Predicting Covid Cases using Google Trends on Symptoms

I realized by my own that it takes quite a while if a positive covid test result makes it into the official numbers.

  1. You start with symptoms
  2. You check where and when to test for corona – it will take some days before the test is made
  3. You wait for the test result – this can also take 1-2 days

If you were in direct contact with a person that was tested positive you have to go to quarantine right away in Germany – you are able to test after 10 days. Because of all those reasons I wanted to start with the symptoms as a early estimator for covid cases that are lagging behind for some days.

What are the “best” symptoms to look at?

I decided to go with “Geschmacksverlust” (Loss of taste) and “Geruchsverlust” (Loss of smell). Yes, there are some other symptoms out there – but both symptoms have the highest growth rates in search volume compared to Pre-Covid times.

We recently build an automated solution using PyTrends to pull data from Google Trends, calculate some time period comparisons like “Weekly Change” and “YTD Change” and render the full result with some JavaScript. This is what we get when running this with the symptoms:

Both symptoms are totally flat and had the first peak in march 2020 – the first covid wave in Germany. In October the search volume was growing heavily again and peaked in the Christmas – the second wave.

This week it seems like we are running into a big third wave – especially “geruchsverlust” (loss of smell) exploded when looking at the weekly change (+350%) – if we average both symptoms we average in a +73% growth week on week. Not good at all!

How does the symptom trends fit to the real cases?

Let’s have a look at the reported cases for that:

The data of the the actual cases cover 1-Year – consider this when comparing the google trends timelines (2-years). In my opinion the curves fit pretty well. If this is the case and the trends for symptoms are a good estimator we will see some exploding numbers the next days 🙁

I hear a lot of opinions that the case numbers are mainly driven by “testing more”. I do not agree with that – why should the search volume for the main symptoms increase then?

Ideas for additional symptoms or search terms we can use?

I would love to hear your feedback about that. We are currently working on a small free tool to create Google Trends reports like shown above easily. Hopefully it will be released next week so you are able to play around with it by your own.

Join the conversation on LinkedIn

I realized by my own that it takes quite a while if a positive covid test result makes it into the official numbers.

  1. You start with symptoms
  2. You check where and when to test for corona – it will take some days before the test is made
  3. You wait for the test result – this can also take 1-2 days

If you were in direct contact with a person that was tested positive you have to go to quarantine right away in Germany – you are able to test after 10 days. Because of all those reasons I wanted to start with the symptoms as a early estimator for covid cases that are lagging behind for some days.

What are the “best” symptoms to look at?

I decided to go with “Geschmacksverlust” (Loss of taste) and “Geruchsverlust” (Loss of smell). Yes, there are some other symptoms out there – but both symptoms have the highest growth rates in search volume compared to Pre-Covid times.

We recently build an automated solution using PyTrends to pull data from Google Trends, calculate some time period comparisons like “Weekly Change” and “YTD Change” and render the full result with some JavaScript. This is what we get when running this with the symptoms:

Both symptoms are totally flat and had the first peak in march 2020 – the first covid wave in Germany. In October the search volume was growing heavily again and peaked in the Christmas – the second wave.

This week it seems like we are running into a big third wave – especially “geruchsverlust” (loss of smell) exploded when looking at the weekly change (+350%) – if we average both symptoms we average in a +73% growth week on week. Not good at all!

How does the symptom trends fit to the real cases?

Let’s have a look at the reported cases for that:

The data of the the actual cases cover 1-Year – consider this when comparing the google trends timelines (2-years). In my opinion the curves fit pretty well. If this is the case and the trends for symptoms are a good estimator we will see some exploding numbers the next days 🙁

I hear a lot of opinions that the case numbers are mainly driven by “testing more”. I do not agree with that – why should the search volume for the main symptoms increase then?

Ideas for additional symptoms or search terms we can use?

I would love to hear your feedback about that. We are currently working on a small free tool to create Google Trends reports like shown above easily. Hopefully it will be released next week so you are able to play around with it by your own.

Join the conversation on LinkedIn

Comments

You must be logged in to post a comment.

Similar posts

Menu