Ambient air pollution and depression: a systematic review and meta-analysis

1 Introduction

Air pollution poses one of the greatest environmental threats to public health worldwide. Globally, each year, air pollution is accountable for approximately seven million premature deaths, in which ambient air pollution accounts for four million. Clean air is a fundamental human right vital for health. In recognition of this, the World Health Organization (WHO) (1) has established guidelines for air quality aiming to address and mitigate the impacts of air pollution. The health implications of air pollution are crucial to understand given that a report from WHO showed that nine out of ten of the world´s population reside in areas where air quality fails to meet these recommended guideline levels (1). Ambient air pollution constitutes a complex mixture of several pollutants, including particulate matter (PM), gaseous pollutants, metallic, and organic compounds (2). Particulate matter (PM) is commonly categorized into PM_2.5 (particles with an aerodynamic diameter smaller than, or equal to, 2.5 µm) and PM₁₀ (particles with an aerodynamic diameter smaller than, or equal to, 10 µm). Sources of ambient air pollution include natural sources, such as forest fires, desert dust and volcanoes, as well as man-made sources, such as agriculture, transportation, factories, and wood burning facilities (3). The concentration of ambient air pollution has increased rapidly throughout the years due to a rapid growth within industrialization, urbanization, and an increase in emissions caused by transportation. However, at a global scale, PM_2.5 exposures have declined slightly from 2000 to 2019, while a mixture of both increasing and decreasing trends for different air pollutants and geographical locations has been shown during the last two decades (4). The concentration and composition of air pollution vary across the globe, but the concentrations are highest in low-to-middle income countries. As a result, the health burden associated with air pollution is most pronounced in these regions (5). Nevertheless, problems with air quality constitute a global issue and it remains to be one of the major challenges of modern days. Moreover, health effects associated with air pollution exposure are also relevant at very low concentrations (6).

The negative consequences of air pollution are greatly recognized as it is considered a risk factor for non-communicable diseases such as cardiovascular diseases, cancers, and respiratory diseases (1, 7). With regard to mental health, public health or policy reports rarely include the effects on psychiatric disorders related to air pollution exposure, partly due to insufficient evidence, but also due to the complexity of establishing an association between mental health and air pollution (8). As part of the broader WHO project “Estimation of Morbidity from Air Pollution and its Economic Costs” (EMAPEC), Forastiere and colleagues relied on the causality determinations provided by the U.S. Environmental Protection Agency's Integrated Science Assessment for PM_2.5, NO₂, and O₃. Although emerging evidence suggests potential links between long-term PM_2.5 exposure and additional outcomes, such as depression, a formal causality assessment for these outcomes was not yet available. Consequently, their review did not include depression (9). Nevertheless, there has been a recent expansion of research within this field, particularly concerning the relationship between air pollution and depression. The concept depression implies a mood disorder with symptoms that affect a person's thoughts, feelings, and ability to perform everyday activities. To be diagnosed with depression, the symptoms must be present for at least two weeks (10). Common symptoms include feelings of hopelessness, irritability, loss of interest, fatigue, sleep disturbances, changes in appetite, physical aches, and thoughts of death and suicide or suicide attempt (10).

Depression is one of the most common mental health disorders and causes the greatest burden of disease within the category of mental illnesses (11). According to WHO, depression is the leading cause of disability and the condition has been shown to have negative impact on educational trajectories, work performance, and social life (12). The burden of depression goes beyond quality of life and functional impairment, as it can potentially affect the somatic health of the individual (13). Depression has been associated with increased risk of suicide, all-cause mortality (14), and chronic illnesses (13, 15, 16). Depression also causes substantial economic burden, whether it is directly through healthcare costs, such as treatment, care, and medicine, or through indirect costs as it can lead to premature death, morbidity, and productivity loss (17). Each year, depression along with anxiety, burden the global economy approximately one trillion U.S. dollars, which is a cost projected to rise even more by 2030 (18). Thus, further research within this topic, identifying the risk factors, would greatly benefit public health and society. Further research would particularly benefit from addressing a distinct subtype of depression, which is post-partum depression (PPD). Although the prevalence of PPD has increased over recent decades, with a recent report from the Centers for Disease Control and Prevention (CDC) estimating that approximately one in eight women are affected (19), research examining the relationship between air pollution and PPD remains limited. PPD can cause negative consequences for the mother's health, but also for the child (20). Consequently, this may contribute to behavioral disturbances, eating and sleep disorders, and delays in infant developmental milestones (21, 22). Incorporating depression into health burden estimates attributable to air pollution would likely result in a marked increase, reflecting the substantial societal and economic burden of this condition.

The mechanism by which ambient air pollution causes psychopathology, such as depression, is not entirely clear. However, air pollution exposure is associated with neurostructural and neurofunctional effects including increased inflammation and oxidative stress, changes in neurotransmitters and their metabolites within multiple brain regions, mainly with regard to hippocampus, the prefrontal cortex, and amygdala, with a consequent increased risk of depression (23). Recent pre-clinical studies have shown that ambient air pollution exposure may increase the risk of psychopathology, by changing frontal brain regions which are responsible for emotional regulation and stress responding (24–31).

Several systematic reviews and meta-analyses have found an association between ambient air pollution and depression (24, 25, 27, 29), including two focusing on PPD and perinatal mental health (32, 33), and one focusing on indoor air pollution (34). Additionally, associations between early-life and childhood exposure to air pollution and impaired mental health, including depression, among children and young adults have been shown based on a systematic review (35).

