COVID-19: CROSS-CHECKING ONS AND NHS DATA FOR NEW INFECTIONS

Takng swab for testing

When we looked at available data on new infections in England  we found that the latest daily estimate from the Office for National Statistics (ONS) was far higher than the figure from NHS Test and Trace, by a factor of about 5.


Let's examine that in more detail.




From the latest ONS report, which is based on sampling of people in the community, from all walks of life, be they symptomatic or not, for the period to 26 July:

So we have 4200 new cases per day across England, but with a wide margin of error.


From NHS Test and Trace, based on actual test results:

The NHS data is unclear whether this data is just “pillar 2” (in the community) or also includes “pillar 1” (in hospitals etc).  But as pillar 2 is typically much larger than pillar 1, it doesn’t make that much difference within the overall estimation.  But this sort of ambiguity is 'regrettable'.

We know NHS data for pillar 2 in the community would be inherently understated due to:

• Only symptomatic people who have volunteered for the test are included
• Pre-symptomatic and asymptomatic people are excluded. 
• Estimates of asymptomatic people are as much as 60-80%.  This study in the Balearic Islands suggests 70%.  That alone means the real level of infections is at least triple the symptomatic level (multiply by 100/(100-70))
• Up to 30% of tests are false negative, higher when self-swabbed.  If symptomatic people are not re- tested, that is another rachet of infection level.

So if 850 people are tested positive:

• To take false negatives into account, let’s multiple by 100/(100-30), which gives 1214
• To take asymptomatic into account, let’s multiple that by 100/(100-70) which gives 4047.  That is nearly 5 times the 850
• To take into account that not all symptomatic people get tested, we need to multiply by another factor, taking us to more than 5 times the NHS figure

The ONS figure is around 5 times the NHS figure.  We might expect it to be more than 5 times the NHS total.  So if anything the ONS figure is understated.  Certainly we should multiply the NHS figures for national and regional totals by at least 5.


IMPACT FOR LOCAL NUMBERS

The NHS data is analysed to some 7000 local areas.  When you look at the weekly data we see lots of areas with zero cases in a specific week.  For example for Oxford:

The 7 for “Churchill” in week 30 has adjacent areas as zero.  If we multiply these by 5 we get 35 and zeroes.  That’s not realistic.  There are probably a small number of cases in adjacent areas, which means we shouldn’t multiply Churchill by as much as 5.  Maybe 4, so the other 7 cases can be allocated to adjacent areas.


COMPARISON OF ONS NUMBERS TO KING'S COLLEGE ZOE APP

The King's College London ZOE app collects symptoms data from  millions of people.  This is from the latest report:

ZOE inherently under-estimates daily infections for two reasons:

• It is restricted to people who can afford smartphones, or otherwise don't want one
• It ignores asymptomatic people 

Both factors would mean it is incomplete, and so we should expect ZOE data to be lower than the ONS data , But we would expect ZOE data to be above NHS, as it is not subject to the false negatives.  Indeed ZOE data does fit in between NHS and ONS.  This suggests the ONS data is not overstated.



IN CONCLUSION

It is 'regrettable' that NHS data is ambiguous.  However we can make some reasonable conclusions.

The ONS estimate of 4200 new infections a day appears reasonable by comparison to both the NHS and ZOE data.

When we compare ONS and NHS data for new daily infections, we might expect the ONS data to be at least 5 times the NHS across wide areas of England.  So if anything, the latest ONS estimates are understated.  As also appears to be the case by comparison to ZOE data.

This means we should routinely multiply NHS national and regional data by a factor of at least 5.
However looking at NHS data for local areas, we need to multiply by a lower factor of say 4.