To ensure methodological rigor and comprehensive coverage, a systematic search was carried out across a set of internationally recognized academic databases: Scopus, PubMed/MEDLINE, Web of Science Core Collection, SciELO Citation Index, and the KCI-Korean Journal Database. These sources were deliberately selected due to their wide geographical reach, the breadth of indexed content in health and biomedical sciences, and the high standards applied to peer-reviewed literature (Gusenbauer and Haddaway, 2020). By prioritizing databases with rigorous inclusion criteria and strong citation structures, the reliability and reproducibility of the analyses were reinforced. The use of these platforms is well-supported in previous studies of similar scope (Gusenbauer and Haddaway, 2020; Lozada-Martinez et al., 2025b; Andreasen et al., 2022), further strengthening the validity of our approach.

To maximize precision and ensure reproducibility, a structured and conceptually grounded search strategy was developed based on Medical Subject Headings (MeSH) and their respective synonyms. The term ‘Mental Disorders’ (MeSH Unique ID: D001523) was used as the foundational descriptor. This third-order concept (Tree Number: F03) encompasses a recognized range of official mental disorders, allowing the inclusion of relevant synonyms. The aim was to systematically identify peer-reviewed publications that explored, analyzed, or evaluated mental health disorders. This approach enabled the retrieval of literature from a wide array of scientific disciplines, where the topic of mental health is relevant and actively studied.

In the initial phase, pilot searches were conducted to iteratively test combinations of descriptors and indexing tags across multiple databases and search engines. This allowed for the refinement and optimization of the search logic based on performance criteria such as relevance, sensitivity, and specificity. The final version of the strategy, which yielded the highest consistency and precision when tested in Scopus, was the following: TITLE-ABS-KEY (“Mental Disorder” OR “Mental Illness” OR “Mental Illnesses” OR “Psychiatric Disorders” OR “Psychiatric Disorder” OR “Psychiatric Diseases” OR “Psychiatric Disease” OR “Psychiatric Illness” OR “Psychiatric Illnesses” OR “Behavior Disorders” OR “Psychiatric Diagnosis”). This version was then adapted to match the structure and indexing frameworks of each additional platform used in the study, ensuring coherence and methodological comparability across sources (Supplementary material 1).

The literature search was conducted on November 28th, 2024, in English, Portuguese and Spanish. The inclusion of these three languages responded to the need for capturing a broader and more representative spectrum of publications. The initial screening phase, focused on titles and abstracts, was carried out between November 30th and February 25th, 2025. This stage allowed for the preliminary identification of studies aligned with the objectives of this analysis.

To ensure the relevance and reliability of the data analyzed, studies were included in the analysis only if they met specific inclusion criteria. These were: (A) being peer-reviewed scientific publications indexed in regularly issued academic journals, which guarantees a minimum standard of editorial and scientific quality; (B) availability of the full-text version, which was essential to allow in-depth examination and data verification; and (C) a clearly defined general objective explicitly focused on the analysis, discussion, investigation, synthesis, or evaluation of any mental health disorder.

To preserve the integrity of the dataset, documents were excluded if they: (A) corresponded to publications with non-regular peer review process such as conference proceedings, books, book chapters, errata, or retracted articles; (B) lacked at least one essential bibliographic metadata (e.g., author names, journal title, or corresponding author information), which limited traceability and verification; or (C) were still in press at the time of the search and therefore lacked a finalized version suitable for analysis.

Although the primary focus was on English, Portuguese and Spanish literature, documents published in other languages were also eligible, provided they included an abstract in either English, Portuguese or Spanish and met all inclusion criteria while avoiding any exclusion criterion. Finally, considering the historical dimension of this study and its scope, no lower limit was applied to the year of publication, allowing for a longitudinal and comprehensive view of the evolution of the field.

The results retrieved from all databases were exported in. CSV format, retaining the complete set of available metadata. This included essential elements such as document titles, author names and institutional affiliations, keywords, publication year, citation count, publication type, and other relevant descriptors. Preserving this information was fundamental to ensure the integrity of the selection and analysis process.

As an initial quality control step, four researchers independently performed a manual screening to eliminate duplicate records and to assess titles and abstracts according to the pre-established inclusion and exclusion criteria. This review was conducted using Microsoft Excel 2016, which allowed for structured visualization and systematic filtering of the retrieved data.

Following this, a second independent review phase was undertaken by the same four researchers to carry out detailed data extraction focused on scientometrics variables, indicators of science and society, and metrics related to global health and global mental health. In cases where discrepancies arose between reviewers in any of the two phases, an external evaluator was brought in to resolve disagreements through discussion and consensus.

To promote consistency across the dataset and strengthen comparability in subsequent analyses, key variables were standardized. For instance, all documents identified as reviews, regardless of their specific methodological format (narrative, systematic, or meta-analysis), were grouped under a unified category labeled “reviews.” Similarly, for the variable ‘country,’ the nationality of the corresponding author was used as the primary reference, ensuring a consistent geographic attribution across all records.

