Effects of climate and air pollution on rates of hospitalization for affective mood disorders in southern Brazil

Abstract

Climate change and air pollution are two of the most pressing global challenges, increasingly recognized for their implications on health. Thus, we aimed to explore the associations between climate, air pollution and monthly hospitalization to Affective Mood Disorders (AMD). Clinical data were collected from the Hospital Morbidity System of the Unified Health System (SUS) for the city of Porto Alegre, Brazil, from 2013 to 2023. Climate and air pollution data were obtained from the ERA5 and the CAMS Global Reanalysis. We use machine learning algorithms, conduct Structural Equation Modeling (SEM), and use Generalized Additive Models to detect the most important variable and to test direct and indirect effects. Insolation (in hours), visibility, and temperature from 15C to 25C, proxies for good weather conditions, decrease the values in monthly hospitalization rates for AMD. Also, when the values of the total column of methane (above 0.00955), ozone (above 0.0008) and PM increase (proxies for air pollution), the AMD rate also increases. Our findings highlight that climate, together with air pollution as a potential predictor of psychiatric admissions, particularly in populations exposed to reduced air quality or extreme weather conditions.

Introduction

Climate change and air pollution represent distinct yet interconnected global challenges, increasingly recognized for their profound implications on mental health¹. Climate change alters environmental conditions, influencing factors such as temperature extremes, natural disasters, and seasonal variations, which can affect psychological well-being². For instance, sunlight exposure is often associated with improved psychological well-being, whereas adverse weather conditions are linked to mood disturbances³. Separately, air pollution has been implicated in neuroinflammation and oxidative stress, mechanisms that may contribute to mood disorders and cognitive decline⁴. The burning of fossil fuels - emitting carbon dioxide, black carbon, and ozone precursors - and agricultural production - releasing methane - are key contributors to both global warming and the release of harmful air pollutants. Weather patterns play a crucial role in shaping the concentration, distribution, and transformation of these pollutants, determining whether they remain close to their sources, are dispersed, undergo chemical reactions, are transported over long distances, or are removed through dry or wet deposition⁵. In addition, natural pollutants such as pollen, windblown dust, and wildfire smoke are also climate-sensitive, further amplifying their potential impacts on human health⁶.

The adverse health effects of air pollution have been extensively studied since the mid-20th century⁷, yet many questions remain unanswered. According to the Organization for Economic Co-operation and Development (OECD) and the World Health Organization (WHO), air pollution is among the leading environmental causes of mortality, reducing its levels could alleviate the Global Burden of Disease, with an estimated net benefit of US$ 135,371 million, yielding significant economic and societal benefits⁸. Climate, in turn, affects exposure to air pollutants by shaping weather patterns, including temperature, precipitation, atmospheric water vapour, wind speed and direction, which influence atmospheric chemical reactions and thereby local and regional pollution concentrations⁹. These environmental conditions impact not only physical health but also mental health outcomes¹⁰. According to WHO data, mental health disorders affect approximately 450 million people worldwide, with expenditures related to mental health estimated at $2.5 trillion in 2010¹¹. Grouped data from 204 countries and territories for 12 mental disorders from 1990 to 2019 reveal that they are the top ten leading causes of burden on the population, with a growing presence and no sign of global reduction since the 1990s¹².

Seasonal variations in mood are well-documented in the literature and provide valuable indicators of population mental health trends¹³. Previous research suggests that ambient temperature influences individuals’ stress levels, cooperation, aggressiveness, and energy levels¹⁴. Globally, studies have linked air pollution exposures and climate to individual mental health in two ways: directly, by damaging the nervous system and increasing anxiety and depression, and indirectly, by compromising overall physical well-being¹⁵, highlighting the complex interplay between environmental and psychological factors related to mood disorders¹⁶. These impacts are often exacerbated by socioeconomic factors like income inequality, poverty, and limited healthcare access, with significant additional economic and societal costs¹⁷. However, it is worth mentioning that contradictory findings have been obtained on how mood is affected by temperature and whether this would depend on the mental conditions of the individual, with some studies indicating that higher temperatures worsen mood¹⁸, while others report improvements or no discernible effect¹⁹.

