Map

Column

Map

Data last updated on Thu Dec 03 16:06 GMT 2020. We define an area to be a hotspot X if weekly reported cases per 100,000 population exceed X. For past weeks we compare the reported cases to the threshold. For future weeks, we give probabilities based on our model, which assumes a situation in which no change in interventions (e.g. local lockdowns) occur. To define weeks we use specimen dates, ie the day on which tests are taken. We consider an area to have increasing new infections if our model estimates that the reproduction number R is greater than 1 with probability of at least 90%. “Likely increasing” indicates a probability between 75% and 90%. Decreasing and likely decreasing are defined analogously, but consider R less than 1.

Table

Data last updated on Thu Dec 03 16:06 GMT 2020. We define an area to be a hotspot X if weekly reported cases per 100,000 population exceed X. To define weeks we use specimen dates, ie the day on which tests are taken. Median and 90% credible interval for weekly cases and Rt can be viewed via the "Column Visibility" button. Colour code for P(hotspot): 0.00 – 0.05 0.05 – 0.15 0.15 – 0.25 0.25 – 0.5 0.50 – 0.75 0.75 – 1.00. Colour code for P(R>1): 0.00 – 0.1 0.1 – 0.25 0.25 – 0.75 0.75 – 0.90 0.90 – 1.00.

Details

Column

Model description

The results on this page have been computed using epidemia 0.6.0. Epidemia extends the Bayesian semi-mechanistic model proposed in Flaxman, S., Mishra, S., Gandy, A. et al. Nature 2020.

The model is based on a self-renewal equation which uses time-varying reproduction number \(R_{t}\) to calculate the infections. However, due to a lot of uncertainty around reported cases in early part of epidemics, we use reported deaths to back-calculate the infections as a latent variable. Then the model utilizes these latent infections together with probabilistic lags related to SARS-CoV-2 to calibrate against the reported deaths and the reported cases since the beginning of June 2020. A detailed mathematical description of the original model can be found here. The model used in this website is an evolution of this original model, incorporating randomness in the infections to account for areas with small number of infections.

\(R_{t}\) for each local authority is parameterized as a linear function of the \(R_t\) for its region/nation as a whole (which we fit too), and a random effect specific to the local authority for each week over the course of the epidemic. The weekly random effects are encoded as a random walk, where at each successive step the random effect has an equal chance of moving upward or downward. For regions/nations, only a region-specific weekly random walk is used.

Model Changes

  • [Fri Oct 09 2020] We have now included survey data from ONS and REACT to help us better estimate the fraction of infections that ens up as deaths or cases.

Limitations

  • Predictions on this page assume no change in current interventions (lockdowns, school closures, and others) in the local area beyond those already taken about a week before the end of observations.
  • An increase in cases in an area can be due to an increase in testing. The model currently does not account for this.
  • Each area (local authority) is treated independently apart from the overall Rt estimate for its region. Thus the epidemic in a region is neither affected by nor affects any other region. It also does not include importations from other countries.
  • The population within an area is considered to be homogeneous, i.e., all individuals are considered equally likely to be affected by the disease progression.

Column

Explanation of terms

  • Dates for cases are specimen dates, ie the date on which tests were taken.
  • An area is defined as hotspot if weekly cases per 100,000 population exceed 50.
  • Change in new infections gives the probability (chance) of new infections increasing or decreasing in the local authority.
  • Columns with column name as Weekly Cases per 100k [Date], in the Table tab, show the reported number of cases per 100,000 population for the week specified by the Date.
  • Columns with column name as P(hotspot) [Date] shows the estimated probability of an area being a hotspot for the week specified by the Date.
  • Column P(R>1) [Date] gives the probability of time varying reproduction number being greater than 1 for the week specified by the Date.

Data sources

  • Daily cases data for all local authorities in England is taken from UK gov site
  • Daily cases data for all local authorities in Wales is taken from Public Health Wales COVID-19 dashboard
  • Weekly deaths data for all Local authorities in England and Wales comes from ONS
  • Daily cases and death data for Scotland is taken from Public Health Scotland
  • Daily cases and death data for Northern Ireland is taken from Department of Health Northern Ireland
  • Lower tier local authority boundaries: Office for National Statistics licensed under the Open Government Licence v.3.0
  • Daily infections from 14th August 2020 taken from ONS Infection Survey
  • Total Infection till end of June 2020 from REACT

Data last updated on Thu Dec 03 16:06 GMT 2020.

FAQs

Column

Frequently Asked Questions

How often do you update the website?

We aim to update the website every evening after the release of the public data. Updates are usually available by the morning.

What do you mean by "Hotspot X"?

Hotspots are areas that have more than X reported cases per 100,000 population.

What do you show on the map?

In the map, we show the probability of an area being a hotspot in the next one, two and three weeks. The projections for hotspot assume no change in interventions and human behaviour since a week before the last observed data.

What does "P(hotspot X)" mean for a given week in the Table?

This is the probability of an area getting more than X cases per 100,000 population for the defined week.

What does "Change in new infections" show in the Map?

Change in new infections shows how confident our model is about the time varying reproduction number (\(R_t\)) being greater than or less than one. We consider an area to have increasing new infections if our model estimates that the reproduction number R is greater than 1 with probability of at least 90%. Likely increasing indicates a probability between 75% and 90%. Decreasing and likely decreasing are defined analogously, but consider R less than 1.

What does "P(R>1)" mean for a given week in the Table?

