Study population

We used data from the Nash-wo-Numa study, a cross-sectional study conducted between January 2019 and February 2020 among early adolescents aged 9 to 15.9 years and their birthmothers living in the Matiari District, Pakistan [20]. Matiari is located in the Pakistani province of Sindh, the second most populous province of Pakistan. According to 2017 Census data, more than 130,000 adolescents aged 9 to 15.9 years live in Matiari, making up 17% of the district’s population [21]. Approximately 77% of them reside in rural areas. The vast majority of the population in Matiari speaks Sindhi.

The primary objective of the Nash-wo-Numa study was to examine multiple aspects of adolescent growth, development, and well-being. To meet this objective and to account for the impacts of pubertal onset, which begins earlier in girls, age criteria for inclusion were 9.0–14.9 years for girls and 10.0 to 15.9 years for boys. The Nash-wo-Numa research team identified eligible households using a list established during a previous household census conducted among 53,000 households covered by Female Health Workers in the catchment population of 26 health facilities in Matiari District [22]. They selected a representative sample of households using computer-assisted random sampling based on sex, age, and the number of occupants in each household [20]. If multiple adolescents from the same household were eligible, only one was randomly selected.

The Nash-wo-Numa Study staff visited a total of 1,873 eligible households. The level of urbanization (rural, peri-urban, or urban) for each household was determined based on household address and census classification. Participants outside the eligible age range (9 to 15.9 years) were not included (n = 151), nor were adolescents whose mothers were not available, did not consent to participate, or were cognitively unable to be interviewed (n = 9), adolescents who suffered from a chronic or genetic illness known to impact growth (e.g., congenital heart disease, diabetes, or Down syndrome) and girls who were or had been pregnant (n = 52). During the data management process, we identified 26 participants whose age did not meet the sex-specific age criteria; those were also excluded. For the purpose of the present study, we removed 239 participants with no data on the depressive and anxiety symptoms scales, leaving a final sample of 1,396 participants.

Both informed written consent from the parent or legal guardian and written assent from adolescents were obtained. Ethics approval for the original study was granted by the Ethics Review Committee at the Aga Khan University, Karachi, Pakistan (5251-WCH-ERC-18) and Research Ethics Board at SickKids Hospital, Toronto, Canada (1000060684). Ethics approval for the current study (secondary use of data) was received from the Centre for Addiction and Mental Health Research Ethics Committee (107/2021). Data were accessed on December 11, 2021. Authors had no access to information that could identify individual participants during or after data collection. Additional information regarding the ethical, cultural, and scientific considerations specific to inclusivity in global research is included in the Supporting Information (S8 Checklist)”.

Procedure

Staff members coordinated appointments and transportation to a field-based clinic. Participants and their mother were interviewed together, with a chaperone present when necessary, using structured questionnaires intended for either mothers (six modules) or adolescents (nine modules). Study psychologists administered mental health measures.

All questionnaires were translated from English into Sindhi and reviewed by psychologists and staff members fluent in both languages. They were pilot-tested in the community before being used in the study. Further details regarding the study protocol can be found elsewhere [23].

Outcomes

Scales and subscales characteristics (number of items, scores range, Cronbach’s alpha and threshold used to define clinically-relevant symptoms) are summarized in Table 1.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of scales and subscales used to assess mental health outcomes.

https://doi.org/10.1371/journal.pone.0350609.t001

Depressive symptoms were self-reported by adolescents using the Short Mood and Feelings Questionnaire (SMFQ), which assesses affective and cognitive symptoms of depression experienced by children and adolescents [24]. Participants were asked to rate how they had been feeling or acting recently (e.g., “I did not enjoy anything at all” or “I felt I was a bad person”), with each item being rated using a 3-point Likert Scale (“not true” = 0, “sometimes” = 1 or “true” = 2). The SMFQ has been validated in community-based samples [25–27]. In the Nash-wo-Numa study, the SMFQ demonstrated strong unidimensionality (Comparative Fit Index = 0.97 in a confirmatory factor analysis of a single factor model) [28], and good internal consistency (see Table 1). In accordance with the original validation study, a total score of 8 or more was considered as indicative of clinically-significant depressive symptoms [24].