The thorough systematic review and meta-analyses by Borroni et al. (24) on the association between ambient air pollution and depression in humans included original studies published until May 20, 2021. The first aim of our review was to update this meta-analysis as several large epidemiological studies on this topic have been published since then. Our second aim was to include post-partum depression (PPD), which this study did not consider. A recent systematic review and meta-analysis on air pollution and PPD found some indication of an association, particularly for PM₁₀ and NO₂, however, with limited statistical power and three or fewer studies that fulfilled inclusion criteria (32). Since the chemical composition of air pollution, PM size and distribution, as well as regional climate vary widely with geographic location, it can be problematic to compare studies from different regions and produce combined effect estimates (36, 37). Therefore, our third aim was to investigate whether the associations between air pollution and depression were different in different regions of the world.

2 Methods

The systematic review protocol was registered and published in August 2024 on International Platform of Registered Systematic Review and Meta-analysis protocol (INPLASY) (38) and in Zenodo (39) in August 2024.

2.1 Search strategy

This systematic review and meta-analysis has followed “The Preferred Reporting Items for Systematic Reviews and Meta-Analyses” (PRISMA) guidelines (40). The primary objective of this study was to examine the association between exposure to ambient air pollution and the risk of depression. Following the PECO (Population, Exposure, Comparator, and Outcome) framework, the research question was formulated as: “Is exposure to ambient air pollution associated with an increased risk of depression?” In line with the PECO framework, a comprehensive search strategy was developed in collaboration with a research librarian. The search strings used for bibliographic databases, including PubMed, CINAHL, Scopus, and Web of Science, are provided in Supplementary S1. Additionally, relevant publications were identified by searching the references of related studies and reports. Reference managing, screening and data extraction was done using the software Covidence (41). The search strategy was developed by combining terms related to ambient air pollution with various terms associated with depression, utilizing controlled vocabularies such as MeSH terms. Additionally, to account for potential variations in the terminology, the search strategy was tailored to each specific database.

2.2 Eligibility criteria

The inclusion and exclusion criteria were formulated based on the PECO framework.

2.2.1 Populations

Any human population.

2.2.2 Exposures

Ambient air pollution, including both short-term and long-term exposure to pollutants including nitrogen dioxide (NO₂), sulphur dioxide (SO₂), carbon monoxide (CO), particulate matter (PM_2.5, PM₁₀), ozone (O₃), and black carbon (BC) examined through either measured data or modelled estimates.

2.2.3 Comparator

Humans exposed to higher levels of ambient air pollution compared to those exposed to lower levels presented in terms of a µg m⁻³ increase, exposure categories with concentration ranges, as well as per (x-fold) unit increase in exposure. This association had to be reported quantitatively, utilizing risk estimates such as odds ratios (OR), hazard ratios (HR), relative risks (RR), or beta-coefficients with 95% confidence intervals.

2.2.4 Outcome

Diagnosis of depression or postpartum depression identified either through clinical or insurance records, self-reported diagnoses (e.g., “Have you ever been told by a medical professional that you have depression?”), or clinical assessments that involve depression test questionnaires. Additionally, studies that examined the exacerbation of depression as an outcome included the number of hospital visits, changes in medication usage, or variations in the severity of depression.

2.2.5 Types of studies

Epidemiological (e.g., follow-up, case-control, and cross-sectional studies) including exposure assessment providing for an exposure effect analysis. Only peer-reviewed published studies were included for data synthesis.

2.2.6 Language

No language or publication year exclusion criteria were used. However, title and abstract searches in the databases required English terms.

2.2.7 Exclusion criteria

Studies that did not constitute primary research, those in which depression was considered only as a component of a broader outcome category (e.g., “mental health disorders”), and those that assessed depression solely through self-reported diagnostic tools without clinical validation were excluded. Additionally, studies focusing exclusively on indoor air pollution or occupational exposure, which complicated the differentiation of ambient air pollution effects from other influencing factors, were not included. Studies that involve individuals with pre-existing depression at baseline were excluded, except when examining the outcome of “depression exacerbation”.

2.3 Selection process

This review followed a structured and thorough process of identifying studies that align with the eligibility criteria. Studies published until December 18, 2023 were selected by two authors (J.H. and H.A.S.M.), and these studies were independently reviewed in Covidence (41). The selection process was based on the eligibility criteria regarding factors such as “Population”, “Study Design”, “Exposure”, “Outcome”, “Comparator”, or other specified criteria. The first phase of the selection process involved initial screening of titles and abstracts. If there was any disagreement regarding a study's eligibility, it was forwarded to the second stage, which involved full-text screening. All studies that progressed to the second phase underwent full-text screening, which was independently conducted by two authors (J.H. and H.A.S.M.). If there was any conflict over eligibility during this phase, a third author was consulted to provide the final decision.

The initial literature search rendered 10,763 articles, out of which 183 remained after screening through title and abstracts. After the full-text review, 90 studies remained for meta-analysis, with 15 studies excluded due to studies using the same cohort, or where the presentation of result estimates rendered standardization of the estimates infeasible. However, all these 90 studies are presented as a result of the systematic literature review. A compilation of studies excluded during the title and abstract screening or the full-text screening stage, along with the rationale for their exclusion, is provided and visualized as a PRISMA Flow Diagram (Figure 1). The search terms are presented in Supplementary S1, Supplementary Table S1.1.

2.4 Quality assessment (risk of bias assessment)

The quality assessment for this review was conducted using the “Office of Health Assessment and Translation” (OHAT), which is a widely recognized tool for evaluating risk of bias (ROB) in observational studies (42). This tool includes six domains consisting of selection, confounding, performance, attrition/exclusion bias, detection, and selective reporting. Each domain has certain criteria and the studies have been rated in terms of “low risk”, “probably low risk”, “probably high risk”, and “high risk” in accordance with these domains. The criteria for each domain were customized to align with this specific research topic, while remaining grounded in the OHAT framework (Supplementary Table S2). Once the ROB assessment for each domain was completed, an overall ROB judgment for the entire study was assigned based on the most severe evaluation of any individual domain (43). A summary of the ROB assessment for each study, along with the overall evaluation, has been presented as a heat map to visualize the various grading levels (Supplementary Table S4).

