Decolonizing mental health: how native languages reshape depression network architecture in 31,276 Peruvians

Abstract

To date, no studies have investigated the effect of native language on depressive symptom networks. To address this gap, we examined depression symptom network structure across four major cultural-linguistic groups in Peru (Castellano, Quechua, Aymara, and Amazonian indigenous languages). We conducted a network analysis of PHQ-9 depression symptoms using nationally representative data from 31,276 Peruvian participants. Networks were estimated using ggmModSelect with Spearman correlations. Native language groups were compared using permutation tests to evaluate network differences. Depression networks exhibited distinct architectures across groups, with Castellano showing the highest connectivity (22 active edges), followed by Quechua (20 edges) and Aymara (12 edges, p < 0.019). While worthlessness/guilt-suicidal ideation was the strongest universal pathway (r = 0.377–0.44), groups exhibited distinct centrality patterns: Castellano networks centered on depressed mood, Quechua on fatigue, and Aymara on suicidal ideation. Aymara networks showed absence of anhedonia-depressed mood connections (r = 0.00) and unique psychomotor-suicidal ideation pathways (r = 0.27), which requires further exploration. It is suggested that native language may shape depression architecture, suggesting distinct cultural-linguistic patterns that challenge current depression models.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-32838-6.

Introduction

Depression affects over 280 million people worldwide, representing one of the most pressing global health challenges of our time^1–34. However, this staggering figure masks a critical complexity: the profound influence of cultural context on how depressive symptoms manifest and interconnect. Western diagnostic frameworks dominate our understanding of depression⁵, yet mounting evidence reveals that cultural worldviews and linguistic structures fundamentally shape both the experience and expression of depressive distress⁶. This Western-centric approach becomes particularly problematic in pluricultural societies, where indigenous conceptualizations of emotional suffering diverge markedly from standardized diagnostic assumptions, potentially leading to misdiagnosis and culturally inappropriate interventions^7–9.

Existing research demonstrates significant cultural variations in depression manifestation across global populations^10,11. While Western conceptualizations emphasize depressed mood as the cardinal feature, South Asian populations predominantly report somatic symptoms, and South American communities frequently present with fatigue as the primary complaint¹². Even within single countries, cultural differences shape symptom expression—indigenous patients in Ecuador, for instance, exhibit more somatic presentations than their mestizo counterparts^13,14. While cultural variations are well-documented, one fundamental question remains unanswered - What is the primary mechanism through which culture shapes phenomenology?. Among all cultural variables, current evidence may pose native language as the primary driver on how individuals conceptualize and experience their world, making it a crucial factor in understanding symptom expression and connectivity^15,16. For example, indigenous languages often lack equivalents of depression, instead emotional suffering is conceptualized as Llaki in Quechua, or as Pena in Spanish¹⁷. These linguistic differences reflect distinct cultural ontologies of mental health that may fundamentally alter how symptoms cluster, manifest, and interconnect within depressive networks.

However, despite these valuable insights into cultural symptom variation, current research approaches fall short of contemporary psychopathological frameworks that emphasize dynamic symptom networks rather than static symptom lists. Network analysis has emerged as a revolutionary paradigm in psychopathology, revealing that mental symptoms interconnect and influence each other, with certain symptoms serving as central hubs that influence entire networks¹⁸. Studies show that guilt dominates networks in traumatic brain injury populations¹⁹,, while sadness proves most central in adolescents²⁰, and suicidal ideation takes precedence in late-life depression²¹.

Furthermore, no studies have examined how native language—arguably the most fundamental cultural determinant—influences depressive symptom networks¹⁰. Peru presents an exceptional opportunity to address this critical gap. With over 47 indigenous languages²² across 25 geographical regions and a population exceeding 30 million, Peru represents remarkable linguistic diversity within a single national context²³. The country’s annual nationwide representative survey captures both demographic data and depression symptoms using the PHQ-9, creating an unprecedented dataset for exploring cultural-linguistic influences on symptom networks^24–26. Moreover, several disparities have been found across prevalence of depression inside the country^27,28, which raises concerns regarding how the depression construct may behave in distinct groups.

