Attitudes About COVID-19 and Health: A Mixed-Methods Smartphone-App and Online Survey Study

Background: Behavioral mitigation strategies to slow the spread of COVID-19 have resulted in sweeping lifestyle changes, with short and long-term psychological, well-being, and quality of life implications. The Attitudes About COVID-19 and Health (ATTACH) study focuses on understanding attitudes and beliefs whilst considering the impact on mental and physical health and the influence of broader demographic and geographic factors on attitudes, beliefs, and mental health burden. uses smartphone-app technology and online survey data collection. Participants completed poll questions twice daily related to COVID-19 and a monthly survey assessing mental health, social isolation, physical health, and quality of life. Poll question responses were graphed using 95% Clopper-Pearson (exact) tests with 95% confidence intervals. Abstract Background: Behavioral mitigation strategies to slow the spread of COVID-19 have resulted in sweeping lifestyle changes, with short and long term psychological, well-being, and quality of life implications. The Attitudes About COVID-19 and Health (ATTACH) study focuses on understanding attitudes and beliefs whilst considering the impact on mental and physical health and the influence of


Table of Contents
72% female, 21% key or essential workers). Nineteen percent of UK participants reported a pre-existing mental health disorder, 31% reported a pre-existing chronic medical illness, and 35% were over 65. Fifty-seven percent of participants reported being more sedentary since the pandemic began, and 41% reported reduced access to medical care. Those with poorer mental health outcomes lived in more deprived neighborhoods, in larger households (ps < .05), had more pre-existing mental health disorders and medical conditions, and were younger than 65 years (all ps < .001).
Conclusions: Communities who have been exposed to additional harm during the COVID-19 pandemic were experiencing worse mental outcomes. Factors including having a medical condition, or living in a deprived neighborhood or larger household were associated with heightened risk. Future longitudinal studies should investigate the link between COVID-19 exposure, mental health, and sociodemographic and residential characteristics. December 2nd, 2020 [3], with Mexico following soon afterwards on December 11th [4]. Even with these pharmacological measures, behavioral mitigation strategies (e.g., physical-distancing, handwashing, face-masks) [5] remain critical to slow the spread of COVID-19 [6]. However, the effectiveness of these behavioral strategies is dependent on adherence to policies and guidelines and on a person's ability to perceive risks associated with the virus and adapt accordingly [7].
As the mitigation guidelines change over time and differ between countries and regions, there are   many areas of uncertainty, including financial and health concerns, employment, and housing, along with fear about the future and social isolation. These sources of uncertainty may impact coping and increase the risk of developing mental health problems, with implications for quality of life in both the short and long term [8], [9]. Families have had to juggle home-schooling children with working remotely or being unable to work at all [10]. Evidence from previous viral disease outbreaks indicates that when the number of stressors is high, there can be a negative effect on mental health, particularly for high-risk persons (e.g., survivors and frontline healthcare workers) [11]- [15].
Currently, individuals are reporting widespread concerns about the effect of social distancing on well-being. There is evidence of increased anxiety, depression, and stress, along with reports of concern about the practical implications of the pandemic response, including for personal finances [9], [16].
Social and medical factors play a significant role in COVID-19 exposure and influence the impact on mental health. It has quickly become apparent that these social and medical consequences of COVID-19 do not affect all people equally. Older adults, people with medical conditions (e.g., asthma, sickle cell disease), as well as those facing longstanding societal inequities (i.e., Black, LatinX, Indigenous, Asian, and traveler communities), face a disproportionate burden [17]. A report from Public Health England indicates that socioeconomic disadvantage, population density, and household composition may increase the likelihood not only of COVID-19 illness and severe disease [18] but also of poorer mental health outcomes [19], [20]. Individuals' concerns about their mental health outcomes will likely result in additional pressure on referral systems. Known barriers to accessing mental health interventions may also increase during the pandemic, such as distance, work commitments, and caring responsibilities.

