Analysing Mental Health Problems of Adults During COVID Using a MIMIC Model: Evidence from Indian Metropolitan Cities

Given reduced social interactions and economic distress, mental health has emerged as an important concern during COVID-19. This study estimates the prevalence of mental health problems during the rst wave of COVID and identies its determinants among the general population of Indian metropolitan cities. The study uses a Multiple Indicators Multiple Causes model to measure depression, anxiety and stress using observed indicators of these latent constructs, and to identify the socio-economic groups at risk of these disorders. The data was collected from 1,275 adults randomly selected from the list of mobile phone users in Bengaluru, Chennai, Delhi, Kolkata, Mumbai and Hyderabad. The Depression Anxiety Stress Scale was administered to the study participants. About 46, 24 and 48 percent of respondents reported symptoms of anxiety, depression and stress, respectively. Single respondents, members of minority groups, less educated and those belonging to large households are identied to be at-risk. Results also indicate a positive relationship between economic stress and mental illness. Although reporting psychological disorders may not While community-based support may be sucient in general, monitoring is to identify risk persons who may require clinical support.


Introduction
Past pandemics had been known to create panic and a sense of threat to individual security, which is manifested in psychological disorders [1]. COVID-19 is no exception. In the absence of vaccination during the rst wave of COVID-19, the measures used by governments to control the spread of the pandemic reduced social interaction. Simultaneously, the unpredictable and uncertain nature of the pandemic, lack of information about appropriate treatment, con icting messages from the authorities, high mortality, stigma and overrun of medical facilities created a mass fear of COVID [2]. 'Coronaphobia' interacted with social isolation, infringement of personal freedom, nancial losses and economic stress to generate mental distress.
A large proportion of these studies have merely reported the prevalence of mental health symptoms among the general public and sub-groups formed based on socio-demographic variables like gender, education level, marital status, and economic status. Statistical analysis, if any, has been restricted to univariate statistical tests of mean differences [11,15,20,21], and tests of association using correlation.
The studies that have undertaken multivariate analysis have generally estimated Ordinary Least Square models [Giallonardo et al., 2020;Varshney et al., 2020], or formed binary or multiple categories using scores of mental health to run logistic [9,17,24] or multinomial regression models [González-Blanco et al., 2020]. Most of the studies ignore the fact that anxiety, depression and stress are latent unobserved variables captured through speci c indicators. Causal relations between observed indicators and the latent variables are best captured through Structural Equation Models (SEM). This has not been attempted to our knowledge in studies on COVID-19.
This study examines the prevalence of mental health problems during the rst wave of COVID and its determinants among the general population in Indian metropolises which had seen a high incidence of active cases and deaths. It is based on primary data collected through a telephonic survey and using the DASS-21 protocol [28]. India had experienced one of the longest and most stringent lockdowns during the rst wave, generating substantial social and economic costs [29], so that considerable mental health issues may be expected. In this analysis we will use a class of SEM, called the Multiple Indicators Multiple Causes (MIMIC) model to model the simultaneous measurement of the latent constructs using observed indicators, and the structural relationship between the latent variables and their determinants.
The paper is structured as follows: Section 2 describes the data collection, sample pro le and method to undertake statistical analysis. Findings are presented in Section 3 and is followed by a section discussing the results. Section 5 summarises the results and states the policy implications of the analysis.

Survey design and instrument
The data used in the study was collected from a randomly selected sample of 1,275 mobile phone users in the six Class 1 metropolises of India-Bengaluru, Chennai, Delhi, Kolkata, Mumbai and Hyderabad.
These cities had seen a high number of active cases (22.30% of all cases), and deaths due to COVID-19 (37.99% of all cases), and the shortage of medical facilities (Supplementary Table S1). Respondents were residents of the city, aged between 21 and 60 years-randomly selected from a list of mobile telephone users. The survey was undertaken during the months of August and September 2020, when the lockdown had ended, but COVID cases and deaths were rising rapidly.
Given the restrictions on social interaction, self-reported scales tailored to detect COVID-related mental health issues [like Fear of COVID-19 scale, Corona anxiety scale, COVID stress scale, and Obsession with COVID scale] are used to measure mental health problems.
"(S)elf-report scales might prove useful as they are short, easy to administer [through paper or a digital platform], and feasible to be used when in self-isolation or quarantine. However, these scales may have limited potential to measure outcome parameters of interventions as the ndings may not be aligned with objective assessment and be more prone to response bias" [32].
Moreover, the new scales generally focus on one dimension, or at most two, and are unable to discern depression related to COVID-19, or incorporate negative socio-economic consequences of the pandemic has also been noted [33]. Finally, their reliability in different socio-cultural contexts, across different age groups, and exposure groups is not known [32]. Therefore, the survey was administered using the abridged shorter version of the DASS questionnaire, . This is an accepted survey instrument used to measures mental health issues related to anxiety, depression and stress, with good internal consistency [34,35]. It has been validated in different geographical and socio-cultural contexts, and for different socio-economic and demographic groups [36][37][38]. It has also been used in research related to SARS [39], and COVID-19 [10,16,21,24,40]. As anxiety, depression and stress symptoms are commonly reported during COVID-19 in the studies cited earlier, DASS-21 appears an appropriate survey instrument to measure mental health during the current pandemic.

