Effect of ambient temperature on overweight: direct and indirect pathways in a multiple mediation model

Efecto de la temperatura ambiente sobre el sobrepeso: vías directas e indirectas en un modelo de mediación múltiple

Efeito da temperatura ambiente no sobrepeso: vias diretas e indiretas em um modelo de mediação múltipla

Abstracts

Global warming and obesity are two major global challenges. The intricate relationships between climate, lifestyle factors, and their combined impact on overweight remain to be fully elucidated. We aim to estimate the effect of ambient temperature on overweight and examine the role of physical activity and fruit/vegetable consumption as indirect mediating pathways in Argentina. This cross-sectional study was conducted using data from the 2018 National Risk Factors Survey. Average air-temperature at 2m height (from ERA5 reanalysis data) was linked with individual-level data. A multilevel logistic generalized structural equation model was applied to examine the mediating effects of physical activity level and fruit/vegetable consumption on the association between ambient temperature and overweight, adjusted for sex, age, and educational level. The raw difference (95%CI) between the indirect effects was estimated using bootstrapping techniques (sample = 10,000, replicates = 5,000). An inverse association (direct effect) was observed between ambient temperature and overweight (c = -0.019, 95%CI: -0.034; -0.004). A one-unit increase in temperature was associated with higher log odds of fruit/vegetable consumption (a₁ = 0.020, 95%CI: 0.005; 0.035) and lower log odds of having moderate (a₂₁ = -0.015, 95%CI: -0.023; -0.007) and high (a₂₂ = -0.059, 95%CI: -0.068; -0.049) physical activity levels, compared to low fruit/vegetable consumption and low physical activity level, respectively. However, the mediating effect of high physical activity level on the temperature-overweight relationship was of greater magnitude. In conclusion, ambient temperature influences fruit/vegetable consumption and physical activity, indirectly affecting nutritional status, with physical activity acting as the key mediator. This underscores the need to prioritize climate change adaptation strategies that promote physical activity.

Keywords:
Climate; Overweight; Mediation Analysis; Physical Activities

El calentamiento global y la obesidad son dos grandes retos mundiales. Las complejas relaciones entre el clima, los factores relacionados con el estilo de vida y su impacto combinado sobre el sobrepeso aún no se han aclarado por completo. Nuestro objetivo fue estimar el efecto de la temperatura ambiente sobre el sobrepeso y examinar el papel de la actividad física y el consumo de frutas y verduras como posibles vías mediadoras indirectas en Argentina. Este estudio transversal se llevó a cabo utilizando datos de la Encuesta Nacional de Factores de Riesgo de 2018. La temperatura media del aire a 2m de altura (a partir de los datos de reanálisis ERA5) se vinculó con los datos a nivel individual. Se aplicó un modelo logístico multinivel de ecuaciones estructurales generalizado para examinar los efectos mediadores del nivel de actividad física y el consumo de frutas y verduras en la relación entre la temperatura y el sobrepeso, ajustado por sexo, edad y nivel educativo. Se calculó la diferencia bruta (IC95%) entre los efectos indirectos utilizando técnicas de bootstrapping (muestra = 10.000, repeticiones = 5.000). Se observó una asociación inversa (efecto directo) entre la temperatura y el sobrepeso (c = -0,019, IC95 % -0,034; -0,004). Un aumento de una unidad en la temperatura se asoció con una mayor chance logarítmica de consumo de frutas y verduras (a₁ = 0,020, IC95%: 0,005; 0,035) y una menor chance logarítmica de tener una actividad física moderada (a₂₁ = -0,015, IC95%: -0,023; -0,007) y alta (a₂₂ = -0,059, IC95%: -0,068; -0,049), en comparación con un bajo consumo de frutas y verduras y un bajo nivel de actividad física, respectivamente. Sin embargo, el efecto mediador del alto nivel de actividad física en la relación entre la temperatura y el sobrepeso mostró una magnitud mayor. En conclusión, la temperatura ambiente influye en el consumo de frutas y verduras y en la actividad física, lo que afecta indirectamente al estado nutricional, siendo la actividad física el mediador clave. Esto subraya la necesidad de centrar los esfuerzos en estrategias de adaptación al cambio climático que promuevan la actividad física.

Palabras-clave:
Clima; Sobrepeso; Análisis de Mediación; Actividad Física

O aquecimento global e a obesidade são dois grandes desafios globais. As relações intrincadas entre o clima, fatores de estilo de vida e seu impacto combinado no sobrepeso ainda precisam ser elucidadas. Nosso objetivo foi estimar o efeito da temperatura ambiente no sobrepeso e examinar o papel da atividade física e do consumo de frutas/vegetais como vias de mediação indireta na Argentina. Realizou-se estudo transversal utilizando dados da Pesquisa Nacional de Fatores de Risco de 2018. A temperatura média do ar a 2m de altura (dos dados de reanálise do ERA5) foi relacionada a dados individuais. Um modelo de equações estruturais generalizadas logístico e multinível foi aplicado para examinar os efeitos de mediação do nível de atividade física e do consumo de frutas/vegetais na relação entre temperatura e sobrepeso, com ajustes por sexo, idade e nível educacional. A diferença bruta (IC95%) entre os efeitos indiretos foi calculada, empregando técnicas de bootstrapping (amostra = 10.000, réplicas = 5.000). Associação inversa (efeitodireto) foi observada entre temperatura e sobrepeso (c = -0,019, IC95%: -0,034; -0,004). O aumento de uma unidade na temperatura foi associado a maiores chances logarítmicas de consumo de frutas/vegetais (a₁ = 0,020, IC95% 0,005; 0,035) e menores chances logarítmicas de ter nível de atividade física moderado (a₂₁ = -0,015, IC95% -0,023; -0,007) e alto (a₂₂ = -0,059, IC95%: -0,068; -0,049) quando comparado ao baixo consumo de frutas/vegetais e ao baixo nível de atividade física, respectivamente. No entanto, o efeito mediador do alto nível de atividade física na relação temperatura-sobrepeso apresentou magnitude maior. Em conclusão, a temperatura ambiente influencia o consumo de frutas/vegetais e a pratica de atividade física, afetando indiretamente o estado nutricional, com a atividade física atuando como o principal mediador. Isso reforça a necessidade de concentrar esforços em estratégias de adaptação às mudanças climáticas que promovam a atividade física.