Goal of This Study
The Attitudes About COVID- 19 and Health (ATTACH) study aims to understand attitudes and beliefs about the COVID-19 pandemic whilst considering the impact on mental and physical health, along with the influence of demographic and geographic factors. We examine how the pandemic and behavioral mitigation strategies influence attitudes and beliefs, which in turn are predicted to affect mental and physical health (e.g., anxiety, social isolation) and therefore influence overall quality of life (see Figure 1 for the theoretical model). Specifically, to assess the impact of COVID-19 related measures (e.g., isolation, physical distancing), our study: 1) tracks attitudes and behaviors daily as the pandemic evolves, and 2) longitudinally monitors mental and physical health symptoms using established measures that are reliable and sensitive to change.
The ATTACH study uses smartphone-app and online survey data collection. Our goals necessitate oversampling those communities who have been exposed to additional harm during the COVID-19 pandemic to guide the development of effective, and ideally, more person-centered interventions.
Data encompasses a sample from the United Kingdom (UK; high-resource country). We also include data from two smaller-scale studies from the United States of America (USA; another high resource country) and Mexico (i.e., low resource country). We chose these countries for inclusion as, before the pandemic, the Global Health Security Index ranked the USA and the UK as the first and second most prepared to manage a pandemic, but by June 2020, they were first and second in excess deaths related to COVID-19 [21]. Mexico has Latin America's second-highest death toll and had 50% excess deaths when data collection began [22]. As Mexico is a primarily Spanish-speaking country with a similarly challenging COVID-19 response as the USA and UK, the comparison could help better understand attitudes, behaviors, and mental health during the COVID-19 pandemic. Although behavioral mitigation strategies changed regularly during the study period, all 3 countries used lockdowns (e.g., shelter in place), social distancing, and relied on individual compliance with the new rules. Additionally, some form of mask mandate had been introduced in all 3 countries, hospitality venues had been closed or required outdoor dining, and stores limited the number of individuals allowed inside [23]- [25]. However, COVID-19 vaccines had not been released in any of these countries at the time of data collection.
Data collection for ATTACH is on-going, and longitudinal aims and specific hypotheses have been preregistered [26]. This paper primarily focuses on a cross-sectional analysis of baseline assessments collected in the UK between June 26 and October 31, 2020. It also considers longitudinal changes in UK daily poll responses through the same period. Descriptive analyses for data from the USA and Mexico are also reported, but we did not conduct statistical analyses. During our first-wave, the sample sizes in those countries were less than 100 participants.

Aims
1. To describe the baseline characteristics of participants in the ATTACH study in the UK, USA, and Mexico.
2. To describe changes in daily UK poll responses over time in relation to specific policy interventions (e.g., mask mandates) and the pandemic's trajectory (e.g., a 7-day rolling average of COVID-19 cases).

Hypotheses
1. We hypothesize that residential population density, socioeconomic deprivation, and household composition will predict self-reported outcomes (i.e., anxiety and depressive symptoms, social isolation, physical health, quality of life) in our UK sample.
2. We hypothesize that determine there will be baseline differences in our UK sample in selfreported outcomes between 1) participants with and without mental health disorders, 2) participants with and without medical conditions, and 3) participants under and over 65 years of age.

Study Design
ATTACH is a prospective cohort study conducted nationwide during the COVID-19 pandemic with arms in the UK, the USA, and Mexico. Our research team developed the study between March and April 2020. Data collection began on June 26th, 2020, in the UK, on July 27th, 2020, in the USA, and on October 10th, 2020, in Mexico. In the UK, the ATTACH study partnered with Air My Opinion (AMO), a smartphone app that enables organizations to gather longitudinal poll data to interpret trends in attitudes and beliefs.

Population
The ATTACH study purposely targets individuals with increased susceptibility to adverse health outcomes for recruitment, focusing on three priority groups: 1) those with a self-reported mental health disorder, 2) those with a self-reported chronic medical condition, and 3) those over 65

Air My Opinion App (UK data only)
The Air My Opinion (AMO) app was customized for the ATTACH study through an iterative design process between the research team and app developer. Before the study launched, the research team appraised a mock-up design, highlighted their satisfaction and dissatisfaction with the design and format, and provided feedback and suggestions for improvement. The AMO app uses the flutter framework, has embedded encryption and is general data protection regulation (GDPR) compliant.