Climatic adversities and extreme weather events are expected to increase in frequency and intensity due to climate change. From 1991 to 2022, Rio Grande do Sul recorded a high frequency of disaster events, ranking as the second-most affected state in Brazil. During this period, the state endured 3,913 hydro-geo-meteorological disasters, resulting in 100 fatalities, displacing over 700.000 individuals, and incurring economic losses exceeding 6 billion Brazilian reais²⁰. According to Clayton²¹, climate change is as much a psychological problem as an environmental issue, once various stressors are associated, including hazardous and potentially traumatic events such as floods, wildfires (and their consequent air pollution), heatwaves, and hurricanes. These disasters are often sudden, recurrent, and can lead to significant losses, damaging property and belongings while posing direct threats to one’s life and loved ones. Beyond the immediate impact of these events, additional stressors arise, such as overcrowding in shelters, disruptions to work and education, difficulties in accessing medical care and increasing mental health problems²¹.

Mental health consequences of such adversities are already evident in Brazil. According to the Unified Health System (SUS), mental health hospitalizations increased by 13% between 2008 and 2023 in the country²². By the study, the rise was particularly pronounced in the North (66%) and Central-West (48%) regions, while the Southeast experienced an 8% decline. The South and Northeast saw moderate increases of 29% and 14%, respectively. When looking locally, the capital of the South leads the diagnoses of depression and suicide cases among the capitals of Brazil, the Brazilian average is about 11.3% cases per 100,000 inhabitants, while the South has a rate of about 17.5%²³. Although global evidence links mental health risks to environmental factors such as high temperatures²⁴, poor air quality¹⁵, and limited urban greenery²⁵, the interplay of these factors remains underexplored in Brazil, particularly in the southern regions.

The WHO has highlighted mental health, which involves the risk for Affective Mood Disorders (AMD), as a growing global concern, projected to become a leading contributor to disease burden by 2030²⁶. Understanding how climate and air pollution influence not only severe outcomes like psychiatric outcomes but also more subtle mood variations is essential for developing early intervention strategies. The insights can inform early intervention strategies, helping policymakers better prepare for the public health impacts of a changing climate. Given the lack of region-specific data, this study aims to investigate the relationships between climate, air pollution, and hospitalization rates for affective mood disorders in Porto Alegre – a predominantly low-to-middle-income city in southern Brazil. Our goals include identifying the most influential variables and examining their direct and indirect effects on hospitalization rates, providing a foundation for more informed and effective public health policies.

Results

During 2013 to 2023, a total of 16,584 (average of 125.63) hospitalizations for affective mood disorders were recorded in Porto Alegre, 2019 being the year with the highest number of hospitalizations, with 1,747 (average of 145.58 by month).

Mapping variables according to their relevance

Figure 1 presents the importance score (IS) of climate and air pollution variables in predicting AMD rates of hospitalization according to the best-fitted RF algorithm (Table 1). The total insolation by month (IS > 1.96 ± 0.37 sd), total column methane (IS >1.56 ± 0.63 sd), temperature at 2 meters (IS > 0.32 ± 0.21 sd) and total column ozone (IS > 0.32 ± 0.14 sd) are the most important predictors of AMD rates.

Fig. 1.

Ranking of climate and atmospheric pollution variables according to their relative importance, as estimated by Random Forests, to predict the affective mood disorders rate of hospitalizations (AMD) in the city of Porto Alegre, South Brazil.

Table 1.

The result of each algorithm was based on the average of these samples and fitted with default parameters.

