Estimating real-world COVID-19 vaccine effectiveness in Israel! Now preprint version. TL;DR: Very effective. 66-85% for infection, 87-96% for severe disease.
This started when I tweeted the numbers provided by the ministry of health (MOH) regarding cases of vaccinated individuals. This tweet had more >250K views, so decided it might be better to formulate the analysis and provide full methodology.
Using the daily incidence of cases and daily vaccination counts and developed a formula to quantify the expected number of cases based on those numbers. I then combined it with observed counts provided by the MOH to calculate the effectiveness.
The problem is that those vaccinated (and especially early) have different disease dynamics than the general population. Different demographics and socio-economic levels. How do we correct for that?
Well, instead of correcting, I present the full range of reasonable scenarios and show how they affect the estimation of effectiveness.
For example, for positive cases of 60+, it looks like this. Without correcting (beta=1) we get 83% effectiveness a week after 2nd dose. If that group has half the chance to be positive, it reduces the estimation to 66%. So, we get a lower and upper bound - 66-83% effectiveness.
Note that half chance is very radical, before vaccinations the beta was found 0.75. What is the correct beta, Idk, but it is most probably in this range (0.5-1)
For positive cases <60, it looks like this. Pretty similar (reasonable beta is ~1). Note that in both plots it seems that the 1st dose is not very effective in reducing cases.
Finally, severe/critical/death cases. Chances for severe cases for older population are about 4-fold of the general population (beta=4), and the full reasonable range (beta=2-6) brings us to effectiveness of reducing severe disease by 87-96%.
Caveat here is that severe cases may be delayed and are expected to increase. We will have to see how those numbers change in the coming weeks.
We see that immediately after the 2nd dose the effectiveness jumps. Two options: 1. the 2nd dose is effective immediately. 2. The 1st dose is effective after ~3 weeks. There is some support for the 2nd option. We will have to wait and see numbers from the UK.
All the data used here is publicly available, and I deposited the code and data to recreate the analysis in github.com/dviraran/covid….
I also created an interactive app to allow users to play with the different parameters and see how they affect the estimation of the effectiveness - I also created an interactive app dviraran.shinyapps.io/VaccineEffectI…. Data will be updated once available.
I also want to cite other efforts that took different approaches to show that the vaccine is working by @eran_segal, @ShalitUri, @GazitDoron, and Maccabi.
In summary, this is the first attempt to estimate the effectiveness of the vaccine in Israel compared to the general population and based on observed counts of cases in vaccinated individuals. The analysis clearly shows that the vaccines are working very well. Go get your shot!
• • •
Missing some Tweet in this thread? You can try to
force a refresh
האנליזה הזו התחילה בציוץ על היעילות של החיסונים לפי הנתונים שמשרד הבריאות פרסם על החולים שהתחסנו. הציוץ הזה הגיע ליותר מ-250 אלף צפיות ודחף אותי לפרמל את האנליזה בצורה קצת יותר אקדמית.
בעזרת הנתונים היומיים על נדבקים וחולים ומספר המחוסנים חישבתי את מספר המקרים הצפויים ללא חיסונים. בעזרת המידע על חולים מחוסנים אפשר לחשב את היחס בין המצוי לצפוי.
How effective are the COVID-19 vaccines? Israel was early to vaccinate its population (~1/3 of pop now with the @pfizer@biontech vaccine) and real-wold data is starting to accumulate. However, as we are learning, calculating efficacy from RWD is complicated. A thread =>
The vaccination campaign in Israel coincided with the beginning of a 3rd wave and a couple weeks later a full lockdown (Israel style). In addition, there are other confounding factors - socio-economic and demographics differences in both the infections and the vaccinations.
This made the real-world data analysis of the effectiveness of the vaccine much more complicated than in the RCT, where all this doesn’t matter because of the randomization. How to tease out all the contradicting factors?
עדכנתי את הפרדיקציה לפי הנתונים החדשים, הפעם מתוקנן גם לכמות החולים היומית. 659 מאומתים שבוע אחרי המנה השניה, לעומת צפי של כ-4967 = 87%~ פחות מהצפוי.
עכשיו שגם יש חלוקה גם לגילאים, אפשר לשפר את התחזית לבני 60+ בלבד.