Introduction

Depression in youth is common worldwide and has been identified as one of the leading causes of disability among those 10- to 24- years old [1]. However, the burden of depression and depressive symptoms in youth living in low- and middle-income countries (LMICs) is still difficult to estimate due to a lack of available data [2]. Depression in adolescence is associated with deleterious behaviours and outcomes. In LMICs, relative to non-depressed adolescents, adolescents with depressive symptoms are more likely to engage in risky sexual behaviour, substance use, delinquent activity, self-harm, and suicidal behaviour [3]. Youth who experience depression or depressive symptoms are also at a greater risk for future affective disorders [4] and ill health during adulthood [5].

Pakistan is classified as an LMIC and is the sixth-largest country in the world, with youth between the ages of 10 to 19 years old accounting for approximately 23% of the total population [6, 7]. In a school-based study of urban youth ages 11 to 18 years old living in Rawalpindi, the estimated prevalence of depression measured by the Hospital Anxiety and Depression scale was 17.2% [8]. Similar results were obtained among urban youth 12 to 14 years old living in Hyderabad: the prevalence of depression as measured by the Children’s Depression Inventory was estimated to be 14.9% in girls and 19.9% in boys [9]. To our knowledge, the only study that has examined the prevalence of depression in youth living in rural Pakistan was conducted in a sample of unmarried 16- to 18-year-old girls living in rural Punjab [10]. The estimated prevalence of clinical depression as measured by the Structured Clinical Interview for DSM-IV Disorders was 4.4% in this sample. However, one-third of the sample had high levels of psychological distress as measured by the Self Reporting Questionnaire, with higher scores being associated with poverty, physical abuse, family stress, and adverse recent life events [10].

Despite the need for mental health care, in Pakistan, the ratio of psychiatrists per 100,000 people is 0.18, which is much lower than the world median ratio of 1.27 [11]. Further, most mental health services for youth are provided in adult psychiatric facilities, located mainly in urban areas [12]. In primary care settings, many health workers have limited time and/or insufficient training to administer diagnostic interviews to all at-risk individuals [13]. Access to free, validated, and easy to administer tools for screening depressive symptoms in youth may thus contribute to the improvement of detection, surveillance, and early intervention of youth depression in LMICs.

Most existing screening tools for youth depression originate from high-income countries (HICs), and few psychometric studies have been conducted in LMICs [14]. However, the clinical presentation of depression and depressive symptoms can vary based on culture and context. Culture may define and create specific sources of distress, which in turn impact the manifestation and self-report of psychiatric symptoms [15]. For example, compared to Western cultures, Eastern cultures have been found to stigmatize mental health problems more and show more concern over the impact that disclosure of a mental illness may have on family social and economic status [16]. In some LMICs, such as Pakistan, suicidal behaviours are considered criminal offences punishable by law, which may further impact disclosure [17, 18]. Low levels of endorsement of items can then affect the validity of the measurement tool they are part of. These considerations are especially salient when attempting to administer screening tools developed in HICs in other contexts and it is unclear whether variation in prevalence estimates across countries may represent actual differences in prevalence or differences in measurement [19]. As such, there is a need for the adaptation of existing scales to cultural specificities and the investigation of the psychometric properties of screening tools prior to their use in different cultures and contexts [20]. Several indigenous and culturally adapted scales that measure adult depression are available in Urdu, Pakistan’s national language [21]. Among scales developed specifically for youth, the Strengths and Difficulties Questionnaire (SDQ) which measures internalizing more broadly [22] and the Centre for Epidemiological Studies Depression Scale for Children have been translated to Urdu [23]. To our knowledge, there are no scales available in Sindhi, the most common language spoken in the province of Sindh [24].

The Mood and Feelings Questionnaire (MFQ) is one of the most useful tests for screening childhood depression in children and adolescents [25]. It has been used and validated in many epidemiological, clinical and non-clinical studies [26–28], and is recommended by the National Institute for Health and Clinical Excellence [29]. The instrument was developed to reflect symptoms of depression specified in the DSM-III-R, in addition to several items of clinical relevance related to loneliness and feeling unloved [30] but has also been validated against DSM-IV(-TR) and DSM 5 depression diagnoses [31, 32]. The validated parent- (MFQ-P; 34-items) and self-report (MFQ-C; 33-items) forms are free of use [31]. Regarding the MFQ factor structure, both single- and five-factor structures (cognitive, vegetative, core mood, suicidality, and agitated distress symptoms) have been supported in HICs [19, 28, 33]. Also available is the short 13-item version of the MFQ (SMFQ), which has also shown good psychometric properties in community and clinical samples [34–36]. Although there are benefits of parents and youth as informants when assessing depressive symptoms, it has been frequently noted that parents tend to underreport their child’s depressive symptoms on screening tools [37–39]. As such, information self-reported by the adolescents themselves is often preferred [40].

