In the wake of a viral outbreak, search trends have been analyzed as a way to measure how aware people are about what is going on. In this paper we take those same data and analyze their correlation with an impending pandemic threat.
Brado has been tracking COVID-19 Google search statistics over the last three weeks. The purpose was to find a solution to the following question:
Is there a substantial link between increases in symptom-related Google searches and an increase in Coronavirus case numbers throughout the United States?
In a nutshell, the response was an emphatic “Yes.” All of the info we discovered with takeout is included here.
A Few Quick Links
We contacted SEMrush for more click-stream data to validate the findings in Google Trends.
We looked at the association in more detail using SEMrush’s March click-stream data. We concentrated on search activity in Oklahoma, Michigan, and New York to save time.
|New York has a population of almost 20 million people.||138,863||5,489|
|Michigan has a population of roughly ten million people.||18,970||845|
|Oklahoma has a population of about 4 million people.||1,472||67|
While New York may be at the peak of the epidemic, and Michigan may be a few weeks away, we selected Oklahoma since they have yet to enact shelter-in-place orders.
Findings from SEMrush’s Click-Stream Data
The spike in searches for phrases like anosmia (uhnaazmeeuh) and loss of taste demonstrates a clear link to the increase in COVID-19 cases. In all three states, the same symptom-based searches are increasing year after year.
SEMrush and Google Trends analysis
We loaded the raw data into Google Data Studio to make it simpler to evaluate. The first graphic depicts SEMrush data, while the second shows Google Trends data. You may compare and pick several symptom-based searches by state and case count increase.
Dataset from Google Data Studio
The graph above demonstrates a significant link between searches for loss of taste and smell and COVID-19 outbreaks. From 2017 to March 2020, however, we found no link between eye-related symptoms and COVID-19 occurrences.
Searches for eye-related symptoms reveal no apparent link to COVID-19.
Is Google Search a Mirror or a Window?
Clearly, correlation does not imply causality. In addition, we must consider how much the media impacts search behavior. Anosmia searches may be useful in answering this issue.
Prior to the media attention, anosmia searches were on the rise.
On March 20th, the media began to extensively report on the loss of smell and taste as possible COVID-19 symptoms.
Searches for anosmia spiked before March 20th, according to data from Oklahoma, Michigan, and New York. In the New York Times, journalist and quantitative expert Seth Stephens-Davidowitz presented surprisingly identical findings. These data show that Google search is more of a window than a mirror reflecting symptoms reported in the media.
Searches for March 2020
In mid-March, compared to the beginning of the month, search data from Oklahoma, Michigan, and New York indicates large rises in anosmia, dry cough, and loss of taste and smell. When comparing data from 2017, 2018, and 2019 for the same areas, the similar surge can be detected.
Each picture below has a link to further information.
Anosmia/Symptom-Related Search Spikes: What Can They Be Used For?
- To assist in the prioritization or rationing of limited medical items like ventilators.
- To predict infection surges in the second or third waves.
- To provide to public health model makers, who may be able to utilize this information to target appropriate control actions (such as when to tighten social-distancing and shelter-in-place restrictions).
Steps to Follow
- Brado will add data from all 50 states to Google Data Studio. Make a note of it!
- Throughout April, we’ll continue to collaborate with SEMrush to look for possible symptom increases by state.
- We’d love to hear about any new ideas or patterns you or your team have discovered that might be useful to the study.
While we can’t establish causality, we feel this substantial association warrants additional investigation by the search and data mining communities, as well as the healthcare sector.
These searches might be critical in directing medical supplies, such as ventilators, to the “hottest” places from a trend standpoint. As previously stated, predictive search data might aid in the prediction of second- or third-wave viral outbreaks.
Perhaps the most successful application isn’t simply medical; perhaps it’s also cultural and political – when developing standards and infrastructure for public health modeling – when deciding whether to implement social distance and shelter-in-place limitations.
So, what’s the point? As Brado data scientists, our work is forever transformed. For our organization and our customers, this is the turning point and a new paradigm. We have instruments to forecast and maybe assist prevent the transmission of future infections, in addition to compassion and care.