2.5 Data extraction

The data was independently retrieved by two authors (J.H. and H.A.S.M.). Information regarding study characteristics (author name, publication year, study design, and location), participant characteristics, list of exposures, exposure assessment method, outcome (definition and collection method), and effect estimates (RR/OR/HR/beta-coefficient) with their confidence intervals was extracted. The information was reported in a table designed in advance by using tools in Covidence (Supplementary Table S5). Any discrepancies were settled with a dialogue, and with a third author if needed. In instances where studies provided both unadjusted and adjusted estimates, the estimate that controlled for the greatest number of confounders was selected.

In instances where depression diagnoses were established through various methods such as a physician's diagnosis, questionnaires, medical records, clinical assessments based on the International Classification of Diseases (ICD), or self-reporting, we prioritized the most objective method, recognizing the ICD as the gold standard. Additionally, some studies employed different lag times to ascertain significant exposure windows, which were reported either as single lag days or cumulative lag windows. For this review, we utilized or reported the most pertinent lag times in terms of effect size and p-value, where lag0–3 has been used in most short-term studies that have been included in the meta-analyses. In cases of uncertainty, or where missing essential information was needed for the inclusion of studies in the meta-analysis, the corresponding authors were contacted for clarification.

2.6 Standardization of data

The studies included in this systematic review and meta-analysis reported effect estimates (risk coefficients) using a mixture of various measures, typically expressed as relative risks (RRs), hazard ratios (HRs), odds ratios (ORs), or beta-coefficients. These measures were collectively referred to as relative risk measures (RRs).

To ensure comparability, each adjusted effect estimate extracted from the included studies, along with its 95% confidence interval (CI), was standardized to a fixed air pollution concentration increase of 1 μg m⁻³ [Equation (1)] except for CO where a standardized concentration of 10 μg m⁻³ was used. For studies reporting risk coefficients based on quartile comparisons, the results were converted to a 1 μg m⁻³ increase using the distance between the quartiles, calculated from the median exposure values, or mean exposure values in cases were median values were unavailable. For studies reporting risk coefficients as increases in air pollution measured in parts per billion (ppb), these increments were converted to μg m⁻³ using the ideal gas law. For studies reporting risk coefficients as beta-coefficients, they were converted into relative risks with 95% CIs [Equation (2)].

StandardizedRR=e[ln(originalRR)originalincrease×standardizedincrease](1)

RR(95%CI)=eβ±1.96SE(2)

2.7 Pooled estimates (meta-analyses)

For each exposure-outcome pair with a sufficient number of studies meeting the eligibility criteria, and given the similarities in study populations, outcome definitions, and exposure assessments, a meta-analysis was conducted. For studies using the same cohort, the study selected was based on factors including risk of bias assessment, outcome measure method, and exposure method. Also, studies excluded from the meta-analysis included those where the presentation of result estimates rendered standardization of the estimates infeasible. In cases where a meta-analysis was deemed inappropriate, results were synthesized qualitatively.

For exposure to NO₂, SO₂, CO, PM_2.5, PM₁₀, O₃, and BC in relation to the outcomes “depression incidence” and “postpartum depression (PPD)”, meta-analyses were stratified by exposure duration: short-term exposure (defined as one month or less) and long-term exposure (more than one month). For the outcome “depression exacerbation”, only short-term exposure periods were considered.

Overall pooled estimates, as well as estimates stratified by region (Europe, North America, and Asia), were calculated using a random-effects model with restricted maximum likelihood (REML). Heterogeneity among studies was assessed using the I² and Tau² statistics (44). The pooled estimates were calculated according to Equation (3).where the included variables are defined as follows:

Θ=∑k=1Kθkωk∑k=1Kωk(3)

Θ = pooled weighted average effect (inverse-variance weighting)

θ_k = the effect size for each individual study

ω_k = the weight of each individual study

K = the number of studies

Publication bias was evaluated by visually inspecting the funnel plots (45). To complement visual assessment, which may be unreliable on its own, Egger's linear regression test was used to formally assess the presence of publication bias according to Equation (4).where the included variables are defined as follows:

θkSEθk=β0+β11SEθk(4)

θ_k = the effect size for each individual study

β₀ = the intercept of regression

β₁ = the weight of regression

SEθ_k = standard error of the effect estimate

An intercept that is significantly different from zero (z-score ≥1.96) indicates publication bias.

All statistical calculations in this systematic literature review and meta-analysis were performed using STATA version 18.0 (StataCorp, College Station, TX, USA).

2.8 Certainty of evidence assessment

To assess the overall quality of evidence across the included studies, the “Grading of Recommendations, Assessment, Development, and Evaluation” (GRADE) approach was employed. The GRADEpro software was utilized to generate an evidence profile table, summarizing the quality of evidence and key information for each outcome (46). The GRADE approach includes five domains through which the quality of evidence may be downgraded: (1) Risk of Bias, as detailed in Section 2.4 of this review; (2) Inconsistency, which refers to heterogeneity in results across studies; (3) Indirectness, indicating limitations in how directly the evidence applies to the specific research question; (4) Imprecision, which relates to factors such as small sample sizes and wide confidence intervals; and (5) Publication Bias, assessed using funnel plot asymmetry and Egger's test. Furthermore, there are three factors that can lead to upgrading the quality of evidence for each outcome: Effect size, dose-response relationship, and confounding (47). The overall quality of the evidence was subsequently rated according to four levels: high, moderate, low, or very low (48).

