The importance of the new daily infections rate is that it fundamentally determines the level of hospitalisations, death and LongCOVID. An increasing trend shows R is over 1, resulting in the danger of a new exponential rise. A decreasing trend would be welcome.
How does the number of new cases look now? Have new cases risen, fallen, or stayed much the same? What about the near future?
DATA AVAILABLE
There are a number of organisations which are monitoring infection levels in England, typically public bodies and universities. The four of most interest, providing data publicly, are:
- Public Health England publish data from the NHS Test and Trace programme. This is the "New Confirmed Cases", and the only data produced daily. But as it only reflects people actually tested, it is only the tip of the iceberg of true new cases, with previous analysis suggesting figures need to be multiplied by about 5. The data is analysed by nearly 7000 local areas across all four nations of the UK. But as community testing is only if the person has symptoms, the data lags by around a week.
- The ONS (Office for National Statistics) with their “Coronavirus (COVID-19) Infection Survey”. This is a randomised study, across everyone in England, be they symptomatic or non-symptomatic, and across all walks of life. It is therefore could be the most reliable data for England. But the samples so far have been too small to provide reliable regional analysis, and indeed the 95% confidence spread for the national figure is rather wide. But the sampling is to be substantially increased to also include Wales, Scotland and Northern Ireland. The data is only produced weekly.
- The King’s College London study using the ZOE app, downloaded by some 4 million people across the whole UK. The data is again reported weekly, principally as a UK total. Some regional analysis is available. The data is based only on symptomatic people, and only for people who can afford a smartphone. So the data is inherently only a subset of the total who have caught the disease, biased away from the lower-paid. The figures seemingly need to be multiplied by 2, and is useful for trends. As the data is for symptomatic, it inherently lags about a week behind new infections.
- The Medical Research Council (MRC) Biostatistics Unit (BSU), at Cambridge University. Their data has not been analysed in this blog before. That will have to wait to be done until the next report is published, as the latest report appears to be right back on 6 August.
Let’s look at each of these in turn.
"NEW CONFIRMED CASES" FROM NHS TEST AND TRACE
The "New Confirmed Cases" arise primarily from testing people with symptoms, a short while after the symptoms start. The data is available daily, but therefore lags about 7 days after new infections actually happen. The data is of two types, usually added together:
- "Pillar 1" carried out by NHS labs for people in hospitals, care homes and some other settings
- "Pillar 2" carried out for people in the general community, by subcontractors
Pillar 2 tests numbers for all four nations were only included from 14 July, which explains the significant jump in July. Data for March, April. and May was just for pillar 1.
Test numbers tend to be upset by weekends, so a 7-day rolling average is usually used. If we back-date the dates by a week to reflect when people were likely infected, we get this trend for England:
Has this rise simply occurred due to more testing capacity and tests? The testing activity has been pretty consistent since mid July, so the rise is basically a reflection of a rising infection rate.
The rise in July implies the R number for England must have been above 1.0 for some weeks now, as I have been saying in previous reviews. The UK infection rates have also been rising, mirroring England as its dominant component. So the R for the UK has also been above 1 for weeks, and only just now officially acknowledged as possibly over 1. We've been misled for weeks, as I have been reporting.
But aren't these New Cases numbers significantly understated?
Yes, because of three main reasons:
- Testing in the community is optional, so not every one with symptoms gets counted.
- Does not include asymptomatic people, who can catch the disease and who can be more than half the total. Whether they can be infectious and subject to LongCOVID is subject to further investigation, but in terms of the spread of the virus, they need to be counted
- Self-swabbing produces 'false negative' results in about 30% of cases, unless a negative test is repeated. Repeats can happen in hospitals (pillar 1) but not so easily for the wider community (pillar 2). As pillar 2 is far greater than pillar 1, the total is somewhat understated
In other words the latest 7-day average for England's new confirmed cases of 878 represents an average total level of daily infections in the week to 14 August of around 4,400. That would be around 8 per day per hundred thousand for a population in England of around 56.3 million
ONS "COVID-19 INFECTION SURVEY" FOR ENGLAND AND WALES
(** awaiting feedback from ONS **)
This report released today, 21 August has failed my 'reasonableness' review. If my observations below are correct, the report appears misleading and it should not have been published.
For England
After an initial review disclosed some odd aspects, I went back to review all reports since 12 June. All aspects coloured yellow look odd:
- They say there was a major change in the reports dated 24 July onwards, including that it looks like the population figure used was reduced to reflect that those aged under two were excluded from testing. This is because "All individuals aged two years and over in sampled households were invited to provide samples for testing."
- Reports prior to 24 July do not have comparable figures for daily infections, and the earlier reports covered a fortnight or longer thereby reducing their usefulness for weekly trends
- The latest report says "new COVID-19 infections ... in the community population in England, equating to around 2,400 new cases per day (95% credible interval: 1,200 to 4,200), for a week that overlaps with the previous reported week by three days
- 2400 average in the week when 3 days were at 3800 per day, means the last 4 days were at 1350 a day. This reduction in not credible.
- This change is inconsistent with the small drop in the total of number infected from 28,300 to 24,600, accompanied by the number of days infected increasing markedly to 10.3.
