The study employs a critical narrative review method, selected because it is appropriate for synthesizing heterogeneous literature and for offering a conceptual critique of intricate phenomena, according to Greenhalgh et al. [¹²]. Our synthesis includes studies using both the original BDI (BDI-I) and the revised BDI-II. While both versions assess depressive symptomatology, they differ substantively: BDI-I reflects symptoms over the past week and includes items on weight loss, somatic preoccupation, and work difficulty, whereas BDI-II assesses symptoms over the past two weeks and replaces several items (e.g., adding “changes in sleeping pattern” and “concentration difficulty”) while removing others. These structural differences may independently influence factor structure, reliability, and optimal cut-offs, regardless of cultural context.

A structured search was conducted in PubMed, PsycINFO, and Scopus covering January 2000 to January 2025. The search strategy, as elaborated in Supplementary File 1, employed: ("Beck Depression Inventory" OR "BDI") AND (adolescent* OR youth OR teenage*) AND (valid* OR reliab* OR psychometric*) AND (LMIC OR "low and middle income" OR "developing countr*" OR "global health"). Reference lists of assessed papers were manually screened to identify additional relevant studies. Inclusion criteria covered empirical studies that demonstrated the validation of the BDI-I or BDI-II in adolescent populations (ages 10–19) living in low- and middle-income countries (LMICs) and that reported at least one psychometric property (e.g., reliability, validity, factor structure). The status of lower- and middle-income countries was defined according to the World Bank Group’s 2023 country income classification. Exclusion criteria were studies that were completely adult-only, non-English, or non-translated papers without an English abstract, and theoretical discussions.

Critical narrative reviews are well-suited to interrogating conceptual assumptions and synthesizing heterogeneous evidence through purposive selection of illustrative cases. A narrative synthesis was performed through thematic analysis organized around cultural, methodological, and developmental challenges. From the 54 full-text studies assessed for relevance, we purposively selected 8 case exemplars based on a pre-specified framework emphasizing: (1) geographic diversity across LMIC regions, (2) variation in cultural adaptation procedures, and (3) depth of psychometric reporting (particularly factor structure or measurement invariance). These exemplars were chosen to maximize insight into cross-cultural validity challenges rather than statistical generalizability. The remaining 46 studies informed contextual understanding but were excluded from detailed synthesis due to insufficient psychometric reporting, methodological limitations, or sample ineligibility, as elaborated in Supplementary File 2. Additionally, we applied a custom quality rubric assessing four criteria: (1) sample representativeness, (2) use of cognitive interviewing or local input in adaptation (beyond back-translation), (3) reliability reporting (e.g., Cronbach’s α), and (4) critical discussion of validity (e.g., item equivalence or cut-off appropriateness). Each criterion scored 1 (met) or 0 (not met); studies scoring ≥2 were selected as exemplars. Two reviewers appraised independently; disagreements were resolved by consensus. Invariance evidence was defined as explicit reporting of multi-group confirmatory factor analysis (MGCFA) for configural, metric, or scalar invariance; claims without statistical testing were not counted. Following Cohen’s general effect size guidelines (often applied in psychological assessment), convergent validity coefficients ≥0.50 are interpreted as strong, 0.30–0.49 as moderate, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$<0.30$\end{document}<0.30 as weak [¹³].

This narrative review summarized the results of eight representative validation studies of the Beck Depression Inventory (BDI-I and BDI-II) conducted in adolescent populations in low- and middle-income countries (LMICs), as shown in . Our exemplar studies span three distinct population contexts: (1) general adolescent populations (school-based samples in Nigeria, India, Brazil, Chile, Iran, and Turkey), (2) clinical or medically vulnerable groups (HIV-affected youth in Malawi [²²]), and (3) institutionally at-risk cohorts (orphaned adolescents in Nepal [¹⁸]). These settings differ markedly in baseline distress levels, symptom expression, and comorbidity profiles. The studies included 181-1,412 adolescent participants, and in most cases, the adolescents ranged in age from 13 to 18 years. Various forms of BDI were also applied, including the original BDI, BDI-II in several translated versions, and culturally adapted versions. Translation-back-translation was a common protocol across all studies; however, the extent and quality of the cultural adaptation process varied significantly from one study to another.