To our knowledge, only two studies have examined the psychometric properties of the MFQ-C (long and short forms) in community samples of adolescents living in LMICs. In Thailand, the full-form MFQ-C was recently validated in a school-based study of youth 12 to 18 years old. The Thai MFQ-C showed excellent internal consistency (a = .92), convergent validity with related measures of depression (the Childhood Depression Inventory and the SDQ-emotional symptoms subscale), and good test-retest reliability [32]. However, the factor structure of the scale was not examined. In Bangladesh, the psychometric properties of the short form (SMFQ-C) were studied in a convenience sample of 9 to 17-year-old adolescents living in rural and urban areas [41]. The unidimensional structure of the SMFQ-C translated in Bangla was supported. The brief scale showed good internal consistency (α = .84) and adequate convergent validity with the Spence Children’s Anxiety Scale-Parent version [41].

Both the MFQ-C and SMFQ-C show limited but promising evidence of good psychometric properties in LMICs. This study aims to investigate the psychometric properties of a culturally adapted version of the MFQ-C, translated into Sindhi, in adolescents living in the rural district of Matiari, Pakistan. Given the benefits that a shorter screening tool may provide in general populations or low-resource settings, the psychometric properties of the original SMFQ-C will also be examined.

Method

Analyses were conducted using data from the Nash-wo-Numa study, a cross-sectional study of adolescents living in Matiari, Pakistan, between January 2019 and February 2020. Information relevant to the current study is presented. Further details regarding sampling strategies, procedures, study measures, and objectives are described elsewhere [42].

Results

Sample characteristics

Only one participant was excluded from our analyses due to missing responses to all MFQ items (see Fig A in S1 File for participant flow-chart). All other participants had complete data on the MFQ. The final sample consisted of 1350 participants (52.3% female).

Participants’ demographics are presented for the whole sample and by sex in Table 1. The mean age of participants was 12.4 (SD = 1.7) years, 12.8 (SD = 1.6) years for males, and 12.0 (SD = 1.7) years for females. The majority of participants attended school and lived in rural areas. Compared to males, females were younger, were less likely to attend school but more likely to live in an urban area. Their mothers were also slightly younger and they were more likely to live in a Muslim family (all p≤0.01, see Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic and family characteristics of participants.

https://doi.org/10.1371/journal.pgph.0000968.t001

Factor structure of the MFQ-C and the SMFQ-C

The overall KMO measure of sampling adequacy was 0.93 for the full scale MFQ and 0.88 for the SMFQ, indicating that the sample was suitable for factor analysis. Fit indices for all models are presented in Table 2. The unidimensional structure from the original MFQ-C (29 items) showed adequate fit (CFI = 0.94, TLI = 0.94, RMSEA = 0.06) with item factor loadings ranging from 0.34 (“I didn’t have any fun in school”) to 0.97 (“Talking more slowly”) (Fig 1). However, the 26-item four-factor model showed an excellent and overall better fit than what was observed for the unidimensional factor model (CFI = 0.97, TLI = 0.97, RMSEA = 0.05). All factor loadings were greater than the specified cut-off of 0.32 [47] (Fig 2). Again, the lowest factor loading was observed for the item “I didn’t have any fun in school” on the ‘cognitive’ subscale (0.35). In a sensitivity analysis conducted in the subgroup of participants who attended school, this item loaded higher but still had the lowest factor loading (0.46), and there was minimal change in model fit indices. Finally, the unidimensional model for the SMFQ-C also showed excellent fit (CFI = 0.97, TLI = 0.96, RMSEA = 0.07), with all factor loadings being greater than 0.49 (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Factor loadings from the unidimensional confirmatory factor analysis of the modified Mood and Feelings Questionnaire (child-version, 29 items) and the Short Mood and Feelings Questionnaire (child-version, 13 items).