Hence, we decided to conduct this study to examine the network structure of depressive symptoms as measured by the PHQ-9 across four major cultural-linguistic groups in Peru (Aymara, Quechua, Amazonian indigenous languages, and Castellano), identifying culture-specific patterns of symptom centrality and connectivity. In this study, we use native language as a proxy measure for cultural worldview, recognizing that language represents one of the most accessible yet imperfect markers of cultural identity. While this approach simplifies the complex reality of cultural identity in Peru, where bilingualism is widespread and many individuals navigate hybrid cultural frameworks that blend indigenous and mestizo elements, the employment of language remains as a meaningful indicator of primary cultural socialization and cognitive-linguistic frameworks for understanding emotional experience. Furthermore, understanding these culture-specific patterns will advance theoretical knowledge of how cultural context shapes depression’s internal architecture while informing the development of culturally-responsive assessment tools and interventions for Peru’s underserved indigenous populations.

Methods

Research design

A cross-sectional analytical study was conducted using secondary data from the 2023 Peruvian Demographic and Family Health Survey (Encuesta Demográfica y de Salud Familiar, ENDES). This study followed the STROBE guidelines (Strengthening the Reporting of Observational Studies in Epidemiology) for reporting observational studies, as well as the reporting checklist for psychological network studies^29,30.

The ENDES is a nationally representative survey conducted annually by the National Institute of Statistics and Informatics (Instituto Nacional de Estadística e Informática, INEI) of Peru. The survey follows the methodology recommended by the Monitoring and Evaluation to Assess and Use Results (MEASURE DHS) program. The ENDES employs a probabilistic, two-stage, stratified, balanced, and independent sampling design that ensures representativeness at national, regional, and urban-rural levels across all departments of Peru²⁴.

The sampling methodology involves two stages: first, the selection of clusters, followed by the selection of dwellings within each cluster. The sample was stratified by department and area (urban-rural), with clusters selected systematically and proportionally to their population size. A total of 50 balanced samples were generated for each of the 250 strata, which were obtained by combining 26 departments, 3 domains, and 6 sociodemographic strata.

The ENDES 2023 datasets were obtained in SPSS format from the official website of the National Institute of Statistics and Informatics (INEI) (https://proyectos.inei.gob.pe/microdatos/Consulta_por_Encuesta.asp). The datasets were downloaded and merged within RStudio using unique household and individual identification codes, as described in a previous paper²⁶.

Participants

The population consisted of 31,276 participants between 15 and 97 years (mean = 39.28, SD = 16.69); 17,903 females (57.25%) and 13,373 males (42.75%). For subgroup analyses, we examined participants who scored 5 or higher on the PHQ-9, representing a “screen-positive subsample” or “elevated depression symptom subsample” rather than a formally diagnosed clinical sample, as participants were not evaluated through structured clinical interviews. In the screen-positive subsample, the mean age was 42.13 (SD = 18.79), with 5237 females (70.79%) and 2161 males (29.21%). The sample size was estimated employing the powerly package, with 9 nodes, a statistical power of 0.80, and a density of 0.40, which indicated a recommended sample of 300 observations³¹.

Native language groups were classified into four distinct populations: Castellano speakers (the primary group), Quechua speakers (Andean highlands), Aymara speakers (Andean highlands), and Amazon speakers (representing various Amazonian indigenous languages). For the screen-positive subsample analysis, sufficient sample sizes were available for Castellano (n = 5,043), Quechua (n = 2,058), and Aymara (n = 172) speakers; however, Amazon groups were excluded in the screen-positive subsample depression analyses due to an insufficient sample size.

Patient health questionnaire 9

Depression symptoms were assessed using the Patient Health Questionnaire-9 (PHQ-9)³². The PHQ-9 is composed of nine items, and is described as an unidimensional scale: however, recent research suggests it is composed of two (affective and cognitive) to three dimensions (affective, somatic, and cognitive/motor)^34–36. Each item is rated on a 4-point Likert scale, ranging from 0 (not at all) to 3 (nearly every day), to assess symptoms over the preceding two weeks. Items cover several symptoms of depressive disorders, such as anhedonia, depressed mood, guilt, and suicidal ideation, to name a few. It has been validated in the Peruvian context with adequate reliability (ω = 0.87)³⁶. Total scores range from 0 to 27, with established cut-off points for severity levels: minimal depression (0–4), mild (5–9), moderate (10–14), moderately severe (15–19), and severe depression (20–27). However, recent evidence has suggested that the cut-off of 8 could improve sensitivity and specificity³⁷. In light of this contradictory evidence and cut-off points, we considered a cut-off score of 5 or higher for depressive symptoms, as described in the original instrument³². The version of the PHQ-9 administered in the ENDES was in Spanish; no translation was performed.

It is worth noting that the PHQ-9 reflects Western depression concepts, potentially limiting indigenous applicability. Indigenous distress manifests as Llaki, susto, and community-centered spiritual expressions. These cultural frameworks may alter how indigenous participants interpret items. Consequently, measurement limitations could explain observed differences across linguistic groups.