Algorithm	MSE		RMSE
Gradient Boosting	5.841	0.321	2.416
Random Forest	5.347	0.378	2.312
Support Vector Machine	7.436	0.135	2.726
Decision Tree	13.791	−0.602	3.713
XGBoost	6.862	0.202	2.619
LightGBM	5.765	0.330	2.401

Direct and indirect effects of air pollution on AMD rates

The multiple metrics of goodness-of-fit and regression parameters obtained from the best-fitting SEMs are summarized respectively, in the Supporting Information (Tables S1). The best-fitting SEM revealed both direct and indirect pathways linking climate and air pollution variables to AMD hospitalization rates, with path coefficients representing the relative strength and direction of associations, with positive values indicating increasing AMD rates and negative values indicating decreasing rates (Fig. 2). The SEM indicates that insolation, visibility, and temperature is positively correlated on AMD rates, as represented by solid arrows pointing toward the outcome. Among these, insolation showed the strongest direct association (standardized coefficient = 0.58), followed by temperature at 2 meters (0.46) and visibility (0.22), highlighting the role of meteorological conditions in shaping hospitalization. In contrast, air pollution and vegetation cover did not show direct effects on AMD, but instead influenced the outcome indirectly through their effects on meteorological mediators. For example, particulate matter and carbon monoxide were associated with insolation and visibility, while methane and ozone were associated with temperature. The vegetation cover was inversely related to visibility. These indirect pathways indicate that those characteristics may contribute to AMD hospitalization risk by altering weather-related conditions rather than acting independently.

Fig. 2.

Best-fitting structural equation models (SEM) for predicting the affective mood disorders rate of hospitalizations (AMD). Continuous and dashed red lines indicate positive and negative significant effects respectively. Numbers adjacent to lines indicate standardized path coefficient estimates. The higher the coefficient, the greater the magnitude of the relationship between variables.

The Generalized Additive Models confirm the effects of the total insolation by month, total column methane, temperature at 2 meters, and total column ozone (Fig. 3, Table S2). There is a negative relationship between AMD rate and total insolation; as the total insolation increases, AMD rate tends to decrease. Also, there’s a slight negative relationship between visibility and the AMD rate; as visibility increases, hospitalization rates appear to decrease. Besides, our results highlight the positive association between total column methane and particulate matter (PM) with the AMD rate. The temperature at 2 meters shows a negative relationship with the AMD rate, where higher temperatures are positively associated with a lower AMD rate, particularly beyond 20C. The plot for the total column ozone indicates a trend to increase above 0.0008 (kg m), which seems to also increase the mood rate. Initially, the AMD rate decreases with increasing ozone levels, but after exceeding the threshold, an increase in this parameter is observed.

Fig. 3.

Smoothing splines for predicting the affective mood disorders rate of hospitalizations (AMD) in response to climate and atmospheric pollution variables. The smoothing spline (continuous black line) and the 95% confidence bands (shaded areas) were obtained by fitting the mood rates as a function of climate and atmospheric pollution variables using Generalized Additive Models.

Discussion

Our study employed machine learning and structural modelling approaches to investigate the associations between climate and air pollution variables with affective mood-related hospitalization (AMD) rate in Porto Alegre. The machine learning analysis identified significant influences of total insolation by month, total column methane, temperature, and total column ozone as the primary predictors of AMD rate. These variables have significant associations, with evident linear and non-linear trends on the predicted variable. Specifically, an increase in the total insolation by a month above 100 hours was associated with a decrease in AMD rates. A similar trend was observed for the total visibility, where higher visibility values (from 45x upward) corresponded to lower values of the mood rates and to the temperature, where values from 15C to 25C were linked to a decline in AMD rates, suggesting that favourable climate variables indicating less extreme conditions may contribute to better mental health outcomes. Our results go straight ahead with data from 40 countries located at different latitudes showing that broad changes in the variation of insolation, especially considering the variations observed between winter and summer and between the minimum and maximum monthly values of these locations, were associated with an increase in the risk of suicide attempts in bipolar I disorder²⁷. Also, recent studies, including participants in the Gulf Long-Term Follow-up Study who completed validated screening questionnaires for depression, suggest a possible association between high levels of solar radiation and fewer symptoms related to mood disorder²⁸.