To assess the scientometrics characteristics of the included publications, data were extracted on two widely recognized journal-level metrics: the quartile ranking and the h-index, each adjusted to the publication’s corresponding reference year. These indicators were sourced from the historical archives of Scimago Journal & Country Rank (available from 1999 onward) and the Journal Citation Reports (available since 1997). For each journal, the highest available metric was selected to ensure consistency and capture the maximum potential visibility or impact associated with the publication venue. This approach allowed for a standardized evaluation of the scientific positioning of each article within the academic landscape over time.

To enable comparisons across contexts, countries were classified into six geographic regions according to the WHO’s regional groups: The Americas, Europe, Western Pacific, Eastern Mediterranean, South-East Asia, and Africa (World Health Organization, n.d.). This classification reflects widely accepted global health frameworks and facilitates alignment with global official reports. In addition, each country was categorized by income level based on the World Bank’s latest 2024 economic classification (World Bank, n.d.), which stratifies economies into four tiers: low-income (LICs), LMICs, upper-middle-income (UMICs), and HICs (World Bank, n.d.). Incorporating both geographic and economic stratifications provided a multidimensional lens for interpreting global publication trends and their associations with structural and contextual factors.

It was collected data on 60 indicators across all available countries and years from official international sources such as the WHO’s Global Health Observatory (World Health Organization, n.d.), the World Bank (World Bank, n.d.), United Nations World Population Prospects (United Nations, n.d.), and the Institute for Health Metrics and Evaluation (Institute for Health Metrics and Evaluation, n.d.). These indicators were grouped into four categories: (1) Economy, Development, and Education; (2) Health; (3) Inequality and Poverty; and (4) Governance and Rights. Data were retrieved via public APIs when available or downloaded manually. To enable comparability between regions for subsequent analyses, the indicators were summarized as shown in Supplementary material 2.

To examine the associations between the selected indicators and scientometrics variables, linear regression analyses were performed. Each indicator was evaluated in relation to the number of publications. Depending on the context of the specific analysis, variables were treated as either dependent or independent. To account for possible regional variation, models were developed separately for each WHO’s regional groups (World Health Organization, n.d.). Only statistically significant models with meaningful effect sizes are presented in the main results, while the full set of outcomes is included in Supplementary material 3. Additionally, we constructed a matrix of regression coefficients by converting scientometrics variables to standardized Z-scores.

To enhance transparency and reproducibility, all variables included in the regression analyses were explicitly operationalized according to their measurement level and source. The scientometrics variables (e.g., annual publication counts, h-index, quartile ranking) were treated as continuous or ordinal variables depending on their inherent scale, whereas country and region classifications were treated as nominal variables. All indicators retrieved from international databases (e.g., Gross Domestic Product [GDP] per capita, literacy rate, life expectancy, Disability Adjusted Life Years (DALYs), poverty rate, governance indicators) were used in their original continuous form, as reported by the respective institutions (WHO, World Bank, and others). No transformations were applied except for standardization (z-scoring) when producing the cross-regional comparative matrices.

In all models, publication volume was used as the dependent variable when assessing whether contextual indicators predicted research output. Conversely, indicators were used as dependent variables when the analytical objective was to examine how research activity related to specific health or societal outcomes. This bidirectional analytic structure reflects the ecological nature of the study, which aims to describe patterns of association rather than infer causation.

A meta-analysis was conducted on the regression results. For each health indicator, the regional regression coefficients and their corresponding standard errors served as input for a random-effects meta-analysis. This method incorporated regional heterogeneity and yielded pooled estimates of the associations. The between-region variance was estimated using the Restricted Maximum Likelihood (REML) method. Only significant findings from the meta-analyses are reported in the main text; comprehensive results are provided in Supplementary material 4.

All statistical analyses were conducted using R software (version 4.3.1). The scripts for these analyses, along with detailed annotations, are available at: https://doi.org/10.5281/zenodo.15161386.

This study was approved by the Scientific Committee of Universidad de la Costa (code GRA.2021-07-002-19). However, no humans, animals, or medical records were used as units of analysis.

A total of 386,671 documents were identified (Figure 1). The largest proportion of publications originated from the Americas (43%) and Europe (35.4%) (Table 1), with a relatively consistent number of publications across years (Figure 2). Although not shown, based on our search strategy, the earliest publication on mental health topics was from the Americas, published in JAMA in 1886 (Carpenter, 1886). The Eastern Mediterranean and African regions contributed the least, with 1.7 and 0.8% of total publications, respectively.

The Americas contributed nearly half of all citations (49.6%) and included the most cited article, published in the Journal of Psychiatric Research, with over 77,000 citations (Folstein et al., 1975). The Eastern Mediterranean region had the fewest citations (10.9%) and the lowest average H-index (64.3). The highest average H-index values were observed in the Americas (148.3), Western Pacific (123), and Europe (113.5). The highest gold/bronze open access ratios were found in the Eastern Mediterranean (7.01) and South-East Asia (3.03), while the Americas (1.05) and Europe (1.6) had the lowest (Table 1).