https://doi.org/10.1371/journal.pgph.0000968.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Factor loadings from the four-factor solution in confirmatory factor analysis applied to data obtained with the modified Mood and Feelings Questionnaire (child-version, 26-items).

https://doi.org/10.1371/journal.pgph.0000968.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Fit indices for confirmatory factor analyses models of the Mood and Feelings Questionnaire (MFQ-C) and Short Mood and Feelings Questionnaire (SMFQ-C).

https://doi.org/10.1371/journal.pgph.0000968.t002

Measurement invariance across sexes was assessed for the MFQ-C four factor and for the SMFQ-C unidimensional models. There was evidence of measurement invariance. The changes in χ² /DF across the configural, metric, and scalar models were non-significant and models had acceptable fit for males and females (see Table A in S1 File).

Reliability

Cronbach’s alphas and corrected-item total correlations are presented in Table 3. Internal consistency for the total MFQ-C score was excellent (α = 0.92) and the corrected item-total correlations were adequate (≥ 0.30) for all items except for: “I didn’t have any fun in school (r = 0.23) and “I slept a lot more than usual” (r = 0.24).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Reliability indices for items of the Mood and Feelings Questionnaire and the Short Mood and Feelings Questionnaire.

https://doi.org/10.1371/journal.pgph.0000968.t003

Regarding MFQ-C subscales, internal consistency was good for the ‘cognitive’ and the ‘vegetative’ subscales (respectively α = 0.87 and α = 0.81) and acceptable for the ‘core mood’ and ‘agitated distress’ subscales (respectively α = 0.62 and α = 0.66). Again, low item-total correlations were found for the items “I didn’t have any fun in school” (r = 0.17) on the ‘cognitive’ scale and “I slept a lot more than usual” on the ‘vegetative’ subscale (r = 0.24). Removal of these items only slightly improved the internal consistency of their corresponding subscales. In sensitivity analyses conducted among adolescents attending school, the item-total correlation for the item “I didn’t have any fun in school” remained modest on the unidimensional scale (r = 0.31) and the “vegetative” subscale (r = 0.28).

The internal consistency of the SMFQ-C was good (α = 0.87). Item-total correlations ranged from 0.36 (“I cried a lot”) to 0.67 (“I hated myself”) and Cronbach’s alpha when items were removed remain comparable (α-i between 0.85 and 0.87).

Convergent and divergent validity

Convergent validity was supported by moderate-to-strong positive correlations between MFQ-C (total and subscale) scores and scores on the SDQ-emotional symptoms and on the SCARED-C subscales (Table 4). Similar results were observed for the SMFQ-C total scores.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Spearman’s correlations between the Mood and Feelings Questionnaire scales and subscales and similar psychometric scales.

https://doi.org/10.1371/journal.pgph.0000968.t004

In terms of divergent validity, the MFQ-C scale, its core mood, vegetative, and agitated distress subscales and the SMFQ-C scale demonstrated a small but significant negative correlation with the SDQ-prosocial behaviour subscale (-.20 ≤ ρ ≤ -.10). However, no significant correlation was found between the SDQ-prosocial behaviour and the MFQ-C cognitive symptoms subscales. Surprisingly, the MFQ-C total and subscale scores, as well as the SMFQ-C scores, demonstrated low-to-moderate correlations with scores on the SDQ-hyperactivity and SDQ-conduct problems subscales (.08 ≤ ρ ≤ .39). The magnitude of the correlations varied greatly across the MFQ-C subscales, with a moderate correlation observed for core mood symptoms and low correlations observed for the vegetative, cognitive and agitated distress subscales.

Discussion

Our study is the first to examine the psychometric properties and factor structure of instruments assessing depressive symptoms in early adolescents from a community sample in rural Pakistan. Though Urdu is considered the national language, Pakistan is a multilingual country with each province linked to a specific ethnic group with its own language [56]. Sindhi is the first language of over 30 million people in Pakistan [24] yet there are no psychometric scales that are available in the language which may be problematic, particularly in rural communities where bilingualism is less common. Our results showed that the modified MFQ-C and the standard SMFQ-C translated to Sindhi have good psychometric properties in a primarily rural population of adolescents.

