What can internet users' behaviours reveal about the mental health impacts of the COVID-19 pandemic? A systematic review

Abstract

Objectives

At the end of 2019, an acute infectious pneumonia (coronavirus disease 2019 [COVID-19]) caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) began in Wuhan, China, and subsequently spread around the world starting a pandemic. Globally, to date, there have been >118 million confirmed cases, including >2 million deaths. In this context, it has been shown that the psychological impact of the pandemic is important and that it can be associated with an increase in internet searches related to fear, anxiety, depression, as well as protective behaviours, health knowledge and even maladaptive behaviours.

Study design

This is a systematic review.

Methods

This review aims to collect, analyse and synthesise available evidence on novel data streams for surveillance purposes and/or their potential for capturing the public reaction to epidemic outbreaks, particularly focusing on mental health effects and emotions.

Results

At the end of the screening process, 19 articles were included in this systematic review. Our results show that the COVID-19 pandemic had a great impact on internet searches for mental health of entire populations, which manifests itself in a significant increase of depressed, anxious and stressed internet users’ emotions.

Conclusions

Novel data streams can support public health experts and policymakers in establishing priorities and setting up long-term strategies to mitigate symptoms and tackle mental health disorders.

Graphical abstract

Introduction

From November to December 2019, the acute infectious pneumonia, termed COVID-19 (coronavirus disease 2019), which is caused by a new highly contagious and pathogenic respiratory coronavirus (severe acute respiratory syndrome coronavirus 2 [SARS-CoV-2]), began in China and subsequently spread causing a global pandemic.¹ To date, approximately 1 year after the World Health Organization (WHO) first declared the COVID-19 epidemic a pandemic,² >118 million cases and >2 million deaths have been recorded.³ The world is facing a complex crisis not only from a health perspective but also from a social and economic point of view.⁴

According to Brooks et al., the psychological impact of the pandemic has been moderate to severe, with 17% of adults reporting moderate-to-severe depressive symptoms, 29% moderate-to-severe anxiety symptoms and 8% moderate-to-severe stress levels.⁵ This was mainly due to isolation or quarantine measures, the scarce possibility of social interactions and the interruption of work for many with consequent economic loss.⁵ Children and adolescents have also experienced profound life changes characterised by emotional isolation and intensive use of the internet, which may increase vulnerability to anxiety, stress and suicidal ideas or exacerbate self-inflicted violence.⁶

The perception of threat associated with a global pandemic may generate fear and lead to negative emotional reactions; fear, in turn, motivates people to initiate behavioural changes and improve their health knowledge.^{7 , 8} Traditional and social media also tend to report on cases of infection and mortality, while rarely mentioning cases of recovery, thereby, facilitating the spread of fear and quickly generating anxiety or distress.⁹ To reduce these effects, the WHO has advised the public to focus less on COVID-19 news, since watching, reading or hearing news about COVID-19 has been shown to exacerbate fear.¹⁰ This is particularly important as a large amount of news (and fake news) is available to the general public who may not always have the appropriate knowledge to understand scientific updates (infodemic).^{11 , 12} Hossain et al.¹³ also show that, despite demographic and social differences, the epidemiological distribution of mental health problems (due to the pandemic and associated factors) has been heterogeneously distributed across the general public, as well as among COVID-19 patients and healthcare professionals. This suggests that there is an ongoing psychiatric epidemic parallel to the COVID-19 pandemic, which requires immediate intervention from the public health community.^{14 , 15}

In this context, it is necessary to understand how people cope with the pandemic. Internet searches could help answer this question since the internet represents a primary channel for finding health-related information.¹⁶ It has been shown that a rise in COVID-19 cases can be associated with a surge in internet searches related to fear, anxiety, depression, as well as protective behaviours, health knowledge and even maladaptive behaviours.¹⁷