Palavras-chave:
Clima; Sobrepeso; Análise de Mediação; Atividade Física

Background

Overweight is a significant risk factor for mortality and morbidity, responsible for about 3 million deaths each year worldwide ¹. The rising prevalence of overweight can be attributed to an increase in individual energy intake, coupled with a decline in energy expenditure resulting from an unhealthy diet and sedentary lifestyle. However, it is important to recognize that genetic, biological, and environmental factors may also contribute to this phenomenon ². Few studies have investigated how climate variables can impact body mass index (BMI), weight gain, and obesity ³.

The recognition of obesity and overweight as two of the greatest public health challenges of this century is accompanied by growing concern over the health impacts of climate change on populations. A global syndemic of obesity, undernutrition, and climate change has been identified ⁴, in which climate variations − particularly rising ambient temperatures − increase the risk of food insecurity by impacting food quality, availability, accessibility, and affordability ⁵. Climate variations also pose increasing risks during physical activity ⁶. However, the mechanisms underlying the complex relationships between ambient temperatures, diet, and physical activity, and their impact on the nutritional status of populations, have not yet been fully explained.

Recent studies have examined the relationship between temperature and food intake ^7,8. Some findings suggest that weather conditions can influence consumer decisions, indicating that high temperatures impair body thermoregulation, reducing food demand ⁷. Additionally, from a climate change perspective, it has been highlighted that long-term climate variations directly impact food systems and, consequently, global food production ⁹. These changes could directly influence human nutrition and health ¹⁰.

Environmental conditions can either provide or limit opportunities for physical activity. Studies have shown that extreme weather conditions, particularly heat stress, can reduce mobility and willingness to engage in physical activity ^10,11. The frequency and duration of extreme temperature events have been shown to directly and negatively impact physical activity patterns over the medium and long term ¹². However, a knowledge gap remains regarding the specific weather conditions that lead to less favorable physical activity patterns, and how these conditions interact with dietary habits.

Recently, epidemiological research has gone beyond the “black box” exposure-outcome paradigm ^13,14, and one effective strategy involves assessing mediation ¹⁵. Mediation analysis enables researchers to probe underlying mechanisms within a causal relationship, providing evidence to test path-specific hypotheses. Moreover, it offers valuable insights for designing and refining targeted interventions ¹⁶.

The application of mediation analysis would contribute to advancing scientific knowledge on the intersection between climate and health, which is particularly important for Latin America, where significant methodological gaps regarding climate analysis and health data have been recently identified ¹⁷. In this regard, it is noteworthy that Argentina has a valuable data source to explore the nutrition-climate relationship further: the National Risk Factors Survey (Encuesta Nacional de Factores de Riesgo − ENFR).

The ENFR is a nationwide survey conducted regularly to examine common and known risk factors for chronic diseases, including physical activity and fruit/vegetable consumption. While the relationships between behavioral factors and obesity have already been studied using this data source ¹⁸, the effects of temperature on them have yet to be explored.

In this context, we hypothesized that the effect of temperature on overweight could be partially mediated by physical activity levels and fruit/vegetable consumption. By using multiple mediator models, it is possible to simultaneously test and compare different indirect effects. Distinguishing the role of these mediating pathways could be particularly useful for helping decision-makers design public health policies ¹⁹.

This study aims to estimate the effect of ambient temperature on overweight and examine the role of physical activity and fruit/vegetable consumption pathways as mediating mechanisms in the Argentine adult population of 2018.

Methods

Study design and data

A nationwide cross-sectional study (n = 16,410) was conducted in Argentina using secondary data from the ENFR, conducted in the last quarter of 2018 by the Argentine Ministry of Health and the National Institute of Statistics and Census. The objective of the ENFR is to provide information on risk factors for noncommunicable diseases (NCDs) in Argentina. The ENFR used a probabilistic, multistage sampling design, the details of which have been described elsewhere ¹⁸. The initial sample comprised 29,224 individuals who completed the first part of the questionnaire, covering all jurisdictions nationwide. For the second stage, which included anthropometric measurements, a probabilistic subsample was drawn from 75% of the selected households. Out of a total of 23,556 households who completed the questionnaire, 16,577 individual interviews were conducted (non-response rate of 21%). This sample is representative of the population aged 18 years and older living in Argentine urban areas with at least 5,000 inhabitants ²⁰. For our analysis, we used a subset of 16,410 individuals aged 18 or older with complete and consistent anthropometric data. We excluded 167 cases that did not meet these inclusion criteria due to missing or inconsistent information.

The ENFR uses a questionnaire integrated into a mobile data collection device, administered by trained interviewers and health professionals responsible for physical measurements. This questionnaire includes individual anthropometric data (weight and height), sociodemographic characteristics at the household and individual level, and lifestyle-related data such as physical activity level (measured in total daily metabolic equivanlent − METs) and fruit/vegetable consumption. Height and weight were measured by a trained health personnel using portable equipment, following the World Health Organization (WHO) STEPS protocol endorsed by Argentine Ministry of Health and Social Development ¹⁸. In this study, BMI was categorized as overweight (BMI ≥ 25, no/yes) based on the WHO criterion ²¹. Physical activity was categorized into three levels according to the International Physical Activity Questionnaire (IPAQ) criterion (low, moderate, and high) ²². Fruit and vegetable consumption was assessed via a closed-ended question administered by a trained interviewer, asking the respondent about the number of portions consumed on a typical day. In this study, we used the consumption variable from the ENFR dataset, categorized as adequate (≥ 5 servings/day) or inadequate, in accordance with the dietary guidelines for the Argentine population ²³.

In addition, ambient temperature data were extracted using the Google Earth Engine (https://earthengine.google.com/) geospatial platform. Specifically, ERA5 images were filtered by date and location, and the average air temperature at 2m height (mean_2m_air_temperature variable) was obtained from the ERA5 Monthly Aggregate temperature data file for 2018 ²⁴. The raster layer representing annual means was imported into QGIS software (https://qgis.org/en/site/), where the mean pixel values were calculated for each province (n = 24 provinces/jurisdictions) to obtain the average annual temperature at the provincial level. This environmental information was combined with individual data, considering the geolocation of residences at the provincial level. Training was provided in the extraction and processing of environmental data, and the research team employed a double control system.

