Though it's only been a week since South Africa alerted the world to the rapid spread of a new variant with many worrisome mutations, the broad contours of what we can expect are already coming into view.
Before continuing, I first want to note that we are at a very early stage of understanding Omicron and there is great uncertainty with respect to many of its characteristics. Much work remains to be done, and in some cases it will take weeks — if not months — to have clear answers to some of the relevant questions. The estimates and forecasts I cite and make below will change as we learn more. In some cases, they will change substantially.
Nevertheless, I argue that given what we've learned in the past few days, any sensible threshold for action should now be surpassed. I assign high confidence to Omicron not only being substantially fitter than Delta, but moreover that it amounts to a multiple-fold greater change in fitness compared to previous evolutionary jumps (i.e., Delta -> Omicron is a much greater and concerning jump than Wu1 -> Alpha and Alpha -> Delta). For the most part, the forecasts of the Metaculus community and of many experts with good track records are largely aligned with my views. As such, Omicron calls for a vastly different response than the world's response to Alpha and Delta. If we don't act quickly, most of the world will soon see a faster and larger peak than they've experienced previously. Tens of thousands of people — maybe hundreds of thousands —will die.
What follows is an overview of what we currently know about Omicron, in order from what we know most to least about: its (1) current epidemiological trajectory, (2) mutation profile, (3) immune evasion, (4) transmission fitness, and (5) lethality. I’ve used the Metaculus community predictions of the ~dozen Omicron questions we currently have to inform my own views on many these questions (for the sake of not making this article longer than it already is, I only explicitly discuss a few). I then (6) bring it all together and make firm recommendations that I am ready to stand behind.
1: Current epidemiological trajectory
On 25 November, South Africa shared that they were tracking the spread of a new variant. This variant was spreading widely in Gauteng province — South Africa's most populous province, with 15.9M people — and appeared to be overtaking other variants while also driving an increase in new cases. On 26 November, WHO gave this new variant the name Omicron — and on the same day, both WHO and ECDC classified Omicron as as a variant of concern.
Now, in their most recent update on 3 December, South Africa notes that almost all provinces are seeing sustained increases — and that Gauteng is now experiencing the highest effective reproductive number and quickest sustained increase in new cases it has had thus far in the pandemic:
Here's a nice graph by John Burn-Murdoch of the FT that summarizes daily cases, test positivity, and hospital admissions in all of South Africa using data available up through yesterday, 2 December:
As you can see in the above images, all South African provinces are experiencing a rapid increase in new cases. Moreover, South Africa as a whole is seeing cases, test positivity, and hospital admissions increase on a steeper trajectory than was the case during the start of South Africa's previous three waves (note: weekly hospital admissions will be backfilled and will soon show a steeper increase than currently appears to be the case). These increases are occurring in the context of almost everyone in South Africa having at least partial immunity against SARS-CoV-2 (either vaccine-induced or due to previous infection). So, taken together, this indicates that Omicron either has a larger R0 (higher intrinsic transmission, irrespective of immunity), substantially evades immunity (reinfections of those with some immunity), or has some combination of both — more on this later.
But do we really know that these increases can't be attributed to a founder effect (which we've seen before during this pandemic)? After all, at the moment South Africa is the only country that is experiencing the start of a new wave in the context of widespread Omicron circulation. While at this point last week — or even just a couple of days ago! – one might have seriously entertained this notion, it's now clear that this isn't the case. The rapid increases in South Africa have continued in tandem with Omicron overtaking Delta as the predominant variant in South Africa, and South Africa continues to experience exponential rises in the above key metrics. I assign a <1% chance to the spread of Omicron being due to a founder effect, and as such I don't even think it's worth running a Metaculus question on it.
OK, but wait. Even if this isn't a founder effect, what if this is something specific to South Africa — as was the case for the Beta variant? We've previously seen Beta become predominant in South Africa and a few other countries in southern African while never taking off anywhere else. This is a good point, and the reasons for this having occurred remain unclear (it probably has something to do with already-high population immunity in South Africa at the time). However, this much is clear: while Beta was able to outcompete Alpha (Alpha was the predominant variant globally before Delta) and prevent it from taking hold in South Africa, it could not compete against Delta. As you can in the below graphic I've taken from CoVariants, Delta was able to almost completely displace Beta in a little over 3 months:
How long has it taken Omicron to almost completely displace Delta? Just one month.
