Many have asked me to comment on the recent ONS deaths report. (THREAD)
https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/causesofdeath/bulletins/coronaviruscovid19relateddeathsbyoccupationenglandandwales/deathsregisteredbetween9marchand28december2020

So I’ll try to discuss disentangling infection rate from death, complicating factors, and uncertainty.

I'll also discuss some potential indicators of higher teacher infection. 1/
First off, the utility of data depends on what we want to ask.

If our question is: “Which 20-64-year-old workers in England and Wales incurred the most direct covid deaths, from the pandemic start up through infections from early Dec?”

then this ONS report is a great answer. 2/
On the other hand, if our question is:

“If we reopened schools right now to all students, with the same measures in place as in the fall, how often would teachers get infected vs workers from other professions?”,

then the recent ONS report informs this answer much less. 3/
That’s due to many reasons:

-Schools were closed to most students for most of 9 Mar-28 Dec.

-60% of covid deaths in this survey are from 9 Mar - 25 May (the period of the only comparable ONS deaths survey)…. 4/
-This survey excludes many deaths from December infections, and predates most B.1.1.7-variant-related deaths that have occurred by now.

-Deaths for 20-64-year-olds involve such small fractions that there are huge confidence intervals for teachers, and many confounders. 5/ https://twitter.com/chrischirp/status/1355119044068864002
What do I mean by confounders? For each person,

Probability of death from Covid
= (Probability of SARS-CoV-2 infection for that person)
x (SARS-CoV-2 Infection fatality rate for that person),

(IFR is the probability an individual dies, given that they were infected.) 6/
For each sex, SARS-CoV-2 infection fatality rate is correlated first with age and next with underlying health burden.

But the same is true for most types of death. Especially for women, most 20-64-yr-old deaths are among the highest ages, and most are due to chronic illness. 7/
So to estimate differences in INFECTION from women deaths data, a good way to cancel out the input of infection fatality rate is to compare Covid deaths to other deaths.

Fortunately, the ONS provided some detailed deaths comparison data in their spreadsheets. 8/
There are different options for comparing deaths, producing slightly different results.

One method is to compare 9 Mar-28 Dec 2020 Covid deaths to 9 Mar-28 Dec 2020 deaths from all causes, since this was an atypical year. 9/
If we take the top 40 highest-worker occupations for women in the UK, and compute the ratio of covid deaths to deaths by all causes,

Childminders+ rank first.
Secondary school teachers rank second.

That’s DESPITE schools being closed to most students most of 9 Mar-28 Dec. 10/
Another option is to compare covid deaths to avg deaths from prior years. This is arguably a better approach, though depends what you’re trying to study.

Here, childminders were 2nd, secondary teachers were still high, and TAs and school midday workers joined the top group. 11/
Or, there are excess deaths. This metric can help identify high-risk groups that might have been under-tested, but is subject to other confounders and to variance in death rates.

Here, both care workers and secondary teachers dropped off our top 10 list, but TAs were high. 12/
Note that I didn’t compute confidence intervals (CIs) for any of the above tables.

So for any of the above categories or metrics, it could well be that none of the occupations in the top 10 are “statistically significantly” different to any others in the top 10. 13/
The ONS did compute CIs for what they called “age-standardised death rates” (ASDRs) out of 100,000, both for Covid and for recent all-causes deaths.

But they didn’t compare the former to the latter. Here are the top 10 Covid ASDRs out of 40 largest occupations for women. 14/
And here we meet complications.

The “care workers and home carers” category has nearly double the Covid ASDR for nurses.

But it ALSO has nearly double the *all-causes* ASDR for nurses. In fact, if you take the ratio, nurses are slightly higher. 15/
In addition to potential differences in health inequity, the 2-fold difference in ALL-CAUSE AS death rate estimates for carers and nurses might also reflect basic counting challenges.