Empirical evidence from exemplar studies demonstrates substantial variability in optimal cut-offs; several criterion-validation studies that underwent structured diagnostic interviews showed a wide range of optimal cut-offs. Adewuya et al., in their study in Nigeria, identified a cut-off of ≥18 as optimal for screening, yielding excellent sensitivity (0.91) and specificity (0.97) against K-SADS/E diagnoses. [¹⁴]. In contrast, the Indian Tamil validation by Basker et al. distinguished between a screening cut-off of ≥5 and a diagnostic cut-off of ≥22, suggesting that different thresholds may be appropriate depending on assessment purpose and clinical context [¹⁵].

This variability in cut-off results seriously challenges clinical practice and epidemiological research. Firstly, when researchers from various countries decide on different cut-offs to apply, or in case they uncritically apply a universal Western-derived threshold, cross-national prevalence estimates are not comparable. Secondly, the Chilean study’s finding of sex-specific optimal cut-offs [¹⁶] Makes it even more difficult to deal with. If optimal thresholds vary not only across countries but also within each country by gender, this would suggest that population-level norms may be insufficient and that gender-specific reference data are needed.

Several studies noted developmental considerations that warrant attention. Toosi et al. even went so far as to suggest cognitive interviewing for younger adolescents, recognizing that early adolescents might interpret the abstract or metaphorical meanings of the items differently than older adolescents. The Basker et al. study, conducted in primary care pediatric settings, emphasized the importance of pediatrician-administered support, suggesting that adolescents may need more structured support than is typically provided in self-administration.

These observations align with developmental neuroscience research documenting that adolescence involves ongoing maturation of prefrontal cortex regions responsible for abstract reasoning, emotional regulation, and metacognitive awareness [³¹]. Younger adolescents who are still mainly dependent on concrete operational thinking may interpret an item such as "I feel I am being punished" literally rather than metaphorically [³²]. Similarly, items assessing self-evaluation and future orientation presuppose cognitive capacities for abstract self-reflection and temporal projection that continue to develop through adolescence [³³]. Administration mode varied across exemplar studies and may contribute to response heterogeneity. Adewuya et al. [¹⁴] and most Brazilian/Chilean studies used standard self-administration. In contrast, Basker et al. [¹⁵] administered the BDI with pediatrician support in a clinical setting, while Kim et al. [²²] and Bhatt et al. [¹⁸] used trained interviewers for orphaned or HIV-affected youth with potential literacy or trauma-related barriers. Clinician- or interviewer-assisted administration may reduce missing data and increase disclosure of sensitive items (e.g., suicidality, guilt), whereas self-report in school settings may heighten social desirability or literal interpretation—particularly in low-literacy contexts.

Considering that literacy and educational heterogeneity may further threaten construct validity in LMIC settings. The BDI assumes a minimum 5th–6th grade reading level in its original English form; translated versions rarely undergo readability validation. In contexts where adolescent literacy is variable—due to school dropout, limited educational access, or language mismatch—item comprehension may be compromised, biasing responses toward literal interpretation, acquiescence, or non-response. This risk is especially acute for abstract or metaphorical items (e.g., “I feel I am being punished”), which demand higher-order cognitive and linguistic processing [³].

Measurement Invariance and the Fallacy of Universal Cut-offs Perhaps the most critical methodological gap across the reviewed studies is the near-complete absence of measurement invariance testing, without which no valid cross-cultural score comparisons can be made. Among the eight exemplar studies, only the Chilean study by Araya et al. and the Brazilian study by Ferreira-Maia et al. explicitly referenced invariance testing—and even these established it only partially or noted gender-specific differences, which made interpretation difficult [¹⁶,²¹]. The finding of Araya et al. that optimal cut-offs differ by sex is a direct indication that scalar invariance—the state where item intercepts are equal across groups—does not hold, because equal cut-offs would only be valid under scalar invariance.

This gap has profound psychometric consequences, as measurement invariance testing proceeds hierarchically through configural, metric, and scalar levels [³⁴]. Configural invariance requires equivalent factor structure; metric invariance, equivalent factor loadings; and scalar invariance, equivalent item intercepts. Only after establishing scalar invariance can latent mean differences be interpreted as reflecting true differences in the underlying construct rather than measurement artifacts [²⁸].