Native languages

Linguistic classification was determined based on participants’ self-reported native language from ENDES, which asks respondents to identify their first language learned during childhood. This approach was selected because native language serves as a fundamental marker of cultural identity and cognitive framework, particularly given that these languages represent distinct genetic families (Indo-European, Quechuan, Aymaran, and Amazonian families) with contrasting morphological types, evidentiality systems, and temporal conceptualizations that structure speakers’ symptom expression patterns^10,15. This was used as a proxy of culture, and terms may be interchangeable during the discussion.

Castellano language is the dominant Indo-European language in Peru, spoken approximately by 90% of the population³⁸, and it is spoken all over the country. Its morphological type is fusional, which leads to larger lexicons, and the basic word order is subject-verb-object, making clear differentiations between the subject and object.

Quechua is the most widely spoken Amerindian language in Peru, spoken by approximately 20% of the population. It is divided into two branches: Quechua I, centered in the Peruvian highlands, and Quechua II, which has a wider geographical extension. It is an agglutinative indigenous language with subject-object-verb structure, and a small lexicon^22,38,39.

The Aymara language is a part of the Aymara family, spoken by approximately 9 to 10% of Peruvians. It is primarily spoken in the Southern Highlands, with a small branch also present in the northern region due to migration. It is an agglutinative indigenous language with a complex system of suffixes. Its basic word order is subject-object-verb, it has a small lexicon, and similarly to Quechua, it requires evidentiality^22,38,39.

Amazonian language families are spoken by approximately 1.3% of Peru’s population across the Amazon basin. These languages belong to multiple unrelated families, including Arawakan, Panoan, Jivaroan, Tupian, and Cahuapanan, with over 40 distinct languages representing some of the world’s most linguistically diverse regions. They are predominantly polysynthetic indigenous languages with subject-object-verb structure and highly complex morphological systems that can express entire sentences within single words^22,40.

Data analysis

Data analyses were conducted using the R programming language (version 4.1.2) within the RStudio environment. The analysis followed a four-phase framework⁴¹, for network analysis: (a) network estimation, (b) network stability assessment, (c) network accuracy evaluation, and (d) network comparison. These analyses were performed for both the general population network (N = 31,276) and the screen-positive subsample network (PHQ-9 ≥ 5; n = 7,398) to examine potential differences in symptom connectivity patterns between population-based and screen-positive subsample presentations of depression.

Phase 1: network Estimation

Network estimation was performed using the bootnet package (version 1.5) employing the ggmModSelect estimator with Spearman correlation matrices⁴². This approach was selected due to the ordinal nature of the PHQ-9 variables and its appropriateness for identifying central symptoms when both positive and negative edges are present⁴³. The ggmModSelect method utilizes stepwise model selection to identify the most parsimonious network structure while maintaining statistical rigor.

Node centrality was calculated using the Expected Influence Index (EI). EI is considered more appropriate than traditional centrality measures when networks contain both positive and negative connections⁴⁴. While depression networks typically show predominantly positive correlations, we observed several near-zero and potentially negative edges (particularly involving suicidal ideation and appetite symptoms), making expected influence the methodologically appropriate choice..

Phase 2: network stability assessment

We evaluated network stability using case-dropping subset bootstrap procedures (1,000 iterations) to calculate correlation stability coefficients (CS-coefficients)⁴¹. CS-coefficients indicate the maximum proportion of cases that can be dropped while maintaining > 95% probability that centrality order remains stable (correlation > 0.7 with original centrality). Following established guidelines, CS-coefficients should exceed 0.25, with values > 0.5 indicating excellent stability. Additionally, we assessed edge weight accuracy through bootstrapped confidence intervals (95% CI) to quantify estimation precision.

Phase 3: network accuracy evaluation

Network accuracy was assessed through repeated model estimation using bootstrapped samples to generate 95% confidence intervals around edge weights. This procedure involved 1,000 non-parametric bootstrap samples to evaluate the precision of network parameter estimates⁴¹. Narrow confidence intervals indicate high precision in edge weight estimation, while wider intervals suggest greater uncertainty in the relationships between symptoms.

Phase 4: network comparison

Native language groups comparisons were conducted using the NetworkComparisonTest package, which employs two-tailed permutation tests to evaluate differences between network structures⁴⁵. The procedure utilized 1,000 permutation replications to assess: (a) network invariance (M statistic) testing whether the overall network structures differ between groups, and (b) global strength invariance (S statistic) examining whether the overall connectivity differs between networks. The null hypothesis assumed no differences between linguistic groups at a significance level of α = 0.05.

