Thread by @andrewlilley_au

NSW is the ideal setting to measure the severity of Omicron, and it is showing a CFR for Omicron that is <1/2 of Delta
Five reasons it's ideal:
*Prior infection is irrelevant (~2-3% of NSW is prior infected)
*Delta cases were steady through Nov, making a stable baseline CHR

1/15

*Omicron became dominant extremely rapidly, so a change in the CHR will be noticeable
*Accurate case surveillance w/ age breakdown and <1 day lag in test to case count.
*Vaccination was largely complete by Omicron's arrival (at ~94% of adults), boosting was minimal.

2/15

NSW detected its first Omicron cases on December 2 and saw rapid genomic domination is likely now that >75% of cases are Omicron. (The gov have sequenced enough to say "it's mostly omicron" & they won't be sequencing anymore). Given the steadiness of Delta cases over the prior

six weeks we can have a fairly good idea that the composition is somewhere between 60% and 90% Omicron.

4/15

I've estimated the daily case hospitalisation rate in NSW by dividing hospital admisions today by cases 5 days ago (to avoid the growth bias that has been discussed elsewhere). The 5 day gap between confirmed case and hospitalisation comes from NSW's reports showing a 5-6 day gap

between becoming ill and being hospitalised.

Each data point is an *estimate* (method described later) of that day's hospital admissions divided by *measured* cases from 5 days ago. The shaded orange region is the period in which Omicron rapidly became the dominant strain. 6/15

Delta had a steady 12.5% hospitalisation rate from Sep through Oct, and it then started to fall a bit in Nov as fully vaccinated breakthroughs + child cases went from 30% to 75% of all cases. Then in December it began falling much more sharply - from 6.9% ten days ago to 3.6% now

It coincides with Omicron becoming dominant, and as such is evidence of it having a much lower measured CFR than Delta.

Unlike other countries, we cannot explain this lower CFR by selectivity towards reinfections - prior infections only make up 2-3% of NSW's population.

We also can't explain it with vaccine evasion - vaccinations were largely completed, and 75% of cases in the last week of November / first week of December were already made up of fully vaccinated breakthroughs or children. In fact, this is likely to bias the change in measured

CFR upwards - vaccine evasion means vaccinated adults will also take over the share from unvaccinated children (with a much higher risk of hospitalisation).

9/15

We have observed the aggregate CFR fall by ~50% over the last 10 days - what does that mean for the CFR for each Delta case v. Omicron case? We can trace out the relative CFR based on the estimated % of cases which were made up of Omicron 5 days ago (& assuming ~0% start of Dec).

Aside pt1: If you are wondering if there might have been a reason that Delta's CFR kept falling and contributed to the fall in the orange region, it's unlikely. Fully vaccinated breakthroughs can't have changed much in the first 2 weeks of December as unvaccinated adults have

stayed constant at about 6% of the population, so they should remain ~20-25% of cases. (Any further drop would imply a very low effectiveness of the vaccines against Delta in the first ~3 months after vaccination).

10/15

A few caveats on method: I don't observe admissions directly (if you're from NSW health, please contact me), so I construct a hazard rate by taking published research on hospital stays in NSW by age, and weighting these by daily cases and their relative probabilities of

hospitalisation. That's surely one of the reasons for the noise you see here. I seasonally adjust daily cases by their day of week and in the lower panel of the graph I adjust daily cases by age composition for their relative risk of hospitalisation. (Top panel is only s.a.).

To pre-empt a well-informed objection to this - even if immunity did not change in this period, I'm measuring the change in CFR and that includes both the difference in virulence and the change in the distribution of vaccinated/unvaccinated cases. I agree.

But the vaccinated share of cases can't increase by much with Omicron - the first week of December already had 75% of cases in fully vaccinated individuals or children - it can't go much higher as 94% of adults were fully vaccinated already.

And even if it tilts towards vaccinated adults, it will also tilt away from partially vaccinated teenagers and unvaccinated children, who are still lower risk than vaccinated adults.

