season; its latest figures are on the forecast page.

UKHSA publishes on Thursdays, so on release day its latest figures are a week old (two weeks for norovirus). The survey and the searches cover the week just ended. The index page says which week each input refers to.

The household survey

Who can take part

Adults aged 18 or over living in the UK, answering for their own household. Anyone can join through sithi.net/survey; it is a volunteer panel, not a random sample.

Questions asked once, when you join

• Your age: 18 to 29, 30 to 44, 45 to 64, 65 to 74, 75 or over.
• Your sex (optional): female, male or prefer not to say.
• The letters at the start of your postcode.
• How many people live in your household, including you? Counts for 0 to 4, 5 to 17, 18 to 64 and 65 or over.

Questions asked each week (one screen)

• In the last 7 days, has anyone in your household had any of these symptoms: fever or chills, tiredness or feeling generally unwell, headache, muscle or body aches, cough, sore throat, shortness of breath, a runny or blocked nose, vomiting or diarrhoea? Yes or no.
• How many people in your household have had any of these symptoms in the last 7 days?
• For each person who was unwell: age group (one of the household age groups); which symptoms? tick all that apply. (the ten symptoms listed above); and did the symptoms come on suddenly, over a few hours? Yes, no or not sure.
• Did anyone who was unwell test positive for COVID-19 or flu? Tick all that apply. No test taken, Tested negative, Tested positive for COVID-19, Tested positive for flu, Not sure.
• Did anyone miss work, school, college or nursery because of these symptoms? Yes or no.
• Did anyone contact a GP, NHS 111 or a pharmacy about these symptoms? Yes or no.

Definitions

• Flu-like illness uses the European Centre for Disease Prevention and Control (ECDC) case definition, as UKHSA's Flusurvey does: symptoms that came on suddenly; at least one of fever or chills, tiredness or feeling generally unwell, headache or muscle aches; plus at least one of cough, sore throat or shortness of breath.
• Any cough or cold: cough, sore throat, shortness of breath or a runny or blocked nose.
• Sickness bug: vomiting or diarrhoea.

Fieldwork

Each survey week runs from Monday 00:00 to Sunday 23:59 UK time and asks about the previous seven days. A household can answer once a week; a second answer in the same week replaces the first.

Weighting

Volunteer panels tend to include more women and older people than the population as a whole. We weight responses so that the panel matches the population on four things at once, using raking (iterative proportional fitting):

• age group of the person answering (18 to 29, 30 to 44, 45 to 64, 65 to 74, 75 or over), from ONS mid-year population estimates population estimates for adults;
• sex, from the same estimates (people who prefer not to say are left out of this step only);
• area: five broad areas (North, Midlands and East, London, South, plus Wales, Scotland and Northern Ireland combined); the nine English regions and three nations separately once a week has at least 2,000 responses and 30 in every area;
• household type, from Census 2021 (England and Wales): one person (30% of households), with dependent children (28%) and adults only (41%).

No household's weight may exceed five times the average; the raking is repeated with this cap until the weights settle. Each week we publish the number of responses and the effective sample size (Kish's formula), which shows how much precision the weighting costs.

Postcode areas that straddle a border (for example SY, CH and TD) are placed in the region that holds most of their population.

Margins of error

The margin of error shown with each survey figure is the 95% range for the household share, based on the effective sample size. For a volunteer panel this is a guide rather than a guarantee, as with Flusurvey and commercial online panels. As a rule of thumb, allowing for weighting:

When the survey enters the index

The survey is shown on its own from 30 weekly responses and counts towards the index only in weeks with at least 300. Because it is new, it has no winter average of its own yet. Its scale is set by matching its average over its first four weeks with at least 300 responses to the rest of the index over the same weeks ("panel: calibrating"). From the fifth such week it carries its 40% weight in both figures. After a full winter it will be rescaled to its own winter average, and that change will be logged.

What we publish and what we hold back

Only weighted totals are published. Any figure for an area with fewer than 10 responses is suppressed, and no rates at all are shown in a week with fewer than 30. We never publish individual answers.

How well the index tracks UKHSA data

We rebuilt the index without the survey for every week from October to March, using only what would have been available on each Tuesday, and compared it with UKHSA's laboratory measure for the week itself as later published. To keep the test fair, the method for coughs, colds and flu was chosen using the winters 2022/23 to 2024/25 only, testing each of those winters with factors fitted on the other two. The 2025/26 winter was held back and used once, at the end, with factors fitted on the three earlier winters. Lower error is better. The full search, with every option tried and its score, is in our records.

In plain terms, version 2.0 roughly halves the error of simply repeating last week's UKHSA figure (8.1 against 16.7 points in 2025/26) and got the direction of the weekly change right in 23 of 26 weeks. Most of that gain comes from UKHSA's own figures: assuming last week's change carries on already does much better than repeating last week's figure. Searches then cut the average error further (10.7 to 8.1 points in 2025/26 and 13.4 to 8.9 in the earlier winters). In 2025/26 they did not improve on that simple rule for direction, which was also right in 23 of 26 weeks; over the earlier winters they did (87% against 83%). The largest errors come at turning points: in 2025/26 the index peaked a week after UKHSA's measure and overshot it (260 against 238). Version 1.0 was further from the outcome than last week's figure, which is why it was replaced (see the corrections log).

For sickness bugs only one winter of norovirus data is available (2025/26), so there is no independent test. On a Tuesday the latest settled norovirus figure is two weeks old. In 2025/26 the sickness bugs figure had an average error of 34.6 points against 30.8 for that two-week-old figure carried forward, so it was not better. From April to September 2026, which was not used to set the method, it was slightly better (10.7 against 12.3). We tested alternatives using searches, UKHSA's five-season average and the syndromic data. None was clearly better in both periods, so the method is unchanged. Version 1.0 of this page compared the sickness bugs figure with last week's norovirus figure, which is not available on a Tuesday; that comparison has been corrected.

The household survey is the one input that measures the week just ended directly. For coughs, colds and flu it should help most at turning points, where last week's change and searches are slowest to react. For sickness bugs it is the only input that is neither two weeks old nor driven by news coverage. The test will be repeated at the end of this winter and the results published here, including where the index was wrong.

Limits

• The survey is a self-selected panel. Weighting corrects the balance of age, sex, area and household type but cannot correct for everything, for example people who are more health-conscious being more likely to join.
• Searches rise with news coverage as well as with illness. For coughs, colds and flu they only adjust last week's UKHSA figure and never replace it; for sickness bugs they carry a fifth of the weight. Google Trends figures are samples and can differ slightly from one download to the next. In testing, random changes of 10% to the search figures typically moved the 2025/26 average error by about half a point.
• UKHSA's first published figure for a week can be revised later (flu admissions are usually first published a little low). The back-test uses figures as later revised, so it may flatter both the index and last week's figure slightly.
• The laboratory and syndromic inputs cover England. The survey covers the whole UK.
• Syndromic levels (used for sickness bugs) are compared with UKHSA's own baselines, which we cannot see in full; the scoring is a simple summary of UKHSA's published table.
• Nothing in the index is a diagnosis or advice. The survey never tells anyone what they have.

Reuse

The Sithi Illness Index figures are published under the Creative Commons Attribution 4.0 licence: please credit "Sithi Illness Index, sithi.net". The UKHSA data used is reused under the Open Government Licence v3.0. Contains public sector information licensed under the Open Government Licence v3.0. Google Trends data is credited to Google Trends. Neither UKHSA nor Google endorses Sithi.

Method version 2.0 (24 October 2026). Questions about the method: hello@sithi.net.