Anxiety symptoms were assessed based on information provided by adolescents using the Screen for Child Anxiety Related Emotional Disorders (SCARED), initially developed by Birmaher et al. [29]. Five SCARED subscales screen for panic disorder or significant somatic symptoms, generalized anxiety disorder, separation anxiety disorder), social anxiety disorder and – among adolescents attending school – school avoidance. The presence of each symptom over the past three months (e.g., “I am nervous” or “I worry about what is going to happen in the future”) was rated as 0 (Not true or hardly ever true), 1 (Sometimes true) or 2 (True or often true). The psychometric properties of the SCARED have been validated across different countries, including LMICs such as South Africa and China [30]. For each subscales, we used international thresholds previously suggested to indicate the presence of clinically-significant anxiety disorders [29]. Although the psychometric properties of the SCARED were satisfactory in LMICs [31,32], its international thresholds have not been consistently validated, with studies reporting different optimal cut-offs values [33]. A total anxiety score was computed using the items of the panic/somatic, generalized anxiety, separation anxiety and social anxiety subscales (37 items). Items from the school avoidance subscale were not included, as they were administered only to the subsample of adolescents who attended school. A similar score was employed in a study conducted among adolescents from a rural community in India [34]. To date, no universally accepted cut-off has been established for this version of the total SCARED score.

Household characteristics

Household characteristics included socioeconomic factors commonly examined in relation to adolescent depression and anxiety in LMICs, including mother’s marital status, mother’s and partner’s employment and school attendance, household wealth, and food insecurity [17,18]. IPV has been reported as a key correlate of adolescent mental health outcomes in LMIC settings [13,35] and was used to capture interpersonal dynamics within the household.

Mother marital status was declared by the mother herself and classified as ‘married’ or ‘widowed, divorced or separated’.

Mother’s and partner’s working school attendance. Mother reported whether she and her partner had ever attended school.

Mother’s and partner’s working status. Mothers were asked about their own occupation as well as their partner’s. Partner’s occupation was categorized as ‘manual labour’ (skilled and unskilled and agriculture), ‘non manual labour’ (i.e., sales and services, professional jobs), and ‘unemployed’. Because most mothers reported being housewives, we created a binary variable to distinguish homemaker mothers from those who had an occupation.

Wealth Index. The Wealth Index is a composite indicator of a household’s cumulative living standard, used as a proxy for socioeconomic position [36]. Information about housing characteristics, ownership of house/land, number of rooms, fuel used for cooking, source of drinking water, type of sanitation, assets and livestock were reported by mothers. The Wealth index was created using principal component analysis on these variables as recommended by Filmer and Pritchett [36]. The first component obtained was considered as a proxy indicator of the level of wealth at the household level and was split into five wealth quintiles (poorest, poor, middle, richer and richest quartiles). The poorest and poor categories were then combined into a single category “poor”, and the middle, richer and richest wealth quintiles were grouped into “non-poor”.

Food insecurity. Food insecurity is described as a direct consequence of insufficient economic access to food and is closely tied to income poverty, especially in rural areas [36]. Mothers were administered the Food Insecurity Experience Scale (FIES), which consists of eight yes/no (1/0) questions referring to the experience of food insecurity at the household level over the last 12 months (i.e., “Was there a time when your household ran out of food because of a lack of money or other resources?”) [37]. The internal consistency of the scale was very good in both boys (α = 0.96) and girls (α = 0.92). The FIES score was categorized as follows: 0–3 = food secure/mild food secure, and 4–8 = moderate to severe food insecure in accordance with the Food and Agriculture Organization of the United Nations’ guidelines [38].

Mother’s exposure to Intimate partner violence (IPV). Mothers reported intimate partner violence using three subscales of the Conflict Tactics Scale short form (CTS2S) [39], each including 2 items: psychological aggression (“My partner insulted or swore or shouted or yelled at me”; “My partner destroyed something belonging to me or threatened to hit me”), physical assault (“My partner pushed, shoved, or slapped me”; “My partner punched, or kicked, or beat me up”) and injuries due to partner violence (“I had a sprain, bruise, or small cut, or felt pain the next day because of a fight with my partner”; “I went to see a doctor or needed to see a doctor because of a fight with my partner”). IPV was coded as present if any of the six behaviours occurred in the past year.

Religion, measured at the household level, was used to describe the sample of boys and girls included in our sample, but was not considered as an explanatory variable in the present study, as our aim was to address factors that could be modified by public health or social interventions.

Adolescent’s characteristics

Date of birth and sex assigned at birth were reported by mothers.

Age was calculated as the number of years between the date of birth reported by the mother and the date of the interview at the field-based clinic.

School attendance was assessed by asking adolescents whether they were currently attending school at the time of the survey.

Mother’s mental well-being

Trained psychologists administered the 14-item Warwick-Edinburgh Mental Wellbeing Scale (WEMWBS) to mothers to assess the feelings and thoughts they experienced over the last two weeks (e.g., “I’ve been feeling good about myself”) [40]. Each item was rated from 1 (none of the time) to 5 (all of the time), with total sum scores ranging from 14 to 70. Higher scores indicated a higher level of mental wellbeing. The WEMWBS demonstrated good internal consistency in our sample (α = 0.91 in both boys and girls).

Analysis

Analyses were stratified by sex to address the evidence gap regarding boys’ mental health, and because age ranges were different for boys and girls. Prevalence estimates were derived based on international threshold for each instrument and its subscales and confidence intervals were obtained using the Wilson score method. Differences in prevalence estimates according to sex were compared within overlapping age ranges using chi-square tests.