There are some caveats I should note here — namely, that sequence data for the end of November is still sparse and will continue to come in over the next few weeks. However, this lag means that most of the sequences currently available are not from samples taken in the past week (almost all those have yet to be deposited) — i.e., most of the already-deposited sequences aren't from the time since the alarm was pulled on Omicron. This has the effect of blunting a potential bias in sequencing that may soon become apparent: preferential sequencing of S-gene dropout samples in an attempt to find cases of Omicron (more on this in a bit). So Omicron genuinely does appear to have overtaken the already-very-transmissible Delta variant, and at this point my guess is that it's responsible for nearly 100% of new cases in South Africa (which will show up in the sequences that are deposited ~2 weeks from now).
So to sum up so far: there is a rapid increase in new cases, test positivity, and hospital admissions in all of South Africa, with the worst increases being seen in Gauteng — which was the first province to be hit by this new wave. These increases are being driven by Omicron, which we can be confident makes up a very large majority of new cases. And this is occurring in the context of a country with very high population immunity.
What's the situation outside of South Africa? Omicron has now been detected in dozens of other countries, which shouldn't come as a surprise. It also shouldn't come as a surprise that there now appears to be community transmission of Omicron in many countries, including the USA, UK, and mainland Europe. Despite the fact that the Omicron variant was detected and identified relatively early, travel restrictions imposed on countries in southern Africa are already too late. And once community transmission is truly underway, they'll basically be useless.
Going forward, countries will have to do a lot of sequencing to track spread of Omicron and try to contain its spread to buy time. Moreover, in the near future — if not already — most cases of S-gene dropout can be treated as Omicron until confirmed by sequencing. S-gene dropout occurs when some PCR assays (namely TaqPath) are used, since these target the 69-70 sequence location of the spike protein. 69-70 is deleted in Omicron — thus causing that part of the assay to fail (it "drops out"). Many are now using this as a signal to preferentially sequence those samples, since this indicates the possible presence of Omicron. So, S-gene dropout on its own can be used as a leading indicator of Omicron frequency (some of you may recall that this was also one way Alpha, which also has del69-70, was tracked in late 2020). Indeed, there already appears to be a rapid increase in the frequency of S-gene dropout in the UK, which probably indicates the start of community transmission of Omicron there:
OK — so Omicron is not only already present in dozens of countries outside southern Africa, but is also already causing community transmission. When, if ever, can we expect it to overtake Delta? The Metaculus community expects this to happen in the USA sometime in early 2022, probably January, February, or March (median: 10 February). Given the Metaculus community's expected % per-day growth rate of Omicron and how long it has taken for previous variants to take off, I largely agree with the community here — except my timeline is a little more aggressive. I think Omicron will become predominant in the USA in either January or February. Crucially, both the community and I assign a small probability to Omicron not overtaking Delta until July (if Omicron is fitter than Delta to even a small degree, it will very likely have overtaken Delta before July) — the community assigns >1 July 2022 an 8% chance, and I give it a 4% chance. In other words: there is a pretty small chance that Omicron won't overtake Delta, meaning Omicron is almost certainly fitter.
I expect that forecasts for Omicron take-off in other Western countries would be very similar to those for the USA — i.e., a >90% chance that Omicron will overtake Delta sometime in early 2022. There will likely be more of delay in Omicron take-off in Latin America and Asia since there are far fewer flights between South Africa and countries in Latin America and Asia.
So, the epidemiological trajectory of Omicron is already beginning to make it clear it's very likely fitter than Delta — but why? Let's start by taking a look at Omicron's mutation profile.
2: Mutation profile
Omicron has ~36 characteristic amino acid substitutions ("mutations") in the spike protein. The spike protein is the target of vaccines and it contains the receptor-binding domain that mediates viral entry into host cells. Before Omicron emerged, results of deep-mutational scanning experiments predicted that changes to multiple spike sequence sites would negatively impact the ability of antibodies to bind to SARS-CoV-2. Omicron has many of these changes. The expectation is thus that many of Omicron's mutations will have some degree of an antigenic effect. From Jesse Bloom:
The sheer number of mutations that Omicron has accrued is substantial, and is clear in this table by CoVariants:
That middle column with the long list of mutations? That's Omicron. And it has dozens of spike mutations that differentiate it from Delta. Delta itself had fewer than 10 key mutations that differentiated it from Alpha.