The digital era has given rise to new data sources and large amounts of data also known as Big Data; according to De Mauro et al., “[Big Data] represent resources/assets of an informative nature characterised by such a high volume, speed and variety as to require technology and analytical methods specific for its transformation into value”.¹⁸ The use of Big Data in scientific research is destined to grow and become increasingly present in daily healthcare practice.¹⁹ Within Big Data, we can include the novel data streams, which are identified as those data whose “content is initiated directly by the user (patient) themselves”.²⁰ Research studies, which are often limited to certain categories of patients, could expand their targets, including several different population subgroups, while the analyses conducted on Big Data could make health care even more personalised at the same time. The magnitude of data available on patients with the same comorbidities and specific rare diseases could facilitate the creation of ad hoc individual-level therapeutic plans in the future.¹⁹

This review aims to collect, analyse and synthesise available evidence on novel data streams for surveillance purposes and/or their potential for capturing the public reaction to epidemic outbreaks, particularly focusing on mental health effects and negative emotions. Indeed, previous research shows that nowadays any mediatic event produces a social reaction on the internet, which, in turn, generates a large volume of data, that can be collected and analysed in order to answer health-related questions and issues.21, 22, 23 For this review, we considered the COVID-19 pandemic as the mediatic event; subsequently, the general public reaction generated a novel data stream, which has been analysed in order to identify any potential impact on mental health.

Methods

The Cochrane Collaboration²⁴ and the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines²⁵ were followed in order to conduct the current systematic review. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses²⁶ guidelines²⁷ were used to report the process and results. The structured literature search was conducted on PubMed/Medline and Scopus on 17th February 2021, combining free text words and medical subject headings (MeSH). Keywords were combined using Boolean operators AND and OR. No time filter was applied. The full search strategy is available in Supplementary Table S1.

Inclusion/exclusion criteria

In order to be considered eligible, articles had to meet the following inclusion criteria: (i) article written in English; (ii) reporting original data; (iii) focusing on mental health; (iv) during the COVID-19 pandemic; and (v) addressing the usage of non-conventional data approaches in capturing public reaction to COVID-19 epidemic outbreaks, particularly focusing on adverse mental health effects and emotions. Exclusion criteria were as follows: (i) no original data (i.e. a review, opinion, book chapter, commentary or letter); (ii) article not published in English; (iii) full text not available; (iv) focusing on mental health but not during the COVID-19 pandemic; and (v) assessing outcomes not related to mental health.

Articles were firstly screened based on title and abstract and then assessed in full text. Both article screening phases and data extraction were conducted independently by two authors (VG and OES). Full texts were downloaded only for potentially eligible studies. Data extraction was conducted only for those articles that met all the inclusion criteria and was performed using a pre-defined and pre-piloted spreadsheet elaborated in Microsoft Excel® for Windows. Extracted data included author and year, data source, type of mental health outcome assessed in the study, study period, country where the study was conducted, study aims, keywords used to perform the search, type of analysis conducted and main results. Any disagreement in data extraction was resolved through discussion among the two authors; if any disagreement persisted, a third author was consulted (SP).

Results

Literature search

A total of 113 papers were retrieved on PubMed and 436 on Scopus. Of these 549 articles, 62 were duplicates and immediately removed, leaving a final sample of 487 unique papers. After preliminary screening, based on the title and abstract, 462 articles were removed for the following reasons: unrelated topic (n = 411), review (n = 25), protocol study (n = 17) and article not published in English (n = 9). Out of 25 eligible articles, six articles were removed after full-text assessment because of the following reasons: two articles did not specifically report mental health data;^{28 , 29} one article was a feasibility study;³⁰ one study referred to pre-COVID-19 times;³¹ one article, although using smartphone app usage data, did not correlate depressive symptoms and anxiety (self-reported symptoms) with app data, but with COVID-19 news;³² and, the last article using passive mobile monitoring platforms, assessed the behavioural changes (i.e. time spent at home) among subjects with psychiatric disorders.³³ Fig. 1 depicts the flow diagram reporting the selection process. At the end of the screening process, 19 articles were included in the systematic review.34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52

Fig. 1.

Flow diagram of the study-selection process.