The results from our models also identify positive relations between methane and particulate matter with the predicted variable, and a slight increasing trend relation with ozone. Those proxies stand out as key pollutants in atmospheric pollution investigations. At present, both CO and particulate matter are the most harmful to human health in urban areas²⁹, with methane responsible for around 30% of the rise in global temperatures since the Industrial Revolution³⁰, and CO beign a common contaminant of indoor and outdoor environments, it can induce neurotoxicity and disrupt critical processes within the brain³¹. For instance, data from 75 cities in China indicate significant positive associations between higher concentrations of particulate matter (PM and PM) and nitrogen dioxide (NO) and increased daily hospital admission rates for depression³². Their results suggest that even short-term exposure to air pollutants exacerbates psychiatric conditions, increasing healthcare burdens. Similarly³³, found that a one-standard-deviation increase in PM concentration over the past month was associated with a 6.67% point rise in the probability of experiencing severe mental illness. These findings highlight air pollution as a potential predictor of psychiatric admissions, particularly in populations exposed to reduced air quality or extreme weather conditions.

Studies in the literature show that meteorological factors can influence everything from hospital admissions for mania to triggers for symptoms related to mood disorders. One of these factors may be the variation in levels of natural light, global radiation and sunshine duration, which may be associated with symptoms of Seasonal Affective Disorder (SAD)³⁴. A possible sunlight effect on mood-modulating is through the vitamin D pathway, which occurs when the human skin is exposed to sunlight, specifically ultraviolet B radiation (280–315 nm), which facilitates the transformation of 7-dehydrocholesterol into previtamin D3³⁵, insufficient levels of this vitamin could directly impact brain cell function and contribute to mood and depressive disorders. Besides vitamin D synthesis, sunlight exposure may also activate additional localized systems that are thought to play a role in mood regulation, including a cutaneous counterpart to the hypothalamic-pituitary-adrenal (HPA) axis³⁶, the serotonin/melatonin system³⁷, and the immune system³⁸. Nevertheless, due to climate change, thermal comfort in urban areas has been increasingly compromised³⁹, increasing temperatures, intensified by the phenomenon of urban heat islands, characterized by higher temperatures in large cities and metropolitan areas due to solar radiation absorption during the day and slow release at night, directly affecting the well-being of the population⁴⁰, also contributing significantly the influences on air pollution levels and air quality.

Nearly the entire global population, particularly in low- and middle-income countries, is exposed to air quality exceeding the WHO guidelines⁴¹, potentially leading to severe health consequences. In Brazil, most studies on air pollution and health impacts are concentrated in southeastern cities. For instance, a study conducted in São Paulo, the largest city in Latin America, assessed the health impact of PM exposure, revealing an association between PM and cardiovascular and respiratory mortality, with a total of 231,691.8 years of life lost, meaning an overall economic loss of more than US$14.1 billion⁴². In another study⁴³, reported that between 2,378 and 6,282 deaths could have been avoided in 2017 across 15 southeastern cities if PM levels had met WHO guidelines. In Rio Grande do Sul, atmospheric pollution is a major risk factor for non-communicable diseases. The atmospheric polycyclic aromatic hydrocarbons (a group of persistent organic pollutants (POPs)), in the Metropolitan Area of Porto Alegre were associated with respiratory and cardiovascular diseases⁴⁴, and at the moment, no studies have reported the impacts on mental health. Here, our results bring these associations as a major concern also for mood disorders. Despite air quality in Porto Alegre having improved concerning particulate matter, a recent study showed that air pollution levels (PM and PM) decreased by more than 45% between 2002 and 2021⁴⁵. On the other hand, ozone (O) levels increased significantly in the same period⁴⁵. Besides some positive results concerning particulate matter levels, efforts are still needed to ensure that the air quality related to other pollutants in Porto Alegre is suitable for human health.