To our knowledge, our study is also the first to examine the multiple-factor structure of the child version of the MFQ. We found the MFQ-C unidimensional structure to be adequate but the four-factor structure comprising core mood, vegetative, cognitive and agitated distress symptoms best fit the data. Past research has examined the factor structure of the MFQ-C in HICs, with all supporting the original unidimensional solution, however, alternative structures have not been considered [19, 57]. The hypothesis of a five-factor solution was developed with data collected using the MFQ-P. It was first suggested by Jeffreys et al. (2016) and has since been supported in a study based on the Spanish version of the MFQ-P, in addition to a modified 33-item single-factor structure [28, 33]. Similar to our findings, both the unidimensional and the five-factor solution fit the data well, however, a slightly better fit was observed for the five-factor solution [28].

Four items of the original MFQ-C, assessing suicidal behaviours, were not included in the modified version of the tool used in the Nash-wo-Numa Study. The exclusion of these items reflects the cultural norm of youth in younger age groups being accompanied by a parent or chaperone when sensitive questionnaires are administered in Pakistan or in other countries where suicidal behaviour is criminalized. Administering these questions in community samples may result in unnecessary distress and may impact accurate disclosure [17]. However, the need for this adaptation is not limited to cultural considerations and can also be necessary when adequate follow-up cannot be provided to participants who endorse suicidal items. In a study conducted in a sample of youth from the United States, three suicidality related-items of the MFQ-C were eliminated for this reason [19]. Similar to our results, the unidimensional solution was still supported for the 30-item scale [19]. Though the exclusion of suicidality items can be justified in various contexts and cultures, their exclusion prevents cross-cultural comparisons from being made when total scores are computed as normative cut-off values cannot be applied. In this case, a four-factor solution is particularly useful when examining the MFQ factor structure as it allows for the exclusion of the four suicidality items contained on a single subscale, but does not prevent reasonable comparisons to be made across the remaining subscales [33]. As such, we recommend that future users of the scale examine the subscale scores as opposed to total scores from the 29-item MFQ.

Factor loadings and item-total correlations for the item “I didn’t have fun in school” in both the unidimensional and four-factor structures of our modified MFQ-C were lower than those observed in HICs [28, 33]. Similar findings were noted in the subgroup of participants attending school. For comparability, we retained the item in our analyses but future research regarding the usefulness of inquiring about school-related symptoms when assessing depression in LMICs is warranted.

The reliability of the full-scale MFQ-C in this sample was similar to previously reported estimates from other LMICs (α = 0.90–0.92) [27, 32]. Our results were also comparable to that of studies examining the MFQ-P, with the exception of the ‘core mood’ subscale in which the internal consistency (α = 0.62) was lower than previously observed (α = 0.78) [33]. This discrepancy may be attributed to the number of items included in the ‘core mood’ subscale, which differs between the MFQ-P (five items) and the MFQ-C (four items).

Our results provide strong support for the unidimensional structure of the SMFQ-C and its reliability and validity in a sample of youth in Pakistan. The internal consistency and unidimensional factor structure of the SMFQ-C was in accordance with results obtained in past studies conducted in LMICs [41, 58]. These findings provide strong support for the use of this brief measure in community-based and/or low-resource settings in which long-form questionnaires are not feasible to administer. However, the mean SMFQ-C scores were lower in our sample compared to what was observed in a similar sample of adolescents in Bangladesh (M = 8.67, SD = 5.61) [41]. This discrepancy may reflect a true difference in the severity of depressive symptoms in adolescents living in rural Pakistan compared to those living in Bangladesh. Alternatively, the variation could be attributed to the younger age range and the use of random sampling in the present study, or because participants were not interviewed alone. Both depressive symptomatology and risky behaviours have been negatively associated with disclosure when an adult is present, thus adolescents may have underreported depressive symptoms in our study [59].

The convergent validity of our modified MFQ-C (full scale and subscales) and the SMFQ-C was supported. The observed positive correlations with scores on the SDQ-emotional symptoms subscale are consistent with past research examining the MFQ-C and MFQ-P [27, 28]. Similarly, correlations between the SCARED-C subscales and the SMFQ-C and MFQ-C total score and subscale scores were consistent with past research examining the MFQ-P [33] and are in agreement with the body of literature emphasizing the high comorbidity between depression and anxiety [60]. The SDQ-emotional symptoms and SCARED-C do not measure the exact same constructs as the MFQ-C and SMFQ-C, which explains the moderate-to-strong correlations observed. However, these correlations reflect the expected associations between depressive symptoms emotional problems and anxiety symptoms, suggesting that the MFQ-C total and subscale scores and SMFQ-C scores effectively capture the characteristics shared between these constructs.

