Scatterplot chart showing how states with more barriers to the COVID-19 vaccine also do fewer “vaccine” searches
Surgo Ventures found that states with higher COVID-19 vaccine barriers also did fewer vaccine searches.

How Google searches signal barriers to COVID-19 vaccine supply and demand

Analysis finds areas of the U.S. that face many vaccine barriers also showed lower demand for the vaccine

Surgo Ventures

--

This is the third in a three-part series deep-diving into what Surgo Ventures has learned about Google Trends as an additional monitoring tool for researchers and policymakers during pandemics and other major national concerns. In Part 1, we explored the broad impact of COVID-19 through national and state-level search trends for economic, social, and health-related queries juxtaposed with the US government’s policies, just as the pandemic was developing. In Part 2, we focused specifically on the timely topic of COVID-19 vaccines and developed a framework of sentinel indicators to track and analyze search interest in vaccine topics of ‘access’, ‘general’ information, and ‘risks’. Today, in our final post, we expand on this indicator framework using modeling to investigate how COVID-19 vaccine searches (especially ‘access’ searches) are related to actual vaccine uptake barriers, demand, and speed of rollout.

In previous work, we showed the utility of handpicked Google Trends indicators which combine several thematically consistent queries for vaccine access, general information, and risks. But how do we build on this indicator framework to monitor changes in COVID-19 vaccine information seeking? Can we connect any of these search indicators to real-life vaccine behaviors? This is our focus in today’s post.

Our findings have implications that could help the government to respond more quickly and effectively to people’s needs and worries — whether for this pandemic or a future one.

Want to read this story later? Save it in Journal.

Nowcasting vaccine rollout with Google Trends

In this post, we compare search volumes for each state with the COVID-19 Vaccine Coverage Index (CVAC), an index of state-level vaccination barriers we developed, which includes indicators such as lack of transportation, not having a family doctor, and low-performing health systems, among others. We also relate search volumes to the speed of the vaccine rollout. We’ll see how the signals from Google Trends complement these other data streams to improve vaccine allocation efforts.

What we found:

  • States with high barriers to vaccination also had lower search rates for vaccine access. This suggests that highly vulnerable states may struggle with public demand.
  • At the same time, however, states with resource-constrained healthcare systems showed more vigorous search patterns, which may reflect uncertainty about how to access the vaccine when healthcare systems are weak.
  • At least through April 2021, a spike in searches in a given week was linked with higher vaccination rates three weeks later, indicating demand which could be monitored.

Vaccine-vulnerable states search for vaccine access at lower rates

We developed our COVID Vaccine Coverage Index to identify counties and states at risk of slow vaccine rollout, due to either supply or demand concerns. The CVAC is directly correlated with vaccinations: states with higher barriers to deployment have seen much slower rollouts. However, the index only captures pre-existing barriers, and doesn’t take into account live demand, so Google Trends can be used as a proxy. (Later, we show how vaccine searches for access precede actual vaccine uptake).

Combining these two concepts, we uncover evidence of a double burden, whereby areas that face many vaccine barriers also faced lower vaccine demand from the public:

Line graph illustrating how states with high and medium barriers to COVID-19 vaccine rollout show consistently lower search rates than states with low rollout risk.
We’ve already seen that searches for vaccine access increase after the first Emergency Use Authorization (EUA) in December 2020. Here, we show that the increase was uneven: states with high and medium barriers to vaccine rollout, as measured by Surgo’s COVID-19 Vaccine Coverage index, show consistently lower search rates than states with low rollout risk. Source: Surgo Ventures, Google Health Trends.

As we saw in the last post, after vaccines were authorized, vaccine access searches increased everywhere. But search demand has been less robust in states that already face more barriers to rollout. The below chart also highlights significant regional clustering, with Northeastern states vigorously searching for COVID-19 vaccine access, whereas Southern states showed the fewest searches and the most structural barriers to vaccination success.

Scatterplot chart showing how states with more barriers to COVID-19 vaccination are searching less for vaccines.
The 50 states, plus DC, with their COVID-19 Vaccine Coverage index score plotted against search rates for vaccine access. States with more barriers to vaccination — those with a higher index score — are also searching less for vaccines. Data include searches after Dec. 11, 2020, the date of the Pfizer EUA. Source: Surgo Ventures, Google Health Trends

States with resource-constrained healthcare systems need clarity around vaccine access

Our CVAC index comprises five themes, each of which describe an axis of potential vulnerability in vaccine delivery:

  • Theme 1: Historic Undervaccination
  • Theme 2: Sociodemographic Barriers
  • Theme 3: Resource-Constrained Healthcare System
  • Theme 4: Healthcare Accessibility Barriers
  • Theme 5: Irregular Care Seeking Behavior

The interested reader can investigate each theme further on our CVAC website.

Using regression models, we unpacked the overall CVAC trend to show how each theme correlates with demand. We found that each of Themes 1, 2, 4, and 5 all embody the overall CVAC trend: States facing higher barriers also show lower search volumes. However, Theme 3 (Resource-constrained Healthcare System), has a positive association with access searches. After the EUA, a state with a 10-point higher Theme 3 score had about 1% more searches than its counterpart.

