1/ The ONS infection survey has come out, and the graph below has led some people to speculate that B.1.1.7 isn't more transmissible any more.

Unfortunately I don't think we can conclude that. Here is why [1/n]

2/ The apparent clincher here is that this would suggest that new-variant-compatible cases are falling *as a proportion of cases*. It was the increase in B.1.1.7 as a proportion of genomes in multiple areas that suggested its increased transmissibility.

3/ But what do the ONS mean by new variant compatible? Each TaqPath test amplifies 3 targets: S, N and OR. Each of these can give a positive or negative result. So any TaqPath test could give one of 7 results:
- OR+N+S
- OR+N
- OR+S
- N+S
- just N
- just S
- just OR

4/ The ONS definition of new-variant-compatible is when just the S gene drops out, and you get OR+N. But we know that amplicons can drop out randomly, esp. if you have few viruses in a sample. So the combinations technically compatible with B.1.1.7 are:
- OR+N
- just N
- just OR

5/ I had a look at ONS data from September (before B.1.1.7 was widespread) to show that this random dropout occurs, and correlates with the general levels of viral load in a region

High Ct values mean low amounts of virus per sample, and lead to more of this random dropout

6/ So what's happened recently in England? If we look at all of these different combinations over time we can see that OR+N (which is how the ONS defines 'new-variant-compatible') has indeed flatlined or fallen lately.

7/ But it hasn't been replaced by OR+N+S (which is essentially incompatible with B.1.1.7), it's been replaced by N only and OR only, which are completely compatible with being either B.1.1.7.

8/ And we can see that mean Ct has risen in this recent period, explaining the change from OR+N to OR only or N only (random dropout of one or other amplicon)

9/ Finally (for now), it's really complicated. Random dropout of S could lead WT viruses to present as "OR+N" (new-variant-compatible). SGTL could lead a proportion B.1.1.7 to present as "OR+N+S", creating an upper limit on apparent B.1.1.7 prevalence..

10/ ..Other variants might have come along that are genuinely as transmissible as B.1.1.7. Other variants might have appeared which always show N or OR dropout due to their genomes. No analysis method is going to be perfect with this type of data, it's lucky we have genomes too.

11/ You can check my workings in this work-in-progress blog post: https://theo.io/post/2021-01-22-ons-data/

12/ Something I should have mentioned in the thread is why Ct values are rising (number of viruses in samples falling). It's because we're controlling new infections better, so a higher proportion of the infections we identify are from people who were infected some time ago.

13/ The ONS data are available here, in sheets 6a and 6b. https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/bulletins/coronaviruscovid19infectionsurveypilot/24december2020/relateddata It is to their very great credit that this level of detail is available.

(My own view is that line-level Ct data would be helpful here, and without demographic data wouldn't compromise privacy)

14/ Just made this, which I think is an easier way to look at these different ways of considering B.1.1.7 compatibility.

15/ And by region: