Tracking online searches for emotional wellbeing concerns and coping strategies in the UK during the COVID-19 pandemic: a Google Trends analysis [version 1; peer review: awaiting peer review]

Background: The coronavirus disease 2019 (COVID-19) pandemic is the largest acute public health emergency of this century. Government intervention to contain the virus focuses on nonpharmacological approaches such as physical distancing/lockdown (stay-at-home orders). As the situation develops, the impact of these measures on mental health and coping strategies in individuals and the population is unknown. Methods: We used Google Trends data (01 Jan 2020 to 09 Jun 2020) to explore the changing pattern of public concern in the UK to government measures as indexed by changes in search frequency for topics related to mental distress as well as coping and resilience. We explored the changes of specific topics in relation to key dates during the pandemic. In addition, we examined terms whose search frequency increased most. Results: Following lockdown, public concerns as indexed by relative search trends were directly related to COVID-19 and practicalities such as ‘furlough’ (paid leave scheme for people in employment) in response to the pandemic. Over time, searches with the most substantial growth were no longer directly or indirectly related to COVID-19. In contrast to relatively stable rates of searches related to mental distress, the topics that demonstrated a sustained increase were those associated with coping and resilience such as exercise and learning new skills. Open Peer Review Reviewer Status AWAITING PEER REVIEW Any reports and responses or comments on the article can be found at the end of the article. Page 1 of 11 Wellcome Open Research 2020, 5:220 Last updated: 22 SEP 2020


Introduction
The coronavirus disease 2019  pandemic is the largest acute, global public health emergency of the century with over 500,000 deaths and 10 million people infected worldwide at the time of writing 1 . In the UK there were an estimated 43,550 deaths from COVID-19 by June 28 th 2020.
Public health measures to control the spread of the virus have had wide ranging effects. In particular, the lockdowns (stayat-home orders) that have been put in place in many countries have resulted in restrictions to everyday life, such as home confinement as well as school and work closures. Concerns have been raised about the impact of these measures, alongside those related to employment, the economy and the virus itself (secondary to infection, fear and bereavement), on population mental health and risk of suicide 2,3 . It is crucial to track the societal impact of all these factors to identify effects and possible unintended consequences in order to respond appropriately. The World Health Organization has released guidelines outlining coping and resilience strategies after COVID-19 was declared to be a pandemic 4 . Other suggested initiatives to avoid the exacerbation of mental health struggles that prolonged isolation may foster include adapted safety plans, and more generalised coping plans tailored to options available during lockdown that identify positive coping skills and strategies 5 .
It is essential to monitor mental and psychological wellbeing of the population in these uncertain times and to understand the coping and resilience strategies that individuals adopt to plan appropriate responses, promote adaptive behaviours and evaluate the impact of public health advice [2][3][4][5][6] . In the absence of immediately available surveillance data on the impact of the COVID-19 pandemic on population mental health and coping strategies, several researchers have used Google search trends data as a potential early proxy measure [7][8][9][10] Google Trends, is a publicly available source of near real-time internet search data which has been used for more than a decade by academics, including suicide prevention researchers, to investigate trends in healthrelated searching and the impact of population-level exposures such as celebrity suicides 11 . In a recent study we used Google Trends to investigate the changing pattern of public concerns (as indexed by internet searches) in relation to sentinel dates during the COVID-19 pandemic worldwide and in specific countries in the early weeks of the pandemic up to 30 March 2020, when there had been around 200,000 global deaths (2,000 in the UK) 8 . In this early phase of the pandemic we found marked increases in relative search volumes (RSVs) for indicators of financial, employment and educational concerns, whereas RSVs for depression and suicide dropped temporarily as the number of COVID-19 deaths increased, with some evidence that RSVs for suicide returned to pre-pandemic levels in subsequent weeks. Findings from other investigations are mixed. Sinyor et al., in their analysis of worldwide and US data, reported falls in searches for suicide, anxiety and hopelessness, but not depression in an analysis up to 4 th April 2020 7 . Another analysis of Google Trend data for the US between 16 th and 23 rd March 2020 reported a rise in searches indicative of anxiety, negative thoughts and suicidal ideation in the weeks before lockdown, but these subsequently levelled off 10 . By contrast, an analysis from India reported positive correlations between the number of COVID-19 deaths and searches depression, anxiety, insomnia and suicide between March 25, 2020, and May 16, 2020 9 .
In this analysis we focus on Google search trends in the UK up to 12 June 2020, a period covering 12 weeks of lockdown that began on 23 March 2020, with some slight relaxations in some UK regions announced on May 10 th and implemented from May 17 th . We extend our previous analysis to investigate trends in searching for topics related to coping and resilience, as well as mental distress, to assess how individuals were responding to the UK lockdown. In addition, we explored the search terms that experienced the most marked rises in search frequency during the pandemic.

Data source
We used Google Trends for this analysis, following the suggested checklist for documentation and development of our analysis 11 .
Within Google Trends, search words are defined either as terms or topics. Search terms include matches for each word in a query entered into a Google search box. Topics, in comparison, are broader and comprised of an aggregated group -defined by Google -of related terms which share the same concept in any language.
The Google Trends explorer allows the user to request trend data for a specific term or topic, location (e.g. Italy, UK) and time period (e.g. last 12 months). It also allows users to specify which Google query categories (e.g. games, health), and types of search (i.e. web, news, Google shopping, and YouTube searches) they are interested in. If the user doesn't specify a term or topic in the explorer, Google Trends will return results for the top and rising terms and topics, for the specified time period and location, and will include data for all categories and for web searches.
Google Trends uses a sampling method to produce a representative dataset of all Google searches each time data are extracted and therefore results may vary each time a query is made on Google Trends 12 . Google does not provide information on absolute numbers of searches. Instead it provides a normalised value to display search activity for a given term or topic at a specific period and location. Each data point is divided by the total volume of searches for a given area at the specified time period. The resulting number is then indexed between 0 and 100, where 100 is the maximum search interest for the topic/ term during that time in that location. These are called relative search volumes (RSVs). RSVs tell us how popular that topic was in a particular location during a specific time period.

Search strategy
Searches were carried out on June 12 th 2020. All searches included all Google query categories and included all web searches in the UK.
Time trends: In our previous analysis, we explored the change in trends of key topics related to mental distress 8 . For the current analysis we excluded previously investigated mental distress topics which had low to moderate RSVs and were stable over time (loneliness and fear) and included the topic of self-harm given that self-harm is a strong predictor of future suicide 13,14 . We also explored topics related to coping and resilience. The topics identified for inclusion in this study were selected based on the considerations highlighted by the World Health Organisation 4 and those that mapped onto Google defined topics. We selected nine topics associated with coping and resilience, divided into two categories: those that were associated with self-agency (exercise, helpline, how-to, videotelephony, volunteering) and abstract themes associated with feelings and long-term goals (coping, hope, kindness and rainbow). The former topics were related to searches associated with learning new skills (how-to), changing behaviours (exercise) and connecting with people (helpline, videotelephony, volunteering). 'Rainbow' was included as a topic to indicate community and social connection during a period of separation. This referred to images of rainbows displayed by many UK households in their windows indicating solidarity with National Health Service (NHS) workers and the community. Hope, coping and kindness were included to examine potential signs of optimism in contrast to the developing pandemic and the restrictions associated with it. These topics were intended as a proxy measurement of individual responses to a period of disruption combined with coping and resilience strategies associated with psychological wellbeing.
In addition to exploring specific search terms, we also examined the related Google-defined search topics (Table 1). Some topics have overlapping related search terms (e.g. "anxiety" is in both anxiety and depression topics).
Daily RSV data between 01/01/2020 and 09/06/2020 were downloaded, and this formed the basis of our main analysis. Periods with very low search volumes (i.e. RSVs <1) are identified as zero activity. Based on previous recommendations, we also repeated all searches on different days using the same search parameters 15 . There is no recommended standard for the number of days on which to repeat the searches, but consistent with our previous analysis, we did so on 7 different days 8 . These additional data downloads were used to create our validation dataset (see sensitivity analysis).

Data analysis
Time trends: We provide graphical presentations of Google Trends data between 01/01/2020 and 09/06/2020 for mental distress topics (depression, anxiety, suicide and self-harm). Two graphs are presented for topics related to coping and resilience based on whether the topics relate to action and/or short-term distractions or abstract themes related to feelings and long-term goals. Key dates related to the pandemic are marked on the graphs and relate to the first UK death (05/03/2020), UK lockdown (23/03/2020), and the easing of lockdown measures (10/05/2020). When peaks in the time trends were observed, we explored what people were searching for by looking at the related search queries.
Rising terms: For each rising search term identified, a percentage value is provided for the increase of that term in the given period of study. However, for terms that experienced growth in excess of 5000% within Google Trends (i.e. breakout terms), data were not available to calculate an exact percentage increase. In order to make meaningful comparisons, we approximated the percentage increase value of each rising term that was a breakout term. For each breakout term we obtained the average RSVs for that term for the period under investigation and multiplied this by 5000. This ensured that the breakout terms maintained approximately their relative weight compared to the other breakout terms for each specified time period and were displayed (see below) appropriately. For example, the rising breakout term "furlough" (UK government's paid leave scheme) had a more sustained increase in search frequency during the post-lockdown period compared to the breakout term "boris johnson hospitalised" (following the UK Prime Minister's hospitalisation for coronavirus) (Figure 2).
Word clouds are a way of graphically representing text data and are often used to provide a visual representation of the frequency of words/text appearing in a text document. Words with a larger font-size in a word cloud have a higher associated frequency. We utilise word clouds to graphically represent the rising terms from each of the specific time periods, where the increase in search popularity (i.e. percentage change) is used as the frequency value typically used in word clouds. For periods where there was a wide range in the percentage changes in the 25 rising search terms, we scaled the word cloud to allow terms with lower values to be visible. Breakout terms appear in red.
All analyses were conducted in R version 4.0 and word clouds were created using the R 'wordcloud' package version 2.6. Data used to generate Figure 1-Figure 7 are available as Underlying data 17 .

Sensitivity analysis
Using data downloaded from seven different days (detailed above), we created an averaged dataset. We took the average value for each search topic (13 topics) on each of the 161 datapoints, as it was recorded on each of the 7 separate days. We then correlated (using Pearson's correlation test) that 7-day average with values for the 161 days (01/01/2020-09/06/2020) used in the main analysis (which used values as they stood when the data were extracted on 12/06/2020) using Excel (Microsoft Office 365 ProPlus). We checked to see whether the use of the averaged dataset would have altered our overall conclusions.

Results
Google searches for the mental distress topics investigated are shown in Figure 3. Relative searches for the topics depression and self-harm were highest in February 2020, then appear to drop and then start to rise again in the post-lockdown period. Relative searches for anxiety have remained fairly stable throughout the time period under investigation, with an apparent peak in relative search interest on the 28 th April 2020. Exploration of the related search queries indicated a sudden rise in searches for "feeling of worry or anxiety crossword" which related to a crossword that appeared in the Scottish Daily Mail. Apart from a spike in relative searches for suicide on the 15 th Feb 2020, the trend line is not indicative of either an increase of decrease in suicide related searches. The peak in searching on the 15 th Feb 2020 is related to an increase in searching for the suicide death of television presenter Caroline Flack.
Prior to the first COVID-19 death in the UK, the relative searches for the search topics related to self-agency (markers of coping and resilience) appear to have remained stable (Figure 4). However, as the number of COVID-19 deaths started to rise, searches related to exercise, how-to, helpline and videotelephony   also started to increase. There was a relative peak in the searching for helplines when lockdown was announced in the UK, and the related searches on this day were related to the "coronavirus helpline". There was also a spike in relative searching for volunteering immediately post-lockdown which was related to searches for "nhs volunteers". Apart from the topics of  volunteering and helpline, the relative searching for the other topics remained elevated in the post-lockdown period. With the exception of volunteering, we observed a "sawtooth pattern" in trends reflecting weekday versus weekend searching.
Abstract topics related to coping and resilience are presented in Figure 5. Relative searching for the topics of coping and hope started to rise as the number of COVID-19 deaths increased in the UK. The rise in relative searches for rainbow started a little later, just before the lockdown was announced. Exploration of the related search queries indicated searching for rainbow was related to the NHS rather than the weather or PRIDE (the promotion of the self-affirmation dignity, equality, and increased visibility of lesbian, gay, bisexual and transgender people). Relative search interest for coping, hope and rainbow remain elevated during the post-lockdown period, with a peak in relative searching on Easter Sunday for the topic of hope -it was  not, however, possible to extract related searches from Google Trends on this day for this topic. Changes in the trend line for the topic kindness does not appear to be related to the pandemic, and remained fairly stable during the time period, with the exception of two peaks. The first peak (16/02/2020) was related to searches for "in a world where you can be anything be kind" and may be associated with the publics' response to the death of Caroline Flack, and the second peak (18/05/2020) for the topic of kindness was related to an increase in searches for "mental health awareness week" where kindness was chosen by the Mental Health Foundation as the focus of the week for 2020.
During the period before lockdown (01/01/2020-22/03/2020) the search terms which experienced the greatest increase in searching were directly related to the virus. After the lockdown was officially announced (23/03/2020-09/06/2020) Google searches in the UK which showed substantial increases were related to practical matters related to the public health measures introduced. The searches with the most substantial growth were "furlough" (government paid leave scheme), "popular google doodle games" (online games), and "george floyd". George Floyd was an African-American man killed during his arrest on May 25 th 2020 whose death sparked protests around the world (Figure 6).
We also explored the searches which experienced considerable growth in fortnightly intervals during the post-lockdown period (Figure 7). Similar to Figure 6, in the two weeks immediately after the lockdown was announced, searching for terms related to practical issues in response to the pandemic started to rise (e.g. "furlough", "home delivery"). As the period of lockdown continued (06/04/2020-09/06/2020) the rising terms changed to be less directly related to COVID-19 (e.g. "captain tom" -a 100-year-old ex-army officer who raised money for NHS charities during lockdown). Towards the end of the period studied, the searches which appeared to have the most substantial searching growth were no longer directly or indirectly related to COVID-19 (e.g. "george floyd"), with the exception of searches for "dominic cummings" (chief advisor to the UK prime minister alleged to have broken lockdown rules).
Our sensitivity analysis confirmed the findings of our main analysis, and the validation dataset was well correlated with our main dataset ( Table 2). The dataset used for senbsitivity analysis is available as Underlying data 17 .

Discussion
In this study we examined trends in mental distress, coping and resilience related searches, and search terms with significant growth in popularity, associated with the COVID-19 pandemic to the 9 th of June 2020, using data from Google Trends in the United Kingdom. While we found dynamic patterns of searching as the pandemic progressed, demonstrating the publics' evolving responses to the unfolding crisis, the increase in searches related to self-agency and coping strategies was the most striking and sustained. Over time, searches with the most substantial growth were no longer directly or indirectly related to COVID-19.
Relative searches for depression and self-harm initially decreased following the first death in the UK but then rose over time to levels found in the early part of 2020, whereas searches related to anxiety and suicide remained relatively stable throughout.
Other studies based on convenience-sampled repeat cross-sectional surveys showed declining anxiety and depression in the weeks following lockdown 18 . In contrast, relative search volumes for topics related to coping and resilience increased notably during the period after the first death in the UK. Those that remained elevated in the following 13 weeks and demonstrated a sustained increase, were the topics 'how-to', 'exercise', 'videotelephony' and 'rainbow'. During the period before lockdown search terms directly related to the virus experienced the greatest increase in searching. After the lockdown, Google searches which showed substantial increases were related to more practical responses to the public health measures introduced, such as 'furlough'. Towards the end of our study the searches which appeared to have the most substantial searching growth were no longer directly or indirectly related to COVID-19 (e.g. "george floyd"), with the exception of searches for "dominic cummings" potentially illustrating adaptation to the pandemic and life circumstances.
The value of understanding population mental health, coping and resilience strategies and public concerns during the COVID-19 pandemic has been highlighted by the World Health Organization 4 . Governments need both to be responsive to evolving population needs through public health messaging 19 and resources but also to monitor the effect of any actions taken. Google Trends data potentially offer a near-real time tool to do this in the absence of immediately available surveillance data and there is an emerging evidence base on its use during the COVID-19 pandemic 7-10 . The patterns we observed in the early stages of the pandemic have been replicated elsewhere 7 . Sinyor et al. 7 suggest that the focus on resilience and survival may mean that suicide rates in the early phase of the pandemic may behave more like wars, in which rates decrease rather than in recessions, where the opposite is seen. However, whether the changes we observed in search patterns are associated with actual rates of suicide or diagnosable disorder and real uptake of exercise and new skills requires further investigation building on initial studies exploring these correlations 20 . Similarly, it will be important to explore whether the topics associated with coping and resilience that demonstrated a sustained rise from pre-lockdown levels reflect a change in population agency and self-management is real and the long term.