Depression and anxiety scores were right-skewed (i.e., more scores at the lower end of the scale). Correlations between scores and between covariates were examined using Spearman rho.

We estimated the association of household characteristics independently of adolescent characteristics (age and school attendance) and mother’s well-being. Negative binomial regression was preferred to Poisson regression because of significant over-dispersion (likelihood ratio test of the dispersion parameter, p < .001 in all models). Associations were described using incident rate ratios (IRR). The IRR corresponds to the ratio of the rates of symptoms in two different groups. The association of each household characteristic with depressive and anxiety symptoms was first estimated in models adjusted for age and school attendance (in models M1). All the adolescent and household characteristics were then introduced simultaneously in a full model (M2). We added mother’s mental well-being in a third model (M3). When a covariate was significant in M2 but not significant in M3, we tested for potential interactions between that covariate and mother’s well-being in predicting the outcome.

All statistical analyses were performed using STATA version MP/16 (Stata Corp., College Station, Texas, USA). A p-value less than 0.05 was set to be statistically significant.

Participants

The characteristics of the 678 boys (mean age = 12.9; SD = 1.6) and the 718 girls (mean age = 12.0; SD = 1.7) included in the study are described in Table 2. Nearly three quarters of boys, but only half of the girls, attended school. More than 90% of mothers were married and were living with their partner. The vast majority of mothers’ partners were working in manual labour and agriculture. Mothers of girls were younger and more likely to have attended school. A greater proportion of girls than boys lived in Muslim families, whereas boys were more likely than girls to reside in rural areas.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Characteristics of participants.

https://doi.org/10.1371/journal.pone.0350609.t002

Compared to participants included in the analysis, those excluded because of missing mental health data (n = 239) were significantly older, their mother’s partner was less likely to work in manual labour and they were less exposed to food insecurity (see S1 Table).

Prevalence

The mean SMFQ score was 2.3 (SD = 3.2; min = 0; max = 21) in boys and 2.7 (SD = 3.5; min = 0; max = 24) in girls. The prevalence of clinically-significant depressive symptoms was 8.1% (95%CI 6.3–10.4) in boys and 10.2% (95%CI 8.2–12.6) in girls (S2 Table).

Boys had a mean anxiety symptom score of 12.5 (SD = 9.9; min = 0; max = 56). The most prevalent clinically-significant symptoms were symptoms of separation anxiety (23.9%; 95% CI 20.8–27.3), followed by symptoms of panic/somatic disorder (13.3%; 95%CI 10.9–16.0), social anxiety (11.2%; 95%CI 9.1–13.8), and generalized anxiety disorder (5.8%; 95%CI 4.2–7.8). In boys attending school (n = 508), the prevalence of school avoidance was 15.8% (95%CI 12.8–19.1).

Girls had a mean anxiety symptom score of 16.7 (SD = 11.8; min = 0; max = 72). The prevalence of clinically-significant symptoms of separation anxiety disorder was, again, the highest (39.1%; 95%CI 35.6–42.8), followed by symptoms of social anxiety disorder (22.3%; 95%CI 19.4–25.5), panic or significant somatic disorders (17.1%, 95%CI 14.6–20.1), and generalized anxiety disorder (7.7%; 95%CI 5.9–9.8). The prevalence of school avoidance was 15.5% (95%CI 12.2–19.5) in the subsample of 375 girls who attended school.

Age-specific prevalence estimates were compared according to sex from age 10 to age 14. Significantly more girls than boys presented clinically-significant symptoms of depression, panic or somatic disorder and generalized anxiety disorder at age 12 (S2 Table). Girls were more likely to experience clinically-significant symptoms of separation anxiety at age 11, 12, 13 and 14, and to experience clinically-significant symptoms of social anxiety at age 12, 13 and 14. No sex difference was found in the prevalence of school avoidance.

Association of symptom scores with household characteristics

Depression and anxiety symptom scores were moderately correlated (Spearman ρ = 0.60; p < .001) and correlations between explanatory variables ranged from negligible to moderate (S3 Table).