Ethical considerations

All procedures were performed in compliance with relevant laws and institutional guidelines. Since this study was based on already available and anonymized secondary data sources, ethical approval and informed consent were not required.

Statistical analysis

Data description

Data were described using means (standard deviation − SD) or proportions (%) and compared with the t-test or chi-squared test, as appropriate, and displayed in Table 1. Moreover, temperature data were graphically represented on a map (Figure 1). These and all subsequent analyses were conducted using Stata software, version 18 (https://www.stata.com).

Table 1
Characteristics of the study population by overweight status (n = 16,410).

Characteristics	Overweight				Total (n = 16,410)		p-value
	No (n = 5,268)		Yes (n = 11,142)
	Mean	SD	Mean	SD	Mean	SD
Age (years)	40.1	18.2	49.0	17.1	46.1	17.9	< 0.001
Variables	Count	%	Count	%	Count	%
Sex
Male	1,988	38	4,972	45	6,960	42	< 0.001
Female	3,280	62	6,170	55	9,450	58
Education
Incomplete primary school	365	7	1,282	12	1,647	10	< 0.001
Incomplete secondary school	1,656	31	4,250	38	5,906	36
Complete secondary school	2,165	41	3,847	35	6,012	37
Tertiary or university	1,082	21	1,763	16	2,845	17
Physical activity
Low	2,125	40	5,549	50	7,674	47	< 0.001
Moderate	1,985	38	3,855	35	5,840	36
High	1,158	22	1,738	16	2,896	18
Fruit/Vegetable consumption (≥ 5 servings/day)
No	4,812	94	10,168	94	14,980	94	0.345
Yes	299	6	676	6	975	6

• SD: standard deviation.

Figure 1
Distribution of the annual mean of 2m-air temperature (°C) by provinces in Argentina, 2018.

Parallel multiple mediation models

The generalized structural equation model (GSEM) was performed in sequential steps (Table 2). The units of analysis comprised 15,955 observations after excluding 455 cases with missing data on fruit and vegetable consumption. First, a variance component model examining the relationship between ambient temperature (exposure as a continuous variable) and overweight (outcome, no/yes), with a latent contextual variable at the provincial/jurisdictional level, was fitted. Subsequently, a second model including sex (male/female), age (continuous), and education (incomplete primary school, incomplete secondary school, completed secondary school, and university/tertiary degree) as covariates (Model 2, Table 2) and a third model incorporating physical activity level and fruit/vegetable consumption (Model 3, Table 2) were fitted. Goodness of fit was assessed using the Akaike information criterion (AIC) and Bayesian information criterion (BIC). A multiple mediator model (Model 4, Table 2) was then adjusted considering physical activity level and fruit/vegetable consumption as mediators. Survey weights were not applied due to technical limitations in fitting multilevel models with GSEM in Stata, prioritizing the hierarchical specification required for the analysis.

Table 2
Sequential models examining the relationship between ambient temperature (continuous variable) and overweight (yes/no).

Overweight (outcome)	Crude model		Model 1		Model 2		Model 3		Model 4
	Coefficient	95%CI	Coefficient	95%CI	Coefficient	95%CI	Coefficient	95%CI	Coefficient	95%CI
Mean 2m-air temperature (continuous)	-0.013	-0.027; 0.001	-0.013	-0.027; 0.001	-0.016	-0.032; -0.0003	-0.019	-0.034; -0.004	-0.019	-0.034; -0.004
Age (continuous)	0.030	0.028; 0.032			0.031	0.028; 0.032	0.029	0.027; 0.031	0.029	0.027; 0.031
Sex (male as reference)
Female	-0.290	-0.358; -0.223			-0.330	-0.399; -0.260	-0.365	-0.437; -0.293	-0.365	-0.437; -0.293
Education (university as reference)
Incomplete primary school	0.743	0.602; 0.883			0.299	0.152; 0.446	0.269	0.119; 0.420	0.269	0.119; 0.420
Incomplete secondary school	0.423	0.327; 0.519			0.338	0.239; 0.438	0.317	0.216; 0.418	0.317	0.216; 0.418
Complete secondary school	0.059	-0.033; 0.152			0.222	0.126; 0.319	0.208	0.110; 0.306	0.208	0.110; 0.306
Physical activity (low level as reference)
Moderate	-0.290	-0.365; -0.152					-0.168	-0.247; -0.090	-0.168	-0.247; -0.090
High	-0.567	-0.658; -0.476					-0.362	-0.460; -0.264	-0.362	-0.460; -0.264
Fruit and vegetable consumption (< 5 servings/day as reference)
≥ 5 servings/day	0.049	-0.092; 0.191					0.025	-0.123; 0.173	0.025	-0.123; 0.173
Moderate physical activity (outcome)
Mean 2m-air temperature (continuous)									-0.015	-0.023; -0.007
High physical activity (outcome)
Mean 2m-air temperature (continuous)									-0.059	-0.068; -0.049
Fruits and vegetables consumption (outcome)
Mean 2m-air temperature (continuous)									0.020	0.005; 0.035
Area-level clustering			Variance	SE	Variance	SE	Variance	SE	Variance	SE
Province			0.023	0.008	0.030	0.011	0.026	0.010	0.026	0.010

• 95%CI: 95% confidence interval; AIC: Akaike information criterion; BIC: Bayesian information criterion; SE: standard error.
• Note: Model 1: AIC = 20546, BIC = 20569.12; Model 2: AIC = 19464.92, BIC = 19526.56; Model 3: AIC = 18858.38, BIC = 18942.83; Model 4: parallel multiple mediation models.

The selected model (Figure 2) considers a direct effect of mean 2m-air temperature (X, continuous) on overweight (X→Y; c coefficient), with Y assumed to follow binomial distribution. Fruit/vegetable consumption and PA level (low/moderate/high) were identified as mediators (M stage: M₁ and M₂, respectively) in this relationship (Outcome stage: X→M₁→Y; a₁.b₁ for the indirect effect of M₁; and X→M₂→Y; a₂.b₂ for the indirect effect of M₂) (Figure 2).