Strengths and limitations
There are a number of limitations in the use of Google Trends data to characterise population responses to the pandemic. Certain population groups, often the more vulnerable, are less likely to access the internet so are potentially underrepresented in our findings, these groups include those such as those in digital poverty, the economically inactive, older adults, and those with disabilities which includes those with mental health conditions and in suicidal distress 21 . There is no way to assess this sort of response bias since Google Trends does not provide demographic data. More practically we found limited data available related to the methodology and contributory factors used to construct topics. More rigorous studies would be possible in the future if Google expands the metadata they make available. Our sensitivity analysis demonstrated that the time series differ from day to day despite the use of identical parameters (time, location or topic). This is likely related to the sampling method Google Trends uses to produce a representative sample each time data are extracted. Additionally, Google Trends do not provide data on actual volume of searching. Therefore, it is not possible to confirm whether changes in relative search volumes in the trend lines during the pandemic are related to real changes in search activity of that topic, or are attributable to changes in the denominator (i.e. total number of Google searches). As such, we did not conduct any formal time-series analysis because the interpretation of any statistical tests would be more meaningful in the presence of search volume data. Right now, it remains unclear if and to what extent total search behaviour has changed in the time period analysed. As such our study is a descriptive analysis aimed at understanding the impact of the pandemic on peoples' searches in the context of societal events and seasonal changes which may be impacting results.

Implications for research/policy
Google Trends is a free, open-access expansive dataset that provides a means of monitoring public concerns, responses and interventions to the COVID-19 pandemic. The reactive nature of search activity and its timely availability are huge strengths.
However, the limitations associated with the methodology, topic construction and inconsistencies in the datasets in Google Trends, mean that findings should be triangulated with other sources and detailed surveys to expand on the social narrative and inform policy currently. Public Health England are currently using weekly Google Trends data, derived from our study 8 , as part of a suite of measures from a variety of difference sources to track the mental health of the UK population. Given the divergence of policies related to the relaxation of lockdown in the nations of the UK there may be potential to explore the impacts of these on public concerns in the future.

Conclusion
Google trends shows promise as a tool to gauge public responses to the COVID-19 pandemic. Over time, searches with the most substantial growth were no longer directly or indirectly related to COVID-19 potentially illustrating adaptation to the pandemic and life circumstances as a consequence of it. In contrast to relatively stable rates of searches related to mental distress, the topics that demonstrated a sustained increase were those associated with coping and resilience.

Data availability
Source data All data are freely available directly from Google Trends (https://trends.google.com/trends/).

Underlying data
Open Science Framework: Tracking online searches for emotional wellbeing concerns and coping strategies in the UK during the COVID-19 pandemic: a Google Trends analysis. https://doi.org/10.17605/OSF.IO/KQ2D6 17 .
This project contains the following underlying data: • Averages UK.xlsx (Averaged dataset for sensitivity analysis).
Data are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).
Author information Duleeka Knipe and Hannah Evans are joint first authors. David Gunnell and Ann John are joint senior authors.