Here are the visualized spike mutations in Delta vs. Omicron:
I won't go into detail here describing all the mutations Omicron has — but suffice to say that it has many mutations that have been previously been linked to immune erosion and/or are predicted to have an adverse impact on immunity.
What's the result of all these mutations for how Omicron compares to other variants, evolutionarily? The magnitude of Omicron's jump, which is evident in this phylogenetic tree constructed via NextStrain, is striking:
Next, let's discuss the implications all these mutations and the resultant evolutionary jump has for Omicron's transmission fitness and ability to evade existing immunity.
3: Immune erosion
Before anything, here's a quick pre-Omicron SARS-CoV-2 immunology 101 lesson:
Immunity against SARS-CoV-2 can be induced via vaccination or infection. A combination of the two elicits the broadest and most potent immunological response. Vaccine-induced immunity is targeted at the spike protein, while infection-induced immunity is broader and targets other parts of SARS-CoV-2 in addition to the spike. Both B cells and T cells are involved in the adaptive immune response against SARS-CoV-2, and some degree of both appear to be required for protection against disease. B cells are more critical in acting like a first line of the adaptive immune defense to prevent SARS-CoV-2 from gaining a foothold in the first place — ~2/3 of vaccine efficacy against disease is thought to be mediated via neutralizing antibodies that are produced by B cells. The T cell response is very multifaceted and complex but, essentially, it is largely responsible for preventing progression to severe disease by stopping the virus from continuing to spread within the body (T cells do this by killing infected host cells, for instance). So basically: antibodies produced by B cells protect cells from getting the virus (maintaining virus-free cells) and thus mostly protect against infection/disease, while T cells can mostly only act against cell-associated virus and thus mostly protect against severe disease/death. Keep all this in mind for later.
OK, now that we generally know what constitutes immunity let's look at how immunity has been holding up against Delta, courtesy of Eric Topol:
Though vaccine effectiveness is still maintained to a significant degree, there is clearly lowered vaccine efficacy — probably due to some combination of waning immunity (mostly because of a drop in neutralizing antibody levels over time) and Delta's partial ability to evade immunity (due to some antibodies being unable to target Delta). The T cell response is thought to be less affected, and so protection against death is mostly maintained.
Fortunately, a booster targeted at the original Wu1 not only brings vaccine effectiveness back up to pre-Delta levels but actually makes for an even better immune response (for the mRNA vaccines, higher than 95% efficacy!). A huge recent study on seven vaccines as boosters also finds that boosters result in a fantastic increase in B cell and T cell responses — both the quality and quantity of which are greatly improved.
So, the general pre-Omicron picture was that boosters would bring immunity at least back up to pre-Delta levels. And the picture was even better for those with hybrid immunity (those who are both vaccinated and previously infected).
OK, now let's try to understand how Omicron might impact immunity. Let's start by quickly looking at how immunity has held up against previous variants. A very large Dutch study came out recently that tried to quantify the odds of non-boosted vaccinated people being infected with Beta, Gamma, or Delta compared to Alpha. They found: 3.1x greater odds of non-boosted vaccinated people being infected with Beta, 2.3x odds for Gamma, and 1.9x for Delta (of note: there appeared to be no increased risk for reinfection with Beta, Gamma or Delta variants relative to Alpha for those with infection-induced immunity — probably because infection-induced immunity is not only targeted at spike and is broader).
Of the variants looked at in this Dutch study, Omicron is most similar to Beta (but still very different from it!), so we can infer that the infection odds for non-boosted vaccinated people is at least 3.1x that of being reinfected with Alpha, and ~80% greater than the odds of being reinfected with Delta. Given this and Omicron's more worrisome mutation profile, I'll go out on a limb and say we at least know for sure that the drop in vaccine efficacy will be greater than this (we do have a few studies that look at vaccine effectiveness against Beta, but they're small and their results are mixed). The extent to which Wu1 boosters will counteract this drop is unknown at this point, but it will surely counteract it to a substantial degree.