Participants could freely install the app on an internet-enabled smartphone (Google Play or Apple
App Store) running Android or iOS operating systems. Responses to daily poll questions were collected via a short message service (SMS) service-centre voting platform. Some participants (N = 344) had difficulty sending their first SMS message (i.e., changed the wording, did not realize it had not sent), which included their registration data (i.e., postcode). As such, the app development team made changes so that registration data was incorporated within the app instead of being included in the first SMS message.
If a participant's phone plan is traditional pay-as-you-go, they pay per SMS message, and if they have a monthly plan or pay-as-you-go bundle, there is no additional cost. Each phone number is associated with a unique one-way encrypted participant key [e.g., 1::747d6f41-2f7a-47fd-a2a0-0cf41b8ca9f2::i2::1::484::], which feeds directly into a secure response firewall-protected database (see Figure 2). Per GDPR, age range, sex, postcode ethnic minority status (yes/no prefer not to say) and parent and chronic medical condition group status were stored in a separate secure firewall protected database. Data protection registration has been obtained for this study (UCL Data Protection Registration Number: Z6364106/2020/04/110). The data are minimized at the first opportunity, with new keys assigned and the original keys stored separately from the rest of the data on the UCL shared drive.

Poll Questions (UK data only)
In March and April 2020, the research team conducted a literature review and held several virtual video conference meetings to develop daily poll questions using prior expertise, survey knowledge, group discussion, and reliable sources of COVID-19 information (e.g., World Health Organization questions were focused on descriptive analyses (see below). These questions were chosen as they were asked continuously from the beginning of the study, represent a question from each category, and were chosen before any analyses were conducted.

Residential Risk Factors (UK data only)
Geographic region, socioeconomic disadvantage, and household composition may influence an individual's ability to follow COVID-19 related guidelines and restrictions [18]. In our study, two measures captured these factors: deprivation and population density. Participants provided the first five characters of their UK postcode (i.e., postcode sector) on the AMO smartphone app. In 2016, there were 12,381 postcode sectors in the UK. Postcode sectors vary in terms of the number of dwellings but typically range between 200 & 5,000 [33].

Socioeconomic Deprivation
The index of multiple deprivation (IMD; 2017-2020) is an official postcode based measure of relative deprivation in England [34], Wales [35], Scotland [36], and Northern Ireland [37] and is available as open-source government data. The IMD defines deprivation to encompass income, employment, education, health, crime, barriers to housing and services, and living environment. All neighborhoods are ranked on a relative rather than absolute scale according to their deprivation level relative to other areas. All zones are grouped into 5 bands (quintiles), each containing 20% of the zone, with "1 = most deprived, 5 = least deprived." In this study, each participants' postcode sector was assessed to determine the level of deprivation. IMD quintile scores are combined from all UK countries.

Residential Population Density
Population density estimates for England and Wales [38], Scotland [39], and Northern Ireland [40] are produced for each country using the cohort component method and cover the "usually resident population" from 2011 Census data. In this study, each participants' postcode sector was assessed to determine the population density (i.e., number of usual residents per hectare), and population density was combined from all UK countries.

Household Composition
Participants responded to the question, "How many people (adults and children) live in your household?" The five response options were "1" through "5 or more".

Procedures
Recruitment strategies for all countries are described in Multimedia Appendix 5. Participants  Figure 3 for a schematic) [41], [42].
If responses to survey questions indicated that participants had severe depression or experienced high levels of stigma due to their medical condition, a pop-up message provided mental health resources along with the study email address, which was monitored regularly by a clinical psychologist (AMH) to provide additional resources or referrals. After the survey, participants were redirected to the study webpages (located on the UCL Child Health Institute website), which contain links to mental health (e.g., Mind), COVID-19 (e.g., government), and authoritative medical information (e.g., NHS).
Participants could see the results of daily poll questions (i.e., pie charts) by clicking on a link in the app or via the study Twitter account.
In the USA, participants completed poll questions and monthly surveys using REDCap™, and in Mexico, participants completed monthly surveys using Qualtrics with open survey links.
Participants in the UK and USA who completed the monthly survey were entered into a monthly prize draw (UK £10/ US $10 Amazon card). In Mexico, most participants did not receive any compensation. A subset of the sample (i.e., undergraduate students) had the opportunity to receive culture credits (obtained through research studies or going to concerts or workshops) for their participation.

