Covid in winter 2020, a worst-case scenario

In pandemics, UK governments plan their response based on an unpublished ‘worst-case scenario’. Below, we are publishing Sage’s classified report for the winter of 2020. It’s intended not as a prediction, but an example of how bad things might get and what HM Government should prepare for. It was published in July 30 and envisaged cases growing in November. Cases started rising in September, so the figures will certainly have been revised. But this shows an example of the thinking behind UK government policy. Its assumption is that restrictions stay in place until March 2021.

Reasonable worst-case planning scenario – 30 July 2020

Purpose: To help local planners prepare for the impact of COVID-19, this document sets out a reasonable worst-case planning scenario as agreed by SAGE (Scientific Advisory Group for Emergencies) on 30 July 2020. This reasonable worst-case scenario (RWCS) is available for the whole of the UK, for devolved nations, and the seven NHS regions of England. The differences between the scenarios lie in the assumptions made about rate of epidemic growth as a result of relaxing and re-implementing behavioural and social interventions (BSIs).

It should be noted that this is a scenario, not a prediction. The nature and precise timings of any peaks in infection and, in particular, demand on healthcare are subject to significant uncertainty. The scenarios are sensitive to initial conditions and any increase in the starting estimates of numbers of infections, hospitalisations, or deaths could lead to a larger peak.

These assumptions will be kept under review and amended as the scientific and medical advice develops, and implications of the current measures are further understood.

For the RWCS, deaths, ICU occupancy, hospital admissions and new infections are modelled from July 2020 until March 2021. The scenario modelled incidence continuing as per current trends until the end of July 2020, with all non-household contacts assumed to be constant with current levels. Incidence is then assumed to double once by the end of August 2020, and double again during the first two weeks of September. At this point, social contacts are reduced that reduce R to approximately 1, keeping infection levels steady until the end of October. Two-week doubling times return throughout November (i.e. incidence quadruples through November), after which policy measures are put in place to reduce non-household contacts to half of their normal pre-March 2020 lockdown levels, while all schools contacts are assumed to be maintained. These measures are sustained until the end of March 2021.

This is only one of several possible scenarios that could lead to a reasonable worst case, all of which could be different and challenging in their own ways; this RWCS has a difficult autumn followed by a large winter peak. It is a reasonable scenario to plan for, however, planners will need to be flexible and be conscious of the limitations of relying on a single scenario.

SAGE provides scientific advice to government. It does not make decisions on what scenario government should be planning for. The Cabinet Office Civil Contingencies Secretariat and COVID-19 Taskforce currently advises that HMG should plan based on the RWC scenario below.

Do not share without explicit approval from Cabinet Office.

Annex A: RWC Scenario Graphs UK

Annex B: Estimating additional deaths to expand the RWCS

RWCS Assumptions

The RWCS is based on a mitigated second wave of the epidemic, however the changes in the rate of growth of the disease from July to November have been modelled based on incidence doubling times, following which, measures are implemented that lead to a reduction in non- household contacts. This profile of increasing incidence to the end of November 2020, was agreed by SPI-M-O co-chair in collaboration with SAGE and Cabinet Office Civil Contingencies Secretariat and COVID-19 Taskforce. No specific assumptions as to what these measures may be were made.

The model makes a variety of assumptions including:

• Mortality in hospitalised COVID-19 patients treated with dexamethasone is reduced by 17%, based on data from the RECOVERY study.

• Case isolation, household quarantine and contact tracing were assumed to reduce transmission outside of the households by 40%.

• Immunity acquired at any point is maintained over the course of the scenario.

The RWCS model explicitly estimates incidence and subsequent metrics based on the community and hospital transmission. From the first wave of the epidemic, it is known that care homes accounted for a significant proportion of deaths, and their effect needs to be considered in this revised reasonable worst-case scenario as far as practicable. A different model suggests that the division between deaths in care homes and hospitals accounted for approximately a one third to two thirds split respectively during the first wave of the UK epidemic; this model now estimates that, in a second wave, the split would likely be one fifth to four fifths. This means a 25% uplift in incidence has been applied to the RWCS output, in order to represent care homes as far as possible. 25% has been chosen as lower incidence in care homes is anticipated than that seen in the first wave due to a smaller susceptible population and improved mitigations but more pessimistic than those estimated in this other model.