This makes sense: If a state has a struggling healthcare system, there is a greater burden on individuals to seek out their own information about obtaining a shot.

One could also hypothesize that those without a primary care doctor would also need to search more for vaccine access. Therefore it is a bit surprising that Theme 5 (Irregular Care Seeking Behavior) shows the opposite to be the case. We think that this supports a story, then, of depressed demand for vaccines in areas with irregular medical care patterns, rather than confusion about supply.

Scatterplot charts illustrating the 50 states, plus DC, with their COVID-19 Vaccine Coverage index theme scores plotted against search rates for COVID-19 vaccine access.
The 50 states, plus DC, with their COVID-19 Vaccine Coverage index theme scores plotted against search rates for vaccine access. Four of the five themes show that with increasing barriers come fewer searches for vaccines. However, Theme 3 shows that with fewer health resources comes more searching. Data include searches after Dec. 11, 2020, the date of the Pfizer EUA. Source: Surgo Ventures, Google Health Trends

Thus far we’ve examined Google searches over time through the context of a static risk index. But how do Google searches relate to contemporaneous vaccine rollout?

Vaccine rollout speed is linked to Google search trends

One hypothesis is that Google searches for vaccine access could be attributable, in part, to a supply-constrained system where vaccines are hard to find. To measure this, we examine Google Trends as a predictor of vaccine doses administered (daily, per 1 million population), at the state-week level. We use several lag periods in different directions:

  • Searches up to three weeks AHEAD of a vaccination week could mean that people are searching for appointments in preparation to receive their dose.
  • We also examine searches for up to 2 weeks AFTER a given vaccination week. A positive link would mean that as more people are vaccinated, more people search for vaccine access, implying a compounding effect of fast vaccinations. A negative link would mean that more vaccinations lead to fewer searches, indicating saturated demand.

Thus far there’s fairly strong evidence that Google Trends searches three weeks ahead have a strong positive relationship with vaccine administration. That is, the higher a state’s search volume for vaccine access, the more vaccines they are likely to administer three weeks from now (adjusted for the size of the population).

Chart showing how Google Trends searches three weeks ahead have a strong positive relationship with COVID-19 vaccine administration.
While holding a vaccination week in January and February as fixed, we examined searches up to three weeks before, and up to two weeks after, the week in question. We found a U-shaped curve: searches three weeks before and two weeks after are positively associated with vaccination rates. However, searches the week before and the week of are negatively associated, likely indicating that appointment slots had filled up and/or eligibility restrictions applied despite search activity. Source: Google Health Trends and Our World In Data.

In January and February, we also observed a significant negative association between searches and vaccinations within about a week of each other. This tells a compelling story that states with the most constrained rollout saw the most searching. As appointments filled up, searching intensified. This squares with reports of “vaccine hunters” traveling long distances to circumvent limited appointments or tight restrictions locally.

Such was the situation in January and February of 2021. Since early March, however, that association appears to have evened out: now, in all time periods, more searches indicate more vaccinations and vice versa. This is even more evidence that Google searches are good signals of vaccine demand. And now that vaccine eligibility is open to anyone who wants one, we expect that the association will strengthen in the short term as both searches and vaccination rates become more variable.

Chart illustrating how more Google searches means faster COVID-19 vaccinations, over all time periods analyzed.
While fixing a vaccination week in January through early April, we examine searches up to three weeks before, and up to two weeks after, the week in question. We now find that more searching means more vaccines, over all time periods. Source: Google Health Trends and Our World In Data.

For full results, including our search segmentation, PCA, and linear regressions, please contact us at COVID19@surgoventures.org.

Conclusion: Google Trends mirror real-world vaccine barriers and demand

We’ve seen that there is useful information in vaccine-related Google searches in predicting advanced demand for appointments, as well as underscoring barriers to vaccine deployment. We suggest that state health departments monitor Google searches for vaccines as one more tool to gauge public demand and sentiment of vaccines; e.g., access-related searches can be part of a broader strategy to monitor changes in vaccine demand. Given that information seeking for vaccine access is highest among the most vulnerable states with insufficient health care systems, policymakers in those states should consider public awareness campaigns for vaccination procedures to proactively respond to information deficits.

Recommendations:

  • If Google searches are high or trending upward, the state should prepare for more appointments in 1–3 weeks.
  • If Google searches are low or trending downward, the state should enhance campaigns to boost vaccine awareness & accessibility.
  • States with highly vulnerable health care systems (Theme 3 in Surgo’s COVID-19 Vaccine Coverage index) should be focused on awareness and accessibility of the appointments system.

This is Part 3 of a three-part series. Read Part 1 and Part 2.

This work was made possible by everyone at Surgo Ventures, including (in alphabetical order): Daisy Chung, Aaron Dibner-Dunlap, Tich Mangono, and Peter Smittenaar. Additional graphic design support from Katie Armstrong.

Enjoyed this post? Subscribe to the Machine Learnings newsletter for an easy understanding of AI advances shaping our world.

--

--

No responses yet