Ethics consideration

This study used publicly available, anonymized secondary data from the ENDES survey. The original ENDES survey obtained informed consent from all participants and was conducted in accordance with national and international ethical guidelines. As this analysis used de-identified, publicly available data, additional ethical approval was not required. However, all data handling and analysis procedures maintained participant confidentiality and anonymity.

Results

Network structure in the whole population

Global network properties

The PHQ-9 depression symptom network analysis revealed 35 non-zero edges out of 36 possible connections, yielding a density of 97.22%. The mean edge weight was 0.106 (range: 0.00–0.38.00.38), indicating moderate-to-strong relationships between depression symptoms.

Estimation of the network and centrality

Figure 1 presents the estimated network structure and centrality indices. The strongest edge weight occurred between Worthlessness/Guilt (PHQ6) and Suicidal ideation (PHQ9) (r = 0.377), followed by Anhedonia (PHQ1) and Depressed mood (PHQ2) (r = 0.364), and Concentration difficulties (PHQ7) and Psychomotor symptoms (PHQ8) (r = 0.270). Additional notable connections included Fatigue (PHQ4) with Anhedonia (r = 0.20), Fatigue with Sleep problems (PHQ3) (r = 0.18), and Depressed mood with Sleep problems (r = 0.17).

Fig. 1.

Network estimation and centrality index. Network structure and centrality of depressive symptoms in the general population Left panel: Depression symptom network in the general population (N = 31,276). Nodes represent PHQ-9 items categorized by symptom domain according to Gunzler criteria for RDoC⁵⁷: Core symptoms (red) - PHQ1: Anhedonia, PHQ2: Depressed mood; Vegetative symptoms (green) - PHQ3: Sleep Disturbance, PHQ4: Fatigue or Loss of Energy, PHQ5: Appetite or Weight changes; Motor/Attention symptoms (blue) - PHQ7: Distractibility, PHQ8: Psychomotor retardation or agitation; Cognitive symptoms (purple) - PHQ6: Guilt, PHQ9: Suicidability. Edge thickness indicates partial correlation strength, with thicker edges representing stronger connections. Blue edges indicate positive associations. Right panel: Strength centrality indices showing PHQ2 (Depressed mood) as the most central symptom, followed by PHQ4 (Fatigue) and PHQ1 (Anhedonia).

Centrality analysis identified “Depressed mood” as the most central node (Strength = 1.73), followed by “Fatigue/Low energy” (Strength = 0.85) and “Anhedonia” (Strength = 0.47). The least central nodes were Suicidal ideation (Strength = −1.84) and Appetite changes (Strength = −0.64).

Stability and accuracy of the network

Bootstrap procedures (1,000 iterations) demonstrated robust edge accuracy with narrow confidence intervals around edge weights. The Worthlessness/Guilt-Suicidal ideation connection showed exceptional stability, maintaining consistent edge weights around 0.38 across bootstrap samples. Stability coefficients (CS) for edge weights, expected influence, and strength centrality all reached 0.75, exceeding the recommended minimum threshold of 0.25.

Native Language network comparison

Network analyses were conducted for four native language groups: Castellano (n = 23,153), Quechua (n = 6,565), Aymara (n = 879), and Amazon (n = 551). Figure 2 displays the network structures by native language group, and detailed network metrics for all linguistic groups are presented in Table 1.

Fig. 2.

Networks according to native language. Depression networks across native language groups Network structures of depressive symptoms stratified by native language in the general population. Top left: Castellano speakers (n = 23,153) showing the highest connectivity with 22 active edges. Top right: Quechua speakers (n = 6,565) with 20 active edges. Bottom left: Aymara speakers (n = 879) demonstrating sparse connectivity with 12 active edges. Bottom right: Amazonian indigenous language speakers (n = 551) with 13 active edges. Node colors and positions are consistent with Figure 1. Right panel: Expected Influence centrality indices by language group, revealing distinct centrality patterns across cultures, with Aymara speakers showing elevated centrality for sleep problems and suicidal ideation.

Table 1.

Comparative network metrics across native Language Groups.