3 Results

3.1 Literature retrieval and study characteristics

The initial systematic search retrieved 10,763 studies, supplemented by one additional record identified through citation searches. Following the removal of 1,840 duplicates, 8,923 unique records underwent title and abstract screening, of which 183 were deemed eligible for full-text review. At this stage, the majority of excluded studies were deemed ineligible due to inappropriate exposure or outcome measures (n = 61). A smaller proportion of studies were excluded because they were available only in abstract form (n = 26). Overall, this resulted in a total of 90 studies that investigated the associations between short- or long-term exposure of air pollution and depression or PPD, which were included in this review (Figure 1). Out of these eligible studies, 76 were incorporated into the meta-analysis, after accounting for duplicates in the study populations and results in which estimates could not be standardized.

The key characteristics of the 90 studies included in this review are presented in Tables 1–3. Table 1 summarizes the 51 studies examining long-term exposure. Table 2 provides an overview of the 37 studies focusing on short-term exposure. Eight studies have examined both long-term and short-term effects, and are thus presented in both Tables 1, 2. Finally, Table 3 presents the ten studies that assessed air pollution exposure and postpartum depression. The majority of the studies (n = 47) were conducted in Asia, followed by 24 studies in Europe and 16 studies in North America. Additionally, two studies were conducted in Australia, and one study included data from multiple countries from Asia, Africa, North America and Europe. Regarding study design, most studies employed a cohort design (n = 42), followed by time-series (n = 23) and cross-sectional studies (n = 18). Additionally, two studies utilized a case-control design, while five studies adopted a case-crossover design. The primary outcome measure in most of the studies was the ICD-10 diagnosis. The remaining studies utilized self-administered questionnaires, such as the Center for Epidemiologic Studies Depression Scale (CES-D) and the Patient Health Questionnaire-9 (PHQ-9). A limited number of studies relied on self-reported physician diagnoses of depression and/or records of antidepressant prescriptions. The diagnostic tools employed in studies on postpartum depression were predominantly the same as those used for general depression. However, a few studies specifically utilized the Edinburgh Postnatal Depression Scale (EPDS) for the diagnosis of postpartum depression. In the majority of studies, air pollution exposure was assessed using data from monitoring stations, while land use regression (LUR) was the second most commonly employed estimation method. The rest of the studies used estimation through chemical transport models, spatial or spatiotemporal dispersion modelling or other.

Table 1. Characteristics of long-term exposure studies included in the systematic review and meta-analysis.

Table 2. Characteristics of short-term exposure studies included in the systematic review and meta-analysis.

Table 3. Characteristics of post-partum depression studies included in the systematic review and meta-analysis.

3.2 Long-term air pollution exposure and depression

Overall, this review included 51 studies on depression and long-term exposure to air pollutants including PM_2.5, PM₁₀, NO₂, SO₂, CO, O₃, and BC. The exposure was considered long-term if the exposure duration was more than one month prior to the diagnosis of depression. The pooled estimates along with corresponding 95% CIs and test for heterogeneity (I²) are presented in Figures 2–8 and Table 4. All pollutants associated with long-term exposure demonstrated substantial heterogeneity across all studies, with all being >97% except for BC with an I²-value of 66%. The meta-analysis demonstrated a statistically significant association with a 1 μg m⁻³ increase in exposure to PM_2.5 and NO₂ and depression. The pooled RR was 1.044 (95% CI: 1.024–1.064) (Figure 2) for PM_2.5, 1.011 (95% CI: 1.003–1.018) (Figure 4) for NO₂, and 1.147 (95% CI: 1.053–1.241) (Figure 8) for BC. No statistically significant association was observed for PM₁₀ where the pooled RR was 1.010 (95% CI: 0.999–1.021) (Figure 3). The results for SO₂, CO and O₃ (Table 4) similarly showed no significant associations, and were derived from fewer studies compared to PM_2.5, PM₁₀, and NO₂.

Table 4. REML-based meta-analyses regarding the associations between long-term exposure to ambient air pollutants with relative risk estimates with 95% confidence intervals and the degree of heterogeneity (I²).

The estimates refer to a 1 μg m⁻³ increase for PM_2.5, PM₁₀, NO₂, SO₂, BC, and O₃. For CO it is per 10 μg m⁻³ increase.

Publication bias was visually assessed by using funnel plots and showed to be present for PM_2.5, and slightly for NO₂ and CO (Supplementary S3, Supplementary Figures S3.1–S3.7). Publication bias based on Egger's linear regression test with their p-values are presented in Supplementary Table S3.1.

The ROB assessments for each included study on long-term exposure are provided in Supplementary S4. The majority of these studies were judged to have an overall ROB that was probably high. Based on the GRADE evaluation, presented in Supplementary S5, the overall certainty of the evidence was rated as very low.

Table 4 summarizes the meta-analyses regarding long-term exposure presented in Figures 2–8.

