Thread by @the_yaniv

What we can learn from 20 waves 🌊 of SARS-CoV-2?
A thread 🧵 1/11
👇👇👇

The figure 👆 shows the strongest wave (⬛️grey lines) in 20 countries around 🌍 with the largest testing per capita. The peak is centered to be at day 0 and the cases are normalized to its height. Black: the averaged wave 2/11

We define a wave to start and finish with ½ of the cases in the peak. For instance, here is an example from Cyprus 🇨🇾. The peak (🔴) at about 1100 daily cases per 1M. The wave's start & finish are labeled w/ an 🟧line at ~550 cases.
Gray: raw data
Blue: smoothed (our input)
3/11

Why are we interested only at ½ of the cases of the peak?
1. This is the time of last doubling.
2. As you well know, not all waves have strong valleys (here 👇is an example from Italy🇮🇹).
3. In signal processing, pulse width is usually defined as ½ of the peak.

4/11

Back to the initial figure!

The average wave takes 49 days (SD=27days) with a median of 37 days 🥴. Waves don't only feel long, but they are long...
5/11

Many people expect that the waves are symmetrical. In other words, they believe that it takes the same amount of time to go from half of the peak to the peak and from the peak back again to half of the cases.
Let's tests it empirically using 20 waves!
6/11

This density histogram (gray bars) shows the difference between the # days post-peak versus # days pre-peak (black: smoothed distribution).

In 75% of the waves, the post-wave was longer. And it was significantly longer (p<0.02, t-test), by 11 days on average!
7/11

📢An intermediate summary: once the sh*t 💩 hits the fan, it takes longer to go back to normal* than it takes to get to the peak from normal*.

*I say normal but our lives are anything but normal.
8/11

So post-waves are longer - but do they generate more cases?
The answer is YES!

The histogram shows the ratio between total cases in the post-wave compared to pre-wave (notice the log10 scale).

On average, post-waves generate 1.4x more cases than their pre-waves!
9/11

Final message:
Even if cases peaked and going down now in your area, watch out 👀. Most people will get Covid19 at the post-peak days than at the pre-peak days...
Stay safe!
10/11

PS:
All data from @OurWorldInData and using up to Oct 1st, 2021 data to avoid Omicron.

Peak and half-peak detection is automatic and uses Python Scipy peak detector.

Input for peak detection = smoothed cases using a Gaussian window of 14 days, std=5, centered.
11/11
END