NOTE: This modelling has been performed at the UK, nation and NHS England region level and does not necessarily reflect the variability that might be observed at a more local level. Care must be taken when applying this scenario and its data to smaller geographies as there will be significantly more variability at more local scales.

SAGE RWCS planning assumptions – 30 July 2020

The data in tables 1, 2, and 3 provided information that underpins the model that has been chosen by SPI-M and SAGE to provide the RWCS. This model makes various assumptions based on the data available about the how long it takes to progress from one stage of time in hospital e.g. ICU admission to another e.g. discharge, including how this differs by age groups. Table 4 provides headline data for the key metrics (deaths, hospital admissions, peak ICU bed occupancy) across the time frame of the RWCS.  

Table 1: Severity estimates for stages of COVID-19

Table 2: Average length of stay for COVID-19 hospitalisation phases

†Includes step-down care in hospital (non-ICU) following ICU stay but prior to discharge of 6.4 days.

Table 3: Key RWCS headline data based on epidemiological modelling, to support HMG planning decisions.

Most popular

Katja Hoyer

How the Magdeburg Christmas market attack will change Germany

RWCS data for the nations and regions

Regional and devolved nation breakdowns are available for the RWCS. This scenario, however, was designed to consider the whole UK. As policy responses to the epidemic are likely to differ between the devolved nations, it would be expected that each of Scotland, Wales, and Northern Ireland would have different trajectories over time; devolved administrations will want to take this into account when considering this paper.

When case numbers are low, due to small populations and/or low incidence as is the case in the devolved nations, uncertainty increases and fluctuations in the data can have a significant impact on estimates. Significant variability across a region, for example due to a local outbreak, can also add to uncertainty. Care must therefore be taken when applying this scenario and its data to these smaller geographies.

Table 4: Key RWCS headline data based on epidemiological modelling for each devolved nation, to support planning decisions from 1st July 2020 to 31st March 2021.

Figure 1: RWCS daily direct COVID-19 deaths for Scotland, Wales and Northern Ireland, actual data to end of June, RWCS estimation from 1st July 2020 to 31st March 2021.

Annex A: RWC Scenario Graphs UK

Figure 1: Number of deaths directly from COVID-19 under RWC planning scenario

Figure 2: ICU occupancy under RWC planning scenario

Figure 3: Number of hospital admissions per week under RWC planning scenario

Figure 4: Number of new infections per week under RWC planning scenario

Annex B: Estimating additional deaths to expand the RWCS

Office for National Statistics (ONS), Department of Health and Social Care, July 2020

Table of contents

1. Summary
2. Adding COVID-19 deaths outside hospitals and care homes
3. Estimating COVID-19 “excess deaths” – i.e. those which would not have occurred anyway in the period in question
4. Adding non-COVID-19 “excess deaths”
5. Combining these effects
6. Calculating total weekly deaths

1. Summary

The Scientific Pandemic Influenza Group on Modelling (SPI-M) have produced a new Reasonable Worst- Case Scenario (RWCS) for government planning for the period August 2020 to March 2021; estimating the deaths which could occur directly due to COVID-19. The estimate is aligned to ONS’s weekly death registration data. These show deaths where COVID-19 is mentioned on the death certificate whether or not a test for COVID-19 was completed. There have also been increases in deaths not involving COVID- 19 visible in those data which coincide with spikes in COVID-19 deaths, assumed to be related to the COVID-19 outbreak and/or government’s interventions to tackle COVID-19 [A possible, alternative explanation is that a proportion of these were undiagnosed COVID deaths but this cannot be proven at present]. To assist in planning, ONS and DHSC have produced the following adjustments to SPI-M’s scenarios, to identify how many are “excess deaths” and to capture the deaths which the SPI-M analysis does not include. The adjustments change the RWCS to account for:

• COVID-19 deaths in private homes, and other settings not covered by the RWCS.