The lack of invariance testing represents a fundamental psychometric gap. A score of 18 on the BDI in a Nigerian, Chilean, Indian, or Iranian adolescent sample cannot be assumed to reflect equivalent depression severity if item intercepts differ due to cultural response styles, differential stigma, linguistic nuance, or divergent symptom conceptualization [³⁵]. The use of Western-derived cut-offs in these circumstances conflates true disorder prevalence with measurement bias—a problem that is raised in Indian school prevalence studies [²⁷].

Empirically, the failure to establish scalar invariance is mirrored in the divergent factor structures reported across validation studies [¹⁵,²¹,¹⁹], sex-specific optimal thresholds [¹⁶], and context-dependent somatic item interpretations [¹⁸,²²]. The findings, in fact, challenge the whole idea that BDI cut-off scores have the same diagnostic value across different cultural-linguistic settings. When scalar invariance is not achieved, the symptoms of two adolescents having the same level of depression -for example, one in Chile and the other in Nigeria- may reflect differently on their respective test scores because of culture-influenced ways of responding, stigma, or the translation meaning [³⁶].

Epidemiological reports based on Western cut-offs in LMIC populations will likely yield an incorrect estimate of prevalence [²⁴], while clinical screening or monitoring based on unvalidated thresholds risks systematic misclassification. Future validation research must therefore prioritize multi-group confirmatory factor analysis to sequentially test configural, metric, and scalar invariance across gender, age, and cultural subgroups [³⁷,³⁴]. When invariance fails, researchers are required to identify items that function differentially, modify or discard them, or create context-specific norms indicative of culturally grounded symptom expression rather than the application of external psychometric universals. Cultural Psychology and the Limits of Beck’s Cognitive Model The poor and inconsistent performance of cognitive-affective items across multiple validation studies [¹⁴,¹⁶,¹⁹] reveals fundamental tension between the BDI’s theoretical foundation and non-Western constructions of selfhood. This reflects deeper ontological differences in selfhood; as Beck’s cognitive model presupposes an internalized, individualized self, in which negative automatic thoughts about personal worth, competence, and the future constitute core depressive pathology [³⁸]. Yet, in collectivist cultures existing in numerous LMICs, the concept of self is more interdependent and is mainly determined by one’s social roles, family relationships, and community status rather than personal characteristics [³⁹].

This ontological difference has profound implications for symptom expression. Chentsova-Dutton and Tsai demonstrate that cultural values fundamentally shape not merely how depression is described but how it is phenomenologically experienced [⁴⁰]. In individualist cultures, depressed individuals emphasize personal failure, worthlessness, and internal negative self-evaluation—precisely the cognitive content the BDI was designed to assess. On the other hand, in collectivist cultures, individuals may feel and communicate their distress primarily through physical symptoms, disruptions in interpersonal relations, or spiritual aspects rather than self-criticism [⁴¹,⁴²].

The sex differences in emotional expressivity items reported by Adewuya et al., and the sex-specific cut-offs identified by Araya et al., likely reflect culturally prescribed emotional display rules that interact with gender socialization [¹⁴,¹⁶]. In many cultures, masculine ideals emphasize emotional stoicism, making male adolescents less likely to acknowledge symptoms related to crying, sadness display, or emotional vulnerability even when experiencing significant depressive distress [⁴³].

These patterns necessitate a fundamental shift from "imposed etics"—treating Western psychological constructs as universal standards—toward combined "emic-etic" approaches that integrate universal dimensions of negative affect with locally salient idioms of distress [⁴⁴,⁴⁵]. This might involve retaining certain BDI items that assess transculturally relevant symptoms, while adding or changing those culturally problematic cognitive items with the local expressions of distress that make sense in the particular culture [⁴⁶].

Ethical Imperatives and Guidance Beyond Psychometric Adequacy Beyond psychometric limitations, ethical concerns require explicit addressing. Item 21 ("Loss of interest in sex") has been repeatedly identified as developmentally inappropriate and culturally problematic for adolescents, particularly in contexts where discussing sexuality is taboo, where many adolescents lack sexual experience, or where sexual activity before marriage is religiously proscribed [⁴⁷,⁴⁸,⁴⁹]. This item routinely produces missing data, systematic under-reporting, and discomfort among respondents [⁵⁰,⁵¹,⁵²].