3.3 Short-term air pollution exposure and depression

In total, there were 37 studies in this review that examined the association between short-term exposure and depression. The pollutants examined included PM_2.5, PM₁₀, NO₂, SO₂, CO, and O₃. Short-term exposure was defined as exposure occurring within one month prior to the onset or worsening of depression, or hospitalization due to depression. The meta-analytic estimates, along with their corresponding 95% confidence intervals and tests for heterogeneity, are presented in Figures 9–14 and Table 5. Substantial heterogeneity was observed for most pollutants, with I² values exceeding 90%, except for PM_2.5 and SO₂ which exhibited moderate heterogeneity with I²-values of 65.5% and 43.7%, respectively. Overall, statistically significant associations for depression were identified between short-term exposures to all assessed pollutants except for CO (Figure 14) with the strongest association for NO₂. However, some associations were borderline in significance, including NO₂ (RR = 1.006; 95% CI: 1.000–1.011) (Figure 11). In contrast, weaker associations were found for PM₁₀ and PM_2.5, where each 1 μg m⁻³ increase was associated with an approximate 0.1% increase in the risk of depression (Figures 9, 10 and Table 5). The meta-analyses were based on data from at least ten studies for all pollutants. Evidence of potential publication bias was observed for the association between PM_2.5 and depression, while the risk of publication bias appeared to be lower for the other pollutants (Supplementary S3, Supplementary Figures S3.8–S3.13). Supplementary S4 presents the ROB assessments for all included short-term air pollution studies. A substantial proportion of these studies were deemed to have a probably high overall risk of bias. Consequently, the GRADE evaluation, also detailed in Supplementary S5, rated the overall certainty of the evidence as very low.

Table 5. REML-based meta-analyses regarding the associations between short-term exposure to ambient air pollutants with relative risk estimates with 95% confidence intervals and the degree of heterogeneity (I²).

The estimates refer to a 1 μg m⁻³ increase for PM_2.5, PM₁₀, NO₂, SO₂, and O₃. For CO it is per 10 μg m⁻³ increase.

Table 5 summarizes the meta-analyses regarding short-term exposure presented in Figures 9–14.

3.4 Long-term air pollution exposure and postpartum depression

A total number of nine studies assessing the association between air pollution and postpartum depression were used in the meta-analyses. The pollutants included in these studies were PM_2.5, PM₁₀, NO₂, and O₃. The meta-analytic estimates, along with their corresponding 95% confidence intervals and tests for heterogeneity, are presented in Figures 15–18 and Table 6. The heterogeneity among studies evaluating the relationship between air pollution and postpartum depression was notably high, with I² values exceeding 80%, indicating substantial variability across study results. Statistically significant associations were observed between exposure to both PM_2.5 and PM₁₀ and the risk of postpartum depression. The pooled RR for PM_2.5 exposure was 1.018 (95% CI: 1.004–1.031) (Figure 15), and for PM₁₀ exposure it was 1.023 (95% CI: 1.006–1.039) (Figure 16), corresponding to a 1.8% and 2.3% increased risk, respectively. These estimates were based on data synthesized from nine studies for PM_2.5 and five studies for PM₁₀. In contrast, no statistically significant associations were found between postpartum depression and exposure to NO₂ or O₃. With respect to publication bias, the greatest risk was observed for studies examining PM_2.5, while the risk appeared to be comparatively lower for the other pollutants (Supplementary S3, Supplementary Figures S3.14–S3.17). The ROB assessments for all included postpartum depression are presented in Supplementary S4. Half of the studies were assessed to have “probably high risk of bias”, while the other half was assessed to have “probably low risk of bias”. Nevertheless, taking the other domains of the GRADE into consideration (Supplementary S5), the overall certainty of the evidence was rated as “very low”.

Table 6. REML-based meta-analyses regarding the associations between ambient air pollutants and post-partum depression with relative risk estimates with 95% confidence intervals and the degree of heterogeneity (I²).

The estimates refer to a 1 μg m⁻³ increase for all pollutants.

Table 6 summarizes the meta-analyses regarding postpartum depression presented in Figures 15–18.

3.5 Stratification by geographic region

Tables 7–9 present meta-analyses divided into different continents including Asia, Europe, and North America. More detailed descriptions are presented in Supplementary S4, Supplementary Figures S4.1.1–S4.3.4.

Table 7. REML-based meta-analyses regarding the associations between ambient air pollutants and depression based on studies conducted in Asia with relative risk estimates with 95% confidence intervals and the degree of heterogeneity (I²).

The estimates refer to a 1 μg m⁻³ increase for PM_2.5, PM₁₀, NO₂, SO₂, and O₃. For CO it is per 10 μg m⁻³ increase.

Table 8. REML-based meta-analyses based on Europe regarding the associations between ambient air pollutants with relative risk estimates with 95% confidence intervals and the degree of heterogeneity (I²).

The estimates refer to a 1 μg m⁻³ increase for all pollutants.

Table 9. REML-based meta-analyses based on North America regarding the associations between ambient air pollutants with relative risk estimates with 95% confidence intervals and the degree of heterogeneity (I²).

The estimates refer to a 1 μg m⁻³ increase for PM_2.5, PM₁₀, NO₂, SO₂, and O₃. For CO it is per 10 μg m⁻³ increase.

4 Discussion

4.1 Summary and main findings

In this systematic review, we conducted a comprehensive synthesis of multiple studies assessing the relationship between both short-term and long-term exposure to air pollution and the onset or exacerbation of depression, as well as postpartum depression (PPD). With respect to long-term exposure, a statistically significant association with depression was identified for PM_2.5, NO₂, and BC. Specifically, an increase of 1 μg m⁻³ in the concentration of these pollutants was linked to an elevated risk of developing depression, with increases of 4.4% for PM_2.5, 1.1% for NO₂, and 14.7% for BC. Regarding short-term exposure, all evaluated pollutants except for CO were statistically significantly associated with depression; however, these associations were generally weak, with confidence intervals in the low end close to the threshold for statistical significance. The four meta-coefficients for post-partum depression exhibit statistical significance in two of these cases, but with considerable heterogeneity in all four cases.