- Assuming the increase in infection days was not intentional, 2400 infections per day is clearly unrealistic. A better estimate might be 3800*24600/28300, which is some 3,300.
- The 95% credible interval would also potentially increase by a factor of 3300/2400, i.e 1,650 to 5,800.
- That range more than covers the estimate of 4,400 from the NHS data
The reported figures in the previous three weeks were around 4,000 per day. That's equivalent to about 0.74 oer 10,000. The graph published this week includes a graph with figures through this period below 0.5. It looks like something has gone seriously wrong with the latest report.
The title for this graph is "There is some evidence of a small increase in people testing positive for COVID-19 in July 2020 after a low point in June, which appears to have now levelled off".
Taking the daily data from section 4 of the latest report "Incidence rate in England" we get this graph, confirming the later section of the graph above in more detail. Rising in August, not "levelled off"!
This more detailed graph shows daily numbers around 1500-1700 in July, rather than figures in the 2800-4200 range previously reported. The ONS reports say that data is "subject to revision" but that extent of change is not credible.
It is possible that ONS calculates daily numbers from the total of swab test results, using a factor for the average number of days someone would be infected. The change from around 7.5 to 10.3 days in the latest figures, if backdated to previous weeks, would help to explain the reduction in total daily numbers. But there is no mention of any such change in the report. Hence the assumption abpve that this change is not intentional.
Regional Analysis
For the regional analysis of "Incidence %" at 10 August, the title for this graphic is "There is no evidence that infection rates differ by region". Whereas there are widely differing rates, with the Northwest being twice the level in the South East, for example.
So not only is key data in the latest report apparently incorrect and badly presented, the commentary also does not match the data.
I suspect the reporting was disrupted by annual leave in August. But even so, given that the "daily new infections" figures in th report are arguably the most important, publishing evidently misleading information is not acceptable.
Indeed the problems highlighted here in the data, presentation and commentary would be unacceptable in business. This report is not credible, indeed apparently misleading, and therefore should not have been published.
Let's hope the quality of the next report improves dramatically.
KING’S COLLEGE LONDON STUDY USING THE ZOE APP
The ZOE data is based on when symptoms appear, so is about 5 days behind new infections.
The ZOE report that appeared yesterday, Thursday reported:
"1,265 daily new cases of COVID in the UK on average over the two weeks up to 15 August 2020 (excluding care homes)"
The latest daily infections graph shows 953 cases on 17 August (i.e. new infections around 12 August), significantly lower than the 1579 on 31 July and the two weeks to 15 August:
Again it is unfortunate that the last few weeks have been squashed into such a small portion of the Y axis. Time to create a second graph that ignores the period before June?
In a previous analysis, we established that:
- The ZOE app is only used by people rich enough to use a smartphone, and then to have downloaded and used the app.
- Based on symptoms, excluding asymptomatic people who ought to be counted for tracking spread of the disease.
In any case, previous analysis has suggested that the ZOE UK figures should be doubled to get a more realistic England figure.
953 in the UK for new infections around 12 August, would then mean around 1900 total for England. Though today (23/8/20) the daily number has increased markedly to 1,265, which would mean around 2,500 total.
MRC BIOSTATISTICS UNIT, CAMBRIDGE
The latest report on 6 August said "Our current estimate of the number of infections arising each day across England is 3,200 (1,700–5,800, 95% credible interval)." That compares reasonably well with 3,700 from ONS for the same period.
COMPARING TRENDS
The three sources have graphs showing very different trends:
NHS, with daily new cases rising from 21 July into early August, but then showing signs of reducing:
ONS showing daily new cases only rising in August, with no sign of reducing:
ZOE showing a general reduction in daily new cases in late July and early August. maybe because of the affluence of their user base:
IN CONCLUSION
Comparing trends, the three sources analysed are so inconsistent in recent weeks as to make conclusions difficult. If we take the NHS data as the most reliable, the trend has been that daily infection has been rising slightly, with R slightly over 1. Whether the daily new cases have been reducing more recently is difficult to say.
On the absolute number of new infections per day a week ago, the differing trends cast doubt on comparability:
- NHS reported 1.56 per day per hundred thousand as the average in the week, equivalent to 878 a day, adjusted to 4,400 if a factor of 5 is still appropriate
- ONS reported 2,400 a day, but this has been 'rectified' as explained above to 3,300 with the 95% credible interval of 1,650 to 5,800.
- King's College ZOE suggests 953 new cases per day, adjusted to 1,900, since risen to 2,500
Given all this uncertainty, it is reasonable to assume that the daily total of new infections is at least 2,000 per day in England, and perhaps as high as 4,400 per day. That's 3.6 to 7.8 per hundred thousand, significantly higher than 2 per day suggested by Professor Whitty, and possibly still rising.
Until next week's figures can be reviewed, with more trust hopefully in the ONS figures, the NHS multiplier looks like it should be at least 3. Though 5 is still suspected given the reasons why the raw reported figures would be too low.
The government is rightly putting its trust in the ONS Infection Survey by increasing the sample sizes significantly. But the ONS must improve the data, its presentation and the commentary in their reports , preferably in the next report this coming week.
No comments:
Post a Comment