Results of the final multivariable regression models describing the association between household characteristics and symptom scores are presented in Table 3 and 4. Results obtained in simple (M1) and intermediate multi-adjusted models (M2) are available in S4-S7 Tables.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Association of household characteristics with depressive and anxiety symptoms in adolescent boys living in Matiari, Pakistan (n = 678).

https://doi.org/10.1371/journal.pone.0350609.t003

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Association of household characteristics with depressive and anxiety symptoms in adolescent girls living in Matiari, Pakistan (n = 718).

https://doi.org/10.1371/journal.pone.0350609.t004

In boys, lower levels of depressive and anxiety symptoms were found in boys whose mother was a homemaker (respectively IRR = 0.74; IC95% = 0.61–0.91; p = .003 and IRR = 0.89; IC95% = 0.79–1.00; p = .047), compared to boys whose mother had a professional activity (Table 3). Higher levels of depressive and anxiety symptoms were found in boys exposed to intimate partner violence against their mother (IRR = 1.61; IC95% = 1.30–2.00; p < .001 and IRR = 1.20; IC95% = 1.06–1.35; p = .003) and to food insecurity (IRR = 1.45; IC95% = 1.16–1.83; p = .001 and IRR = 1.22; IC95% = 1.07–1.40; p = .003).

In girls, living in a rural area was associated with decreased symptoms of anxiety (IRR = 0.88; IC95% = 0.79–0.98; p = 0.022) (Table 4). As found in boys, levels of depressive and anxiety symptoms decreased in girls whose mother was a homemaker (respectively IRR = 0.72; IC95% = 0.60–0.86; p < .001 and IRR = 0.90; IC95% = 0.81–0.99; p = .034) and increased with exposure to intimate partner violence (IRR = 1.42; IC95% = 1.18–1.71; p < .001 and IRR = 1.18; IC95% = 1.06–1.30; p = .001). The significant positive relationship between food insecurity and depressive symptoms, originally found in model M2 (p = .008, see S6 Table), was no longer significant after adjustment for mother’s mental health well-being (in model M3, p = 0.553). A significant negative interaction was found between food insecurity and mother mental health (IRR = 0.98; IC95% = 0.66–1.00; p = 0.024) and was included in the final model predicting depressive symptoms. Moderate to severe food insecurity was associated with higher anxiety symptoms (IRR = 1.14; IC95% = 1.01–1.28; p < .001). As shown in Fig 1, among girls, the association between moderate-to-severe food insecurity and depressive symptoms was stronger when mothers had low mental health well-being scores, and weakened as mother’s mental well-being score increased. No significant interaction was found between food insecurity and mother’s mental health well-being when predicting anxiety symptoms (p = .127).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Predicted depressive symptoms score according to mother’s mental health well-being score and food insecurity in girls living in Matiari, Pakistan (n = 718).

https://doi.org/10.1371/journal.pone.0350609.g001

Strengths and limitations

Despite the rigorous selection of participants being a key strength of our study, a selection bias may still have occurred, as those excluded due to the absence of mental health data were slightly older and less exposed to food insecurity. This may have affected the prevalence estimates, but does not undermine the validity of the associations between household characteristics and depressive and anxiety symptoms.

Our large sample size allowed us to robustly examine the independent associations of multiple household characteristics with depression and anxiety symptoms. But small cell counts in some categories (such as mother’s marital status or partner’s unemployment) may have affected the precision of our estimates and the significance of our results. For these variables, limited statistical power may have biased our results toward non-significance. Moreover, the selection of household characteristics was constrained by the available data, with many variables centered around the mother. Some of these variables, such as the mother’s “homemaker” status, may reflect diverse household arrangements or caregiving practices, which were not measured in our study. The characteristics included may represent only part of the broader household context and should be interpreted alongside their limitations, including the lack of comprehensive measures of caregiving practices and more paternal characteristics.

Due to cultural sensitivities, the child and the mother were interviewed together, sometimes in the presence of a chaperone. While this approach improved the feasibility of the study, it may have increased social desirability bias. This could have led to under-reporting of anxiety and depressive symptoms, as well as other sensitive information, such as IPV. If so, such misclassification could have resulted in the underestimation of the prevalence estimates and of the associations observed.

Psychometric interviews in Sindhi by trained psychologists were employed to reduce language and comprehension barriers and the scales used demonstrated acceptable to excellent reliability in our sample [28]. Unfortunately, clinical measures of depression and anxiety disorders were not collected, which prevented us to assess the validity of international thresholds in our population. Although psychometric assessment tools cannot be used to diagnose clinical disorders, they are useful to quantify the severity of symptoms, and to screen for ‘probable’ cases of depression. Their use as proxies for clinical diagnosis in prevalence studies has important advantages for estimating population’s needs and enabling comparisons, but they have been criticized for having a high chance of false-positive results and thus, for overestimating the burden of diseases [73].

Due to our cross-sectional design, the associations observed should not be interpreted as causal. Longitudinal studies are needed to better understand the temporal relationships between household characteristics, mother’s mental health and anxiety and depression in adolescents. To date, no such study has been conducted in Pakistan.

Finally, the generalizability of our findings to other locations and settings (particularly urban areas) should be considered with caution. The data were collected prior to COVID and the 2022 floods and may not reflect the current situation. However, our estimates may serve as a useful reference for understanding how these events have subsequently impacted adolescents’ mental health, particularly in the district of Matiari.