Regarding divergent validity, surprisingly significant correlations were found between scores on the MFQ-C (full scale and subscales), the SMFQ-C, and scores on the SDQ-conduct problems and SDQ-hyperactivity subscales. However, the associations observed between depressive symptoms and conduct problems are in agreement with previous results obtained in community samples from HICs and could be explained by the overlap of symptoms of both conditions, by their mutual influence on one another, or by shared underlying etiological factors such as socioeconomic factors [61]. In LMICs, this association may be even stronger as lower socioeconomic status has been associated with both higher levels of depression [62] and conduct problems [63]. The significant correlations observed between hyperactivity and depressive symptoms has also been noted in past research [27]. This association could be attributed to the similarity in items between the two scales. The SDQ-hyperactivity subscale enquires about problems such as restlessness, fidgeting, being easily distracted, and finishing tasks [43] much of which is reflected in the MFQ-C. However, divergent validity was supported through low negative correlations between all MFQ-C scales and SDQ- prosocial behaviour, analogously to results of past research examining the MFQ-P [28].

The main strength of our work lies in the use of a representative sample and of standardized methods that limit selection and information biases and allow us to generalize our results to the population of adolescents living in Matiari. Nevertheless, our findings should be considered in light of several limitations. Firstly, because our modified version of the MFQ-C excluded four suicidal items, the total scores were computed by summing the score of 29 items, rather than 33. As such the mean total MFQ-C score obtained in our sample cannot be compared to scores obtained in other studies. Secondly, in accordance with Jeffreys et al. (2016)’s analysis, our four-factor solution did not include the following three items: “I ate more than usual”, “I worried about aches and pains” and “I did not want to see friends”. Modest item-total correlations were found for “I ate more than usual” and “I worried about aches and pains” when assessing the unidimensional structure of our MFQ-C (0.30 and 0.32, respectively), with similar findings also noted in a Danish school-based study [33, 38]. Further, in a study conducted in Thailand, the item-total correlation for the item “I ate more than usual” was also low (r = 0.20) [32]. This item may not be as relevant in low socioeconomic status populations as it is in HICs when assessing depressive symptoms, particularly when the prevalence of food insecurity is high. In the rural districts of the province of Sindh, food insecurity is estimated to affect 52.4% of adolescents [64]. However, we cannot exclude the possibility of these three items being clinically relevant in this population. Thirdly, due to the cross-sectional design of the study, we were unable to assess the test-retest reliability of the MFQ-C and SMFQ-C. Fourthly, since a mostly rural community sample was examined, it is unclear if these findings are generalizable to clinical or urban populations in Pakistan. Lastly, as no formal diagnostic interviews were conducted, the discriminant validity of the modified Sindhi MFQ-C and SMFQ-C could not be analyzed and meaningful cut-off scores could not be determined in our sample. Consequently, we recommend future users of these scales to consider continuous scores, in addition to dichotomized measures of depression.

Conclusion

Overall, both the Sindhi MFQ-C and SMFQ-C appear to be reliable measures of depressive symptoms in early adolescents living in rural Pakistan. While the unidimensional structures of the MFQ-C and SMFQ-C provide a global measure of depressive symptom severity, the symptom-specific subscale scores of the MFQ-C allow for the assessment of specific dimensions of depressive symptoms. The brevity of the SMFQ-C further emphasizes its usefulness in low-resource settings, where limited time and/or trained staff prevent long-form questionnaires from being administered. Our findings also support the use of these instruments in future population-based studies, which may contribute to the surveillance and the identification of covariates of adolescent depressive symptoms in rural Pakistan. Research regarding reasonable cut-off scores that indicate probable depression in this population would provide further insight into its usefulness as a diagnostic or screening tool. More generally, the MFQ questionnaires may also be used in other LMIC populations that share similar characteristics and cultures.

Supporting information

Acknowledgments

We appreciate the ongoing assistance and support provided Fariha Shaheen, Imran Ahmed, Arjumand Rizvi and the entire team at the Aga Khan University. We express gratitude for the participating youth of Matiari and their families to advance adolescent research in rural Pakistan.