	Castellano	Quechua	Aymara	Amazon
Sample size	23,153	6,565	879	551
Active Edges	22	20	12	13
Network density	61.1%	55.6%	33.3%	36.1%
Mean Edge Weight	0.112	0.108	0.095	0.089
Strongest Edge	PHQ6-PHQ9	PHQ6-PHQ9	PHQ6-PHQ9	PHQ6-PHQ9
Strongest Edge Weight	0.38	0.39	0.44	0.37
Most central symptom	PHQ2	PHQ4	PHQ9	PHQ2
Expected Influence	1.73	0.92	0.88	1.45
Network correlation with Castellano	1.00	0.89	0.67	0.44

Castellano speakers showed the highest network connectivity with 22 active edges, followed by Quechua (20 edges), Amazon (13 edges), and Aymara (12 edges). Statistical comparisons revealed significant differences between Amazon and Castellano networks in edge weights (M = 0.23, p = 0.019) and connectivity patterns (S = 0.26, p = 0.009). Amazon and Quechua networks also differed significantly (M = 0.196, p = 0.108; S = 0.326, p = 0.019). Castellano and Quechua networks showed high structural similarity (M = 0.08, p = 0.069; S = 0.65, p = 0.99; r = 0.89).

Network correlations ranged from 0.435 (Castellano-Amazon) to 0.89 (Castellano-Quechua). Amazon groups demonstrated unique pathways, including Sleep problems to Suicidal ideation (r = 0.23) and Concentration difficulties to Suicidal ideation (r = 0.18), which were absent in other groups. Aymara speakers showed the strongest Worthlessness/Guilt-Suicidal ideation connection (r = 0.44).

Expected influence analysis revealed the highest average centrality in Aymara (0.847), followed by Quechua (0.845), Castellano (0.837), and Amazon (0.779). Sleep problems showed dramatically elevated centrality in Aymara speakers (EI = 1.064) compared to other groups (EI = 0.773–0.843). Suicidal ideation centrality increased progressively from Castellano (0.632) to Aymara (0.818).

Network structure in the screen-positive subsample

Global network properties

Among participants with PHQ-9 scores of 5 or higher, the network analysis revealed 22 non-zero edges out of 36 possible connections, resulting in a density of 61.11%. The mean edge weight was 0.073 (range: 0.00–0.40.00.40).

Estimation of the network and centrality

Figure 3 presents the screen-positive subsample network structure. The strongest edge weight occurred between Worthlessness/Guilt and Suicidal ideation (r = 0.398), followed by Anhedonia and Depressed mood (r = 0.337), and Concentration difficulties and Psychomotor symptoms (r = 0.260). Multiple symptom pairs showed zero connectivity, including Anhedonia-Suicidal ideation and Anhedonia-Appetite changes.

Fig. 3.

Network estimation and centrality index in the screen-positive subsample. Network structure and centrality in the screen-positive subsample Left panel: Depression symptom network among participants with PHQ-9 scores ≥ 5 (n = 7,398). Network density decreased to 61.11% (22 non-zero edges) compared to 97.22% in the general population. Node representations follow the same color scheme as Figure 1. Notable changes include stronger isolation of vegetative symptoms and emergence of distinct symptom clusters. Right panel: Strength centrality showing PHQ2 (Depressed mood) maintaining highest centrality, but with PHQ8 (Psychomotor symptoms) rising to second position, indicating altered symptom hierarchy in clinical presentations.

Centrality analysis identified “Depressed mood” as the most central node (Strength = 1.30), followed by “Psychomotor symptoms” (Strength = 0.95) and “Worthlessness/Guilt” (Strength = 0.55). Appetite disturbances (Strength = −1.88) and Sleep disturbances (Strength = −1.15) showed the lowest centrality.

Stability and accuracy of the network

Bootstrap analysis (1,000 iterations) confirmed network stability. The Worthlessness/Guilt-Suicidal ideation connection maintained consistent edge weights around 0.40 across bootstrap samples. Stability coefficients reached 0.75 for edge weights, expected influence, and strength centrality.

Native language network comparison.

Screen-positive subsample analyses included Castellano (n = 5,043), Quechua (n = 2,058), and Aymara (n = 172) speakers. Amazon groups were excluded due to insufficient sample size. Figure 4 displays the screen-positive subsample networks by native language.

Fig. 4.

Networks according to native language in the screen-positive subsample. Clinical depression networks across native language groups Depression symptom networks in the screen-positive subsample (PHQ-9 ≥ 5) stratified by native language. (A) Castellano speakers (n = 5,043) maintain high connectivity with 22 edges; (B) Quechua speakers (n = 2,058) show 18 edges; (C) Aymara speakers (n = 172) demonstrate sparse connectivity with only nine edges. Notable findings include the complete absence of anhedonia-depressed mood connections in the Aymara network and the emergence of unique psychomotor-suicidal ideation pathways (r = 0.27). Suicidal ideation showed dramatically elevated centrality in Aymara speakers (EI = 0.876) compared to other groups.