Moreover, models were fitted using the median 2m-air instead of the mean, without changes in results.

Figure 2
Parallel multiple mediator model of the relationship between ambient temperature and overweight.

Metrics for comparing indirect effects

To compare the magnitude and effects of mediator variables, the metrics proposed by Coutts & Hayes ²⁵ were applied, particularly the raw difference, estimated as the difference between the indirect effects to be compared, in this case: a ₁ b ₁ − a _2i b _2i .

Since physical activity level (M₂) was categorized into three levels, two mediation pathways (i) were created corresponding to moderate (M₂₁) and high (M₂₂) physical activity levels, with low physical activity as the reference category. These pathways were then compared to the fruit/vegetable consumption (≥ 5 servings/day) mediation mechanism.

A bootstrap confidence interval was applied to the sampling distribution of the estimated direct and indirect effects, as well as to the difference between indirect effects ²⁶ and total effects. The 95% confidence interval (95%CI) was built by bootstrapping the raw difference with a 10,000-unit random sample and 5,000 replications. The histogram of the bootstrap distribution was graphically displayed. Evidence of a difference in magnitude between indirect effects was found if the confidence interval did not include zero ²⁵.

To further probe the effects of direct and indirect pathways on the outcome, obesity (BMI ≥ 30) was analyzed using the same methodological steps (Table 3).

Table 3
Coefficients and effects estimated, and comparisons of indirect effects using bootstrapping techniques for obesity as outcome.

Mediation paths	Coefficients	Value	SE	95%CI
X → M1	a1	0.020	0.007	0.005; 0.035
M1 → Y	b1	0.018	0.072	-0.123; 0.160
X → M21	a21	-0.015	0.003	-0.023; -0.007
M21 → Y	b21	-0.234	0.038	-0.310; -0.159
X → M22	a22	-0.059	0.004	-0.068; -0.049
M22 → Y	b22	-0.413	0.051	-0.514; -0.311
Direct effects
X → Y	c	-0.021	0.005	-0.032; -0.009
Indirect effects
X → M1 → Y	a1 b1	0.0003	0.002	-0.003; 0.004
X → M21 → Y	a21 b21	0.003	0.001	0.0007; 0.006
X → M22 → Y	a22 b22	0.024	0.004	0.015; 0.033
	Estimators	Value	Boot SE	95% Boot CI
Total effects
Direct plus Path 1	c + a1 b1	-0.020	0.005	-0.031; -0.009
Direct plus Path 21	c + a21 b21	-0.017	0.005	-0.028; -0.006
Direct plus Path 22	c + a22 b22	0.003	0.006	-0.009; 0.016
Metrics of difference
Raw difference M1 vs. M21	a1b1 − a21b21	-0.030	0.002	-0.001; 0.008
Raw difference M1 vs. M22	a1b1 − a22b22	0.024	0.005	0.014; 0.034

• 95%CI: 95% confidence interval; 95% Boot CI: 95% bootstrap confidence interval; Boot SE: bootstrap estimate of standard error; M₁: fruit/vegetable consumption of ≥ 5 serving/day; M₂₁: moderate level of physical activity; M₂₂: high level of physical activity; X: 2m-air temperature; Y: obesity.

Results

Table 1 shows participant characteristics (n = 16,410). The mean age (SD) was 46.1 (16.9) years. Women represented a higher proportion in both groups, without (62%) and with overweight (55%). Statistical associations were observed between overweight status and both sex and educational level (p < 0.001), with a higher percentage of individuals with low educational attainment (incomplete secondary school or less) among those with overweight compared to those without (50% vs. 38%, respectively) (Table 1). This is consistent with the adjusted models, in which higher education was associated with lower odds of overweight, while female sex showed an inverse association (Table 2).

A statistically significant association was also observed with physical activity: 60% of participants without overweight had moderate or high physical activity levels, whereas about half of those with overweight had low physical activity levels (Table 1). Most participants who consumed fewer than five servings/day of fruits/vegetables were overweight (94%), although these percentages were not statistically different. Figure 1 shows a map illustrating the unequal distribution of mean ambient temperature across Argentina.

Results from the parallel multiple mediator model, adjusted for age, sex, and education, are summarized in Table 2 and graphically displayed in Figure 3. A direct statistically significant negative effect of 2m-air temperature on overweight was observed (-0.019, 95%CI: -0.034; -0.004). A unit increase in 2m-air temperature increased the log odds of fruit/vegetable consumption (a₁ = 0.020, 95%CI: 0.005; 0.035) while decreasing the log odds of moderate (a₂₁ = -0.015, 95%CI: -0.023; -0.007) and high (a₂₂ = -0.059, 95%CI: -0.068; -0.049) physical activity levels, compared to low fruit/vegetable consumption and low physical activity, respectively. Moreover, consuming ≥ 5 servings/day of fruits/vegetables showed a non-significant increase of the log odds of overweight (b₁ = 0.025, 95%CI: -0.123; 0.173), while moderate (b₂₁ = -0.168, 95%CI: -0.247; -0.090) and high (b₂₂ = -0.362, 95%CI: -0.460; -0.264) physical activity categories were inversely and significantly associated.

Figure 3
Results of the multiple mediator model of the temperature-overweight relationship in Argentina, comparing mediation pathways of fruit/vegetable consumption and physical activity levels, using a generalized structural equation model (GSEM).

The specific indirect effect of a unit increase in 2m-air temperature for fruit/vegetable consumption had only a minimal effect on overweight (a₁b₁ = 0.0005; 95%CI: -0.003; 0.004) (Table 4). The specific indirect effects of 2m-air temperature for physical activity were a₂b₂₁ = 0.002 (95%CI: 0.0002; 0.0004) for moderate physical activity and a₂b₂₂ = 0.021 (95%CI: 0.012; 0.030) for high physical activity, both of which increased overweight, particularly high physical activity (Table 4). Additionally, in provinces with higher average annual temperatures, these effects were more pronounced (data not shown).

Table 4
Coefficients and effects estimated, and comparisons of indirect effects using bootstrapping techniques for overweight as outcome.