Across all regions, original articles were the most common document type, particularly in the Eastern Mediterranean. Reviews were second, with the highest proportion in the Americas. Letters were less frequent, though South-East Asia had the highest share (6%). Of the total publications, 329,796 had a reported journal quartile. The Americas had the highest proportion of Q1 articles (67.4%), South-East Asia led in Q2 (27.1%), and the Eastern Mediterranean had the highest proportions of Q3 (31.4%) and Q4 (20.1%) publications. Q1 articles predominated in all regions except the Eastern Mediterranean.

The correlations between the five domains of indicators and the different geographical regions are presented in Figures 3a–e.

Among the mental health disorders included, measured either by prevalence or by DALYs, we observed a tendency toward protective effects in regions with high publication rates (i.e., the Americas, Europe, and the Western Pacific), in contrast to regions with low publication rates (Figures 3f,g). Among these disorders, eating disorders showed the largest effect sizes, both in terms of prevalence and DALYs. Notably, this effect shifted markedly from a protective association in high-publication regions to a null or even contradictory positive association in low-publication regions, when using the number of mental health publications as the independent variable in the models. Specifically, the coefficients ranged from a 0.017-point decrease in the prevalence of eating disorders per additional mental health publication (SE = 0.0010, R² = 0.899, p < 0.001) in the Americas, to a 0.048-point increase in prevalence (SE = 0.0266, R² = 0.098; p = 0.081) in low-publication regions.

Concerning global health indicators, we observed an inverse relationship between child mortality rates and regional publication rates, with regions exhibiting lower annual publication volumes showing the greatest reductions in child mortality (Figure 3h). Additionally, we used physician density (physicians per 1,000 people) as an independent variable to approximate the average scientific output per physician in each region. The Americas showed the highest association, with each additional physician per 1,000 people corresponding to approximately 3,292 more publications (SE = 323.07; R² = 0.638; p < 0.001). Other regions also showed positive coefficients exceeding 1,000 publications, except for the Eastern Mediterranean and Africa, which showed 289 (SE = 60.88; R² = 0.414; p < 0.001) and 10.6 (SE = 51.71; R² = 0.001; p = 0.839), respectively, the latter being non-significant (Figure 3i).

Among economic indicators, Research and Development Expenditure (R&D) (% GDP) exhibited an inverse relationship in the group of regions with high publication rates. The highest coefficient was observed in the Western Pacific Region, where each additional percentage point of GDP allocated to research and development was associated with an increase of 2,899 publications (SE = 286.05; R² = 0.811; p < 0.001). In contrast, the remaining regions showed positive but smaller coefficients, with the exception of the African Region, where a one-point increase in this indicator was associated with a decrease of 69.1 publications. However, this association was not statistically significant (SE = 85.77; R² = 0.032; p = 0.429) (Figure 3j).

Interestingly, no observable effect was found for the share of the population living in extreme poverty on publication rates in the Americas and Europe, when using this indicator as the independent variable (Figure 3k). In all other regions, the coefficient was negative, indicating a reduction in publication volume with each percentage point increase in this poverty indicator.

As shown in Figure 3e, the majority of governance-related indicators used as independent variables exhibited positive coefficients. However, Indicators A and I [Functioning Government Index and Percentage of Territory Effectively Controlled by Government (Supplementary material 2)] showed predominantly negative and statistically significant coefficients. Similarly, the Western Pacific, Eastern Mediterranean, and South-East Asia regions displayed mostly negative coefficients across the governance indicators analyzed. In contrast, the Political Corruption Index showed positive and significant coefficients for nearly all regions, with the exception of Africa (Figure 3e).

Following the meta-analysis of all regional regression coefficients, 11 indicators retained statistically significant effects (Table 2). Among the education-related indicators, both the average years of schooling and the adult literacy rate (% of individuals aged 15 and above) demonstrated positive and significant associations with publication rates. The average years of schooling yielded a coefficient of 526.05 (95% CI: 30.5–1,021.5), while the literacy rate showed a coefficient of 71.37 (95% CI: 1.73–141). Regarding economic indicators, R&D expenditure as a percentage of GDP and GDP per capita (current US$) emerged as positive stimulus for publication output.

Among general health indicators, physician density exhibited a stronger association with publication rates compared to nurse and midwife density. Specifically, the coefficient for physician density was 1,289.6 (95% CI: 322.2–2,257.1), while that for nurse and midwife density was 472 (95% CI: 18.4–925.6). Additionally, when life expectancy for both sexes was used as a dependent variable, it showed a positive coefficient of 213,848.4 (95% CI: 50,784.6–376,912.2). Within mental health indicators, only self-reported life satisfaction retained a significant but small effect after the meta-analysis (Table 2).

Inequality and poverty indicators, measured through lifespan inequality via the Gini coefficient for both women and men, also maintained significant positive associations. The coefficients were 77.1 (95% CI: 55.13–99.08) for women and 65.6 (95% CI: 48.5–82.6) for men. Finally, among governance indicators, only the functioning government index remained significant after meta-analysis. It was negatively associated with publication output, showing a coefficient of −821.7 (95% CI: −1,624.7 to −18.7).