• Out of the COVID-19 direct deaths, the number which would have been expected to die from other causes within the 38-week scenario reference period (1 July 2020 to 30 March 2021). This adjusts the estimates down.

• The number of non-COVID-19 excess deaths expected. In most cases this adjusts the estimates up; for younger age groups there is no effect.

The first two of these calculations roughly cancel each other out when applied to all deaths in England and Wales, so we expect ONS COVID-19 excess death values to be approximately equal to SPI-M COVID-19 total deaths, before non-COVID-19 excess deaths are added in the third adjustment.

For each of these calculations, the adjustments provided are based on the information we have so far about COVID-19 deaths and non-COVID-19 excess deaths, which remains a limited time-series. Further, we still have a lot to understand about COVID-19 fatalities and how close the individuals were to end of life. SPI-M have provided estimates of deaths in care homes separately from deaths in other settings, which allowed some adjustment for the lower life expectancy of care home residents for the same age/gender grouping. This has been adjusted for in the calculations.

The relationship between deaths involving COVID-19 and not involving COVID-19 has been unstable over time. In the absence of better understanding, we have compiled a set of instructions to create an adjustment to the RWCS to present all excess deaths, based on cautious assumptions that err on being a potential over-estimate, rather than under-estimate, of total deaths. The estimates cover deaths in all settings; these could not be separated out as there are insufficient data to apply uplifts separately for different settings.

As for SPI-M’s estimates themselves, these represent scenarios and not forecasts of all deaths.

SPI-M acknowledge that their modelling has been performed at the UK, nation and NHS England region level and does not necessarily reflect the variability that might be observed at a more local level. Care must be taken when applying this scenario and its data to smaller geographies as there will be significantly more variability at more local scales.

2. Adding COVID-19 deaths occurring in locations other than hospitals and care homes

The RWCS provides estimates of deaths occurring in hospitals and care homes. ONS weekly death registration data present COVID-19 deaths also occurring in private homes, hospices and other communal establishments. The proportion of COVID-19 deaths occurring in hospitals and care homes to date is 92.75% of all COVID-19 deaths.

To adjust the RWCS in order to present 100% of all COVID-19 deaths, the RWCS is multiplied by 1.078.

3. Removing deaths which would have occurred anyway within the reference period

ONS’s internal modelling has estimated how many of the SPI-M scenarios’ deaths were expected to occur anyway within the 38-week period due to non-COVID-19 related causes, so would not represent additional activity for mortality-related services such as funeral homes. This estimate is broken down by age. Overall, the expected change in total deaths is -7.0%. So 93.0% of COVID-19 deaths (both in the RWCS, and additional added in Section 2) are not expected to occur otherwise within the 38-week period.

Table 1 below provides the adjustment factor to apply to the total number of COVID-19 deaths (calculated in Section 2). For example, 94.2% of deaths involving COVID-19 for 70-79 year olds are excess deaths – so 5.8% of COVID-19 deaths within that age group would have occurred anyway within the RWCS reference period, and do not represent additional deaths during this period.

Analysis does suggest some deaths in age groups younger than 40 years would have occurred anyway within the reference period in the absence of COVID-19; but the analysis is not sufficiently robust at these ages to include an adjustment.

When producing values for regions, we recommend only using the ‘all ages’ factor.

Table 1: adjustment factor for COVID-19 excess deaths

Note the “all ages” adjustment approximately reverses the adjustment in Section 2. COVID-19 excess deaths = RWCS COVID-19 total deaths * 100/93 * 93/100

So: 

COVID-19 excess deaths = RWCS COVID-19 total deaths (approximately)

These first two adjustments are both still important because the second adjustment will vary if adjusting results for a specific age group; and these values could change in future when more data are available.