Castellano showed the highest connectivity (22 edges), followed by Quechua (18 edges) and Aymara (9 edges). Statistical comparisons revealed significant differences between Castellano and Quechua networks (M = 0.118, p = 0.009), though connectivity patterns remained similar (S = 0.121, p = 0.227). Aymara networks differed from both Castellano (M = 0.265, p = 0.138; S = 0.0585, p = 0.94; r = 0.55) and Quechua (M = 0.267, p = 0.128; S = 0.062, p = 0.920; r = 0.304).

Aymara speakers showed absent Anhedonia-Depressed mood connections (r = 0.00) and unique Psychomotor-Suicidal ideation pathways (r = 0.27). The Worthlessness/Guilt-Suicidal ideation connection remained moderate across all groups, strongest in Aymara (r = 0.45).

Expected influence analysis revealed the highest average centrality in Quechua (0.607), followed by Aymara (0.601) and Castellano (0.584). Suicidal ideation showed dramatically elevated centrality in Aymara speakers (EI = 0.876) compared to Castellano (0.568) and Quechua (0.480). Depressed mood centrality was highest in Quechua (EI = 0.809), while Psychomotor symptoms showed highest centrality in Castellano (EI = 0.705).

Discussion

Summary of findings

Here, we conducted a network analysis to explore the structure of depression symptoms across native language groups based on data from 31,276 Peruvians using the PHQ-9 questionnaire. Our major findings reveal: (1) Depression manifests with distinct cultural-linguistic patterns shaped by native language, with network density decreasing from general population (97%) to screen-positive samples (61%), suggesting that symptom interdependence varies by severity and cultural context; (2) While core universal pathways exist—worthlessness/guilt to suicidal ideation and anhedonia to depressed mood—each cultural group exhibits unique architectural patterns: Castellano networks show high complexity, while Aymara demonstrates selective connectivity with suicide-centered organization; (3) Neurovegetative symptoms (sleep, appetite) remain isolated across all networks, suggesting these operate through culturally-invariant biological pathways distinct from cognitive-affective symptoms; (4) The Aymara network represents a potentially unique depression presentation characterized by suicidal ideation as the most central symptom, absent anhedonia-mood connections, and exclusive psychomotor-suicide pathways; and (5) Centrality patterns shift between general and screen-positive populations—psychomotor symptoms emerge as second-most central in elevated depression symptoms, replacing fatigue—suggesting that symptom hierarchies evolve with severity differently across cultures.

Interpretation of findings

Our analysis revealed significant differences in depression network density between screen-positive subsample and general population samples, with the first demonstrating markedly higher connectivity patterns. This finding is consistent with previous research indicating that screen-positive samples exhibit selective connectivity characteristics within symptom networks^45–47. Furthermore, emerging evidence suggests that severe forms of depression arise not merely from elevated symptom severity scores, but rather from complex interactions and interconnections among symptoms⁴⁸. The observed differences between general and screen-positive populations may indicate that clinically significant depression emerges not simply from the presence of individual symptoms, but from the dynamic interactions and reciprocal relationships among these symptoms within the network structure. This pattern supports the hypothesis that depression states may arise after a critical threshold of connectivity—characterized by a constellation of highly interconnected symptoms—is reached, distinguishing clinical depression from well-documented subclinical presentations⁴⁹. These findings have important implications for future research and clinical practice, suggesting promising directions toward developing network-specific treatment patterns, implementing management strategies tailored to central symptoms across the depression spectrum, and potentially designing prevention strategies aimed at interrupting network reorganization processes before they reach pathological thresholds. Such approaches could fundamentally transform how we conceptualize, assess, and intervene in depression across different stages of severity.

Our findings suggest that the native language fundamentally shapes the architecture of the depression network, with particularly pronounced differences between Aymara speakers and Quechua/Castellano speakers. This pattern aligns with an extensive body of research documenting diverse cultural expressions of depression across global populations¹², suggesting that native language serves as a significant determinant of depression phenomenology within our study context¹⁵. Most notably, the Aymara population exhibited the highest centrality of cognitive symptoms and suicidal ideation—a network configuration not previously reported in the depression literature. While the Castellano population’s symptom pattern closely resembled the Western depression model, the Quechua network demonstrated greater centrality of somatic symptoms and fatigue, which bears striking resemblance to other cultural syndromes such as Susto, where the soul’s departure from the body results in both physical (somatic symptoms) and spiritual (energy) depletion⁵⁰. A compelling explanation for these divergent patterns may lie in the fundamental differences in mental symptom experience, which appears to be shaped by underlying worldview orientations. From an Andean perspective, the worldview is inherently holistic, transcending the Western mind-body dualism that characterizes traditional psychiatric frameworks⁵¹. Moreover, the cosmovision from the Andes which center on the collective rather than the individual, may also influence the depression experience, as from the Andean philosophy, no being can be completely devoid of relations and every being is immersed in multiple relationships, hence, this worldview may alter how the narrative self organize the mental phenomena⁵². Therefore, rather than representing measurement invariance issues, these distinct network patterns likely reflect profound worldview differences and may suggest the presence of fundamentally different underlying constructs of psychological distress across these linguistic and native language groups.