Statistical Analysis
Analyses were conducted using R version 4.0.3 [43] and SAS version 9.4 [44]. Descriptive statistics summarized demographic and clinical characteristics. The PROMIS [29] anxiety, physical health, and the meaning and purpose scales were scored using the online HealthMeasures scoring service [45] that utilizes information from each item to calculate a T-score (Mean = 50, SD = 10). For the PHQ-9 and the UCLA-10, raw scores were calculated in line with published norms. Mean imputation was used when a participant completed at least 80% of items [46], [47]. The frequency of somewhat and yes poll question responses from July 15th, 2020, to October 15th, 2020, were graphed based on the timing of response using 95% Clopper-Pearson (exact) tests with 95% confidence intervals (CI), alongside the 7-day average counts of daily COVID-19 cases (/100 persons) and deaths. Pearson correlations examined the relationship between household composition and self-reported outcomes.
Hierarchical linear regression analyses assessed whether residential socioeconomic deprivation and population density independently predicted self-reported outcomes after controlling for age and sex.
Generalized linear models were used to produce adjusted least-square mean scores and differences to compare groups (participants with and without mental health and medical conditions, those under and over 65 years). Analyses controlled for demographic factors (i.e., age, sex, level of residential socioeconomic deprivation) and comorbidities (i.e., mental health and medical condition). All analyses were conducted using pairwise deletion as variables generally contained less than 1% of missing data. Adjusted p-values were based on the model t-statistics. Semi-partial eta squared ( 2 ) was used as a measure of effect size to describe the proportion of total variation accounted for by the effect being tested. Statistical significance was determined at an alpha level of p < .05 (two-tailed).

Data Exclusion
The smartphone app and online survey data were examined for duplicates by matching unique app identifiers and participant numbers. Through this process, it was determined that 15 participants had downloaded the app more than once. Duplicate data were excluded before analyses.

Power
The ATTACH study was powered to detect longitudinal effects and not baseline group differences [26], so power calculations were not performed for these largely descriptive first-wave baseline assessments. However, for our longitudinal analyses with 3 subgroups, 5 covariates, 5 or more repeated measures at .80 power, an error probability of .05, and an effect size of .3, the sample size required for between participants analyses is 153 and is 50 for the within-participants analyses.

UK Preliminary Analyses
By October 31, 2020, 1405 individuals had downloaded the smartphone app and consented to participate in the UK ATTACH study. As of October 31, 2020, the study link that takes participants to the study recruitment page had been clicked on 5068 times, with Facebook and Twitter being the most common referrers. No participants reached out via email or social media to indicate that they had a mental health concern.
One hundred and twenty-three participants (8.8%) answered poll questions on at least one occasion but did not complete the monthly survey (see Table 1). Sensitivity analyses indicated that participants who did and did not go on to complete the survey were similar in age (p = .24), sex (p = .62), level of deprivation (p = .83), and UK country of residence (p = .74). However, those who did not complete a monthly survey were significantly more likely to identify as non-White (p < .001) and were from a more populated geographic region (p = .02) than those who did. Analyses assessed whether there were differences between participants who did and did not (N = 344) send their first SMS message correctly. They did not differ in terms of identified race (p = .54), but they were significantly older (p < .001) and more likely to be female (p = .02).