Mediation paths	Coefficients	Value	SE	95%CI
X → M1	a1	0.020	0.007	0.005; 0.035
M1 → Y	b1	0.025	0.075	-0.123; 0.173
X → M21	a21	-0.015	0.003	-0.023; -0.007
M21 → Y	b21	-0.168	0.040	-0.247; -0.090
X → M22	a22	-0.059	0.004	-0.068; -0.049
M22 → Y	b22	-0.362	0.049	-0.460; -0.264
Direct effects
X → Y	c	-0.019	0.007	-0.034;-0.004
		Value	Boot SE	95% Boot CI
Indirect effects
X → M1 → Y	a1 b1	0.0005	0.002	-0.003; 0.004
X → M21 → Y	a21 b21	0.002	0.001	0.0002; 0.0004
X → M22 → Y	a22 b22	0.021	0.004	0.012; 0.030
	Estimators	Value	Boot SE	95% Boot CI
Total effects
Direct plus Path 1	c + a1 b1	-0.019	0.005	-0.030; -0.007
Direct plus Path 21	c + a21 b21	-0.016	0.005	-0.027; -0.005
Direct plus Path 22	c + a22 b22	0.002	0.006	-0.010; 0.015
Metrics of difference
Raw difference M1 vs. M21	a1b1 − a21b21	0.002	0.002	-0.002; 0.006
Raw difference M1 vs. M22	a1b1 − a22b22	0.020	0.004	0.001; 0.030

• 95% Boot CI: 95% bootstrap confidence interval; Boot SE: bootstrap estimate of standard error; M₁: fruit/vegetable consumption of ≥ 5 serving/day; M₂₁: moderate level of physical activity; M₂₂: high level of physical activity; X: 2m-air temperature; Y: overweight.

Subsequently, the raw difference metric was applied to both indirect effects to test the equality of magnitude and value of the effects. The results of contrasting indirect pathways including fruit/vegetable consumption and moderate physical activity level were 0.002 (95%CI: -0.002; 0.006), whereas indirect pathways including fruit/vegetable consumption and high physical activity level were 0.020 (95%CI: 0.001; 0.030) (Table 4). Figure 4 shows the confidence intervals of the bootstrap distributions of the difference between fruit/vegetable consumption and moderate (Figure 4a) or high (Figure 4b) physical activity. As shown, the confidence interval of B (fruit/vegetable consumption vs. high physical activity) does not include zero.

Figure 4
Bootstrap distributions of the raw differences between indirect effects of fruit/vegetable consumption vs. moderate level of physical activity and fruit/vegetable consumption vs. high level of physical activity.

Discussion

The increasing prevalence of overweight and obesity worldwide requires a thorough examination of their multi-causal etiology ^27,28. While the relationship between nutrition, physical activity, and overweight has been extensively studied ²⁹, research exploring the relationship between climate variables and body weight remains limited, albeit with promising findings ³. Increased obesity rates and accelerated climate change represent global health challenges driven by lifestyle changes and human environmental modifications ³⁰, with significant economic implications ⁴. This study estimated the effect of ambient temperature on overweight in the Argentine population aged 18 and older of 2018, while exploring the role of physical activity and fruit/vegetable consumption as mediating pathways in a parallel mediator model. Our findings showed that the effect of 2m-air temperature on overweight operates simultaneously through direct and indirect effects. However, the indirect pathway involving high physical activity levels was stronger and differed in magnitude from the fruit/vegetable consumption pathway.

As previously reported ^31,32, our sample also revealed 67.9% of overweight participants, with higher prevalence in men than in women. Moreover, 38.14% of overweight people had not completed secondary education, similar to previous studies ^31,33, and greater impact of extreme temperatures on more vulnerable socio-educational groups was also noteworthy ³⁴. Moreover, from a syndemic perspective the negative effects of climate change are exacerbated by the interaction between chronic diseases and socioeconomic factors ^4,35.

The impact of ambient temperature on energy expenditure and dietary intake has been explored, but its relationship with overweight remains unclear ^36,37. We identified an inverse direct effect of 2m-air temperature on overweight, which contrasts with previous research. For instance, one study − which employed a design that did not consider mediating factors − suggested that elevate ambient temperature was linked to an approximately 5% increased likelihood of obesity ². Another study reported a similar impact, particularly among girls and women ³. It has been proposed that higher atmospheric ambient temperature reduces adaptive thermogenesis and leads to less physical activity ³⁸. Conversely, growing evidence indicates that humans expend energy on thermoregulation ³⁹, and recent studies have shown that adults have functional brown adipose tissue depots ^40,41,42,43 that respond to environmental temperature and exhibit seasonal variation. This result suggests a negative energy balance may occur at high ambient temperature, as appetite and intake decrease while energy expenditure increases ^42,44,45. Differences in the climatic conditions of the studied populations could explain these contrasting results. We found an inverse relationship between temperature and moderate and high physical activity levels (versus low physical activity), though some studies have reported opposite results ⁴⁶. A systematic review on climate change and physical activity indicated that this relationship is complex and multifaceted ⁴⁷, showing that rising temperatures were associated with a net increase in active travel and leisure-time physical activity worldwide. However, this trend may reverse once a certain temperature threshold is exceeded. This latter point aligns with our results, considering our cross-sectional design and the use of geographically aggregated environmental temperature data. Therefore, the temperature trend in our study can be interpreted not as a continuous increase, but as a transition from cooler to warmer geographic areas. Argentina has significant climatic variations across regions, with annual mean temperatures ranging from 4.6-14.8°C in the south to 17.4-23°C in the north.

Some studies have examined the ambient temperatures food intake relationship using purchase records. While no specific association was found with fruit/vegetable consumption, results indicated a general negative effect of temperature on food intake ⁷. Although the connection between food choices and climate is not yet fully understood ⁴⁸, it is well established that food choices are constrained by local food availability. The complexity of food selection process is related to its availability and accessibility ⁴⁹. The ongoing global climate change and diet-related health crises are linked to food systems, environments, and consumer choices. Understanding these choices and the effects of climate change is essential for transforming the food system to benefit human and planetary health. In Argentina, contextual variables can determine the accessibility and availability of fruits/vegetables within food systems ⁵⁰. Moreover, fruit/vegetable consumption is generally low, with few people reporting adequate consumption ^20,51. This could explain the lack of association found between fruit/vegetable consumption and overweight. Incorporating additional dietary components in future analyses would offer a more comprehensive understanding of the intersection between climate, diet, and overweight.