In Fig. 3, we demonstrated the direct and indirect effects of air pollution on AMD. The city of Porto Alegre, besides having heavy vehicular traffic, receives pollutants from petroleum refineries, metallurgical industries, chemical industries, petrochemicals, and coal-fired thermoelectric power plants⁴⁶. The most recent Greenhouse Gas Emission Inventory for Porto Alegre⁴⁷ shows that 67.7% of emissions originated from the transportation sector, 23.0% from stationary energy sources, 8.8% from the waste sector, and 0.5% from agriculture, forestry, and other land uses. Humans, when exposed to this kind of pollution, may experience a range of neurobiological effects. Evidence suggests that air pollution can trigger inflammatory and oxidative stress responses in the central nervous system, largely driven by elevated levels of pro-inflammatory cytokines, factors that have been implicated as contributors to the onset of various psychiatric disorders⁴⁸. In parallel, disruptions in the hypothalamic-pituitary-adrenal (HPA) axis and elevated cortisol levels, commonly associated with stress, have also been linked to the development of mental health conditions⁴⁹. Furthermore, pollutants such as particulate matter are believed to contribute to damage in brain structures, including the white matter, cortical gray matter, and basal ganglia, with potential consequences for both cognitive function and emotional regulation⁵⁰. Studies suggest that long-term exposure, rather than short-term pollution spikes, may be more closely associated with these changes⁵¹. Other possible mechanisms for this decompensation of psychotic and mood disorders may lie in the neuroinflammation induced by continuous exposure to particulate matter. A study with mice exposed to PM for 4, 6, and 8 weeks led to behavioral changes related to Depression-like behavior⁵². A systematic review pointed out that biological pathways other than inflammation, such as changes in gene expression, oxidative stress and anatomical changes in nervous tissue, may also be implicated⁵³. Despite growing evidence, further research is needed to clarify the mechanisms by which both acute and chronic exposure to air pollutants influence the risk and progression of specific psychiatric disorders, particularly mood and psychotic conditions.

A few limitations to our study should be acknowledged. First, individual factors related to mood or psychotic exacerbation, such as stressful life events, socioeconomic status, drug consumption, or inflammatory events and physical comorbidities, could not be analyzed within the scope of our study. The availability of such data is often limited, particularly in low-income communities. In Brazil, the Unified Health System (SUS) operates through its Department of Information Technology (DATASUS), which is responsible for the development, research, and integration of information technologies to support healthcare systems. DATASUS manages the systems and applications used to collect, process, and disseminate health data; nevertheless, despite broad national recognition of its role in health surveillance, approximately 17.7% of the information remains unreported⁵⁴, posing a challenge for studies such as ours. Environmental factors such as noise levels or access to and use of green space were also not within the scope of our study^52,55. Moreover, indoor pollution, daily exposure to other air pollutants (such as automobile traffic, and/or tobacco use), which may have a large impact on the level of exposure, could also not be analyzed. It is important to emphasize that these types of studies, including the present study, are not able to establish causal relationships between the variables observed, but they do offer biological plausibility for more robust future studies to analyze whether acute and prolonged exposure to particulate matter from the environment, which is related to worsening air quality, can be a causal element in decompensation or increased incidence of mood disorders.

In addition, the use of monthly aggregated data for both exposures and outcomes limited our ability to capture short-term associations and extreme temperature events, such as heatwaves. While this temporal resolution is appropriate for assessing medium-term seasonal patterns and broader associations between climate variables, air pollution, and hospitalization rates, it does not allow for the evaluation of acute effects related to short-duration climatic extremes. Future studies using daily or sub-daily exposure data and event-based definitions of extreme temperature events are warranted to better characterize their short-term impacts on mental health outcomes.

To the best of our knowledge, for the first time in Brazil, we analyzed for a low- to middle-income city the relationships between recent exposure to climate and air pollution and hospitalization rates for affective mood disorders. Our multiple analyses, using machine learning and structural modeling techniques, show a significant association between total insolation, column methane, temperature, column ozone and PM.The major strengths of this study are the ten-year period of data collection, the adjustment for climate and air pollution variables, and the use of innovative methods for data analysis. The ecological time-series design also represents a strength of this study, as it enables the assessment of large-scale and cost-effective population-level associations between environmental exposures and psychiatric hospitalizations, capturing temporal patterns relevant to public health monitoring and policy-making. These results move us toward a better understanding of the pathophysiology of psychotic disorders and reinforce arguments for striking climate change and reducing pollutant emissions, especially in metropolitan areas.