Characteristics of included studies

Among the 19 retrieved articles, 15 used a single-data source, whereas four used a combination of two or more data sets. In particular, Twitter was the source most frequently used alone (n = 4),^{36 , 46 , 51 , 52} and in combination (n = 3) with YouTube and a forum,³⁷ Geographic information system (GIS)³⁹ and Weibo.⁴⁷ Google Trends was the second most frequently used data source (n = 6);^{38 , 41 , 43 , 44 , 48 , 49} one study investigated Weibo⁴⁰ data and another investigated Baidu³⁴ data alone. Forums and chats were both used alone (n = 3)^{35 , 42 , 45} and in combination with other social networks.³⁷

Regarding the country where the studies were conducted, four were internationally based,41, 42, 43 ^{, 52} whereas the rest were carried out in a single country. Specifically, the US was the most frequently explored country (n = 5),^{36 , 38 , 39 , 46 , 50} followed by China (n = 3)^{34 , 40 , 47} (including one study that compared data between Wuhan [China] and the Lombardy region [Italy]),⁴⁷ India (n = 3),^{44 , 48 , 49} Australia,^{35 , 51} Canada⁴⁵ and Spain.³⁷

The mental health outcomes assessed were frequently general mental or emotional health, or a combination of selected mental health symptoms (e.g. depression, anxiety, suicide ideation, sleep disturbance, fear and stress), while only three articles focused on a specific mental health disorder (one study focused on stress,³⁹ one on eating disorders⁴⁵ and one on depression⁵¹).

Almost all included studies performed an analysis of the internet activity content/sentiment in order to explore the predictive value of these searches in forecasting mental health disorders. Some studies assessed how online searches changed throughout the pandemic,^{37 , 38 , 47 , 49} whereas, six articles assessed the association between COVID-19 cases or deaths and internet search volume related to mental health.^{39 , 41 , 43 , 44 , 48 , 51}

All included articles found a significant association with at least one of the explored mental health disorders/symptoms; however, three studies found conflicting results for specific mental health outcomes. In particular, the study conducted using data from Baidu (China) found a significant association between the COVID-19 pandemic and panic and fear, but failed to detect an association with depression, sadness and obsessive-compulsive disorders.³⁴ An international study performed on Google Trends found an association between insomnia and COVID-19 deaths, but not with depression and suicidal thoughts.⁴¹ Lastly, even an international (24 countries) study found a significant association between COVID-19 cases and deaths and suicide and depression, but no association with insomnia.⁴³ These differences may be explained by considering the different countries and the length of the study period. Characteristics of included studies and main results are provided in Table 1 .

Table 1.

Characteristics and the main results of included studies (reported in alphabetical order of first author).