A particularly striking finding was the centrality of suicidal ideation in the Aymara network, a pattern not previously documented in the depression literature. This represents a significant departure from established Western frameworks, where mood symptoms typically occupy the most central positions in depression networks⁵³. This finding gains additional significance when considered alongside epidemiological evidence demonstrating that indigenous populations exhibit substantially higher suicide rates compared to non-indigenous groups⁵⁴. The notable absence of centrality among widely recognized core depression symptoms—specifically anhedonia and depressed mood—in the Aymara network suggests that the phenomenological experience of depression in this population may differ fundamentally from Western conceptualizations. One possible explanation is that suicidal ideation may represent either an ancient threat response mechanism or reflect the authentic manifestation of depression within this cultural context, rather than being a secondary consequence of other depressive symptoms. These findings carry profound therapeutic implications: given the high centrality of suicidal ideation in the Aymara network, immediate clinical management becomes paramount, and comprehensive suicide risk assessment should be considered mandatory for all individuals within this population group. However, further research exploring the lived experience of depression⁵⁵ in indigenous populations is essential to fully understand these distinct presentations and develop culturally appropriate interventions.

Despite the distinct differences observed across language groups, our analysis revealed that guilt and suicidal ideation represent universal high-risk connections across all three networks. This finding aligns with previous research demonstrating the robust relationship between these cognitive and behavioral symptoms across diverse populations⁵³. Multiple theoretical frameworks provide compelling explanations for this universal pathway: guilt and worthlessness can be conceptualized as cognitive equivalents to defeat and perceived burdensomeness, which serve as established pathways to suicidal behavior under the integrated-motivational-volitional model and the interpersonal theory of suicide, respectively⁵⁶. These consistent patterns across culturally and linguistically distinct groups support the hypothesis that suicide pathways within depression networks are fundamentally determined by maladaptive cognitive processes rather than culturally specific symptom presentations^47,53. This finding has important clinical implications, suggesting that interventions targeting these specific cognitive vulnerabilities—particularly guilt-related cognitions and their connection to suicidal ideation—may represent promising cross-cultural therapeutic approaches that transcend linguistic and cultural boundaries in depression treatment.

To synthesize the clinical and theoretical implications of our analysis, Box 1 outlines a comparative framework of the distinct depression network architectures observed. While the Castellano and Quechua patterns demonstrate adequate alignment with the Western depression model, the Aymara pattern exhibit poor fit. It displays isolated symptom clusters, suggesting a fundamentally different underlying construct or lived experience of depression. The cultural-linguistic patterns show distinct symptom centrality patterns: the castellano pattern centers on psychomotor retardation/agitation and depressed mood, indicating a dual mind-body experience; the quechua pattern focuses primarily on low energy; and the aymara pattern is characterized by suicidal ideation as the most central feature. From a dimensional perspective, these patterns appear to represent different aspects of depressive symptomatology—the Castellano pattern reflecting affective and core symptoms, the Quechua pattern emphasizing somatic and vegetative symptoms, and the Aymara pattern highlighting cognitive and high-risk symptoms.

Box 1. Cultural specificity of depression network architecture

These distinct cultural-lingual patterns presentations, although theoretically based, may have important implications for treatment approaches: the Castellano network may benefit from combined psychotherapy and psychotropic interventions, while the Quechua network may respond optimally to psychotherapy alone. In contrast, the Aymara network may require careful monitoring with community-integrated approaches specifically designed to address the elevated burden of suicidal ideation.