Similarly, Item 9 evaluating suicidal ideation raises ethical challenges in contexts where suicidality is stigmatized from a moral point of view or where talking about such thoughts may result in negative consequences. This demands ethically grounded instrument design: ethical principles for research require that instruments be not only psychometrically robust but also culturally sensitive and that they take into account participants’ developmental stage and the social/environmental context [⁵³]. To address these concerns, we recommend: (1) omit or reword Item 21 in developmentally or culturally inappropriate contexts; (2) administer Item 9 only with a safety protocol (trained staff, debriefing, and referral pathways); (3) pilot-test all items with local adolescents to assess acceptability and comprehension prior to use; and (4) tell participants they may skip any item—no coercion, no full-score pressure.

Advancing cross-cultural adolescent mental health research requires methodological rigor balanced with contextual feasibility and ethical sensitivity. We propose a three-component framework:

First, adopt an invariance-first approach as standard practice. Before any cross-group comparison, a multi-group confirmatory factor analysis should test configural, metric, and scalar invariance across gender, age, and cultural groups [²⁸,³⁴]. If scalar invariance fails, the researchers are advised to highlight the items that display differential functioning and then decide whether to modify, eliminate, or create group-specific scoring algorithms.

Second, integrate mixed-methods approaches. Quantitative validation must be complemented by qualitative inquiry, particularly cognitive interviews and focus groups with adolescents from target populations [⁴⁶]. Cognitive interviewing is an approach that can reveal semantic misunderstandings, identify culturally inappropriate content, and even detect developmentally challenging phrasing, all of which can be done before large-scale validation.

Third, incorporate criterion validity assessment. Even when gold-standard diagnostic interviews (K-SADS, MINI) are resource-intensive, conducting them with purposefully selected subsamples can establish locally appropriate cut-off scores. The Nigerian study’s approach offers a feasible example: carrying out diagnostic interviews with extreme scorers to figure out the best sensitivity-specificity trade-offs [¹⁴].

Fourth, embrace decolonizing co-creative frameworks. Rather than merely adapting Western instruments, future research should involve local researchers and adolescents in the generation of items and pretesting [⁵⁴]. Co-creative methods can initially involve questions that are directed towards adolescents and local mental health stakeholders, asking them how they identify, narrate, and comprehend depression in their context.

Fifth, as mobile health (mHealth) screening expands in LMICs [⁵⁵], future validations must assess whether BDI’s psychometric properties—established predominantly via paper-pencil—are preserved in digital formats, where interface design, literacy demands, and privacy perceptions may alter response behavior.

This narrative review has several important limitations. First, our selective synthesis of eight exemplar studies, while appropriate for narrative review methodology, as Green et al. discussed [⁵⁶], necessarily excludes potentially relevant validation research and cannot provide precise quantitative effect-size estimates that meta-analysis would afford. Second, the restriction to English-language publications constitutes a major limitation, likely biasing regional representation and underrepresenting culturally grounded validation work from non-Anglophone LMICs. Third, publication bias likely affects the literature we reviewed, as exemplar-based synthesis limits generalizability. Fourth, our synthesis did not stratify findings by urban/rural setting, socioeconomic status, or time of data collection—factors that may substantially influence symptom expression and instrument validity in rapidly changing LMIC contexts, and did not evaluate alternative instruments (e.g., PHQ-A, CES-D, or locally developed tools) that may offer superior cross-cultural validity or feasibility for adolescents in LMICs.. Additionally, although two reviewers independently appraised study quality using our rubric and resolved disagreements by consensus, the initial exclusion of 46 full-text studies from in-depth synthesis was not independently verified. To enhance auditability, we now provide a supplementary table listing all excluded studies with their rubric scores and exclusion reasons. These version-specific differences (between BDI-I and BDI-II) may also independently influence psychometric properties, potentially confounding cultural and instrument-level variance. Future systematic reviews with meta-analytic synthesis offer more detailed quantitative summaries by utilizing tools such as COSMIN [⁵³].