This corresponds to the 'change in new infections' option on the Map. This is the probability of the reproduction number for the given week exceeding 1, which is the threshold of determining whether the number of cases will continue to grow (R>1) or not (R<1). For example, P(R>1)=50% means that there is an equal chance of the rate of new infections increasing as there is of it decreasing.

Which is the most significant figure - the probability of Rt exceeding one, or the percentage risk of the borough becoming a hotspot?

Becoming a hotspot is arguably more important, as it combines both the current state of the epidemic as well as it's future direction.

Area X has a high probability of becoming a hotspot. Should a (local) lockdown be imposed?

Whatever is on this website should be considered in conjunction with other more detailed evidence before making decisions. At best the website should be used to inform which areas to look at in more detail.

Area Y has a high probability of R being above 1. Should I be concerned?

Generally, we think that the probability of becoming a hotspot is a more useful measure. R only gives an indication whether or not the number of new infections is increasing (R>1) or decreasing (R<1). If R is indeed above 1, and case numbers are low, then this could come down through effective track and trace and/or medical support, robust testing and behaviour change. It is more of a concern if case numbers are already higher.

In local authorities with high R, why is the R dropping in the projections?

The model for local authorities assumes that individuals, once infected, become immune. The effective reproduction number R gets adjusted in proportion to the number of individuals not immune. In a relatively small population, such as a local authority, these effects can be strongly visible in the projections.

Why are the infections plots sometimes "jagged"?

The reproduction number in the model R can only change every week, giving jagged effects in estimated infections if there are large changes in the R.

The model is projecting a decrease in a certain area. Is it possible that his is due to the epidemic only going through a subpopulation (e.g. students)? What will most likely happen afterwards?

The model assumes the population to be homogeneous, so it does not treat subpopulations separately. Hence, it is possible that a quick rise in cases in a subpopulation is coming to an end, overlaying what happens in the general population (where there might be a smaller rise). Eventually, the trend of the general population will likely reassert itself, so the decline could turn into an increase again.

Downloads

Downloads

Latest Model Estimates

The latest model estimates can be downloaded here.

Data Description

The format is currently experimental and subject to change.

There are 9 fields in the csv. The names and details of each field are as follows:

  1. area: Name of the area for which predictions are made. This includes local authorities, regions, nations and a UK-wide estimate.
  2. type: Type of the quantity that is being projected. There are currently nine different values this cell can take:
    1. Cases: Number of total cases (positive tests) expected in a given period.
    2. P(Cases>x) Probability of cases being greater than x in a given period.
    3. R The value of the time varying reproduction number Rt.
    4. P(R>1) Probability of Rt being greater than 1 in a given period.
    5. Infections: Estimated number of infections in a given period.
  3. value: Predicted median value of the quantity being estimated. In case of probabilities it is a single value.
  4. CIlow: The lower end of the credible interval for the value being estimated. This field is NA for types where this does not apply (e.g. probabilites).
  5. CIup: The upper end of the credible interval for the value being estimated. This field is NA for types where this does not apply (e.g. probabilites).
  6. period_start: The starting date of the projection period.
  7. period_end: The end date of the projection period.
  8. coverage: The coverage probability of the credible interval, if present. For example, a value of 0.9 represents that the credible intervals nominally contains the true value with 90% probability.
  9. data_updated: The date on which the data was downloaded from various public sources listed here.

Rt Estimates

The estimates of Rt over the time period the model is fitted to can be downloaded here. This is currently an experimental output - please treat with care

Data Description

The format is currently experimental and subject to change.

There are 6 fields in the csv. The names and details of each field are as follows:

  1. area: Name of the area for which predictions are made. This includes local authorities, regions, nations and a UK-wide estimate.
  2. date: The date projected value is valid for.
  3. CIlow: The lower end of the credible interval for the Rt value being estimated.
  4. Rt: The median of the Rt value being estimated.
  5. CIup: The upper end of the credible interval for the Rt value being estimated.
  6. coverage: The coverage probability of the credible interval. For example, a value of 0.9 represents that the credible intervals nominally contains the true value with 90% probability.

Please cite us if you use these downloads. DOI

About

Column

Imprint

Publisher

Authors: Swapnil Mishra1, Jamie Scott2, Harrison Zhu2, Neil Ferguson1, Samir Bhatt1, Seth Flaxman2, Axel Gandy2

1MRC Centre for Global Infectious Disease Analysis, Jameel Institute for Disease and Emergency Analytics, Imperial College London

2Department of Mathematics, Imperial College London

Website Development: Aided by Fabian Valka

This research was partly funded by the The Imperial College COVID19 Research Fund.

DOI

Licence

The results, maps and figures shown on this website are licenced under a Attribution 4.0 International (CC BY 4.0) Licence.

Column

Privacy Policy

Privacy Policy

This website uses privately hosted, fully anonymized, web-analytics tools so that we can optimize the website according to your needs. No personally identifiable information is stored in the tracking process. For more details please refer to the Matomo anonymization in web analytics description. Our web-analytics tools also don't store cookies and respect do not track settings.

This website uses Google Fonts hosted on the Google CDN which is essential for ensuring the websites performance.
Please refer to the Google Font developers FAQ regarding privacy.

For further information please refer to the Imperial College London privacy policy.

Column

Contact

Contact

For any enquiries please contact:

Corresponding Author
Axel Gandy
a.gandy@imperial.ac.uk

External Relationships and Communications Manager
Sabine L. van Elsland
+44 (0)20 7594 3896

s.van-elsland@imperial.ac.uk