A pre-Omicron lab study backs up that Omicron will likely result in a large drop in vaccine effectiveness and result in a big increase in re-infection risk. Researchers looked at the impact of a polymutant spike with more than 20 mutations, many of which it has in common with Omicron. They found a significant reduction in neutralization in pseudovirus neutralization assays: "we show that 20 naturally occurring mutations in the SARS-CoV-2 spike protein are sufficient to generate pseudotypes with near-complete resistance to the polyclonal neutralizing antibodies generated by individuals who are convalescent or recipients who received an mRNA vaccine" (however, again note that those with hybrid immunity are better off: "plasma from individuals who had been infected and subsequently received mRNA vaccination neutralized pseudotypes bearing this highly resistant SARS-CoV-2 polymutant spike").
Here is the polymutant spike constructed by researchers vs. Omicron's spike, courtesy of Fabian Schmidt:
Omicron looks even more concerning than their polymutant spike!
Finally, let's look at the initial estimates of reinfection with Omicron based on data coming out of South Africa. The best estimate I've found is from a recent pre-print: 2.39 (95% CI: 1.88–3.11). However this, if anything, is a substantial underestimate since it includes cases from a time period in which Omicron was still in the process of overtaking Delta — so many of these reinfections are caused by Delta, not Omicron. My guess is this number should be at least the number at higher end of the confidence interval, so ~3.
So we have an at least ~3x increase in reinfection hazard on top of the 1.9x reinfection odds for Delta estimated by the Dutch study. We can't(?) directly combine these since one looks at the hazard ratio and the other looks at the odds ratio, but it is pretty clear this adds up to a very substantial increase in reinfection risk. For example, if we assume that non-boosted mRNA-vaccinated individuals have 80% protection against Delta (down from ~95% against Wu1), then a ~3x higher infection risk with Omicron vs. Delta would mean protection drops from ~80% to 40%. This drop would entail a multiple-fold drop in neutralizing antibody activity (expect to see some big numbers from lab studies in the next ~2 weks). And this drop is conservative, since some of this greater infection risk would be relative to Beta and not just Delta. Wu1-boosted people will fair better, but to an unknown degree.
Protection against severe disease (hospitalization/death) will likely fare better due to the broader range of the T cell response. But it will also likely take a big hit:
Mutations occurring on specific epitopes (part of virus recognized by antibodies) will probably affect the recognition of the epitopes. Omicron mutations impact 348 T cell epitopes (~27% of the total) and 550 B cell epitopes (~31% of the total), which is far greater (~2-3x) than the number epitopes impacted by Beta, Gamma, and Delta. (4 December clarification: this specifically discusses linear spike epitopes)
Taking together all of the above, it's little wonder that the Metaculus community assigns a ~3-in-4 chance to the UK, EU, or USA authorizing an Omicron-specific booster before 2023. I give it a 91% chance. Note that such an authorization would indicate that regulatory agencies see a need to update vaccines in response to Omicron, which is a high bar to meet — this option was never even seriously considered in response to Delta.
The next two (short!) sections, (4) transmission fitness and (5) lethality, are more speculative and uncertain than what you've read so far. As such, feel free to skip ahead to "6: Bringing it all together & recommendations"
4: Transmission fitness
There is a big question mark regarding why Omicron appears to be more fit than Delta. I hope I've already convinced you that this is at least in part due to its ability to dodge immunity, but we don't really know if this explains most of the additional transmission or just some of it. This is because it's possible that Omicron also has a greater intrinsic ability to spread irrespective of immunity — i.e., a higher R0. Trevor Bedford has a nice thread that explains this better than I can and where he also explores some of the possibilities. Basically, there are many different scenarios regarding how large Omicron's R0 might be, which is tied to the extent of immune erosion (higher R0 = lower immune evasion, and vice versa, given the Rt of ~2-3 we're seeing).
However, despite the uncertainty here I want to emphasize that it does not really make sense for the R0 of Omicron to be lower than that of Delta (I've seen multiple suggestions of this!). If R0 of Omicron were lower than that of Delta, then complete or nearly-complete immune escape would have to be occurring — I think this is extremely unlikely. And in any case, a recent analysis that uses data from South Africa indicates that Omicron's R0 is indeed not smaller than that of Delta. Moreover, the Metaculus community mostly agrees that Omicron's R0 is greater than that of Delta to some degree (i.e., greater than ~5).