Another important finding is the high heterogeneity between the studies included in these meta-analyses, which may be due to several different factors, but likely largely due to the methodological difficulties of diagnosing and quantifying depression in a completely objective way. The exposure assessment methods that were used in the studies included in the meta-analyses include a mix of land-use regression models, chemical transport models, satellite-based estimates, and monitoring station data, among others. Differences in these methods with respect to differences in spatial resolution may cause different levels of exposure measurement error, which in turn may contribute to the high heterogeneity of the effect estimates. Differences in exposure to indoor air pollutants may also have an impact, given that they were not considered in this meta-analysis. Additionally, the certainty of evidence for each meta-coefficient was considered to be very low based on five assessment criteria consisting of the following: 1) Risk of bias; 2) Inconsistency; 3) Indirectness; 4) Imprecision; and 5) Publication bias.

4.2 The effects related to different air pollutants

With regard to the air pollutants that have been analyzed in this meta-analysis, PM_2.5 constitutes the only pollutant that exhibits statistically significant meta-coefficients for both long-term exposure, short-term exposure, and post-partum depression. NO₂ exhibits statistically significant meta-coefficients for both long-term and short-term associations with depression, but not for post-partum depression. Additionally, the meta-coefficient associated with long-term exposure to black carbon shows the greatest increase per 1 μg m⁻³ among all pollutants included in this meta-analysis. This indicates that combustion-related air pollution appears to be most important in increasing the risk of depression, as NO₂ and black carbon constitute markers of traffic-related air pollution. From a general point of view, there are clear indications that exposure to combustion-related air pollutants has more serious health effects compared to pollutants from other sources, based on a review study including many studies where health effects from particles from different sources were compared and evaluated (139). Moreover, when the global health costs attributable to ambient PM_2.5 were modeled, for most countries, targeting combustion sources might offer greater health benefits compared to non-combustion sources (140). With regard to depression, and based on the results from this meta-analysis, it appears that combustion-related air pollution is most associated with depression, and may therefore also be most important in terms of intervention strategies.

4.3 Differences in results with respect to different geographical areas and urbanicity

In Tables 7–9 and in the supplementary section (Supplementary S4), we present results stratified by geographical region (North America, Asia, and Europe). For those outcomes where there is sufficient basis to conduct separate meta-analyses, the results from these continents can be compared. With regard to short-term exposure, several air pollutants are presented for both North America and Asia. For short-term exposure to CO, meta-analyses are presented for both North America and Asia (Supplementary Figures S4.1.2., S4.2.6) with meta-coefficients in the range of 1.0001 (95%: 0.9989–1.0014) and 1.0021 (95% CI: 0.9997–1.0045), respectively, where the point estimates differ by a factor greater than twenty. For short-term exposure to NO₂, meta-analyses are presented for both North America and Asia (Supplementary Figures S4.1.3, S4.2.7) with meta-coefficients in the range of 1.0006 (95%: 0.9996–1.0016) and 1.0072 (95% CI: 0.9988–1.0155), respectively, where the point estimates differ by a factor greater than ten. For short-term exposure to O₃, meta-analyses are presented for both North America and Asia (Supplementary Figures S4.1.4, S4.2.8) with meta-coefficients in the range of 1.0007 (95%: 0.9994–1.0019) and 1.0024 (95% CI: 1.0001–1.0047), respectively, where the point estimates differ by a factor greater than three. For short-term exposure to PM_2.5, meta-analyses are presented for both North America and Asia (Supplementary Figures S4.1.5, S4.2.9) with meta-coefficients in the range of 1.0012 (95%: 1.0006–1.0017) and 1.0011 (95% CI: 1.0007–1.0015), respectively, where the point estimates are almost the same.

For post-partum depression associated with exposure to NO₂, meta-analyses are presented for both North America and Asia (Supplementary Figures S4.1.6, S4.2.12) with meta-coefficients in the range of 1.0053 (95%: 0.9926–1.0180) and 1.0190 (95% CI: 0.9896–1.0484), respectively, where the point estimates differ by a factor greater than 3.5. For post-partum depression associated with exposure to PM_2.5, meta-analyses are presented for both North America and Asia (Supplementary Figures S4.1.8, S4.2.13) with meta-coefficients in the range of 1.0457 (95%: 0.9990–1.0924) and 1.0089 (95% CI: 0.9969–1.0209), respectively, where the point estimates differ by a factor just over five.

With regard to long-term exposure and depression, meta-analyses associated with exposure to NO₂, O₃, PM_2.5, and PM₁₀ are presented for both Asia, Europe, and North America. For long-term exposure to NO₂, meta-analyses are presented for both Asia and Europe (Supplementary Figures S4.2.2, S4.3.1) with meta-coefficients in the range of 1.0165 (95%: 0.9998–1.0333) and 1.0096 (95% CI: 0.9968–1.0223), respectively, where the point estimates differ by a factor of 1.7. For long-term exposure to O₃, meta-analyses are presented for both Asia and Europe (Supplementary Figures S4.2.3, S4.3.2) with meta-coefficients in the range of 1.0057 (95%: 0.9923–1.0191) and 1.0205 (95% CI: 1.0147–1.0262), respectively, where the point estimates differ by a factor of slightly below four. For long-term exposure to PM_2.5, meta-analyses are presented for both Asia, Europe, and North America (Supplementary Figures S4.2.4, S4.3.3, 4.1.1) with meta-coefficients in the range of 1.0433 (95%: 0.9987–1.0878), 1.0885 (95% CI: 1.0359–1.1411), and 1.0041 (95%: 1.0023–1.0059) respectively. For long-term exposure to PM₁₀, meta-analyses are presented for both Asia and Europe (Supplementary Figures S4.2.5, S4.3.4) with meta-coefficients in the range of 1.0131 (95%: 0.9948–1.0313) and 1.0285 (95% CI: 0.9808–1.0761), respectively, where the point estimates differ by a factor of slightly above two.