To estimate number of workers per occupation, the ONS used the Annual Population Survey. 16/
Much APS data is safeguarded, but their worker number estimates are partly based on a Labour Force Survey (N = ~500K), from which the ONS has provided publicly available estimates for UK worker numbers by occupation.
https://www.ons.gov.uk/employmentandlabourmarket/peopleinwork/employmentandemployeetypes/datasets/employmentbyoccupationemp04
17/
They estimate 640K women UK care workers+home carers. Rescaling to England + Wales makes ~560K. But from other online sources, I estimated about 1 million women England + Wales care workers and home carers,

or ~1.8 times higher.
https://www.kingsfund.org.uk/projects/time-think-differently/trends-workforce-overview 18/
There’s also evidence of huge flux in this and other worker categories, so I don’t know how good my estimate is.

But in either case, the ONS AS Death Rate confidence intervals don’t take this worker number uncertainty into account. 19/
That’s another reason I think comparing to all-cause deaths, whether in the past year or from 5-year averages, is a more reliable method. At least when you compare deaths, the occupation counts are coming from the same source. 20/
Back to our top 10 list for these ONS Age Standardised Death Rates, carers were the only group to be “statistically significantly” different from any other occupation in this top 10 list—again, with confidence intervals that don’t account for worker number uncertainty. 21/
Some groups, like nurses, had a narrow enough Age Standardised Death Rate CI to be considered “statistically significantly” different from the *mean* ASDR, but

1. The Nurses ASDR CI fits entirely inside the Secondary Teachers ASDR CI, and… 22/
2. The Secondary Teacher CI is so wide that even if Secondary Teachers had had a 10% higher ASDR than the Nurses ASDR, they STILL wouldn’t have had a “statistically significantly” higher ASDR from the mean ASDR. The sample size is just too small. 23/
Back to death comparisons: if we re-rank this top-10 ASDR group by the *ratio* of Covid ASDR to recent all causes ASDR, we again see childminders and secondary teachers rise to the top.

But again, I think the covid/5-yr-avg death ratio is more meaningful. 24/
I’ve so far focussed on women, since

1. Far more school teachers are women.

2. For young men, Covid/all-cause death ratios are much more confounded by accidents and violence. 25/
But for what it’s worth, men Secondary Teachers were very comparable to Large goods vehicle drivers by nearly all metrics except occupation size.

LGV drivers were considered to have an Age Standardised Death Rate statistically significantly above the mean ASDR. 26/
And again, this is all DESPITE schools being closed to most students for most of this period, and DESPITE excluding many deaths due to December infections, when student infection was highest. 27/
Even if secondary teachers had had a lower death rate than the mean (they didn’t), that could still be consistent with much higher teacher infection than the mean in late November and December, when child infection was high. 28/
And primary teachers?

1. The ONS death report combined primary and nursery teachers into one group.

2. Due to a delayed rise in primary student infection, primary teacher death counts were more impacted by excluding many Dec infections and averaging in many pre-Nov months. 29/
But DfE confirmed case school staff absence data show that

in November & December, primary school teachers had nearly exactly the same mean confirmed case rate as secondary school teachers.

Based on thousands of infections, so much higher certainty. 30/
For the aim of guiding policy to reduce community spread among under 64s,

I think infection data are usually much more useful than deaths data, especially for women. For most occupations, deaths just have too much uncertainty and too many non-infection-related confounders. 31/
Still, one reassuring thing about this report is that—even in the very highest covid-risk occupations, for women OR men—

20-64-year-old Covid deaths made up less than a third of deaths from all causes. 32/
So really, the greatest Covid risks to non-clinically-vulnerable 20-64-year-olds are

-Non-fatal long-term complications such as long Covid; &

-Amplifying community spread, which
a) increases risk of harmful new variants,
b) risks reaching clinically vulnerable individuals. 33/
Nevertheless, every single death recorded in this ONS report—from Covid or not—came at terrible human cost.

It is important to honour that loss, and I am grateful for the tremendous efforts that went into assembling this ONS report. 34/
You can follow @SarahDRasmussen.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled:
Those who think consciousness raising & movement building are disconnected, or even a distraction from electoral politics have not paid attention to the history of social change.In this thread, we
Read more
The 2020s will be known as the Remote Work decadeA few predictions of what is likely to emerge[ a thread ] Third Space: Office and Working from Home will
Read more
1/29: Have you ever had a concept explained to you that helps frame complex issues you’ve been wrestling with and opens your eyes to new possibilities? A concept that I
Read more
By continuing to use the site, you are consenting to the use of cookies as explained in our Cookie Policy to improve your experience.