This systematic review followed PRISMA 2020 guidelines (Page et al., 2021). The review was not registered, and no protocol was prepared.

Studies were included based on the following criteria: (1) community-dwelling older adults (mean age ≥60 years); (2) administration of GDS (any version) or psychosocial intervention; (3) gold standard clinical interview for diagnostic accuracy studies or control condition for intervention studies; (4) psychometric properties or depressive symptom outcomes; (5) peer-reviewed journal articles or dissertations; (6) English, Portuguese, or Spanish; (7) database inception to April 2026.

Exclusion criteria: institutional settings; younger adults; qualitative studies; pharmacological interventions only; studies without original data; non-validated outcome measures.

A comprehensive and systematic literature search was conducted on April 1, 2026, across eight electronic databases: PubMed (MEDLINE), PsycINFO (EBSCOhost), SciELO, LILACS, Redalyc, Dialnet, Web of Science, and ProQuest Dissertations & Theses Global. No publication date restrictions were applied to any of the searches.

The search strategies were developed by combining controlled vocabulary terms (e.g., Medical Subject Headings [MeSH] in PubMed and LILACS, and Thesaurus terms in PsycINFO) with free-text keywords using the Boolean operators AND and OR. The search strategy encompassed three principal conceptual domains.

Two independent reviewers screened all records using Rayyan software (Ouzzani et al., 2016). Inter-rater agreement was excellent (title/abstract: κ = 0.87; full-text: κ = 0.92).

A standardized data extraction form was developed in Microsoft Excel. Two reviewers independently extracted data, with discrepancies resolved by consensus. Extracted variables included study characteristics, sample characteristics, GDS version or intervention type, psychometric properties or effect size data, and risk of bias information.

Part A (psychometric studies): COSMIN Risk of Bias checklist (Mokkink et al., 2018). Part B (intervention studies): Cochrane Risk of Bias 2 (RoB2) tool (Sterne et al., 2019). Two reviewers independently assessed risk of bias (COSMIN: κ = 0.84; RoB2: κ = 0.81).

Justification of Methodological Assessment Tools: PRISMA was applied as the overarching reporting guideline. COSMIN was selected for psychometric studies as it provides domain-specific ratings for each psychometric property. RoB2 was selected for RCTs as it addresses key sources of bias relevant to psychosocial interventions (randomization, deviations from interventions, missing outcome data). For included meta-analyses, AMSTAR-2 (Shea et al., 2017) was used to assess methodological quality (results in Appendix B).

A narrative synthesis was conducted following SWiM guidelines (Campbell et al., 2020). For intervention studies, Hedges' g was calculated as the standardized mean difference between intervention and control groups at post-treatment. Random-effects models using the DerSimonian-Laird estimator were employed for pooled effect sizes (DerSimonian & Laird, 1986), conducted in R version 4.2.1 (R Core Team, 2022) with the meta package (Schwarzer et al., 2026). Heterogeneity was assessed using I ^² (Higgins et al., 2003). Sensitivity analyses excluded high-risk studies and those with >20% attrition, a threshold commonly used to identify studies with concerning attrition levels. Publication bias was assessed via funnel plots, Egger's test (Egger et al., 1997), and trim-and-fill procedures (Duval & Tweedie, 2000a, 2000b).

A hierarchical approach was adopted to avoid duplication when both meta-analyses and primary studies were included: meta-analyses provided pooled estimates, while primary studies provided unique data not captured in meta-analyses. Specifically, Cuijpers et al. (2026), Gould et al. (2012), and Mitchell et al. (2010) were included as meta-analyses, with their constituent studies included only when providing additional moderator or subgroup data.

The systematic search identified 1,847 records across seven databases. After removing 438 duplicates, 1,409 records proceeded to screening. Of these, 1,281 were excluded, leaving 128 full-text articles assessed for eligibility. Of these, 100 were excluded, yielding a final sample of 28 studies (12 psychometric, 16 intervention).

Figure 1. Figure 1. PRISMA 2020 Flow Diagram

Figure 1 presents the PRISMA 2020 flow diagram. After removing 438 duplicates, 1,409 records were screened. Following title/abstract screening (1,281 excluded), 128 full-text articles were assessed, of which 100 were excluded for reasons listed in the figure. The final sample included 28 studies (12 psychometric, 16 intervention).

The 28 included studies spanned 1983 to 2026. Psychometric studies included 5 diagnostic accuracy studies, 4 validation studies, 2 factor analysis studies, and 1 meta-analysis. Intervention studies included 9 individual RCTs and 7 meta-analyses. The most common GDS version was GDS-15 (10/12 psychometric studies). CBT was the most frequently tested intervention (6/16 intervention studies). Of the 28 included studies, only 3 (10.7 %) came from Latin America, all from Brazil.

This systematic review found that the GDS-15 demonstrates adequate psychometric properties in populations where it has been studied, with internal consistency (α = 0.79 – 0.91) and diagnostic accuracy (sensitivity 0.80 – 0.92; specificity 0.75 – 0.88) meeting conventional thresholds. Psychosocial interventions, particularly Behavioral Activation (g = 0.62) and CBT (g = 0.56), produced clinically meaningful reductions in depressive symptoms (pooled g = 0.48). However, these conclusions are geographically limited, as the evidence base for the GDS excludes all Spanish-speaking Latin American countries, and no psychosocial intervention RCT has been conducted with older adults in any Latin American country.