Compliance with ethical standards
The Institutional Ethics Committee, Presidency University (the institute hosting the study) approved human ethical clearance to conduct the study (PU/IEC(H)/PROV CL/M-01/2020 dated 12/9/2020). The study did not use clinical trials or experiments; nor did it seek any sensitive information. The study and analysis was undertaken adhering to the Indian Council of Medical Research guidelines relevant for ethical research on humans: respect for participants, informed consent, voluntary participation with the right to withdraw, disclosure of funding sources, no harm to participants, avoidance of undue intrusion, no use of deception, preservation of anonymity, participant's right to check and modify a transcript, con dentiality of personal matters and data protection. The authors have no con ict to declare. The study was not funded.

Sample pro le
The sample pro le is given in Table 1. The mean age of the respondents is 37 years, with most of them in the 21 to 50 years' age group. Females outnumber male respondents marginally. About 40 percent of the respondents belong to the Hindu General category, while Hindus Other Backward Castes comprise 30 percent of the sample. About 81 per cent of the respondents are currently married. Only 15 percent per cent of the respondents are graduates. About 56 percent of the respondents are from households whose main earning member earns a xed income; about a third of the respondents are from households with self-employed main earners. The average monthly per capita expenditure is about USD 80, and median value is USD 59. About 68 percent respondents have monthly per capita expenditure less than USD 80. While 40 per cent of the respondents reside in bungalow type houses, 17 per cent are slum dwellers. 40 percent of the respondents belong to the households comprises of 4 members, while more than onefourth of the respondents are from the households consisting of less than 4 members. About 27 per cent of the households have an aged member (viz. aged 60 years or more). Every third respondent has a family member, relative or friend who had COVID, while one out of ten respondents have suffered bereavement due to COVID.

Statistical analysis
Estimates of Cronbach's alpha indicate a high level of reliability of responses. The value of alpha is 0.8947 for anxiety, 0.8802 for stress, and 0.8720 for depression. The statistical analysis of the data starts with an exploratory analysis of the mean scales across different socio-demographic groups to identify vulnerable groups. It is followed by a con rmatory analysis using the SEM model. It is a type of con rmatory factor analysis used to measure the value of latent constructs (measurement model); in addition, the model identi es the exogenous causes of the latent variables (structural values). Thus, the observed variables (Y) result from the latent factors (F), while the latent factors themselves are caused by other exogenous variables denoted by X. The model may be described using a path diagram ( Figure 1).
The model is described as follows: when b and d are matrices of appropriate dimensions, and e and v are vectors of error terms. Equation 1 describes the relationship between the latent variable and the observed indicators and is called the measurement model. As the 'indicators' are ordered categorical variables, the observed indicators have been linked to the latent dimension using an ordered probit model. It is depicted on the right-hand side of Figure 1. The structural part expresses the relationship between the exogenous determinants of the latent variables. It is portrayed on the left-hand side of the path diagram and is described in equation 2.
The vector Y gives the responses to appropriate questions in the DASS-21 schedule. The vector X comprises of the following exogenous variables: Age of respondents, gender of respondent, whether currently married, socio-religious identity, whether respondent has at least graduate level of education, household size, employment status of main earning member, whether resides in owned house, residence type, and whether faced COVID cases.
Occupation of respondent (or main earning member) and per capita expenditure were not included as they were highly correlated with education, employment status and residential ownership and type. The details of variables are given in Supplementary materials.

Prevalence levels of mental health problems
The mean of the raw scores for anxiety, stress and depression are 10, 12 and 12, respectively. The mean of the sum of the three raw scores (referred to as DASS-21 score subsequently) is 11. Table 2 reports the summary statistics of these scores after normalising. Normalised mean scores have, by construction, mean values of zero; the standard errors are similar and quite small. The con dence interval ranges from -0.05 to 0.05 in all cases. The median value is the highest for anxiety, and the lowest for the aggregate DASS-21 score. All scores are positively but mildly skewed [have a long right tail], and mesokurtic (having value < 3).
Another way of examining the distribution of the normalised scores is by using kernel densities. It is a non-parametric method used to estimate the probability density function. Kernel densities are useful if we wish to use a nite sample to make inferences about the population. Kernel densities are plotted for each score in Figure 2. They reveal a similarity in the nature of distribution of the anxiety, stress and depression scores; the DASS-21 score is also similar but is shifted to the right. The densities reveal that the distribution of scores is bimodal. There are two peaks-one at a low value of the scores (representing normal scores), and another much smaller peak at a higher value of score (just over 1 for the three scores, and over 2 for the DASS-21 score). The latter peak indicates the presence of moderate to severe mental health issues among a sub-section of the population.
The ndings are con rmed by the analysis of distribution of respondents across categories of mental health status ( The majority of the respondents have a 'normal' mental health score. The proportion of respondent's stress and anxiety is higher than that of depression. About 6-10 percent report mild symptoms of mental health. About one in ve respondents report mild or moderate depression, and 15 percent report severe or extremely severe depression. The proportion of respondents reporting moderate (11 percent) and extremely severe (23 percent) anxiety issues is high. While 16 percent report moderate stress levels, about 22 percent reports severe or extremely severe stress. The levels of anxiety and stress reported in the study is higher than that reported on an average in other countries; the reported level of depression, however, is similar. Table 4 reports the mean normalised scores of anxiety, depression and stress for sample groups grouped on the basis of socio-demographic and economic characteristics. Respondents who are middle aged [with age between 31 and 40 years], belongs to the Hindu General caste group, has per capita monthly expenditure levels above USD 100 and whose family members, relatives and friends had COVID but had recovered have lower scores. Economic status is also an important determinant of mental health. Respondents with xed income, monthly per capita expenditure levels in the range USD 25.01-40.00, residing in bungalows or residential complexes, and having at least two bedrooms report lower scores for all three indicators of mental health. The results for the structural model, identifying the determinants of mental health scores, is given in the bottom panel of Table 5. Mental health scores are not associated with the age of respondents, nor with marital status. Females are likely to have a lower depression and stress score, but there is no gender difference in anxiety scores. Respondents from the minority groups signi cantly more likely to have higher scores compared to respondents belonging to different Hindu social groups. Depression scores are signi cantly lower for respondents with at least graduate level of education. Respondents belonging to the households whose main earning member is self-employed are signi cantly more likely to report higher depression, anxiety and stress cores compared to families whose main earner has a xed or variable income. Respondents belonging to larger households are more likely to report higher level of anxiety and stress compared to their counterparts residing in smaller households. Respondents residing in owned ats [vis-à-vis residents of rented ats], and in slums or in bungalows (compared to residents of standalone ats) are more likely to have high scores for all the three dimensions of maternal health problems assessed here. Respondents are also more likely to have a high depression, anxiety and stress score if any household member, relative or friend had been affected by COVID, or died from the infection.

Con rmatory analysis
City xed effects are observed to be signi cant.
We have also estimated a MIMIC model relating DASS-21 scores to its indicators (Depression, Anxiety and Stress scores) and determinants. The results are reported in Table 6. The measurement model reveals that mental health is measured satisfactorily by depression, anxiety and stress scores. The results of the structural model indicate the presence of gender differences in mental health, with males being more at risk. It is also observed that risk of mental health problems is higher among members of minority communities, self-employed respondents, those residing in standalone ats, and who have household members, relatives of friends affected by COVID, and is lower among respondents with at least graduate level of education, currently married and those residing in self-owned ats.

Discussion
A national level study undertaken by National Institute of Mental Health and Neurosciences in 2016 [41] had reported that 10.6 percent of the sample had any mental health disorder. Using the International Classi cation of Diseases (ICD-10), the study estimated that 2.7, 3.0 and 0.2 percent of the population was suffering from depression, anxiety and stress, respectively. Given the home isolation, economic uncertainties, and fear of being infected (with associated treatment related problems), the incidence of mental health problems reported during the pandemic is expected to be higher. Our study, in line with similar studies undertaken in India [20,21,23,24,42,43], found the incidence and severity of mental illness to be higher during COVID. A review of studies on mental health during COVID [44] also reports high levels of mental illness during the pandemic in other countries. Moreover, the high level of co-morbidity found between different forms of psychological disorders [45] is also observed in our study [ Table 7]. It implies that the psychological disorders may reinforce each other.
Our study indicates that male respondents, members of minority communities, respondents with low levels of education, currently unmarried, belonging to larger households and those facing economic vulnerabilities, indicated by self-employed face high risk of being psychologically affected by COVID. By and large this nding corresponds to those of other studies in India and other countries. While most Indian studies report males to be more at risk of psychological distress [21,42,43], studies in Spain [46], China [17,40] and UK [6] report otherwise. Similarly, studies report currently unmarried respondents [20,42] and minorities [6,42] to be more likely to report higher mental health scores. Although age has been identi ed to be an important predictor of mental health [47], our study could not nd any such association.
The negative relationship between economic stress and mental health is one of the "most consistently replicated ndings in the social sciences" [48]. Economic stress impacts mental health through multiple mechanisms-unemployment, nancial threat, indebtedness, economic hardship, volatility in income, deterioration in living standard and decreased levels of welfare support [49][50][51][52]. The psychological impact of economic stress increases under times of economic recession and macroeconomic shocks [49,52], manifesting itself in increased levels of depression, anxiety and depression [53]. However, this relationship has been mainly examined in the context of European and North American countries [50].
The outbreak of COVID-19, the associated lockdown and disruptions in global supply chains had accentuated the slowdown of the global economy, while the failure to provide a social security net in developing countries like India had created economic insecurity [29]. These circumstances may be expected to increase the incidence of mental health problems, particularly among vulnerable sections [6,54].
The result of our study reveals that households whose main earning member is self-employed is more likely to report higher scores for depression, anxiety and stress. Given that the economic shock of the national lockdown took time to cascade throughout the economy, the impact on mental health is likely to be higher at the time of the survey, compared to the onset of the pandemic. It may explain the higher incidence of anxiety and depression reported in this study compared to that in other countries (Table 3). Respondents residing in rented houses vis-à-vis owned houses are also likely to face psychological problems.
On the other hand, slum dwellers are less likely to report high values of depression, anxiety and stress scores. This is somewhat surprising as slum dwellers are considered to be a high risk group [55] given the di culties in maintaining social distancing or hygiene standards and shared communal facilities including toilets in slums [56]. Positional objectivity [57] may be a possible reason. Slum dwellers had been badly hit by demonetarisation and the slowdown of the Indian economy; the incremental impact of the pandemic was marginal. Further, "The severity and importance of health problems were controlled more by the capacity to deal with problems than by the problems themselves. This ability was associated with retaining traditional ruralurban dual social structure such as strong family support, traditional family-based caring, well-built social and kin network ties" [58].

Conclusion
Our study estimates that the onset of COVID increased the levels of psychological disorders in India compared to normal times. We found that 34, 46 and 48 percent of respondents reported symptoms of depression, anxiety and stress. We also nd presence of co-morbidity between depression, anxiety and stress, with correlations ranging between 0.85 to 0.95. The study identi es single respondents, members of minority groups, less educated and those belonging to large households to be signi cantly at risk. Results indicate that the relationship between economic stress and mental illness may be more complex than reported in studies for Western societies.
While studies show a peak in the incidence of psychological problems, all such people may not require clinical attention. People are resilient and nd new strengths in adversity [59]; community level efforts may also provide the necessary level of support. However, certain sections are more vulnerable and need to be monitored for possible clinical support. However, COVID has curtailed the clinical monitoring and supporting of at risk persons [59]; telemedicine is also limited in India. Therefore, we recommend: i. scaling up emergency services for at risk groups; ii. creating mechanisms to continue delivery of such services on a long term to combat Post Traumatic Stress Disorders arising from exposure to COVID-related trauma, and iii. use local authorities to identify and deliver services to groups most at need [60].
Further, given the psychological impact of economic distress, social safety measures targeting the economically vulnerable households should be made a key component of post COVID recovery programmes [6].
Finally, we point out certain limitations of our study. The survey was undertaken over telephone. Despite the pervasive spread of mobile phones, it is possible that we have missed certain sections of the society; further, investigators were unable to cross validate responses. Studies have also noted that the level of psychological distress changes over time-peaking at the onset of the pandemic, it tends to decrease as people start to adjust to the 'new normal'. Such temporal uctuations need to be captured through a longitudinal survey.    Source: Estimated from primary data.  ***, ** and * denotes Prob. < 1%, 5% and 10% level, respectively.
Source: Estimated from primary data.  Kernel densities of normalised scores for mental health

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryTable.docx