First author, year (Reference)	Data Source	Mental Health outcome	Study period	Country	Study aims	Keywords	Analysis	Main results
Chen, 202034	Baidu	Depression, anxiety, sleeping problem, obsessive-compulsive disorder	24 Jan to 22 Feb, 2020	China	To understand psychological public reaction to the COVID-19 pandemic	Depressed, sad, panic, fear, insomnia, obsessive-compulsive disorder, psychological counselling	Non-parametric Mann–Whitney U test	Internet search for panic/fear and psychological counselling were more frequent in 2020 than in 2019 (P < 0.05), depressed/sad and obsessive-compulsive disorder were the opposite (P < 0.05). No differences for insomnia.
Chivers, 202035	“New mum forum” in Google (831 posts)	Distress, despair	27 Jan to 12 May, 2020	Australia	To examine the public discourse in the context of COVID-19 pandemic	COVID, corona, pandemic	Thematic analysis, sentiment analysis and word frequency calculations	Distress related to external high risk; despair for lack of social and family support. Half of the posts have a negative sentiment. The most frequent word was worry (and similar).
Guntuku, 202036	Twitter	Stress, anxiety and loneliness	Jan to May 2020 vs the same period in 2019	US	To analyse the content of tweets with a specific focus on mental health during the COVID-19 pandemic	Sentiment, stress, anxiety, loneliness and COVID-19-related symptoms	Pre-trained data-driven machine learning models	Stress, anxiety and loneliness were more frequent in 2020 (P < 0.001), whereas sentiment was lower in 2020 (P < 0.001).
Iglesias-Sánchez, 202037	Twitter, YouTube, Instagram, official press websites and Internet forums (80,091 posts)	Emotional health	Mar to May 2020, split in three stages (characterised by enforcement of containment measures of the first two, and mitigation of the last one)	Spain	To understand emotional health during the COVID-19 confinement time	Anger, fear, joy, sadness, disgust and uncertainty	Content analysis	Anger, fear, sadness and uncertainty were statistically significantly different throughout the three studied periods, but not joy and disgust.
Jacobson, 202038	Google Trends	Mental health symptoms	16–23 Mar 2020 (stay-at-home orders in 11 states)	US	To examine if COVID-19-containment measures produced changes in mental health symptoms	Anxiety, depression, obsessive-compulsive disorder, hopeless, angry, afraid, apathy, worthless, worried, restless, irritable, tense, scattered, tired, avoiding, procrastinate, insomnia, suicidal, suicide	Generalised additive mixed models	Topics related to anxiety, negative thoughts about oneself and the future, sleep disturbances, and suicidal ideation increased prior to stay-at-home orders.
Li, 202039	Twitter and GIS (Geographic Information System)	Stress symptoms	Jan to Apr 2020	US	To detect COVID-19 related stress symptoms at a spatiotemporal scale	Patient Health Questionnaire lexicon	Machine learning	Stress-related tweets rapidly increased from March 8th until April 5th when the number of COVID-19 cases started to decrease.
Li, 202040	Weibo	Psychological traits	13–26 Jan, 2020	China	To explore the predictive value of Weibo in forecasting mental health	Words of positive emotion, negative emotion, anxiety, anger, health, leisure, family, friend, money, death and religion	Sentiment analysis	Between 20 and 26 January, negative emotional indicators of psychological traits increased in anxiety, depression, and indignation, while positive emotional indicators decreased.
Lin, 202041	Google Trends	Insomnia, depression, and suicide	20 Mar to 19 Apr, 2020	International	To explore the predictive value of Google Trends in forecasting mental distress	Insomnia, depression, suicide	Mathematical model	Iran, Spain, US and Italy were the countries with the highest insomnia research volume. COVID-19 death numbers were associated with days at higher insomnia searches, but not for depression and suicide.
Low, 202042	Reddit Mental Health Dataset	General mental health	1 Jan to 20 Apr, 2020	International	To assess changes in the natural language of public commentary posted to mental health support groups on the Reddit website	Lexicon built on Linguistic Inquiry and Word Count	Machine learning	Attention-deficit/hyperactivity disorder, eating disorders and anxiety showed the largest sum of change in negative semantic features during the pandemic. Anxiety emerged as a general theme across Reddit.
Misiak, 202043	Google Trends	General mental health	18 Feb to 13 Apr, 2020	International (24 countries)	To assess if COVID-19 cases and death are associated with mental health-related internet search volume	Suicide, depression, anxiety, insomnia	Spearman rank correlation	Suicide and depression were both associated with COVID-19 deaths and cases, anxiety was only associated with deaths and insomnia was not associated with cases neither deaths.