The mediator pathways of fruit/vegetable consumption and physical activity level in the temperature-overweight relationship showed a significant positive effect for the physical activity paths. It is difficult to compare our results with others since the literature using this technique in health sciences is limited. Other approaches have been applied instead, such as analyses of the interaction between physical activity and diet, which suggest that a combination of increased physical activity and a healthy diet reduces obesity risk ⁵². Nevertheless, it remains unclear which of these factors has the strongest effect on the risk of obesity ^53,54, as the association between obesity and sedentary behavior, low phyzical activity, alcohol consumption, and smoking habits is still inconclusive.

To characterize the two indirect pathways, we used the raw difference to assess the magnitude of their effects. High physical activity was stronger in magnitude than the fruit/vegetable consumption. The application of mediation analyses, combined with metrics to compare indirect effects, is a valuable methodological strategy for investigating the underlying mechanisms in the complex ambient temperature-health relationship. This approach also provides important insights for public policy interventions ⁵⁵.

It is worth noting that the total effect of both ambient temperature and physical activity, as well as fruit/vegetable consumption, is closer to zero than the direct effect, with mediator paths of opposite signs. This constitutes an inconsistent mediation, indicating that the indirect pathways counteract the direct effect ⁵⁶.

Some methodological issues must be considered. Our study addresses the climate-health relationship, emphasizing the limited development of science-based climate policies in South American countries and a lack of studies on nutrition ¹⁷. The ENFR employs a complex random sampling scheme and a hierarchical structure integrating environmental and individual data. However, this survey only collected fruit/vegetable consumption data focused on urban populations, making it challenging to thoroughly analyze diet-mediated pathways or understand how temperature-overweight associations might differ in rural areas. Moreover, the low prevalence of adequate fruit and vegetable consumption in the sample further constrained the detection of meaningful associations. A prospective design would strengthen this cross-sectional study to assess the causal relationship between exposure and outcome, considering mediation variables. The effect of ambient temperature may have been underestimated in certain regions, as satellite-derived air temperature data was aggregated at a broad geographic scale (province) and averaged annually to align with individual-level data. However, including a contextual latent variable in our final model allowed us to account for unobserved heterogeneity at the provincial scale.

Misclassification bias of the exposure, particularly for fruit/vegetable intake, could be present. However, if present, this misclassification would likely be non-differential and bias toward the null hypothesis. Ambient temperature, physical activity, and BMI were assessed in a reliable way. Although the reliability of measured ambient temperature, it was averaged by month and year. We conducted several analyses by grouping provinces by geographical and climatic zones but did not find significant estimated differences.

Moreover, there may be uncontrolled confounding factors, particularly large spatial variations in poverty levels, a known key risk factor for obesity ⁵⁷. To minimize this, we included “education” as a proxy for socioeconomic status. Nevertheless, residual confounding may persist, especially from unobserved factors such as characteristics of the built environment or disparities in access to healthy food and opportunities for physical activity.

Conclusion

Our study showed that ambient temperature impacts overweight by two indirect effects that contribute to a decrease in overweight. The indirect high physical activity pathway was stronger and different in magnitude from the fruit/vegetable consumption pathway. Although further research is required to consolidate these findings, our results indicate that the identified risk factors significantly impact the development of overweight. Given that obesity is a major public health issue due to its association with NCDs and all-cause mortality, public policies aimed at preventing overweight ought to consider climate change and prioritize the most vulnerable populations ⁵⁸. As highlighted, there is a need to improve food availability and access, especially in the most vulnerable regions ⁴⁹, while recognizing the health and environmental co-benefits of promoting sustainable diets and food systems. The evidence obtained also suggests placing greater emphasis on physical activity patterns, focusing interventions on promoting physical activity in warm climates by increasing opportunities and improving environmental quality through safety, amenities, and facilities ⁵⁹.

• Data availability
  The research data are available upon request to the corresponding author.

Acknowledgments

This research was supported by the Argentine National Science and Technology Agency, Scientific and Technological Research Fund (ANPCYT-FONCyT, grants PICT-2020-A-03283 and PICT-2019-04594).

References

• ¹ Ahmed B, Konje JC. The epidemiology of obesity in reproduction. Best Pract Res Clin Obstet Gynaecol 2023; 89:102342.
• ² Yang HK, Han K, Cho JH, Yoon KH, Cha BY, Lee SH. Ambient temperature and prevalence of obesity: a nationwide population-based study in Korea. PLoS One 2015; 10:e0141724.
• ³ Trentinaglia MT, Parolini M, Donzelli F, Olper A. Climate change and obesity: a global analysis. Glob Food Sec 2021; 29:100539.
• ⁴ Swinburn BA, Kraak VI, Allender S, Atkins VJ, Baker PI, Bogard JR, et al. The global syndemic of obesity, undernutrition, and climate change: the Lancet Commission report. Lancet 2019; 393:791-846.
• ⁵ Himmelgreen D, Romero-Daza N, Heuer J, Lucas W, Salinas-Miranda AA, Stoddard T. Using syndemic theory to understand food insecurity and diet-related chronic diseases. Soc Sci Med 2022; 295:113124.
• ⁶ Hartinger SM, Palmeiro-Silva YK, Llerena-Cayo C, Blanco-Villafuerte L, Escobard LE, Diaz A, et al. The 2023 Latin America report of the Lancet Countdown on health and climate change: the imperative for health-centred climate-resilient development. Lancet Reg Health Am 2024; 33:100746.
• ⁷ Zhang C, Liao H, Wang F, Li R. Ambient temperature and food behavior of consumer: a case study of China. Weather, Climate, and Society 2021; 13:813-22.
• ⁸ Halawa A. Effects of subjective feeling of environmental temperature on appetite and food intake in Chinese sociodemographic and ethnic groups. J Ethn Foods 2019; 6:28.
• ⁹ Romanello M, Walawender M, Hsu SC, et al. The 2024 report of the Lancet Countdown on health and climate change: facing record-breaking threats from delayed action. Lancet 2024; 404:1847-96.
• ¹⁰ Owino V, Kumwenda C, Ekesa B, Parker ME, Ewoldt L, Roos N, et al. The impact of climate change on food systems, diet quality, nutrition, and health outcomes: a narrative review. Front Clim 2022; 4:941842.
• ¹¹ Ferguson T, Curtis R, Fraysse F, Olds T, Dumuid D, Brown W, et al. Weather associations with physical activity, sedentary behaviour and sleep patterns of Australian adults: a longitudinal study with implications for climate change. Int J Behav Nutr Phys Act 2023; 20:30.
• ¹² Bernard P, Chevance G, Kingsbury C, Baillot A, Romain A-J, Molinier V, et al. Climate change, physical activity and sport: a systematic review. Sports Med 2021; 51:1041-59.
• ¹³ Skrabanek P. The emptiness of the black box. Epidemiology 1994; 5:553-5.
• ¹⁴ Susser M, Susser E. Choosing a future for epidemiology: II. From black box to Chinese boxes and eco-epidemiology. Am J Public Health 1996; 86:674-7.
• ¹⁵ Hafeman DM. A sufficient cause based approach to the assessment of mediation. Eur J Epidemiol 2008; 23:711-21
• ¹⁶ Susser E, Schwartz S, Morabia A, Bromet EJ. Psychiatric epidemiology: searching for causes of mental disorders. New York: Oxford University Press; 2006.
• ¹⁷ Palmeiro-Silva YK, Lescano AG, Flores EC, Astorga Y, Rojas L, Chavez MG, et al. Identifying gaps on health impacts, exposures, and vulnerabilities to climate change on human health and wellbeing in South America: a scoping review. Lancet Reg Health Am 2023; 26:100580.
• ¹⁸ Pou SA, Diaz MDP, Velázquez GA, Aballay LR. Sociodemographic disparities and contextual factors in obesity: updated evidence from a national survey of risk factors for chronic diseases. Public Health Nutr 2021; 25:1-13.
• ¹⁹ Ho SS, Peh X, Soh VW. The cognitive mediation model: factors influencing public knowledge of the H1N1 pandemic and intention to take precautionary behaviors. J Health Commun 2013; 18:773-94.
• ²⁰ Instituto Nacional de Estadística y Censos. 4º Encuesta Nacional de Factores de Riesgo. Resultados definitivos. https://www.indec.gob.ar/ftp/cuadros/publicaciones/enfr_2018_resultados_definitivos.pdf (accessed on 14/Jan/2025).
  » https://www.indec.gob.ar/ftp/cuadros/publicaciones/enfr_2018_resultados_definitivos.pdf
• ²¹ World Health Organization. Obesity and overweight. https://www.who.int/news-room/fact-sheets/detail/obesity-and-overweight (accessed on 14/Jan/2025).
  » https://www.who.int/news-room/fact-sheets/detail/obesity-and-overweight
• ²² Craig CL, Marshall AL, Sjöström M, Bauman AE, Booth ML, Ainsworth BE, et al. International physical activity questionnaire: 12-country reliability and validity. Med Sci Sports Exerc 2003; 35:1381-95.
• ²³ Ministerio de Salud de la Nación. Guías alimentarias para la población argentina: manual de aplicación. https://www.argentina.gob.ar/sites/default/files/bancos/2020-08/guias-alimentarias-para-la-poblacion-argentina_manual-de-aplicacion_0.pdf (accessed on 14/Jan/2025).
  » https://www.argentina.gob.ar/sites/default/files/bancos/2020-08/guias-alimentarias-para-la-poblacion-argentina_manual-de-aplicacion_0.pdf
• ²⁴ Copernicus Climate Change Service. ERA5: fifth generation of ECMWF atmospheric reanalyses of the global climate. https://developers.google.com/earth-engine/datasets/catalog/ECMWF_ERA5_MONTHLY (accessed on 14/Jan/2025).
  » https://developers.google.com/earth-engine/datasets/catalog/ECMWF_ERA5_MONTHLY
• ²⁵ Coutts JJ, Hayes AF. Questions of value, questions of magnitude: an exploration and application of methods for comparing indirect effects in multiple mediator models. Behav Res Methods 2023; 55:3772-85.
• ²⁶ Preacher KJ, Hayes AF. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behav Res Methods 2008; 40:879-91.
• ²⁷ Pou SA, Wirtz Baker JM, Aballay LR. Epidemia de obesidad: evidencia actual, desafíos y direcciones futuras. Medicina (B Aires) 2023; 83:283-9.
• ²⁸ Wirtz Baker JM, Pou SA, Niclis C, Haluszka E, Aballay LR. Non-traditional data sources in obesity research: a systematic review of their use in the study of obesogenic environments. Int J Obes (Lond) 2023; 47686-96.
• ²⁹ Shatwan IM, Almoraie NM. Correlation between dietary intake and obesity risk factors among healthy adults. Clin Nutr Open Sci 2022; 45:32-41
• ³⁰ Gildner TE, Levy SB. Intersecting vulnerabilities in human biology: synergistic interactions between climate change and increasing obesity rates. Am J Hum Biol 2021; 33:e23460.
• ³¹ Yusuf FM, San Sebastián M, Vaezghasemi M. Explaining gender inequalities in overweight people: a Blinder-Oaxaca decomposition analysis in northern Sweden. Int J Equity Health 2023; 22:159.
• ³² Jaacks LM, Vandevijvere S, Pan A, McGowan CJ, Wallace C, Imamura F, et al. The obesity transition: stages of the global epidemic. Lancet Diabetes Endocrinol 2019; 7:231-40.
• ³³ Hermann S, Rohrmann S, Linseisen J, May AM, Kunst A, Besson H, et al. The association of education with body mass index and waist circumference in the EPIC-PANACEA study. BMC Public Health 2011; 11:169.
• ³⁴ Vishwakarma PK, Vaghmare S, Banerjee S, Vishwakarma AP, Waghmare A, Agrawal A, et al. Association of public awareness and knowledge of climatic change with sociodemographic factors. Cureus 2023; 15:e47381.
• ³⁵ Di Ciaula A, Krawczyk M, Filipiak KJ, Geier A, Bonfrate L, Portincasa P. Noncommunicable diseases, climate change and iniquities: what COVID-19 has taught us about syndemic. Eur J Clin Invest 2021; 51:e13682.
• ³⁶ Johnson F, Mavrogianni A, Ucci M, Vidal-Puig A, Wardle J. Could increased time spent in a thermal comfort zone contribute to population increases in obesity? Obes Rev 2011; 12:543-51.
• ³⁷ Mavrogianni A, Johnson F, Ucci M, Matmot A, Wardle J, Oreszczyn T, et al. Historic variations in winter indoor domestic temperatures and potential implications for body weight gain. Indoor Built Environ 2013; 22:360-75.
• ³⁸ Koch CA, Sharda P, Patel J, Gubbi S, Bansal R, Bartel MJ. Climate change and obesity. Horm Metab Res 2021; 53:575-87.
• ³⁹ Daly M. Association of ambient indoor temperature with body mass index in England. Obesity (Silver Spring) 2014; 22:626-9.
• ⁴⁰ Nedergaard J, Bengtsson T, Cannon B. Unexpected evidence for active brown adipose tissue in adult humans. Am J Physiol Endocrinol Metab 2007; 293:E444-52.
• ⁴¹ Cypess AM, Lehman S, Williams G, Tal I, Rodman D, Goldfine AB, et al. Identification and importance of brown adipose tissue in adult humans. N Engl J Med 2009; 360:1509-17.
• ⁴² van Marken Lichtenbelt WD, Vanhommerig JW, Smulders NM, Drossaerts JMAFL, Kemerink GJ, Bouvy ND, et al. Cold-activated brown adipose tissue in healthy men. N Engl J Med 2009; 360:1500-8.
• ⁴³ Virtanen KA, Lidell ME, Orava J, Heglind M, Westergren R, Niemi T, et al. Functional brown adipose tissue in healthy adults. N Engl J Med 2009; 360(15):1518-25.
• ⁴⁴ Moellering DR, Smith Jr DL. Ambient temperature and obesity. Curr Obes Rep 2012; 1:26-34.
• ⁴⁵ Institute of Medicine; Committee on Military Nutrition Research; Marriott BM. Nutritional needs in hot environments: applications for military personnel in field operations. Washington DC: National Academies Press; 1993.
• ⁴⁶ Belloc I, Gimenez-Nadal JI, Molina JA. Weather conditions and physical activity: insights for climate emergency policies. Essen: Global Labor Organization; 2024. (GLO Discussion Paper Series, 1385).
• ⁴⁷ Franco Silva M, Favarão Leão AL, O'Connor A, Hallal PC, Ding D, Hinckson E, et al. Understanding the relationships between physical activity and climate change: an umbrella review. J Phys Act Health 2024; 21:1263-75.
• ⁴⁸ Rampalli KK, Blake CE, Frongillo EA, Montoya J. Why understanding food choice is crucial to transform food systems for human and planetary health. BMJ Glob Health 2023; 8:e010876.
• ⁴⁹ Pérez-Ferrer C, Auchincloss AH, Menezes MC, Kroker-Lobos MF, Cardoso LO, Barrientos-Gutierrez T. The food environment in Latin America: a systematic review with a focus on environments relevant to obesity and related chronic diseases. Public Health Nutr 2019; 22:3447-64.
• ⁵⁰ Ballesteros MS, Zapata ME, Freidin B, Tamburini C, Rovirosa A. Social inequalities in fruit and vegetable consumption by household characteristics in Argentina. Salud Colect 2022; 18:e3835.
• ⁵¹ Pou SA, Del Pilar Díaz M, De La Quintana AG, Forte CA, Aballay LR. Identification of dietary patterns in urban population of Argentina: study on diet-obesity relation in population-based prevalence study. Nutr Res Pract 2016; 10:616-22.
• ⁵² Pérusse L, Jacob R, Drapeau V, Llewellyn C, Arsenault BJ, Bureau A, et al. Understanding gene-lifestyle interaction in obesity: the role of mediation versus moderation. Lifestyle Genom 2022; 15:67-76.
• ⁵³ Silveira EA, Mendonça CR, Delpino FM, Souza GVE, Rosa LPS, Oliveira C, et al. Sedentary behavior, physical inactivity, abdominal obesity and obesity in adults and older adults: a systematic review and meta-analysis. Clin Nutr ESPEN 2022; 50:63-73.
• ⁵⁴ Lee KX, Quek KF, Ramadas A. Dietary and lifestyle risk factors of obesity among young adults: a scoping review of observational studies. Curr Nutr Rep 2023; 12:733-43.
• ⁵⁵ Palmeiro-Silva Y, Aravena-Contreras R, Izcue Gana J, González Tapia R, Kelman I. Climate-related health impact indicators for public health surveillance in a changing climate: a systematic review and local suitability analysis. Lancet Reg Health Am 2024; 38:100854.
• ⁵⁶ MacKinnon DP, Krull JL, Lockwood CM. Equivalence of the mediation, confounding and suppression effect. Prev Sci 2000; 1:173.
• ⁵⁷ Drewnowski A, Specter SE. Poverty and obesity: the role of energy density and energy costs. Am J Clin Nutr 2004; 79:6-16.
• ⁵⁸ Ward N, Nichols M, Moodie M, Brown V. Is climate change action present in obesity prevention policy? Aust N Z J Public Health 2023; 47:100015.
• ⁵⁹ Wirtz Baker JM, Aballay LR, Haluszka E, Nicles C, Staurini S, Lambert V, et al. Exploring the quality of urban green spaces and their association with health: an epidemiological study on obesity using Street View technology. Public Health 2024; 237:44-50.

Edited by

• Associate Editor
  Evaluation coordinator: Washington Leite Junger (0000-0002-6394-6587)

Data availability

The research data are available upon request to the corresponding author.

Publication Dates

• Publication in this collection
  01 Dec 2025
• Date of issue
  2025

History

• Received
  05 May 2025
• Reviewed
  28 July 2025
• Accepted
  18 Aug 2025