A general pattern when comparing the meta-coefficients divided into North America and Asia is that the meta-coefficients in Asia are in five out of the seven cases substantially larger in comparison with North America. When comparing Asia and Europe, the meta-coefficients are in three out of the four cases larger in Europe, but the differences are not as great as in the comparisons between North America and Asia. When comparing different regions, several factors can influence the results. When comparing the impact of a specific increase in air pollution concentration (in this study in most cases 1 µg m⁻³), differences in the baseline concentrations with respect to different geographical regions may play a role. Relatively higher baseline concentrations may imply a relatively smaller effect related to a specific increase (1 µg m⁻³) compared to places with relatively clean air and low concentrations of air pollutants. The air pollution concentrations in Asia are in general much higher in comparison with Europe and North America, with annual population-weighted-average values of PM_2.5 in China in the range of 51–66 µg m⁻³ based on the time period 2014–2016 (141). Differences in vulnerability and susceptibility to air pollution exposure among populations in different regions could also play a role. A general pattern with higher air pollution concentrations among groups with lower socioeconomic status has been shown in Europe and North America, while in China, no corresponding relationship, or in some cases even an opposite relationship, has been found (142). Lower socioeconomic groups can be assumed to be more vulnerable to air pollution exposure as their physical and mental health is generally worse compared to higher socioeconomic groups. Moreover, households with lower socioeconomic status have also been shown to have generally poorer indoor air quality (143). Consequently, the large differences between the meta-coefficients in different regions, but with the lack of any clear pattern, can be explained by differences in baseline air pollution concentrations, differences in indoor air pollution, as well as differences in the degree of exposure and vulnerability among different socioeconomic groups.

Several factors linked to living in an urban environment, including noise, emotional stress, and the absence of green areas, can influence the risk of depression. Although some studies included in this meta-analysis have included living in urban vs. rural environments as adjusting factors, the meta-coefficients may be affected by such confounding factors. Urbanicity appears to be associated with a higher prevalence of depression in developed countries, but not in developing countries (144). There is a tendency for people with depression to move from rural to urban environments due to better access to care and treatment, which in addition to the above-mentioned factors may explain the higher prevalence of depression in urban areas (144).

4.4 Differences in results depending on diagnostic methods and type of depression

Three different types of outcomes including depression associated with long-term exposure, depression associated with short-term exposure, and post-partum depression have been used in this meta-analysis. The results differ between these outcomes, with generally larger values of the meta-coefficients for long-term studies and post-partum studies, but with a larger proportion of statistically significant meta-coefficients for short-term studies. However, the meta-coefficients associated with short-term exposure are in general weak, with confidence intervals in the low end that are close to the threshold for statistical significance. The methods for diagnosing depression vary between these three outcomes. The diagnostic methods for the studies on long-term exposure constitute a broad spectrum including mostly either ICD-10 F31-F33 (ten studies), ICD-9 F31-33 (one study), CES-D (14 studies), PHQ-9 (five studies), PHQ-4 (one study), PHQ-2 (two studies), self-reported (six studies), DSM-IV (two studies), and prescription of antidepressants (three studies). The abbreviations mentioned above constitute diagnostic codes, diagnostic manuals, and self-assessment scales regarding depression. Since there are no reliable biomarkers that can measure and quantify depression in a consistent way, the diagnoses entail a certain subjectivity in terms of different methods and interpretations. This may also explain the large heterogeneity between different studies, which is particularly prominent in the long-term studies (Table 4).

Regarding the studies on short-term exposure and depression, the diagnostic methods are in many cases the same as for the long-term studies including among others (ICD-10) F31-F33, CES-D, and PHQ-9. However, a majority of the time-series studies included in the meta-analyses are based on hospital admissions, emergency department visits, and outpatient visits related to the depression diagnosis, which may make these more uniform in comparison to the long-term studies. This may at least partly explain the lower heterogeneity among the short-term studies compared to the long-term studies (Table 5).

Regarding post-partum depression, with a total of ten studies (nine included in meta-analysis), and thus significantly less statistical basis in comparison to the long-term and short-term studies, some diagnostic methods (ICD-10 F31-F33, CES-D, and PHQ-9) are similar to the long-term and short-term studies. The exposure duration in the post-partum studies refers to different gestation periods and post-partum periods which means several months, which is defined as long-term exposure according to our one-month dividing line between long-term and short-term exposure. Similar to the long-term studies, the heterogeneity among the post-partum studies is large (Table 6).

To sum up, the heterogeneity among the included studies in this meta-analysis is very high. The fact that depression is relational and hard to quantify in a consistent way creates some uncertainty compared to health outcomes that can be quantified more precisely with some reliable biomarkers. This means that high heterogeneity is to be expected as different diagnostic methods and individuals' subjective perceptions of their mood state can vary both between studies and even within the same studies. However, with a total number of 17 meta-coefficients (Tables 4–6), ten meta-coefficients are statistically significant, and with point estimates that are greater than one in 15 cases. Consequently, despite very high heterogeneity, there is a clear association between air pollution exposure and the occurrence and onset of depression.