Methods

The city of Porto Alegre

Porto Alegre is the capital of the state of Rio Grande do Sul in southern Brazil, as shown in Fig. 4. The city covers an area of approximately 495.4 km², encompassing 94 neighborhoods, and has a population of about 1,332,570, according to the 2023 demographic census⁵⁶. According to the climatological normal of 1991-2020, the average annual temperature is 19.9C, with monthly maximums exceeding 30C in summer, while the wet-bulb temperature averages 18C annually and rises to around 22C in the summer months, having a humid subtropical climate with a wide temperature range throughout the year and four distinct seasons⁵⁶. Rainfall is well distributed throughout the year, with an annual average of 1,494.5 mm of precipitation and 76.8% relative humidity. Due to its geographical position, winters have around 40% fewer hours of sunshine than summers, with an average annual insolation of 2,153 hours. As reported by the Brazilian Institute of Geography and Statistics, Porto Alegre’s Human Development Index (HDI) stands at approximately 0.805⁵⁶, classified as very high by the United Nations Development Program and is a low- to middle-income city.

Fig. 4.

Map of the study area. Upper left, location of Brazil within South America. Lower left, location of the state of Rio Grande do Sul within Brazil. Right, satellite image of the city of Porto Alegre, with municipal boundaries delineated in red. The map was generated in ArcMap® version 10.6.1⁶⁰.

Climate and air pollution

To investigate the influence of climate and air pollution on mood disorder hospitalization rates, a Geographical Information System (GIS) was employed to analyze spatial associations between the mood rate and a suite of variables. This study utilized remotely sensed data from multiple sources. Climate variables were obtained from the ERA5 Global Reanalysis, provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) through the Copernicus Climate Change Service (C3S). These included the leaf area index of high vegetation (m m), Leaf area index of low vegetation (m m), 2 meters dew point temperature (ºC), Temperature at 2 meters (ºC), Surface pressure (Pa), Total precipitation (m), High vegetation cover (0–1), Low vegetation cover (0–1) and Total cloud cover (0–1). Air pollution variables were sourced from ECMWF Atmospheric Composition Reanalysis 4 (EAC4) also maintained by C3S through the CAMS Global Reanalysis, and included particulate matter < 2.5 m (PM, kg m), particulate matter < 10 m (PM, kg m), as well as total column concentration of carbon monoxide (kg m), methane (kg m), ozone (kg m) and nitrogen monoxide (kg m). Additionally, total insolation (in hours) and the total mean visibility (x) were obtained from the Brazilian National Institute of Meteorology (INMET). All variables were retrieved monthly, spanning from January 2013 to December 2023.

Rates of hospitalizations of Affective Mood Disorders

The absolute number of monthly hospitalizations for affective mood disorders in Porto Alegre from 2013 to 2023 was extracted from the Hospital Morbidity System of the Unified Health System (SIH/SUS)⁵⁷. All hospitalizations within the broad category of affective mood disorders were considered, including maniac episodes, bipolar affective disorders, and depressive episodes, according to the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10 codes F30-F39)⁵⁸. Data on the resident population of Porto Alegre for the reference years were obtained from the epidemiological information system of the Unified Health System. The rates of hospitalizations of Affective Mood Disorders was calculated as follows:

1

where HR is the hospitalizations rate due to schizophrenia (per 100,000 inhabitants), Th umber of hospitalizations due to schizophrenia in Porto Alegre in the month/year of reference and Tp is Population residing in Porto Alegre in the year of reference.

Data analysis

The core statistical analyses were preceded by a data exploration phase aiming at detecting outliers and zero inflation. Collinearity was assessed using the Variance Inflation Factor among predictor variables (see Supporting Information Table S3).

