1/ The MHRA has approved a longer gap between doses for both the AstraZeneca vaccine and the Pfizer vaccine. The latter has concerned some people. Specifically many are citing a figure of 52% for protection after the first dose.

Here is why this 52% figure isn't useful [1/n] Image
2/ The 52% value is a real figure, which comes from the Pfizer trial, for the period between the first and second doses. Here is what that period of time looked like (fda.gov/media/144325/d…):

In red are people who received placebo and in blue are those who received the vaccine Image
3/ We can see that until day 3 we have near identical results in both groups. This is *expected* - no vaccine has an effect until days later when the immune system has had time to develop a response against it.
4/ As an immune response builds the lines slowly diverge from each other. By day 10 they have completely different trajectories. We do have to be careful here, analysing the data post-hoc. But it is clear that (as one expects a priori) full protection isn't realised immediately.
5/ *The 52% figure is the average protection over these 21 days*, so it includes that initial time before the immune system has had time to create a response.
6/ If one instead looks at the day 10 to day 22 period one instead gets an efficacy value of 86% (there will be confidence intervals around that).

Analysis here: theosanderson.github.io/adhoc_covid/pf…
7/ Whereas if one looks at the day 0 to day 10 period there is an efficacy of 10%.
8/ The 52% figure is a mush of those two completely different scenarios. It's not useful. People shouldn't be citing it in this context.
9/ The Moderna vaccine documents (similar vaccine class) actually break down the results after the first dose into 14 day periods and show a very similar effect fda.gov/media/144434/d… Image
10/ There's lots to reasonably discuss about single-dosing. The big question is what the efficacy against severe COVID is in days 21-90. But please stop using this 52% figure.
11/ Document from the JCVI featuring essentially the same analysis: app.box.com/s/uwwn2dv4o2d0…

• • •

Missing some Tweet in this thread? You can try to force a refresh

Keep Current with Theo Sanderson

Theo Sanderson Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!


Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @theosanderson

Aug 5
Anti-nucleocapsid seropositivity (i.e. antibodies that can be acquired only from infection) has gone up from 25% at the start of the year to almost 70% now.

(This underestimates cumulative infection due to waning, and some people not producing measurable anti-N antibodies.)
Still frustrating that the ONS has this data but won't release it.
Read 4 tweets
Aug 4
in my continuing quest to describe SARS-CoV-2 sequencing artifacts.. nt:nt:A14960T / ORF1b:N498I / ORF1ab:N4899I looks like it is caused by a primer artifact that affects a small number of sequencing sites sometimes.
Looking at the reads it looks like it can sort of be explained by a homodimer, (although I know the mechanism below doesn't quite work)
Should be resolved by checking all reads start at a known primer binding site
Read 4 tweets
Jan 13
Today's ONS Infection Survey antibody data breaks up people who are positive into those who have lower (>42 ng/ml) and higher (>179 ng/ml) levels of antibodies, and reveals the effect of boosting
A) It's great that they've been agile and are providing this analysis
B) It shows to an extent a limitation of the public data previously available: a large set of continuous datapoints revealing *levels* of antibodies, and how they vary across a population have until now
been binarised simply into "positive/negative" in any public release (I think?). And now this continuous data is only being broken down into three categories. It would be really valuable to have more of this data in the public domain
Read 7 tweets
Jan 10
Rise and fall of SARS-CoV-2 dynasties
Methodological details: estimated cases are from covid19.sanger.ac.uk, which (thanks @richgoater) combines genome proportions by local authority with case counts from coronavirus.data.gov.uk (UKHSA). Estimates under 10 cases per week are not plotted.
Also: AY.4.2 is here not included within "Delta" to avoid double-counting. And all sequencing in the UK is part of the @CovidGenomicsUK collaboration.
Read 5 tweets
Jan 5
Interesting paper comparing LFD and PCR performance and finding low sensitivity in first days after positivity even where saliva Cts are low medrxiv.org/content/10.110…

On the face of it, results confusing because sensitivity appears better at higher Cts, but.. [1/3]
The Ct data are from saliva, and the LFD results are from nasal swabs. For a small proportion of samples they have paired data from nasal and saliva samples, and those show much higher Cts for nasal than saliva samples very early in infection. [2/3]
So seems like it's more about sample type than test method, and the graph would look quite different with nasal PCRs. But still matters! Would be interesting to know what would happen for nose/throat LFDs. [3/3]
Read 7 tweets
Jan 5
Latest release of ONS Infection Survey in London suggests rates in younger people might have peaked (though accompanying text cautions that it is too early to be sure).

Plotted below are central modelled estimates. See ONS data for confidence intervals. ons.gov.uk/peoplepopulati…
Data in pivotted form for analyses: gist.github.com/theosanderson/…
Version with fewer ages and the CIs plotted:
Read 5 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!


0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy


3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!