Rana, 202044	Google Trends	Mental health indicators	4 Mar to 25 Jul, 2020	India	To estimate the correlation between COVID-19 cases and internet search	Suicide, autism, anxiety, depression, dementia	Pearson correlation	Suicide, anxiety, depression and dementia were highly correlated with the number of daily infectious cases.
Richardson, 202045	National Eating Disorder Information Centre instant chat service	Eating disorders	1 Mar to 30 Apr, 2020	Canada	To describe the impact of the COVID-19 pandemic on help-seeking behaviours	Over-eating/binge eating, over-exercising, dieting/restriction, weight preoccupation, perfectionism, purging, anxiety, and depression	Content analysis	The number of contacts during the pandemic period was significantly higher compared to the two previous years. Eating disorder symptoms, anxiety, and depression were higher in 2020. Content analysis detected four themes: 1) lack of access to treatment, 2) worsening of symptoms, 3) feeling out of control and 4) need for support.
Saha, 202046	Twitter (59,096,694 tweets in 2020 compared with 40,875,185 in 2019)	Anxiety, depression, stress, suicidal ideation and emotional support	24 Mar to 24 May, 2020	US	To study the temporal and linguistic changes in symptomatic mental health expressions during the COVID-19 pandemic	Classification based on Diagnostic and Statistical Manual of Mental Disorders, 5th Edition	Machine learning	Anxiety tweets showed the most significant increase in 2020 compared to 2019, followed by suicidal ideation, depression and stress, with a steady decline during the study period.
Su, 202047	Twitter (3,650,380 tweets) and Weibo	Psychological states	23 Feb to 21 Mar, 2020 (Twitter); 9 Jan to 5 Feb, 2020 (Weibo)	Wuhan (China) and Lombardy (Italy) region	To examine and compare the impact of COVID-19 lockdown on individuals' psychological states in China and Italy	Language Inquiry and Word Count dictionary	Psychometric analysis	In Weibo posts, (China) users increased the use of first-person plural pronoun, religion, social, negative emotion and home significantly after the lockdown. Twitter (Italy) users increased the wording discrepancy and home and decreased anxiety.
Talbot, 202152	Twitter (192 tweets)	Mental health of pregnant women	1 Mar to 31 May, 2020	International mostly from US and UK	To characterise the content of an international sample of tweets related to pregnancy and mental health	Several keywords related to COVID-19, pregnancy and mental health	Sentiment and thematic analysis	Individual and company tweets had the same negative tonality. Among the individual tweets, stress about being isolated, depressive symptoms and sleep difficulties were the most frequent. Among the tweets from companies, stress and depressive symptoms that pregnant women may experience during the pandemic, as well as the services offered were the most frequent.
Uvais, 202048	Google Trends	General mental health	12 Mar to 13 Jun, 2020	India	To explore the association of internet search volume and COVID-19 cases	Depression, anxiety, insomnia, suicide	Spearman rank correlation	COVID-19 cases and deaths in India were significantly associated with searches for depression, anxiety and suicide. Moreover, suicide and depression were the most frequently searched terms.
Uvais, 202049	Google Trends	General mental health	26 Jan to 24 May, 2020	India	To understand the changes in patient interest in psychiatric search terms during the lockdown	Depression, anxiety, suicide	T-test	Depression, anxiety, and suicide search increased from the pre-lockdown to the lockdown period.
Zhang, 202050	Google Search and YouTube	Depression and anxiety	Jan 2020 vs May 2020	US	To explore the predictive value of YouTube and Google Trends for forecasting depression and anxiety	Not available	Machine learning	PHQ-9 and GAD-7 scores increased during the study period. Some internet behaviours such as late-night online activity, short YouTube intervals between videos and videos containing anxiety and sadness keywords were associated with scores increment.
Zhou, 202051	Twitter (94,707,264 tweets)	Depression	1 Jan to 22 May, 2020	Australia	To explore the predictive value of Twitter in forecasting depression	Classification based on Diagnostic and Statistical Manual of Mental Disorders, 5th Edition	Mathematical model	Depression level increased following the increasing COVID-19 cases and reinforcement of government containment measures.