Initially, we tested six different algorithms to detect the most important variable in predicting mood rate hospitalization concerning climate and atmospheric pollution variables, using four machine-learning techniques: Random Forest (RF), Gradient Boosting Machine (GBM), Support Vector Machine (SVM), and Decision Trees (DT). We also used two deep learning methods: a Light Gradient-Boosting Machine (LightGBM) and Extreme Gradient Boosting (XGBoost). In each model, 10-fold cross-validation was set, and data were split into 70% portions for calibration (training) and 30% for the testing, fitted with default parameters, then evaluated using the Mean Squared Error (MSE), the Coefficient of Determination (R²) and the Root Mean Squared Error (RMSE). The result of each algorithm was based on the average of these samples (Table 1), and from the best-performed algorithm, we identified the most important variables related to the mood rate. After the initial analysis and the identification of the most robust algorithm, we fine-tuned the Random Forest (RF) to improve its performance to find combinations of hyperparameters that yielded the best fit. The optimal combination of parameters included five splits a node, ensuring that each leaf node contains enough observations to prevent overfitting and improve generalization, the maximum depth of each tree was set to 10, to balance model complexity and prevent overfitting, bootstrap sampling was enabled, helping to build diverse trees and enhances the generalization capabilities of the model, while the total number of trees was set to 500. The variable importance scores (IS) were evaluated based on permutations of the area under the curve, which is a robust method to analyze unbalanced response variables. The variables are considered significant if their values are greater than the absolute value of the variable with the lowest negative value⁴.

To evaluate the direct and indirect effects of variables on hospitalizations through mood changes, inferred via mood rates, we conducted Structural Equation Modeling (SEM). This approach is particularly suitable for examining chain processes within systems, as it enables the definition and exploration of relationships or pathways between predictors⁵⁹. Initially, SEMs were fitted with mood rates as the response variable, based on climate and air pollution variables that showed significant importance scores in the Random Forest analysis (described above). To further explore the indirect effects of variables on mood rates and enhance model performance, the complexity of the SEM was increased by including predictors with non-negative Random Forest importance scores. The quality of model fit was assessed using chi-square tests along with other relevant metrics⁵⁹. The models were fitted using the diagonally weighted least squares method.

To account for spatial correlations and capture potential nonlinear relationships between mood patterns and predictor variables, we used Generalized Additive Models (GAMs) to analyze mood rates. GAMs extend linear models by incorporating smoothing functions linked to predictor variables, allowing for both linear and nonlinear associations to be effectively modelled. Unlike traditional linear models, GAMs provide multiple slopes rather than a single fixed slope, which facilitates the identification of varying responses of mood rates to different predictors. We fitted a GAM that included significant predictors with direct effects on mood rate, as determined through the SEM analysis. Model parameters were estimated using restricted maximum likelihood. Smooth splines were plotted to illustrate the relationship between mood rate and each predictor variable. Average slope values for the models were derived by calculating the mean of the slopes estimated by the GAM.

Author contributions

L.D.A. and J.N.S. conceived and designed the study. L.D.A. and V.A.A collected and preprocessed the data. L.D.A. performed the statistical analyses and machine learning modeling. J.V.S.F., K.B.L., M.F.D., and J.N.S. contributed to the interpretation of results and contextualization in the mental health literature. V.A.A., L.D.A., and J.N.S drafted the manuscript. All authors reviewed and approved the final version of the manuscript.

Data availability

The clinical and environmental datasets used in this study were obtained from publicly available sources. Mental health data related to the Affective Mood Disorders rate of hospitalizations (AMD) are available from the Hospital Morbidity System of the Unified Health System (SIH/SUS) at https://datasus.saude.gov.br/acesso-a-informacao/morbidade-hospitalar-do-sus-sih-sus/. The climate and air pollution data supporting this research is available via the European Centre for Medium-Range Weather Forecasts (ECMWF) through the Copernicus Climate Change Service (C3S) at https://www.ecmwf.int/. While these data are publicly accessible, the processed and merged dataset generated for the current analyses is available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-43916-8.