4. Adding non-COVID-19 deaths

The following uplifts in Table 2 can be applied to the England and Wales RWCS estimate of COVID-19 deaths; the regions of England, and Wales independently estimates; and estimates broken down by age. These should be applied to the total number of deaths produced in Section 2. The same proportional increase should be applied to all weeks.

Table 2: weekly non-COVID-19 excess deaths to add to the RWCS

These percentage uplifts are calculated using ONS’s weekly registration data.

• Non-COVID-19 deaths are calculated by subtracting death registrations involving COVID-19 from “all causes” total deaths;

• The non-COVID-19 excess deaths are calculated by subtracting the five-year average from the non-COVID-19 deaths;

• The average (mean) of this non-COVID-19 excess deaths result for Weeks 14 to 20 (28^th March to 15^th May) is divided by the total number of COVID-19 deaths in that period. This is used to give a suitable estimate for the number of additional non-COVID-19 deaths to expect for this group, per week. This can be added to the RWCS for that breakdown for every week of the reference period.

We recommend applying the same change to every week of the RWCS reference period, as the evidence available from the first COVID-19 spike does not support using more precise adjustments. A proportional increase has been applied, rather than an absolute number, to account for the RWCS (and other scenarios) presenting peaks and troughs in numbers of deaths.

For some regions and age breakdowns, notably younger age groups, the result for the above calculation is negative. This is because the non-COVID-19 deaths observed in latest data are below the five-year average. In these cases for the RWCS the additional number of deaths to expect has been treated as 0, rather than a reduction in deaths.

Due to the small number of deaths per week in ages 0 to 59, we have summed these into three age bands: 0-19, 20-39, 40-59. The older age groups have sufficient numbers of deaths to not require this change.

Weeks 14 to 20 are used for the average, as these are the weeks in which non-COVID-19 excess deaths were most significant so far, matching the future RWC estimates. This period has been used to represent the first spike. While Week 19 presented non-COVID-19 deaths below the five-year average, this was due to bank holidays affecting registration, and Week 20 again presented an excess for most breakdowns.

Death registrations are used rather than occurrences because there is usually a delay between a death occurrence and its registration, with that delay varying depending on cause of death. As such registrations are a more meaningful measurement when looking at deaths data in very recent periods.

There are alternative methods for calculating the expected number of deaths in the absence of COVID- 19, other than the five-year average. However, the five-year average is the simplest and relies less on additional assumptions, such as adjusting for how the population size and age distribution has changed over time.

The published data used to produce these estimates cover England and Wales. There is no consistent trend over time for non-COVID-19 excess deaths or the relationship between them and COVID-19 deaths– and time series are even more unpredictable for individual regions. For this reason, we are not in a position to recommend what uplift to apply to Scotland or Northern Ireland COVID-19 deaths.

5. Combining these adjustments

All of these adjustments multiply the RWCS to create a proportional change. The first in Section 2 is applied to the RWCS directly, and the adjustments in sections 3 and 4 should be applied to the result of Section 2 separately. The calculation is as follows:

(RWCS * 1.078 * adjustment from Table 1) + (RWCS * 1.078 * adjustment from Table 2)

The RWCS’s weekly values should therefore be adjusted as in Table 3 below to produce a weekly estimate of COVID-19 and non-COVID-19 excess deaths. The full time-series for each of these breakdowns is presented in the accompanying workbook.

Table 3: Weekly excess deaths (COVID-19 and non-COVID-19)

6. Calculating total number of deaths each week

The RWCS deaths, adjusted as in Table 3, can be added to the weekly expected deaths (five-year average of 2015-2019) to produce the total number of deaths expected each week under the RWCS. This is presented below with excess COVID-19 and non-COVID-19 deaths separated. Note the “Excess COVID-19 deaths” element does not equal the SPI-M RWCS deaths estimate, as it has been adjusted up using the methodology in Section 2, and down using the methodology in Section 3.

Table 4: Total deaths over RWCS period, England and Wales