Google search trends reveal vaccine impact

Google searches for ‘chickenpox’ drop sharply in countries with government-mandated vaccinations.

Today’s Proceedings of the National Academy of Sciences (PNAS) study is the first to use a technique called digital epidemiology to observe the effectiveness of a vaccine. The idea behind the study is that Google searches would spike in response to outbreaks, as parents would use Google to find information about symptoms and treatment.

Lead author on the paper, Kevin Bakker, University of Michigan, speaks to us about the study.

ResearchGate: What inspired the use of Google searches to track vaccine impact?

Kevin Bakker: About a year ago Tyler Stevenson, senior author on the paper, wrote me with an interesting figure showing that Google searches in the UK for a few different childhood diseases were seasonal. My research focuses on seasonal outbreak dynamics of childhood diseases, so I quickly noticed that the 'peak season' of searches for these diseases matched the timing of the actual outbreaks. This led us to examine whether or not searches were seasonal at a global scale.

RG: What did you find? Can you walk us though how you came to these results?

Bakker: Our research team, which also included Micaela Martinez-Bakker and Barbara Helm, focused on chickenpox, since the symptoms are clear and rarely confused for other diseases. We studied 36 countries across five continents and found springtime peaks on the global scale. Second, we needed to ground-truth these data – basically does the seasonality of searches correspond with the seasonality of outbreaks? This was the tough part, as very few countries make these clinical case reports available, we were only able to find Mexico, Estonia, the US, Australia, and Thailand. Using these clinical cases, we found a strong correlation between Google searches and clinical cases in countries that do not have nationwide immunization (Estonia, Mexico, and Thailand), but a weaker correlation for the US and Australia – both countries that vaccinate. This confirmed that Google Trends could be used as a proxy for clinical cases, with greater strength in countries without the vaccine.

These results led us to wonder whether we could examine the impact of immunization on Google searches. To do this we needed to find a country that implemented vaccination after 2004, the first year of Google Trends data. Since the US and Australia started vaccination in 1996 and 2001 respectively, they were out. Both the US and Australia have noisy data with regards to Google searches (meaning there is no significant seasonal pattern in searches). Many of the searches in these countries were related to ‘chicken pox vaccine’.

We found that Germany implemented the chickenpox vaccine in late 2004, then nationalized payment for the vaccine in 2007, and finally required a second, booster, dose in 2009. At the beginning of the data in 2004, the searches for ‘chicken pox’ in Germany were very seasonal, but as they instituted stronger vaccine requirements, the seasonality decreased consistently until 2011, when it became random noise. Linking Google searches to immunization, we can surmise that the drop in search seasonality is tied to the implementation of vaccination in Germany. Independent studies in Germany, looking at the vaccine effectiveness, have found similar results – that the chickenpox vaccine has greatly reduced the incidence of chickenpox cases.

Fig 3
Google Trends data over time. Data are weekly; x axes indicate time, and y axes are the detrended Google data (same scale for all panels). Countries with universal (national) immunization are highlighted in red, countries with select (regional or municipal) immunization are highlighted in blue, and countries lacking any mandatory immunization are highlighted in black. Fig courtesy of Kevin Bakker.


RG: What is the relationship between vaccinations and a reduction in Google searches?

Bakker: In countries that vaccinate and report case data, like Australia and the US, the Google search seasonality is weak to non-existent – indicating that there are very few cases, but in order to prove this, we needed a country that implemented the vaccine, cue Germany. We demonstrated that the reduction of Google searches is correlated with increased vaccination, which is backed up by primary research that identified the sharp drop in clinical chickenpox cases in Germany.

RG: The lack of widespread internet access is one limitation with the use of Google search results in vaccine research, are there any others?

Bakker: Definitely the lack of global internet access is the biggest drawback to using digital epidemiology. I do want to stress that digital epidemiology is not a replacement for the actual reporting of clinical cases. Our study would not have been possible without ground-truthing the Google data with actual clinical case reports. But it is definitely a supplement that can be used with clinical data to optimize research. For example, we correlated Google searches for ‘chicken pox’ with clinical cases of chickenpox in five countries, but by demonstrating that Google searches are an appropriate proxy for clinical case reports, we were able to extrapolate and suggest that the peak seasonality of Google searches for ‘chicken pox’ in the other 31 countries we studied are likely strong indicators of outbreak detection. Again, this is important because clinical cases are not reported in those countries – meaning we previously had little idea what the dynamics looked like.

RG: What other applications do you see Google searchers or ‘digital epidemiology’ being used for? Do you have any plans for future research?

Bakker: I think the most beneficial use for digital epidemiology is in forecasting outbreaks. With clinical reporting, it takes time for a reported case to get into an online record, which researchers can then access. Using Google Trends, anyone can go online and see what the most popular search terms are this morning, anywhere in the world. Alternatively, previous studies have looked at trending Twitter hashtags, and Wikipedia page views.

Fast reporting can help public health officials predict when an outbreak may occur, and whether or not it will be a ‘bad’ year, with a larger number of cases. We built a forecasting model that could predict years with an increased number of cases a month in advance. The increase in searches for ‘chicken pox’ indicated when an outbreak was starting, and an abnormally high number of searches predicted that the outbreak was going to be worse that year. This method has also been termed ‘nowcasting’, using real-time or very recent data to predict outbreak dynamics. It’s all very exciting!

Our research team does have a few other ideas for digital epidemiology. We were able to demonstrate that searches for other childhood diseases – croup, fifth disease, and hand, foot, and mouth disease were also seasonal. However, we don’t have the clinical data to corroborate those findings. A simple search on Google Trends with other diseases likely indicates seasonality. A quick word of warning though, you need to be careful – an example is if you type in ‘breast cancer.’ There is a strong peak in November – does that mean it’s seasonal? No, it’s just breast cancer awareness month here in the States, so more people are Googling it.

Image courtesy of Pan American Health Organization-PAHO / World Health Organization-WHO.