This review's most striking finding is the near-complete absence of Latin American research. Of 28 included studies, only 3 (10.7 %) came from Latin America, all from Brazil. No studies from Spanish-speaking Latin American countries met inclusion criteria. This gap is consequential for several reasons. First, without local validation studies, clinicians lack evidence-based cutoff scores. Using cutoffs derived from European or North American samples may produce high false positive rates. Second, without intervention RCTs, clinicians have no evidence on whether psychosocial interventions are effective in Latin American settings. Third, this absence represents a failure of research equity.

The variability in optimal GDS-15 cutoffs across settings (≥5 to ≥7) raises three conceptual questions that extend beyond methodology.

First, does the GDS measure the same latent construct across populations? The factor analytic evidence is mixed: four studies support a one-factor solution, while three support a three-factor solution (dysphoric mood, withdrawal/apathy, cognitive concerns). The three-factor solution may better capture how older adults—particularly those with mild cognitive impairment—experience depression as a multidimensional phenomenon rather than a unitary mood disorder. For Latin America, where cognitive screening is not routine in primary care, a three-factor interpretation could lead to different clinical decisions depending on which domain is elevated.

Second, why does the optimal cutoff increase from Europe (≥5) to Brazil (≥6) to India (≥7)? Three non-mutually exclusive explanations exist: (a) true prevalence differences (higher prevalence requires higher cutoffs to maintain specificity), (b) differential item functioning (certain items are endorsed more or less frequently due to cultural norms about emotional expression), or (c) reference group effects (the gold standard clinical interview itself may have variable cross-cultural validity). The current literature cannot adjudicate among these explanations because no study has simultaneously tested measurement invariance and diagnostic accuracy in a Latin American sample.

The variability in optimal GDS-15 cutoffs across settings may reflect meaningful differences in how late-life depression is experienced across cultural contexts. However, alternative explanations—including differences in study methodology or sampling strategies—cannot be ruled out. The available evidence is insufficient to confirm cultural differences as the primary explanation.

Third, what does the exclusion of somatic symptoms imply for Latin American populations? The GDS deliberately omits somatic items (fatigue, appetite change, sleep disturbance) to avoid confounding depression with physical illness or normal aging. However, in many Latin American cultural contexts, older adults may express psychological distress primarily through somatic complaints—a phenomenon documented in Mexican, Colombian, and Central American populations (Arrieta et al., 2017). If somatic presentations are common, the GDS may systematically under-identify depression in Spanish-speaking Latin America relative to populations where cognitive-affective symptoms are more readily endorsed.

Our findings are broadly consistent with prior reviews: similar pooled sensitivity (0.82 vs. 0.80 in Mitchell et al., 2010), specificity (0.81 vs. 0.81), and effect sizes for CBT (g = 0.56 vs. g = 0.52 in Cuijpers et al., 2026). However, our review extends previous work by explicitly examining the intersection of measurement and intervention literatures, revealing that while each literature is robust in high-income countries, the combined evidence base for Latin America is essentially absent.

We also identified a graded pattern of optimal cutoffs (Europe ≥5, Brazil ≥6, India ≥7) that diverges from Pocklington et al. (2016), who concluded a single cutoff (≥5) is appropriate across populations. This discrepancy may reflect differences in inclusion criteria (primary care vs. community samples) or may indicate genuine cross-cultural differences in depression presentation. Additionally, our finding that Behavioral Activation (g = 0.62) showed the largest effect among intervention types extends the work of Cuijpers et al. (2026), who did not separately analyze BA from CBT interventions. This suggests that behavioral mechanisms may warrant greater attention in future intervention development.

The finding that effect sizes varied by control condition type suggests that approximately half of the observed treatment effect may be attributable to non-specific factors (therapist attention, expectation of improvement, structured routine) rather than intervention-specific mechanisms. This pattern has implications for intervention design in Latin America, where culturally adapted interventions maximizing therapeutic alliance and credibility may be as important as specific techniques.

Behavioral Activation (g = 0.62) outperformed Reminiscence (g = 0.38), suggesting behavioral mechanisms may be more central to LLD than cognitive or autobiographical mechanisms. For Latin America, culturally adapted interventions should prioritize BA with consideration of collectivist values (e.g., family-centered activities rather than individual walks).

In the absence of local validation studies, clinicians in Spanish-speaking Latin America may consider adopting a provisional cutoff of ≥6, while recognizing this recommendation is based on Brazilian data and international literature, and that local validation is needed to confirm its appropriateness.

Clinicians should: (1) use GDS-15 with provisional cutoff ≥6, (2) supplement positive screens with clinical interview, (3) document local performance, and (4) consider PHQ-9 as alternative (Arrieta et al., 2017; Diez-Quevedo et al., 2001).

Immediate priority (0-2 years): GDS-15 validation studies in Mexico, Colombia, and Argentina. Short-term (2-4 years): Measurement invariance testing across Latin American countries. Medium-term (4-6 years): Culturally adapted psychosocial intervention trials.

Methodological limitations include: only three Latin American studies identified (all Brazil), substantial heterogeneity (I^² = 72 %), and 69 % of intervention studies with high risk/some concerns. Scope limitations: community-dwelling older adults only; findings may not generalize to institutional settings. The review was not prospectively registered, and no protocol was prepared; therefore, adherence to prespecified methods and the possibility of selective reporting could not be independently assessed.