The current Metaculus community's median matches my own for Omicron's R0: 6.5. Using current estimates of Rt, this translates to immune escape of ~35% under conditions of high population immunity (which is the case for South Africa).
We know so little about Omicron's relative lethality that I almost didn't include this section, but thought it was necessary to ground expectations. Many have suggested that Omicron might be milder than previous variants, while others have suggested it may be deadlier. I don't expect strong indications that either will be the case, though this is in disagreement with the Metaculus community:
The Metaculus community thinks there's a 29% chance that convincing evidence will emerge of a statistically significant increase in Omicron's mortality relative to Delta. I give this a 20% chance. On the other hand, the Metaculus community assigns a 52%(!) chance that convincing evidence will emerge of a statistically significant decrease in Omicron's mortality relative to Delta. I give this an 11% chance.
My thinking here is that the emergence of previous variants has resulted in, if anything, an increased risk of severe disease. However, evidence is mixed and so I don't think there will be conclusive evidence either way — but if there is, I think there's a ~2x higher likelihood of greater lethality than lower lethality. The Metaculus community strongly disagrees, largely on the basis of initial reports coming out of South Africa and Israel. However, I've usually been wrong when I've veered far away from the community prediction. I hope Omicron is indeed more mild!
Time will tell — little is known at the moment since most of South Africa's Omicron cases have not progressed to severe disease yet. A clear picture with respect to lethality won't emerge until sometime in 2022.
6: Bringing it all together & recommendations
Bringing it all together: Though it's only been a week since South Africa alerted the world to Omicron, I believe the following is very likely (>90% confidence) to be the case:
- Omicron's epidemiological trajectory, mutation profile and anticipated immune erosion all point in the same direction. This variant is much fitter than Delta and, as such, will overtake it almost everywhere by sometime in early 2022. Omicron will result in a much greater decrease in vaccine efficacy than anything we've seen so far. Vaccines will probably end up being updated to target Omicron, but even under optimistic scenarios Omicron-specific boosters won't arrive until sometime after the first Omicron-driven wave hits. Omicron-driven waves will peak fast and high. There is good reason to think the drop in vaccine efficacy against severe disease won't be nearly as drastic as the decrease in efficacy against infection or symptomatic disease, but it will still be a large drop.
For the time being, I am very uncertain as to what exactly explains Omicron's transmission fitness (though immune evasion is certainly a big part of it) and even more uncertain as to its lethality relative to Delta. Stay tuned.
By far my most important recommendation is the one that is unfortunately the least likely to be heeded: given what we already know, vaccine manufacturers should immediately switch all production to begin making Omicron-specific doses — and governments should commit in advance (now!) to buying these doses. However, this doesn't seem like it'll happen — it appears as if both vaccine manufacturers and governments will wait until there is conclusive evidence from lab studies and real-world efficacy data that there is a need for Omicron-specific doses. This will take many weeks, if not months. (4 December clarification: the UK government has recently ordered more boosters, though these aren't necessarily Omicron-specific)
The next best recommendations I can make are the following:
- Authorize and widely distribute Paxlovid, an oral antiviral that has demonstrated 89% reduction in risk of hospitalization/death. Given its mechanism, the efficacy of Paxlovid will almost certainly not be compromised by Omicron.
- Accelerate distribution of Wu1 boosters. These will provide critical extra protection against Omicron, though they won't be anywhere near perfect (don't worry about original antigenic sin, I think the risk here is greatly overstated — even if you get a booster now and go on to get an Omicron-specific booster in the first half of 2022, I'll bet your immune system will respond quite nicely).
- Accelerate distribution of first doses to the rest of the Global South! It's insane that this hasn't happened yet, especially given that the age-specific IFRs in low-income countries are 1.3-2.5x higher than in high-income countries.
- Do bidirectional contact tracing to slow the spread of Omicron and buy as much time as possible.
- Wear a high-quality mask in indoor spaces. Ideally, governments will encourage use of high-quality masks before they resort to more stringent measures (and if they do end up resorting to stringent measures, then they should start with the most effective ones).
- Continue to scale-up genomic sequencing. This will be be key to identifying cases of Omicron and tracking its spread — and, in the future, will be key to identifying any new variants of concern. It is thanks to South Africa's fantastic genomic sequencing program that we've identified and begun to characterize Omicron this early on. Let's not squander the opportunity we've been given.