How do we study the severity of a new variant?

A guide to help you organize your thinking

Want to better understand the research coming out on the severity of Omicron? This is my guide to help you organize your thinking. Consider it a companion to the emerging research, with an emphasis on intuition rather than numbers.

We can think about an individual’s risk of severe disease as reflecting three components: (1) their risk of exposure, (2) their risk of infection given exposure, and (3) their risk of severe disease given infection. We can multiply these three components together to calculate their risk of severe disease.

One scientific question we are interested in is whether the intrinsic severity (virulence) of the Omicron variant is different from that of the Delta variant. This refers to the risk of severe disease given infection (the far-right component).

How are studies addressing this question?

First, let’s step back and think about the types of populations that relate to each of the above components. For example, to estimate the risk of infection given exposure, we might study the contacts of known infections to see who gets infected. To estimate the risk of severe disease given infection, we might study the set of known infections to see who develops severe disease.

Though not exclusively, many of the recent studies on Omicron severity have focused on the last column – outcomes of known infections. Do infected people end up requiring hospitalization or even an ICU stay?

To compare Omicron with Delta, we study the outcomes of known Omicron infections with the outcomes of known Delta infections.

What are some key considerations in our analysis?

• If our outcome is hospitalization, we know that criteria for admission to a hospital can change over time (particularly during periods where hospitals are strained). Thus, we want to compare infections happening at the same time in the same areas. This means looking within time periods where Omicron and Delta are co-circulating.

• We also know that risk of severe disease depends greatly on individual-level factors such as age and comorbidities. Thus, we aim to compare LIKE to LIKE by forming smaller groups with a similar risk profile. We can then compare Omicron and Delta outcomes within smaller groups. An example is provided below with age (younger adults, older adults).

In analyses, this is achieved with a stratified analysis or a multivariable model adjusting for age and other covariates. We can look for similar patterns within subgroups, although this can be challenging for lower risk groups (younger adults) because severe disease occurs less frequently.

As an aside, ultimately, an individual’s risk depends on their risk of exposure and infection along with their risk of severe disease given infection. So if individuals are, say, half as likely to develop severe disease given infection with Omicron as with Delta, this is negated if their risk of getting infected with Omicron is twice as high.

Okay, back to severity. Recall that our goal is to examine the intrinsic severity of the variant. Well then there are a few other important factors we need to consider. These are (1) prior infection, and (2) vaccination.

In the table below, the second row is individuals who are unvaccinated but have been previously infected. For simplicity, assume everyone has a similar risk of exposure, but prior infection reduces their risk of infection by a factor alpha, and it reduces their risk of severe disease given infection by a factor beta.

The third row is individuals who are vaccinated but have not been previously infected. Using the notation of Halloran et al., vaccination reduces their risk of infection by VE_S (where S refers to Susceptibility to infection), and vaccination reduces their risk of Progression to severe disease given infection by VE_P. Individuals who have been both infected and vaccinated form another category, not shown.

In keeping with our strategy to compare LIKE with LIKE, when comparing the outcomes of Omicron versus Delta infections, it is important to adjust for both vaccination status and prior infection.

This is what studies are doing now – looking within subgroups – again either with stratification or multivariable models. This is on top of the other factors included in the models.

A few challenges arise in the analysis:

1)      We may consider the not previously infected/not vaccinated population as the best for an analysis of intrinsic virulence (just like how R0 measures spread within a naïve population, what would the virulence be in a naïve population?). But depending on the population, there may not be many individuals in the not previously infected/not vaccinated population.

2)      Vaccination status is well-documented, but we know that prior infections are under-detected. It means that many individuals who have been infected before are misclassified as NOT PREVIOUSLY INFECTED. When we form our like groups, they are in the wrong group. Because re-infection is more common with Omicron than with Delta, this misclassification leads to a predictable bias that makes Omicron appear milder (since re-infections tend to be milder).

In their analysis, Imperial makes some assumptions about the rate at which these are misclassified to “correct” this bias. The corrected analysis yields higher severity of Omicron than the uncorrected analysis.

Note that this analysis is in an infected population and focuses on the last step in a cascade of steps (from exposure to infection to severe disease). This enables us to study the risk of severe disease given infection, but recall that individual risk is also influenced by exposure and risk of infection. Yet we know that Omicron is more infectious, and more likely to cause re-infection and breakthrough infections. So results must be interpreted with this in mind, that the larger impact reflects other factors.

Hopefully this helps you organize your thoughts as you interpret these studies. The important points I want to make are:

• The importance of adjusting for individual-level covariates;

• The importance of adjusting for vaccination status and documented prior infection;

• The potential for bias due to missed prior infections;

• That analyses that look at outcomes within the infected population are focusing on the end of a larger cascade.

I am testing out this format as a way to share information without the restrictions of a Twitter thread. Stay tuned, as I may create a similar post on assessing vaccine effectiveness…

~ Natalie