4.5 Biological mechanisms behind air pollution exposure and depression

With results pointing at clear associations between air pollution exposure and depression, the biological mechanisms behind will be the next question to investigate. Unlike neurological disorders like e.g., Parkinson's and Alzheimer's disease, where air pollution exposure is associated with exacerbations (145), depression lacks any clear consensus in terms of pathophysiological mechanisms, with a more difficult starting point regarding the understanding of the molecular mechanisms (146). However, certain brain regions including the prefrontal cortex, hippocampus, amygdala, raphe nucleus, and locus coeruleus are considered to be particularly important when it comes to mood disorders (147, 148). There are several potential biological mechanisms behind depressive disorders. The HPA-axis (the feedback system between the hypothalamus, pituitary gland, and adrenal glands) constitutes a stress regulation system involving several hormones, neurotransmitters, and neuropeptides, and dysfunctions of this feedback system may play a crucial role in the pathogenesis of depression (149, 150). The hormones, neurotransmitters, and neuropeptides involved in the HPA-axis are linked to several metabolic processes (151, 152) and immunological processes (153, 154). Metabolic alterations and markers of metabolic syndrome have been shown to be associated with depression (155–162), probably mediated by insulin resistance and low-grade inflammation (163, 164). Associations between immunological processes and depression, possibly mediated by inflammation affecting the brain, have also been shown (165, 166).

In the context of air pollution, exposure can affect the above-mentioned mechanisms including the HPA-axis, metabolism, and immunological processes within the brain (167–169). Influence on the HPA-axis through the gastrointestinal tract microbiota pathway associated with exposure to PM_2.5 has also been proposed as a possible mechanism for neurological dysfunctions (170). Based on a systematic literature review on air pollution and depressive disorders, exposure to air pollution was consistently associated with neuro-structural and neuro-functional effects in terms of increased inflammation and oxidative stress, as well as changes in neurotransmitters and neuromodulators and their metabolites within multiple brain regions implicated in risk of psychopathology (23). Exposure to particulate matter has also been shown to have a particularly large impact on the hippocampus, with harmful effects including oxidative stress, microglial activation, neuronal death, and differential expression of genes (171).

The biological mechanisms behind air pollution exposure and depression have also been analyzed in animal models. Based on a study where rats received intranasal PM_2.5, they displayed depressive-like behaviors caused by stress effects on the endoplasmic reticulum in the hypothalamus, which caused impaired dopamine synthesis with dopaminergic dysfunctions (172). In another study, where mice were exposed to real-world PM_2.5 exposure, anxiety and depression-like behaviors were observed. This was caused by the dopamine signaling being perturbed with specific dopamine receptors as the potential target, and with phenyl-containing compounds binding to dopamine receptors interfering with dopamine signaling, contributing to mental disorders (173). Additionally, in a study where mice were exposed to PM_2.5 for a time period of several weeks, depression behavior was observed, probably related to pro-inflammatory and anti-inflammatory cytokines, as well as impact on protein signaling pathways in the hippocampus and olfactory bulb (174).

To sum up, the biological mechanisms behind air pollution exposure and depression are not fully understood, but different kinds of influence on metabolism, inflammation, and function of neurotransmitters within certain brain regions have been suggested.

4.6 Strengths and limitations of this study

A strength of this study is that it includes 90 selected studies based on a careful and rigorous review following the PRISMA guidelines. Each study has also been evaluated based on bias assessment, and each meta-coefficient was evaluated according to the “Grading of Recommendations, Assessment, Development and Evaluation” (GRADE). The high degree of heterogeneity for almost all air pollutants and depression outcomes may restrict the generalizability of our findings. A limitation of this meta-analysis is that indoor air pollution exposure has not been taken into account. Based on a systematic review and meta-analysis, exposure to air pollution from indoor sources, mainly originating from incomplete combustion of solid fuels and secondhand smoke, has been shown to be linked to an increased risk of depression (34). Differences in air pollution exposure in the indoor environment can have a large effect due to the fact that a large part of the time is spent at home, but these differences are not taken into account as the focus has only been on outdoor air pollutants. Studies that have focused specifically on non-exhaust related air pollutants have also not been included in this meta-analysis. Gender-related differences have also not been taken into account, but a recently published study has shown noticeable differences between the sexes when it comes to the impact of air pollution on dispended antidepressant medications (175). Finally, a great challenge when analyzing the impact of air pollution exposure, based on a large number of studies, is the lack of uniform methods for diagnosing and quantifying depression.

4.7 Future research needs and policy implications

In order to increase the generalizability of the results, future research in this area needs to focus on air pollution and depression based on strict and uniform methods for diagnosing depression. Although there are indications that combustion-related pollutants from traffic are particularly important when it comes to depression, the importance of specific particle-bound components needs to be further investigated. Moreover, the impact of non-exhaust emissions (mechanically generated particles) needs to be investigated. Gender-related differences in the effects of air pollution on depression would also need to be further addressed in future studies. However, despite some uncertainties regarding the link between air pollution exposure and depression, there are clear associations with statistically significant meta-coefficients. Consequently, from a policy point of view, the impact of air pollution on depression could be considered in health impact assessments and cost-benefit analyses that can form the basis for taking the most appropriate measures.

5 Conclusions

In this systematic literature review and meta-analysis, based on 90 epidemiological studies that have examined the association between air pollution exposure and depression, statistically significant meta-coefficients were found in ten out of the total 17 calculated meta-coefficients, indicating a clear association between air pollution exposure and depression. The associations were most pronounced for long-term exposure to PM_2.5, NO₂, and BC, indicating the importance of combustion-related air pollutants from traffic. However, high heterogeneity throughout almost all meta-analyses may restrict the generalizability of our findings.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenvh.2026.1809298/full#supplementary-material

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Footnotes

Edited by: Tadhg Eoghan MacIntyre, Maynooth University, Ireland

Reviewed by: Pierluigi Catapano, University of Campania Luigi Vanvitelli, Italy

Shenghao Yuan, Shanghai Jiao Tong University, China