Corona in Israel: The impact of the vaccine

After I predicted that Israel would have solved the pandemic by the “end of February” in my previous post, it’s time to check the stats.

Unlike in December 2020, when I had to extract the data by copying it from the graphs, the official data site offers now options for direct download:

Option for direct data download on the governmental Corona tracking site

Furthermore, many additional datasets are made publicly available in this portal.

Using these two data sources, I performed a simple analysis, looking for the impact of the vaccine.

To refresh our memory, Israel pursued a very aggressive vaccination plan starting in late December 2020.

The vaccine was rolled out by age group: open initially to seniors and people with preexisting conditions. Over time younger citizens could access the vaccine as well, so that by early February over 35% of the general population and the vast majority of seniors had already received at least one dose of the Pfizer vaccine.

Although the setup seemed great, the tension remained high — mainly due to the steadily high count of new cases.

7-day moving average of new cases in Israel (source)

Over 6,000 cases on average is a very high number. It would be equivalent to some 200,000 cases a day in the US or over 50,000 in either Germany, UK or France. Of course there was a reason to worry.

As we now know, the vaccine not only prevents symptoms from developing, it also prevents people from passing on the disease. In addition to that, among vaccinated individuals, Corona tests usually return negative even after a direct exposure.

In other words, we should see a decrease in everything:

  • Deaths
  • Hospitalizations
  • Cases

The problem is that the cases did not decrease as quickly as I would imagine (see the chart above).

On the other hand, the deaths seem to decline after peaking on January 25th:

7-day moving average of new cases in Israel (source)

Maybe all these declines were vaccine-unrelated and there will be a rebound like with the previous waves?

The answer to this looks like a very clear No.

Let me explain.

I put in one chart the number of deaths and the number of cases over time until January 7, approximately the latest date with zero impact from the vaccine.

7 day rolling sum cases vs. deaths, R2 0.66

As you can see the deaths (red line) follow the blue (cases) with some delay. The measurement of similarity between graphs, R2, stands at 0.66, as 1.00 indicates a perfect match. It means there is a strong correlation but you cannot just take the cases and directly predict the deaths.

Therefore I delay the deaths by 14 days. Meaning for example that on Nov 1st I show 7 day moving sum for

  • cases from Oct 25 to Nov 1
  • deaths from Nov 8th to Nov 14

The resulting graph looks like this:

7 day rolling sum cases vs. deaths, 14 day delay, R2 0.94

R2 jumps to a jaw-dropping 0.94. It means that knowing the case count we can predict the deaths that will occur in the following 14 days with almost perfect accuracy.

Now look at the development of the curves after mid-January:

Purple line indicates the presumable beginning of the “vaccine effect”

Starting mid-January there is a significant discrepancy between the cases and the deaths. The deaths are much lower given the number of cases. Fewer people dying is great news, but what exactly happened there?

To answer this, I show another simplified calculation that predicts deaths-in-two-weeks.

As you probably know, Corona impacts different age groups very differently:

Chart from my previous post on this topic; as of March 2021, little change occurred

Assuming these fatality rates and a 14-day delay between cases and deaths in the respective age group, I get this nice graphic:

Weekly deaths, 14 day delay, R2 0.91

The age adjusted simple prediction is doing overall a terrific job: R2 of 0.91 is pretty solid. Very important here: despite the vaccine effect, up to the end of February the age-adjusted prediction is very accurate.

This means, if you are above 70 and tested positive, your chances to die from Corona are roughly the same in February 2021 or at any point in 2020. Remember, vaccine prevents not just the symptoms but also the transmission and infection.

Inevitably it all means the thousands of daily cases in February are not the same as before that:

Distribution of new cases by major age groups, purple line indicates beginning of the vaccine-effect

The chart above shows that the vast majority of new cases was “always” produced by younger people (blue and red area), under age of 50, who are also much less affected by the virus. However, while their total share fluctuated between 75 and 80%, since January it continuously increased to a combined 90%. On the flip side it means, that highly vulnerable seniors’ share in total infections (green area) dropped from 7% over the course of the year to less that 2% in early March. It is hard to explain this drop other than with the vaccination of the seniors that started in late December.

Just look how the cases for the mostly never vaccinated youth (blue line) compares to mostly vaccinated seniors (green line). The numbers are normalized to 100% at the beginning of January when the new cases count was the highest for both groups.

Weekly cases by age group, normalized to 100% at the peak of the pandemic

While both groups’ new cases peak in the beginning of January, the seniors’ infections plummet to 10% of the peak within 4 weeks. At the same time youth’s new cases decreased only to 50% of the peak.

Everything said indicates that something happened in January that has a lasting and significant impact on the elderly population, while diffusing with younger age, so that we see little to no effect on the teenagers and children.

I think the only possible explanation is that we are observing the effect of the vaccine. According to another simplified calculation, in the first ten weeks, the vaccine saved over 1,000 lives in Israel.

Don’t hesitate to get a vaccine, especially if you are above 50 or have a pre-existing condition.

Data Scientist @Weel. Into Machine Learning, Data analysis using Python, GCP, G Suite. Love to get my hands dirty