The Israeli ministry of health (MoH) releases periodic vaccination reports on its Telegram site. Today, the following table was presented that breaks down vaccination status by age groups as of September 14, 2021 (thanks to Ran Israeli for sending to me and translating):
The tables, going right to left (as is read in Hebrew), are:
Age group
Total population (in 1000s)
Total number receiving at least 1 dose (in 1000s)
Total number receiving at least 2 doses (in 1000s)
Total number receiving booster (3 doses, in 1000s)
Total number recovered from previous infection and unvaccinated (in 1000s)
From this information, we can calculate the total number unvaccinated, given 1 dose, given 2 doses, and given 3 doses, and can construct the following table summarizing the state of Israeli vaccination as of September 14, 2021:
Looking at the 3 dose column, you can see how quickly the Israeli booster program has progressed. In just 5 weeks, they have given third dose boosters to nearly 40% of the 12yr+ population, with >70% in the 60+ group, >50% of the 40-59 group, >25% of the 20-39 group, and even >15% of those in the 16-19 group. However, this post is not about the boosters -- it is about the unvaccinated who were previously infected. As I have repeatedly emphasized, Israel initially did not make vaccines available to previously infected at all, and only made a single dose available starting in March, and it was optional since they awarded "Green Pass" immunity passports to anyone previously infected, whether vaccinated or not. Thus, I have argued that a substantial proportion of the unvaccinated are very likely to have been previously infected.
Indeed, this is true. The last column of the table computes the proportion of the unvaccinated individuals that were previously infected. We see >30% of the total unvaccinated Israeli residents 12+ are previously infected. It is >1/3 in all age groups 20-59yr, a full 38% of the 20-39 age group. Why is this a big deal? Because as we know from numerous publications (as highlighted in a blog post in April), those previously infected as strongly protected against re-infection even if unvaccinated. One paper out of Israel released in April found previous infection was 94.8% protective vs. reinfection and 96.8% protection vs. severe COVID-19, of the same order of magnitude as mRNA vaccination at that time. Many other papers found similar 90-95% protection numbers, and these included many who were previously infected 9-12m prior. While the exact numbers may have changed, at this point it is incontrovertible that previously infected have substantial protection vs. reinfection whether vaccinated or not. (Note: other studies show this risk of reinfection can be further reduced by 1/2 with a single dose of vaccination, so vaccination is still recommended for previously infected, but the point here is that even without vaccination the protection is substantial).
As a result, any vaccine effectiveness estimates that are done on the Israeli data will be strongly attenuated if the previously infected are not removed from the unvaccinated before calculation.
Unfortunately, the current Israeli MoH dashboard does not split these out, but lumps them in with all of the other unvaccinated. Israeli researchers with more complete access to the data routinely remove these previously infected before doing their modeling to estimate vaccine effectiveness (see here and here and here and here and here), but it is not clear whether this is always done for the reports coming from Israeli data.
Waning Efficacy Reports out of Israel
There were a series of press releases that came out of Israel that raised alarm about waning vaccine effectiveness, reporting reduction of vaccine effectiveness to 64% on July 5, and down to 39% on July 22nd. These press releases lacked context -- they did not mention the precise data they used or how they did their analysis. A report was released that showed that vaccine effectiveness vs. infection for those vaccinated in February had declined to 44%, and for those vaccinated in January to 16%. These reports sounded major alarm around the world, and led the emergence of a narrative that "Israeli data shows the vaccines don't work any more."
However, subsequent published papers presenting analyses of Israeli data, while adjusting for confounding factors, found that, while there was waning effectiveness vs. infection, it was not nearly as dire as those reports suggested.
Three preprint papers posted in August 2021 (Mizrahi et al., Israel et al., and Goldberg et al.) found an increased risk of breakthrough infection of 1.6x-2.2x in those vaccinated early (January-February) vs. those vaccinated later. Of these, only the Goldberg paper compared with unvaccinated controls to estimate the vaccine effectiveness vs. infection. They found that the vaccine efficacy vs. infection decreased from 75-80% for those vaccinated in April and May down to 50-65% for those vaccinated in January and February, a steep decline for sure, but nowhere near the 16%-44% in the MoH report. These analyses adjusted for various confounders, but importantly also separated out previously infected before doing their analysis.
If the MoH analyses did not separate out the previously infected, it is possible this is one major source of discrepancy that made the magnitude of waning appear so much greater. (BTW, based on the MoH methodology document linked below pointed out to me by Nurit Baytch, they do routinely remove previously infected from both unvaccinated and vaccinated groups -- if they did that here then there must be some other reason for the discrepancy).
A Hypothetical Example. To demonstrate the effect that failing to remove previously infected from the unvaccinated group can have, I have constructed the following hypothetical example. Suppose we want to estimate Vaccine Effectiveness vs. infection for a given time frame, say for the month of August. We make the following assumptions:
Population of 1 million
75% vaccination rate
1/3 of unvaccinated are previously infected
Infection rate of 0.01 for in the unvaccinated, not previously infected group in August.
Previous infection reduces risk of infection by 90%
Vaccines reduces risk of infection by 65% (i.e. vaccine efficacy 65%)
From this we have the following table:
The top row shows the numbers producing a vaccine effectiveness of 65% vs. infection.
The bottom row assumes we lumped the previously infected with the rest of the unvaccinated when computing the vaccine effectiveness. Note that this attenuates the estimated VE from 65% to 50%. If we had an actual vaccine efficacy of 50%, the same hypothetical situation would find that the VE estimate would be attenuated to 28.6% from 50%.
The key point here is that when computing vaccine effectiveness, it is crucial to separate out the previously infected from unvaccinated or the estimate can be severely attenuated.
This is true anyplace, but is of special concern in Israel given the way the handled vaccination for previously infected individuals.
For Israel, previously infected individuals were not given two doses of vaccine, so there should not be individuals in the fully vaccinated group who got two doses of vaccine and were previously infected. However, in other countries like the USA, full vaccination was done independent of previous infection status, meaning that there would be large numbers of previously infected people in the "fully vaccinated" group, as well. These should also be pulled out of the "fully vaccinated" group before computing vaccine effectiveness, since the previous infection may provide extra protection that could affect their groups.
In the Israeli data there could be some previously infected in the vaccinated data, since those given a single dose of vaccine and previously infected are listed in the full vaccinated category. One could argue this also causes some distortion but it is not nearly the same magnitude. This is because the infection rates in those given one dose and previously infected should be similar to the infection rates in those given two or three doses of vaccine but not previously infected, and certainly not 10x lower as is the infection rate in unvaccinated with previous infection vs those without. Further, while here we see ~33% of all unvaccinated are previously infected, it is clear that <<33% of the vaccinated are previously infected.
However, this is why ideally the data should be stratified based on both previous infection and vaccination status, so that vaccine effectiveness can be computed separately for the previously infected and not. This would allow a comprehensive analysis providing estimates of reduction of risk of infection from vaccination, previous infection, or previous infection plus vaccination relative to SARS-CoV-2 naive controls, i.e. unvaccinated and not previously infected. These summaries would provide the most complete information to inform policymakers in terms of their recommendations (or requirements) for vaccination after previous infection. Here is the spreadsheet where I calculated the table of vaccination and previous infection proportion by age: