The changing link between pandemic-related stress and child and adolescent mental health during the COVID-19 pandemic – Scientific Reports

Sample study with participants

Our study uses data from the Indian Consumer Pyramids Household Survey (CPHS), combined with data on COVID stress and health outcomes collected in the Survey of Health Trends (SEHAT) module among CPHS family group. The CPHS is a long-term survey that is conducted three times a year in 175,000 households, and about 675,000 of their members.⁴⁰. The survey, compiled by the Center for Economic Monitoring of India (CMIE), covers a wide range of topics related to economic and social indicators. The sample, which is based on spatial sampling, is spread across the geography of India, and represents 98.5% of the country’s population.⁴⁰. Each wave of CPHS is conducted over a period of four months, families are re-interviewed after four months, and new families are added to replace those lost to follow-up. The survey was first administered in January 2014 and data collection is ongoing.

The SEHAT was developed as a CPHS module to assess the mental health effects of COVID-related stress. The SEHAT module sample included households interviewed in the first month of each CPHS wave (February, June and October). Data for the current survey are from the February 2022 and October 2022 waves. The SEHAT module was administered to approximately 32,000 households in both February 2022 and October 2022. Other interesting data from the main CPHS , either in the first wave or simultaneously when appropriate, were integrated into the SEHAT module for analysis. Although a portion of the families were also interviewed in multiple waves, the present analysis treats the two data as two two-dimensional waves since more than 20% of the children’s level data were are in households that have dropped or are included in the sample in recent waves. In each interview, the main respondent, usually a woman in her 40s, answered a child mental health module about a child (ages 5-19) of lives in their family.

Our analysis sample included children and youth aged 5 to 19 years with missing covariates, and had household income information from at least one wave prior to (May 2018-February 2020) and after (October 2020-October 2022) COVID to help calculate the difference in currency conversion described below. About 3% of the sample was excluded due to missing data on covariates and an additional 10% was excluded due to missing income data. This led to a sample of 25,683 in February 2022 and 22,356 in October 2022.

Steps

Epidemic concerns

Known COVID-19 illness or death on a social networking site

Participants were asked to indicate for each member of the household list if each person had COVID, became seriously ill from COVID, or died from COVID. Additionally, participants were asked if their neighbors, friends/relatives, or others they know have had COVID, become seriously ill from COVID, or have died from COVID. During the January 2022 wave, the reference period was ‘from the start of the COVID-19 epidemic in 2020’; during the October 2022 wave the reference period was ‘since February’ for those who participated in the February wave and ‘since the start of the COVID-19 pandemic in 2020’ for those who did not of the February wave. For the October wave, a general step has been taken using the most severe form of the COVID-19 pandemic, creating a common reference period ‘since the start of the COVID-19 pandemic in 2020’ across waves and respondents.

Using these measures, we developed a five-stage model to characterize the illness and death of COVID-19 in social networks. The categories were unique and ordered from least (1) to most severe (5):

1. (1)

   Not fully exposed: Those who did not know anyone who was infected (including themselves), or who were seriously ill or died from COVID-19;

2. (2)

   A simple community, it’s not close: Those who only knew of neighbors or others infected with COVID-19. Among this group, none of the participants, nor their friends/relatives were infected with COVID-19; nor did they know anyone who became seriously ill or died as a result of COVID-19;

3. (3)

   A strong community, it’s not close: Those who only knew of neighbors or others who were seriously ill or died from COVID-19. Among this group, none of the friends/relatives had COVID-19 or died from COVID-19;

4. (4)

   A little closer: Those who reported that they or their friend/relative has been infected with COVID-19 but were not seriously ill and did not die due to COVID-19; and.

5. (5)

   It is very close: Those who reported that a friend/relative was seriously ill or died due to COVID-19.

Lockdown stringency index

We use data from the Oxford COVID-19 Government Response Tracker (OxCGRT)⁴¹ to measure the coverage of COVID at the level of the respondents on the day of the interview. Specifically, we use a stringency index that includes 9 measures including school closures, workplace closures, cancellations of public events, restrictions on the size of collections, closures of public transport, demands stay-at-home orders, internal movement restrictions, international travel restrictions, and the existence of social media campaigns. The index is a score of the following symptoms normalized between 0 and 100.

Changes in income

The income variable regression was intended to compare (a) the average household income during the pre-epidemic wave and (b) the average household income during the data collection waves since the outbreak and the wave of data collection (either in February or October 2022). We calculate average household income before the pandemic using six waves of data between May 2018 and February 2020. For the January 2022 samples, we calculate household income during the pandemic using waves of six waves collected between April 2020 and February 2022. For the 2022 samples, we calculate the average household income during the pandemic using six waves collected between December 2020 and October 2022. We created an average income pre-COVID-19 to COVID income, limited only to those with data from the pre- or earlier wave of COVID, as well as those with data from the post-COVID wave or wave after that. Average income was divided into three groups: decreased, remained the same, and increased. A threshold of 0.2 was used to indicate a 20% increase/decrease in income during the COVID period compared to pre-COVID income.

Internalizing symptoms

Internal markers were analyzed in two steps. The first step included 5 items from the parent report of the Pediatric Symptoms Checklist (PSC).⁴². Adults were asked if the child ‘feels scared’, ‘feels hopeless’, ‘isolates’, ‘worries a lot’, or ‘seems not having much fun’. Each question was on a 4-point Likert scale (coded 0–3), with a total score of 0–15. The internal consistency of the items in this sample was 0.84. For regression analyses, these data were standardized to a mean of zero and a standard deviation of 1.

The second step was the general anxiety section of the SCAS parent report^43,44. For 5- or 6-year-old children, the primary school version was used, with 5 items: the child ‘has difficulty controlling anxiety’, ‘is restless, restless or restless’ irritable because of anxiety’, ‘he has trouble sleeping because of anxiety. ‘, ‘he spends a large part of each day worrying about various things’, ‘he asks for reassurance when it seems unnecessary’. For children and adolescents aged 7 to 19, the 6-year school version was used: ‘my child worries about things’, ‘when my child has a problem, he complained of having a funny feeling in his stomach. ‘, ‘my child complains of being scared’, ‘when my child has a problem, he complains of his heart beating very fast’, ‘my child is worried that something will happen to him bad’ my child has a problem, he feels shaking’. Questions in the preschool and school-age versions were assessed using a 4-point Likert scale (never, sometimes, often, always). Responses were summed, resulting in a range of 0–15 for the preschool model and 0–18 for the preschool model. The internal consistency of the preschool items was 0.87 and the school-age version was 0.88. In order to model SCAS scores for all children at the same time, scores for preschoolers and preschoolers were standardized with a standard deviation of 1.

Other covariates

Other covariates of interest used in the analysis include youth age, youth gender, urban/rural residence, status, family education, number of people in the household, income before COVID (see “income changes ” measure for information on measurement), and the adult’s gender and age. Additionally, in the October 2022 wave, a covariate indicated whether the household was newly added to that wave. Area categories are based on the 2011 Census and are defined by CMIE as rural if the family is in a village and urban if the family is in an urban area. Caste was divided into four groups: scheduled caste (SC) / scheduled tribes (ST), other backward classes (OBC), middle classes, and upper classes (UC). The household education variable was produced by the CMIE by including literacy data from all adults in the household.⁴⁵they are divided into low (illiterate or educated adults), moderate (some literate) and high (including college graduates).

Math tests

We used ordinary least squares (OLS) to assess whether epidemiologic stress was related to child and adolescent mental health during February and October 2022. Models were run on sections for wave each of the two waves, together with the result. Results are presented both for the full sample as well as for the age-stratified sample. Age groups were chosen to best capture the distinct developmental periods of childhood (5-7), emerging adolescence (8-12), and adolescence (13-19). All variables controlled for youth age, gender, urban/rural residence, status, family education, family size, pre-COVID income, and adult gender and age. Additionally, for the October 2022 wave, a control was added indicating whether the respondent was new to the sample since the January 2022 wave. Models were also grouped by household ID to account for youth living in the same household, as well as including conversational fixed effects. For ease of interpretation, coefficients and robust confidence intervals pooled at the original sample level were plotted.