In my opinion this a much cleaner estimate of CFR that is uniquely possible in NSW. I will continue to update this thread every few days and begin to construct ICU and fatality etimates as more data rolls in.

Credit to @migga for the scraped daily series of hospitalisations.

I wrote CFR a few times in this thread. Anywhere I wrote it I meant to write "CHR". Sorry, force of habit. It is too early to infer the CFR.

To deal with a common query (sorry I cannot reply to everyone individually), the analysis includes the adjustment for the age composition of cases (I would not have considered doing it if this data weren't available, so many thanks to health NSW for the data). See the labels on

the top and bottom panel of the graph here. The bottom is age adjusted. The top is not. I only show the top because when I'm reading analysis I like to see the difference the author's adjustment made. Note the key piece of info is the recent reduction and cases have

skewed young for all of the last 30 days (ie pre omicron) not just the last 2 weeks (the change from Delta to Omicron), so the adjustment doesn't end up mattering very much for the drop of the last two weeks.

To be crystal clear, the analysis includes the empirical fact that 500 cases in 20-29yo has fewer predicted hospitalisations than 100 cases in people 70+.
The limitation is I can't observe any further granularity in the 70+ bucket but hospitalisation is mostly flat there

The second common query- this assumes that the time between test and hospitalisation is an average 5 days. That's true, I took that from the NSW epidemiological reports which say 5-6 days between feeling ill and being hospitalised. (Assumed a 1/2 day lag to test result so 5d).

The other concern some have raised is what if Omicron just takes 10+ days to be hospitalised but Delta takes 5? This has always been a concern with analysing every variant, but we a) should have seen the relative delay in South Africa data and b)goes against clinician expectation

Anything is possible but it can't be ruled in or out until the Omicron wave turns downwards and then we wait as many days as the expected lengthening of lag. (South Africa is the only place where the wave has ended and we haven't seen any evidence of this.)

As for the questions of why it would be less virulent but more transmissible if that wasn't true for Delta, I have no expertise in this area whatsoever, but I strongly recommend following Francois Balloux - an expert who has had incredibly good judgment

https://twitter.com/BallouxFrancois/status/1473240778734546946?t=gCKE96ImdI2oOgEZ-nl4RQ&s=19

A great suggestion from a friend @Rustinpartow about the noise in the data. Cases are noisy and I deliberately don't smooth them because you never want to do that when you have a rapidly changing trend at the endpoint.

112 observations of 8d changes, Omicron chgs in bottom 10%.

@Rustinpartow So there is definitely a level change amidst the volatility, but we have some reasonably large standard errors for a few more days as you can see from the noisiness of the time series.

Scotland study: two thirds reduction compared to Delta. Controls for age, vaccination status, and reinfections.

research.ed.ac.uk/en/publication…

Update here for the many people requesting to use a distribution of lag times from cases to hospitalisations rather than a basic 5 days.

https://twitter.com/andrewlilley_au/status/1474161427057463299?t=YARsEQbM_2eYKESHNTmUeg&s=19

In this analysis I made the claim that the selectivity of omicron to fully vaccinated people wouldn't bias the CHR much, because almost all adults in NSW are already vaccinated.
With the new vax status breakdown, looks like it lowers the CHR by only ~10%.

Ignoring the "under investigation"
Dec 4 66% of adult cases fully vaccinated
Dec 11 75% of adult cases fully vaccinated
Putting "under investigation" into unvaccinated (suggested by @RichardfromSyd1 who has looked into this)
The breakthrough case shares would be 58% and 63% resp.

Going from 58% to 63% would be a rounding error. If we use the former, going from 66% breakthroughs to 75% breakthroughs would lower the CHR by 11%.
The imperial college report from yesterday says a breakthrough case has a CHR of 1/3 of an unvaccinated: 89%=(75÷3+25)÷(66÷3+34)