Future research should prioritize qualitative phenomenological studies exploring the lived experience of depression in indigenous communities through narrative medicine approaches and participatory action research to understand cultural meaning-making around mental distress. Longitudinal cohort studies that track symptom network stability across developmental stages, seasonal variations, and life transitions are essential for distinguishing stable cultural traits from temporal fluctuations and examining intergenerational transmission patterns. Randomized controlled trials comparing treatment outcomes between culturally adapted and standard protocols, including the integration of traditional healing and community-based interventions, will determine whether distinct symptom networks require differentiated therapeutic approaches. Cross-cultural validation studies across diverse indigenous populations globally should establish taxonomies of cultural depression patterns that can inform international diagnostic frameworks. Implementation science research examining real-world adoption barriers and facilitators for culturally adapted assessment tools, including provider training requirements and community acceptance factors, will ensure that research findings are effectively translated into improved clinical practice.

Box 1.

Cultural Specificity of Depression Network Architecture.

	Castellano	Quechua	Aymara
Diagnostic compatibility with the Western model	Good fit	Good to excellent fit	Poor fit
Network type	Highly connected	Highly connected	Isolated symptom clusters
Most Central Symptoms	Psychomotor retardation and depressed mood	Depressed mood and fatigue/low energy	Suicidal ideation.
Core Depression Symptoms	Standard pattern: Loss of interest connects to sad mood	Standard pattern: Loss of interest connects to sad mood	No connection between loss of interest and sad mood
Physical Symptoms	Sleep and appetite problems cluster together	Strong clustering of sleep, appetite, and energy symptoms	Weak connections between physical symptoms
Cognitive/Motor Symptoms	Concentration problems are moderately linked to motor symptoms	Concentration problems are moderately linked to motor symptoms	No connection between concentration and motor symptoms
Unique pathways	Multiple interconnected symptom loops create a complex presentation	Guilt strongly connects to motor symptoms. Appetite problems are linked to concentration issues	High-risk pathways Motor symptoms → Suicidal thoughts. Loss of interest → Concentration → Guilt
Risk profile	Severe psychomotor disturbance with mood symptoms	Severe mood symptoms with profound energy depletion	High suicide risk with cognitive impairment

Limitations

Several methodological considerations guide the interpretation of our findings. The cross-sectional design prevents the establishment of causal relationships or temporal sequences between symptoms across linguistic groups. Our cultural classification through native language, while methodologically necessary, inevitably simplifies complex identities—many Peruvians navigate multiple linguistic worlds that create hybrid cultural frameworks our categorical approach cannot capture. Sample size disparities pose additional challenges, particularly for Amazonian language speakers and Aymara speakers in the screen-positive subsample, potentially limiting statistical power to detect certain network differences or edge weights in these groups.

Furthermore, the PHQ-9, though validated in Peru, remains rooted in Western psychiatric concepts that may inadequately capture indigenous distress expressions like susto or spiritual imbalances central to Andean healing systems. The absence of formal clinical diagnosis in our screen-positive subsample (defined by PHQ-9 ≥ 5) represents a measurement limitation, as we cannot distinguish between individuals with clinically significant depression versus transient distress or subclinical symptoms. Network analysis assumes temporal stability in symptom relationships, potentially missing dynamic fluctuations across illness phases or cultural cycles. Finally, we did not control for potentially important confounding variables as current methods do not take this into consideration.

Conclusions

This study provides the first large-scale network analysis of depression symptoms across Peruvian native language groups, used as a proxy for cultural worldview and cognitive-linguistic frameworks, revealing that language and culture fundamentally shape depression architecture in Peru’s diverse population. While some pathways appear universal, significant variations suggest that depression may manifest as distinct constructs across cultural contexts. These findings challenge the assumption of universal depression phenomenology and highlight the need for culturally informed assessment and treatment approaches, particularly for Peru’s underserved indigenous populations. The identification of distinct cultural-linguistic patterns offers a foundation for developing targeted interventions that respect cultural authenticity while maintaining therapeutic efficacy. Future research employing longitudinal designs, qualitative methodology, and formal clinical diagnosis is essential to validate these network patterns and establish their clinical implications for prevention, assessment, and treatment across cultural contexts.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

JAFC: Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. SSP: Conceptualization, Writing – review & editing. CLA: Supervision, Writing – review & editing. CCC: Funding acquisition, Supervision, Writing – original draft, Writing – review & editing. BMC: Supervision, Visualization, Writing – original draft, Writing – review & editing. CASV: Conceptualization, Methodology, Supervision, Writing – review & editing. JHV: Conceptualization, Methodology, Writing – review & editing. JB: Conceptualization, Writing – original draft, Writing – review & editing.

Data availability

Data and code employed for this project can be accessed at: [https://doi.org/10.17605/OSF.IO/4XM5F](https:/doi.org/10.17605/OSF.IO/4XM5F).

Declarations

Competing interests

The authors declare no competing interests.