GIS, Geographic information system; GAD-7, General Anxiety Disorder-7; PHQ-9, Patient Health Questionnaire-9.

Discussion

Results of our systematic review revealed that the COVID-19 pandemic has significantly impacted internet searches on mental health of entire populations. Indeed, in almost all included articles, a statistically significant increase in negative (seen as depressed, anxious and stressed) internet and social network users’ emotions was recorded. These data were confirmed both by associating the number of COVID-19 cases or deaths with internet search volume for specific keywords, as well as by content and sentiment analysis of posts published on social networks. This phenomenon was confirmed in all studied countries, including in those articles where several countries were studied together.

It is interesting to note the parallels drawn by Su et al.,⁴⁷ who assessed the psychological states of social network users (Twitter and Weibo) in Lombardy (Italy) and Wuhan (China), the two regions first affected by SARS-CoV-2. Chinese users significantly increased the use of first-person plural pronouns, religion, social, negative emotion and home after the lockdown. Italian users increased the use of the words ‘discrepancy’ and ‘home’ while searching less frequently for anxiety. These results show how the same event can impact differently on communities based on cultural and societal aspects. Despite the posts' common negative and fearful content, the wording between cultures differed, at least partially.

Moreover, it should be considered that, in the e-health era, the diffusion of information by the internet and social media is faster than ever before, even surpassing the velocity of spread of a highly infectious virus such as SARS-CoV-2. Furthermore, information is not always correct or validated, promoting the spread of misleading or incorrect facts on unfiltered internet platforms, which include all major social networks (i.e. Facebook, Twitter, YouTube and several others).

In addition, it is important to note that the COVID-19 pandemic is sustained by an infection with a completely novel and still partially unknown virus.^{53 , 54} This characteristic highly impacts several aspects. Firstly, due to the unknown nature of the virus, even trustworthy sources made mistakes in communication at the beginning of the epidemic, fuelling confusion and a hyperreactive information search among the general public.¹¹ Secondly, the lack of appropriate therapies and preventive measures (such as vaccines) during the first year of the pandemic incited a sense of powerlessness and fear in the population, which, in turn, drove the search for information and fuelled a widespread sense of stress, uncertainty, anxiety and depression.⁵⁵ Moreover, in pandemic times, social deprivation or lack of a well-structured communication of information may strengthen prejudice and stigma towards vulnerable individuals, which, in turn, reinforces a sense of despair, loneliness, depression and anxiety.¹⁵ Also, the implementation of containment measures, such as social distancing, national lockdowns and absence from the workplace, may have pushed people to spend a lot of time on the internet, not only to search for information but also to connect with friends and relatives.⁵⁶

All of these human behaviours generate a significant amount of data on numerous different platforms (Big Data and novel data streams), which contribute to a newly emerging research field aimed at supporting traditional surveillance systems (especially for infectious diseases)^{22 , 57 , 58} and at understanding human behaviours,^{21 , 23} with the aim of guiding public health decisions.^{59 , 60}

Systematically collecting and examining health-related data generated by these new sources might carry the potential to understand, investigate and interpret events relevant for public health purposes.⁶¹ Several cross-sectional studies, based on online surveys administered to convenience samples, have been published throughout 2020, without any information on the real generalisability and representativeness of results.62, 63, 64, 65, 66, 67, 68, 69 In contrast, these novel data streams might reduce several biases typical of surveys, such as social desirability and recall bias, as well as improving the above-mentioned representativeness of the sample, since thousands or millions of users’ data can be analysed together. Moreover, Big Data and novel data streams reduce the time lag between traditional epidemiological data collection and data analysis, as well as the data entry burden and potential errors that can occur with traditional epidemiological systems. Furthermore, novel data streams and Big Data can improve spatial resolution and forecast unanticipated outcomes of interest (such as the current study looking at the mental health impacts of the pandemic), especially for outcomes not routinely surveyed.²⁰ Big data and novel data streams can help to better understand many disease transmission/process aspects that are not usually captured by traditional surveillance systems. Last, but not least, Big Data and novel data streams are important for improving data dissemination.²⁰

Results of this systematic review are extremely relevant because they offer an overview of the general public's psychological reaction to the COVID-19 outbreak, showing how fear, anxiety, stress, suicidal thoughts and depression are strongly associated with the pandemic. Previous systematic reviews on cross-sectional studies assessing the mental health status of the general public during the COVID-19 pandemic showed that the prevalence of depression increased from 7.2% to 14.6–48.3% during the pandemic.⁷⁰ Moreover, in addition to the previously established risk factors for depression, such as female gender, presence of chronic/psychiatric illnesses and unemployment, frequent exposure to news concerning COVID-19 was identified as one of the most important risk factors for depression during the pandemic.⁷⁰