UK Poll Questions
Sixty-two percent of participants answered each poll question separately at 10 am and 2 pm, with the fastest response at 8 seconds and the slowest response at 23 hours and 58 minutes. The average time to answer one question was 2 hours and 17 minutes, with 11.8% of participants answering within 10 minutes. Longitudinal data from 6 poll questions indicated that participants had generally followed social distancing measures, although there was variability across time (responses varied between 50 and 99%). Regarding whether the COVID-19 pandemic had a negative impact on the family and whether participants were worried about the pandemic, both increased over time. Over 80% of participants responded "somewhat or yes" as the pandemic progressed and as the 7-day rolling average of cases and deaths increased. Although participants generally reported feeling that the reasons for current measures had been made clear (70-85% across the study period), there was much less trust that the government was doing everything in their power to meet public needs, with less variability in responses over time (25-50% across the study period). Increased social media use remained consistent, with 30-40% of participants reporting spending more time than usual (see Figure 4).

Descriptive data
Participants at baseline had a mean age of 57 years, with the majority identifying as white

Impact of the COVID-19 Pandemic
Participants reported how the COVID-19 pandemic had changed their lives and impacted their household members. Generally, fewer participants reported physical health changes, though many had received less routine or preventative medical care. At least a third of the sample reported being less physically active and eating less healthy foods. In fact, 57.5% of participants reported being more sedentary since the pandemic began (see Table 3). 6.9% 5.4% derly or disabled family member not in the home unable to get the help they need 5.5% 6.9% able to access medical care for a serious condition (e.g., dialysis, chemotherapy) 2.3% 1.3%

Residential Risk Factors
Pearson correlations demonstrated that residential socioeconomic deprivation was significantly related to increased population density (r = .-24, p < .001) and household composition (r = .07, p = .04). Higher household composition (i.e., more people residing in the home) was however related to less residential population density (r = .-06, p = .09), although this result did not reach significance.

Socioeconomic Deprivation
Younger age, female sex, and living in a more deprived neighborhood (i.e., quintile 1 vs quintiles 2 through 5) were predictive of reporting more anxiety symptoms, F ( Overall, these analyses indicated that living in the most deprived neighborhoods (quintile = 1) predicted worse mental health, increased social isolation, and poorer physical health.

Residential Population Density
Hierarchical linear regression analyses assessed whether residential population density significantly predicted self-reported outcomes after controlling for the age and sex of participants. We found that age (ps < .01) but not sex (ps > .05) was a significant predictor in all 5 regression models, indicating that younger participants were more likely to live in neighborhoods with higher population density.
With regard to self-reported outcomes, the severity of anxiety and depressive symptoms and reporting more social isolation and poorer physical health was not predictive of living in a more

Household Composition
Pearson correlations found small, but significant relationships between higher household composition and more anxiety (r = .07, p = .01) and depressive symptoms (r = .07, p = .01). Higher household composition was, however, related to less social isolation (r = .-06, p = .07), although this result did not reach significance. Household composition was not related to physical health or quality of life (ps > .05).

Differences between groups
We found that participants with mental health disorders reported significantly more anxiety and depressive symptoms, more social isolation, worse physical health, and poorer quality of life than those without a mental health disorder (all ps < .001) after controlling for age, sex, level of residential socioeconomic deprivation, and co-morbid medical conditions (see Table 4). Tables 3-5 focus on specific group contrasts; however, the covariates in each model are identical. Thus, the After controlling for age, sex, level of residential socioeconomic deprivation, and co-morbid mental health disorders, we found that participants with medical conditions reported significantly more anxiety (p = .003) and depressive symptoms (p = .002), and worse physical health (all p < .001) than those without a medical condition. Social isolation and quality of life were similar for both groups (ps > .05) (see Table 5). Table 5. Differences in self-reported outcomes for those with and without medical conditions  Table 6). Table 6. Differences in self-reported outcomes for those with and without mental health disorders

Mexico
During the first-wave of baseline assessments, the Mexico ATTACH study (n = 80) had been collecting data for less than one month. Participants were more likely than the UK sample to be

Principal Findings
This paper offers a description of longitudinal trends in attitudes and behaviors related to the COVID-19 pandemic. Additionally, we report descriptive data from two smaller-scale mirror studies conducted in the USA and Mexico for comparison. Descriptive data for the UK daily poll questions indicated that participants generally followed social distancing measures, but worry and negative impact on families increased as the pandemic progressed. Our cross-sectional baseline assessment in a UK adult population indicated that those with poorer mental health outcomes lived in more deprived neighborhoods, in larger households, had more pre-existing mental health disorders and medical conditions, and were younger than 65 years.
In terms of the UK pandemic trajectory, cases and deaths were relatively stable for most of this period, with more rapid increases observed only in the final month. UK longitudinal smartphone data from 6 of the poll questions indicated that as the pandemic progressed and the 7-day rolling average of cases and deaths began to increase rapidly, participants became more worried about the pandemic with a corresponding negative impact on their families. Although most people felt that information about the COVID-19 pandemic had been conveyed clearly, their trust in the government response was much lower. Nevertheless, participants reported generally following social distancing measures (responses varied between 50 and 99%), and 30-40% of participants also reported spending more time than usual using social media, even though the average age of our UK sample was nearly 60 years.
Cross-sectional data from our first-wave baseline assessments in the UK yielded several important findings. Participants with pre-existing mental health disorders reported worse outcomes across all mental health and psychosocial indicators. Unsurprisingly, those with pre-existing mental disorders reported more symptomatology. However, those with medical conditions and younger participants (less than 65 years) also had increased psychological symptomatology. Specifically, 25% and 17% of participants with chronic medical conditions reported currently experiencing moderate-to-severe anxiety and depressive symptoms, respectively. Similarly, 35% and 25% of younger participants reported currently experiencing moderate-to-severe anxiety and depressive symptoms, respectively.
These data parallel other UK cohort studies conducted during the COVID-19 pandemic as we found that depressive and anxiety symptomology was higher in our UK sample relative to epidemiological data collected before the COVID-19 pandemic. [48]- [51]. We had hypothesized that older adults (> 65 years) in our sample would experience worse mental health outcomes than younger participants because older people are at higher risk for worse COVID-19 related outcomes. However, after controlling for demographic factors and co-morbid mental health and medical conditions, this was not the case. Instead, our data suggest chronic medical conditions are a greater risk factor than older age for poor mental health outcomes. The relationships identified between deprivation, population density, and household composition suggest that they may be surrogates for poorer housing, overcrowding, or the need to use public transportation, which increase the risk of COVID exposure.
Consistent with other online surveys assessing mental health outcomes, women were overrepresented in our study sample [50]. Although this overrepresentation limits some of our conclusions, women are experiencing a disproportionate economic and employment burden related to COVID-19, along with mothers having increased childcare responsibilities [52]. These factors could have long-term deleterious effects on mental health. As our study moves into longitudinal analysis, we can determine how mental health symptoms are related to attitudes and behaviors associated with COVID-19 and whether robust, rather than uncertain public health measures, ameliorate mental health difficulties [53]. Future interventions will need to be tailored to individual and community needs whilst tackling entrenched pre-existing mental health inequities.
Other notable findings from this study were the relationship between mental health, psychosocial outcomes, and potential risk factors, although they likely existed before the COVID-19 pandemic.
Data revealed that those with poor quality of life were more likely than those with excellent quality of life to live in more populated neighborhoods. Further, higher household composition was related to more anxiety and depressive symptoms but appeared protective with regards to social isolation.
Most prominently, living in the most socioeconomically deprived neighborhoods (quintile = 1) was predictive of worse outcomes for all indicators, except quality of life. These results are concerning because they indicate the potential for an exacerbation of pre-existing inequities, particularly if future interventions are not tailored with a consideration of these factors. Evidence shows that those who live in the most deprived neighborhoods are hospitalized more frequently for COVID-19 infections [54]. Our results indicate that the impact of neighborhood deprivation may not only encompass physical health but may be more wide-ranging and include mental health and quality of life.
Many participants reported physical health changes since the pandemic began. Strikingly, nearly 60% of participants reported being more sedentary, and over 40% reported engaging in less physical activity. These results are in conjunction with 40% of participants reporting receiving less preventative or routine medical care. Lockdown measures have reduced physical activity opportunities, and these restrictions to limit the spread of COVID-19 infection may result in a less healthy populace [50]. Given the importance of early detection for many diseases, the impact of reduced preventative care combined with poorer engagement in health-promoting behaviors may be profound. As the risk of infection may remain high for some time, medical providers, administrators, and policymakers should continue to assess procedures to optimize care for all patients as well as identify safe avenues for increasing daily physical activity and reducing time spent sedentary [55].

Limitations
There are limitations of the findings reported in this study. A potential limitation of our survey study is self-selection bias, as participants may have joined the study as they were particularly interested in the topic. Although we used various recruitment methods, targeted efforts to improve representation, and placed a strong emphasis on anonymity and confidentiality, our UK sample was more likely to be White and from higher educational backgrounds, which is not representative of the UK population and so limits the generalizability of our findings to non-White and lower-income communities.
Efforts to engage and form partnerships have begun to increase the representativeness of our sample for future longitudinal analyses. Additionally, data from Mexico ATTACH currently includes over 1200 baseline assessments, so diverse between country analyses will be possible for future waves.
However, for the current UK baseline assessment, cultural factors and behavioral strategies specific to the UK, such as strong public messaging to protect the National Health Service (NHS), may mean our results do not generalize outside the UK.
This study began recruiting after the first UK lockdown (March 2020), so we lack real-time data from before the COVID-19 pandemic and cannot know if symptomology has changed. Further, indicators of pre-existing mental health disorders and chronic medical conditions were based on selfreport rather than clinical diagnoses. Given the unprecedented nature of the pandemic, we prioritized rapid deployment and data collection and so did not pilot test poll questions. However, many of the questions mirror those of other newly developed studies [56], [57]. The fast changes related to the pandemic meant that we could not consider all possible predictors before study initiation, so there are no measures related to quarantine status, fear of COVID-19, and duration of exposure to COVID-19related information, and infection risk perception. Although 87% of the UK population has access to a smartphone, just 53% of those aged 65+ have internet access via smartphone ownership [58].
Access to digital devices and the internet are also limited among other marginalized communities.
Therefore, those who are digitally excluded are likely to be underrepresented in our sample.

Strengths
Strengths of our study include the mix of cross-sectional and longitudinal data collection methodology with the ability to assess daily attitudes and behaviors during the rapidly changing COVID-19 pandemic. Through the use of smartphone technology and mHealth, we can capture realtime responses and bridge physical distance. Participants completed most measures with little missing data, indicating that items were not burdensome. Finally, although our UK sample needs to increase representation related to ethnic and racialized identities, our Mexican sample includes a cross-section of communities. Through targeted recruitment, the USA sample includes an overrepresentation of participants who identified as Black or African American.

Implications
Smartphone apps are increasingly recognized as potentially powerful tools for mobile health studies and interventions [59], [60]. They have been utilized successfully in adult and pediatric chronic illness populations [61]- [64] and in ecological momentary assessment studies and interventions [65], [66]. Data from the UK ATTACH study demonstrates that people are also prepared to answer daily questions and provide mental health data in a longitudinal study using smartphone technology.
Optimizing digital approaches and integrating them into the public health response is possible whilst considering logistical and technological barriers. However, currently, there is limited evidence that mHealth apps and interventions are cost-effective or cost-saving [67], [68].This engagement increases the feasibility of undertaking potentially sensitive longitudinal research and reduces retrospective judgments that tend to be affected by recall bias [66]. Because of this engagement, moving forwards, the Mexico ATTACH study will use the Telegram messaging system to collect daily poll data, as it is widely accessible with an encrypted point-to-point connection. As we move forwards, it will be necessary to including participants input early in the research process to engender long and short term engagement [69].

Conclusions
Our data indicate that those with mental health disorders and chronic medical conditions are experiencing increased anxiety and depressive symptoms. Socioeconomic deprivation also appears to be a considerable risk factor for poor mental health. Although these challenges are not new, they could become more deep-rooted and challenging to tackle in this new COVID-19 era and beyond.
Along with a renewed focus on mental health, investment to increase access to preventative medical care and messaging to encourage health-promoting behaviors, including physical activity, will be critical.

Data Sharing
Data will be shared upon reasonable request and with permission according to the Attitudes About COVID-19 and Health Group data release policy.