Thread by @dgurdasani1

A brief discussion on the impact of omicron we're seeing in some countries currently - e.g. HK - which may seem surprising given the impact observed in South Africa and even perhaps the UK - which while v. significant, seems lower on the face of it.🧵

Some of this will no doubt stem from different vaccination rates by age in different regions - but there is possibly another important factor that may not be immediately apparent.

I want to revisit the discussions on the severity of omicron relative to delta here. As many of us discussed earlier - there are two aspects to severity:
1. intrinsic severity - what is the severity if omicron compared to delta if they infected the same people

2. observed severity- what is the impact of the fact that due to higher escape omicron is far more likely to infect someone with prior immunity (prior infection/vaccination) than delta is?

Prior infection with another variant, while not protecting against omicron infection may still protect against severe disease. The competitive advantage of omicron against delta was greater in those who had previous immunity that was *more protective* against delta than omicron.

So omicron disproportionately infected people who were vaccinated/had prior infection compared to delta. This meant that lower severity could be either due to a virus that would cause less severe disease in anyone, or because those infected were more likely to have prior immunity

Teasing these 2 effects became important- because one is an intrinsic characteristic of the virus, so can be generalised to any context- because it doesn't matter who the virus infects. But the second (prior immunity) is a characteristic of the population & will differ by region.

Studies in the UK and SA did try to do this. How did they do this? When the severity of omicron was compared to delta, past infection and vaccination status was *adjusted* for. But how does this adjustment happen?

It's a statistical process that breaks up the population into people with known prior infection, and those without, and then tries to estimate what severity relative to delta would've been in the group *without prior infection/vaccination*- because this is intrinsic severity.

Although vaccine data is usually accurate, prior infection data is often hugely flawed. Remember we didn't even have community testing till June 2020. And even after many people didn't test, or were asymptomatic, or ineligible due to restricted symptom lists for COVID-19.

So we know re-infections are *massively* underestimated. What does this mean? It means that many the population we consider as not having had prior infection may well have had it without us knowing, which makes it impossible to get to a 'pure' estimate of intrinsic severity.

Estimates of intrinsic severity become mixed in with effects of prior infection, because many prior infections are undetected, so in effect we cannot adjust for them statistically. This means we *underestimate* intrinsic severity.

The level of underestimation is dependent on the level of mixing in of the effects of undetected prior immunity - so it depends on the level of undetected re-infection among people who're unvaccinated.

The higher the level of population immunity due to infection & the lower the detection of past infections, the greater the bias in the estimate. If intrinsic severity estimates were accurate we could generalise these to other contexts because they depend on virus, not environment

The problem is we're not capturing intrinsic severity. Our estimates are mixed in with effects from prior immunity. This means we cannot take estimates from SA, or even the UK, where there is high level of population immunity and apply them to countries with much lower exposure.

'Intrinsic severity' estimates from countries with high population immunity & high undetected prior infections will generally be lower than intrinsic severity estimated in countries that have much lower levels of prior immunity from infection.

This shouldn't happen because conceptually intrinsic severity is a characteristic of the virus and shouldn't depend on external factors. But because we cannot estimate this accurately, our estimates are a mixture of intrinsic severity of the virus & who the virus infects.

This means that all else being the same (e.g. levels of vaccination), impact in terms of severity may be higher in countries with low prior exposure than that observed in countries with high prior exposure. Because intrinsic severity was underestimated in these latter contexts.

It is interesting that a study from Denmark (which had much lower exposure to SARS-CoV-2 pre-omicron) showed ~30% reduction in hospitalisation risk compared to delta (compared to other studies that suggested 50-70% reduction).

This also accords with an early study from Imperial that tried to derive an estimate of intrinsic severity by adjusting for the fact that many re-infections were undetected. They also found 30% or less reduction in hospitalisation risk relative to delta.

To be clear, Intrinsic severity of omicron is certainly lower than delta, but possibly not to the degree that studies from SA & UK suggested. To get less biased estimates, these studies will need repeating in regions with lower levels of prior infection or better detection

It's important to get to these estimates because separating out these components helps us understand & predict impact in other countries. In the meantime, it's worth remembering that estimates of intrinsic severity from the UK may not be entirely transferable to other contexts.