In light of the disease burden already caused by mental disorders,⁷¹ and considering the prolonged exposure to this stressful pandemic event, we might expect long-term psychological consequences. This is especially true considering that, during the pandemic, many healthcare activities have been postponed⁷² or substituted by telemedicine.⁷³ Indeed, telemedicine offers a great opportunity to deliver mental health services remotely, without borders, on a large scale and in a safe way.^{73 , 74} However, telemedicine, in certain cases, has shown some limitations.⁵ Even if telemedicine is a more readily accessible option, overcoming the problems related to physical accessibility to services (e.g. remote location, lack of transportation or funds, stigma associated with seeking treatment and physical mobility-related health issues),⁷⁵ not all patients have the necessary technology, even just a simple internet line, to be able to use telemedicine services, and some individuals may not have the skills to use these tools.⁷⁶ From this perspective, there is a risk that a service designed to reach a large number of people creates inequalities in the population, with the possibility of ‘leaving behind’ some already disadvantaged groups, such as the poorest, the elderly or those who live in disadvantaged social contexts.⁷⁷

Moreover, evidence has found that during the COVID-19 pandemic, a reduced number of mental health services has coincided with an increased mental health burden, potentially indicating an escalating number of untreated individuals with mental health problems.⁷⁸

We should be aware of the above-mentioned elements in planning future healthcare services. From this perspective, Big Data and novel data streams can support public health experts and policymakers in establishing priorities and setting up long-term strategies to mitigate the symptoms and tackle mental health disorders, with the final aim of facilitating the implementation of future preventive interventions.

Strengths and limitations

Before generalising the results of the current systematic review, some limitations need to be taken into account. Firstly, we limited our search to articles published in English. This might have reduced the total number of potentially eligible studies; however, since English is the most commonly used language in the scientific community, we believe this did not significantly affect our results. Secondly, some authors divided their analyses across several short papers, multiplying the evidence volume but reducing the quality of their discussion and the interpretation of results. Lastly, internet search spikes and contents of tweets may be influenced by different triggers, such as the increased number of cases, increased attention given by mass media or as a reaction to the containment measures adopted by governments, making it difficult to precisely identify a cause–effect association. In light of this, it is not possible to differentiate between the mental health effects of fear of the virus itself from the containment measures adopted, neither to differentiate between negative feelings and diagnosis of mental health disorders. However, previous cross-sectional studies conducted during the pandemic confirmed a variable increase in the prevalence of depression, anxiety and stress in the general population.^{70 , 72 , 79 , 80}

Despite the above-mentioned limitations, our review has some important strengths. Firstly, this is a systematic and extensive review offering an exhaustive overview of available evidence on Big Data and mental health during the COVID-19 pandemic. With this overview, we not only synthesised and analysed results obtained by each included study but also provided a picture of the data sources used, the aims of the original studies, as well as the types of analyses performed. Secondly, our search strategy was developed considering several keywords, including both MeSH and free text terms. Lastly, as far as we know, this is the first systematic review addressing the usage of non-conventional data approaches in capturing public reaction to the ongoing COVID-19 pandemic, particularly focusing on mental health and emotions.

Conclusions

To conclude, the results of this systematic review found a statistically significant increase in searches for information about mental health issues by the general public throughout the COVID-19 pandemic. This was observed in both single- and multi-country studies, which analysed data in up to 24 different national contexts. The most-searched-for terms included anxiety, fear, suicide, depression, despaired, stress, solitude and loneliness. While some papers aimed to examine how online expressions of mental health changed throughout the pandemic, others specifically investigated whether containment measures or COVID-19 cases and deaths were associated with mental health internet search volumes. Internet searches and users’ behaviour on social networks generated a large amount of data (Big Data) and novel data streams that were used by numerous researchers all over the world. Results obtained from these analyses might prove extremely useful in informing policymakers and health authorities on the implementation of healthcare services and policies, with a focus beyond treating COVID-19 patients, to also provide care to those affected by direct and indirect mental health consequences.

Author statements

Ethical approval

Not required.

Funding

None declared.

Competing interests

None declared.

Author contributions

VG conceptualised, designed the study and performed the literature search. VG and OES performed resource analysis and data extraction. VG and OES wrote the first draft. All authors have read and agreed to the published version of the manuscript.

Appendix A. Supplementary data

The following is the Supplementary data to this article: