Space-time analysis of the effect of air pollution on children’s health

Abstracts

The study aimed to investigate the short-term association between air pollution and emergency treatments for respiratory diseases in children 0 to 6 years of age. This was an ecological space-time study in Greater Metropolitan Vitória, Espírito Santo State, Brazil. A Poisson regression general additive model (GAM) used the number of daily treatments for respiratory diseases as the dependent variable, and the independent variables were daily concentrations of air pollutants (PM₁₀, SO₂, NO₂, O₃, and CO), temperature, humidity, and precipitation. Average daily concentrations were used to make estimates for the entire metropolitan area and in loco analyses considering children residing in a 2km radius around 8 air quality monitoring stations. An increase of 10μg/m³ in the concentration of air pollutants increased the risk of emergency treatment for respiratory disease. In the overall area, for PM₁₀, the increase was 2.43%, 2.73%, and 3.29% in the cumulative values at 5, 6, and 7 days, respectively. For SO₂, the increase was 4.47% on the day of exposure, 5.26% two days later, and 6.47%, 8.8%, 8.76%, and 7.09% for the cumulative values at days 2, 3, 4, and 5 days, respectively. CO showed a significant association for residents around two stations, and O₃ for only one. Even within the limits set by the World Health Organization, the pollutants PM₁₀, SO₂, NO₂, and O₃ are associated with increased risk of treatment for respiratory diseases in children 0 to 6 years of age, and some effects were only identified when disaggregating by neighborhood, i.e., in loco, which allows capturing greater variation in the data.

Keywords:
Air Pollution; Respiratory Tract Diseases; Child

O objetivo foi investigar a associação de curto prazo entre a poluição do ar e atendimentos em emergências por doenças respiratórias, em crianças de 0 a 6 anos. Estudo ecológico, espacial e temporal realizado na Região Metropolitana da Grande Vitoria, Espírito Santo, Brasil. Utilizou-se o modelo aditivo generalizado (MAG) de regressão de Poisson, com a variável dependente o número diário de atendimentos por doenças respiratórias, e as variáveis independentes, concentrações diárias dos poluentes atmosféricos (MP₁₀, SO₂, NO₂, O₃ e CO), temperatura, umidade e precipitação pluviométrica. Por meio das médias diárias das concentrações, foram feitas estimativas para toda a região e análises in loco com a consideração de crianças residentes no entorno de 2km de oito estações de monitoramento da qualidade do ar. O incremento de 10μg/m³ nos níveis de concentração dos poluentes atmosféricos aumentou o risco de atendimento em emergência por doença respiratória. Na região geral, para o MP₁₀, o aumento foi de 2,43%, 2,73% e 3,29% nos acumulados de 5, 6 e 7 dias, respectivamente. Para o SO₂, o acréscimo foi de 4,47% no dia da exposição, 5,26% dois dias após, 6,47%, 8,8%, 8,76% e 7,09% nos acumulados de 2, 3, 4 e 5 dias, respectivamente. O CO apresentou associação significativa para residentes no entorno de duas estações, e o O₃ somente em uma. Mesmo dentro dos limites estabelecidos pela Organização Mundial da Saúde, os poluentes MP₁₀, SO₂, NO₂ e O₃ estão associados ao maior risco para atendimento por doenças respiratórias em crianças de 0 a 6 anos, e alguns efeitos só foram identificados nas localidades desagregadas por região, isto é, in loco, o que possibilita captar maior variabilidade dos dados.

Palavras-chave:
Poluição do Ar; Doenças Respiratórias; Criança

El objetivo fue investigar la asociación de corto plazo entre la contaminación del aire y la atención en urgencias por enfermedades respiratorias, en niños de 0 a 6 años. Estudio ecológico, espacial y temporal realizado en la Región Metropolitana de la Grande Vitória, Espírito Santo, Brasil. Se utilizó el modelo aditivo generalizado (MAG) de regresión de Poisson, con la variable dependiente que es el número diario de consultas por enfermedades respiratorias, y las variables independientes: concentraciones diarias de los contaminantes atmosféricos (MP₁₀, SO₂, NO₂, O₃ y CO), temperatura, humedad y precipitación pluviométrica. Mediante las medias diarias de las concentraciones, se realizaron estimativas para toda la región y análisis in loco, considerando a niños residentes en un entorno de 2km con 8 estaciones de monitoreo de la calidad del aire. El incremento de 10μg/m³ en los niveles de concentración de los contaminantes atmosféricos aumentó el riesgo de atención en urgencias por enfermedad respiratoria. En la región como un todo, en el caso del MP₁₀, el aumento fue de 2,43%, 2,73% y 3,29% en los acumulados de 5, 6 y 7 días, respectivamente. En el SO₂, el incremento fue de 4,47% durante el día de la exposición, 5,26% dos días después, 6,47%, 8,8%, 8,76% y 7,09% en los acumulados de 2, 3, 4 y 5 días, respectivamente. El CO presentó asociación significativa para residentes alrededor de dos estaciones, y el O₃ solamente en una. Incluso dentro de los límites establecidos por la Organización Mundial de la Salud, los contaminantes MP₁₀, SO₂, NO₂ y O₃ están asociados a un mayor riesgo en relación con la atención por enfermedades respiratorias en niños de 0 a 6 años, y algunos efectos sólo se identificaron en las localidades desagregadas por región, esto es, in loco, lo que posibilita captar una mayor variabilidad de los datos.

Palabras-clave:
Contaminación del Aire; Enfermedades Respiratorias; Niño

Introduction

Respiratory diseases are the principal cause of morbidity and mortality in children and can be caused or aggravated by exposure to air pollutants emitted on a large scale throughout the planet ^1,2. Emissions of particulate matter (PM), nitrogen oxides (NOx), volatile organic compounds (VOCs), sulfur dioxide (SO₂), and photochemical pollutants such as ozone (O₃) have increased in recent decades due to growth of the motor vehicle fleet and increasing industrialization ². Studies in large urban areas prove a significant association between levels of these pollutants and the number of emergency treatments and hospitalizations from respiratory causes ^{3,4,5,6,7,8,9,10,11,12,13,14}.

A recent systematic review on Brazilian children’s health and the environment identified only 17 time-series studies that correlated air pollution and respiratory causes, all conducted in Southeast Brazil. These studies showed a positive association between levels of pollutants and emergency treatments, hospitalizations, and decreased lung function ¹⁵. Two of these studies were conducted in Greater Metropolitan Vitória, Espírito Santo State, Brazil ^12,16.

Greater Metropolitan Vitória is predominantly urban and highly industrialized and is undergoing urban sprawl. Air quality is affected by the emission of pollutants from motor vehicles and large-scale industrial activities, in addition to the logistics sector, with the existence of a large port and airport complex ¹⁷. Emissions from these mobile and fixed sources directly influence the area’s air quality ¹⁷.

Detection of the association between concentrations of pollutants and respiratory conditions requires statistical techniques that allow isolating the effects of air pollution, since there may be various confounding factors such as temperature and humidity. Time-series models have allowed more precise analyses of such associations.

Epidemiological time-series studies generally use a single fixed monitoring station or the average of stations to represent the entire population’s exposure to pollutants. However, this approach may not reflect people’s true exposure ¹⁸. Thus, a more realistic analysis might consider a small geographic area close to the monitoring station ¹⁹. Still, a limited number of epidemiological studies use this type of assessment, although there is some evidence that exposure to classification error in time-series analysis tends towards a downward bias in estimates and in this sense does not limit the results’ importance for public health ¹⁹.

Thus, methodologies in which the areas covered by measurement are better discriminated spatially are important for obtaining more precise inferences. In this context, the current study aimed to assess, via GAM, the short-term relationship between the number of emergency treatments for respiratory problems in children under six years of age and levels of air pollutants in Greater Metropolitan Vitória, considering the temporal variables measured on site (area around monitoring stations) and the average between stations (overall average for Greater Metropolitan Vitória).

Methods

Study area

This was an ecological space-time study conducted in Greater Metropolitan Vitória from January 1, 2005, to December 31, 2010. Greater Metropolitan Vitória consists of 7 municipalities (Vitória, Vila Velha, Cariacica, Serra, Viana, Guarapari, and Fundão), with a territory of 2,318.917km² and approximately 1.7 million inhabitants, as one of the main urban and industrial development hubs in the State of Espírito Santo (Instituto Brasileiro de Geografia e Estatística. Censo populacional 2010. https://ww2.ibge.gov.br/home/estatistica/populacao/censo2010, accessed on 19/Oct/2018). The principal polluting activities in Greater Metropolitan Vitória include roadways, various industries (steel, pelletizing, mining, cement mills), ports, airports, and residential and commercial emissions ¹⁷.

Air quality monitoring in Greater Metropolitan Vitória is done by nine stations which jointly comprise the Automatic Air Quality Monitoring Network (RAMQAr, in Portuguese) managed by the Espírito Santo State Institute for the Environment and Water Resources ¹⁷. The spatial analysis considers each station’s area of influence, as proposed by Santolim ²⁰. The area corresponds to a circle around the station with a radius of approximately 2km. At the time of the study, eight of the nine RAMQAr stations (Figure 1) were functioning.

Figure 1
Neighborhoods with air quality monitoring stations.

Health outcome

Data on emergency treatments for respiratory conditions in children under 6 years of age were obtained from the emergency departments of two hospitals, one public and the other private, the Nossa Senhora da Glória Children’s Hospital and the Unimed Vitória Integrated Healthcare Center, respectively. Respiratory diseases were coded according to the 10^th revision of the International Classification of Diseases (ICD-10: J00-J99). We selected the treatments of children that lived in the neighborhoods located in the area of influence of each of the RAMQAr stations.

Environmental pollutants and meteorological variables

The concentrations of particulate matter with aerodynamic volume up to 10 microns (PM₁₀), SO₂, nitrogen dioxide (NO₂), O₃, and carbon monoxide (CO) for the target period were provided by the Espírito Santo State Institute for the Environment and Water Resources (IEMA), which is responsible for the nine RAMQAr stations: Laranjeiras, Carapina, Cidade Continental, Jardim Camburi, Enseada do Suá, Vitória-Centro, Ibes, Vila Velha-Centro, and Cariacica. These automatic stations continuously collect and analyze air samples and process the data as hourly averages, on-site and in real time. The measurement method for PM₁₀ is Tapered Element Oscillating Microbalance (TEOM), which measures the mass concentration continuously. Measurement of SO₂ uses the principle of ultraviolet ray fluorescence. Measurement of NO₂ uses a combination of dual crossflow modulation type, also used to measure CO and O₃, with the principle of chemiluminescence and the differential calculus method ¹⁷. We thus calculated the daily averages for PM₁₀ and SO₂ and used maximum hourly concentrations of NO₂ and maximum daily concentrations of the eight-hour moving averages for CO and O₃. All pollutants in all the stations were measured in micrograms per cubic meter (µg/m³).

The target meteorological variables were: temperature in degrees Celsius (ºC); relative humidity in percentages (%), recorded by the Carapina and Cariacica stations; and precipitation in millimeters (mm), recorded by the Carapina monitor, which are the stations that record these measurements. We used these data to calculate the arithmetic means of the recorded measurements of humidity and temperature (maximum, average, and minimum) and precipitation represented by daily rainfall. We also included for each day the temperature and humidity values from the current medical care day (lag0), the previous day (lag1), two days previously (lag2), and moving averages of the two and three previous days (mm01, mm02).

Missing data

Failures in monitoring of pollutants in the RAMQAr network during the period, both on single days and on consecutive days, caused gaps in the records of concentrations and were corrected by imputation, according to the methodology described by Junger ²¹. The estimates obtained with this method are explained by the spatial correlation between the different levels of the same pollutant in the different monitors and by the autocorrelation in the pollutant’s levels in the same monitor, over time. The temporal patterns are modeled with a Gaussian autoregressive integrated moving average (ARIMA) model. This algorithm is especially adapted to weather data with missing measurements in some monitors across a given region. The upper and lower limits in the imputation of missing data were the maximum and minimum concentrations, respectively, observed in the historical series for each pollutant’s concentrations.

Statistical analysis

The modeling strategy consisted of defining a central model with all the known information (trend, seasonality, days of the week, holidays, and meteorological conditions) in order to explain the variability in the number of treatments for respiratory diseases, except for the concentration of pollutants.

The choice of variables and covariables for the model was based on tests and diagnoses in each stage of the modeling process. The diagnoses were based on residual analysis and the Akaike criterion (AIC) ²².

Generalized additive model

The daily number of medical treatments represents a counting process, and the generalized additive model (GAM) with Poisson marginal distribution was the statistical tool used to estimate the shape of the curve in the relationship between the health outcome and air pollution ^5,13,14.

Let {Υ_t} ≡ {Υ_t}_t∈Z be a counting series, that is, y_t ∈ {0,1,...}. The conditional distribution Υ_t, given the past F_t-1 that contains all the available information thus far t - 1, is denoted by:

$p\left(y_{\mathrm{t};}{\mu }_t\right|F_{t-1})=\frac{e{-}^{{\mu }_t}{\mu }_t^{y_t}}{y_{\mathrm{t}!}}$ (1)

in which μ_t represents the expected value (average) of Y_t. Thus, given a sample Y,...,Y_n, consisting of “n” mutually conditional independent random variables belonging to “Y_t”, the conditional log-likelihood function is given by:

$l\left(\mu \right)=\sum _{t=1}^n\mathrm{l}\mathrm{n}p\left(y_{t;}{\mu }_t\right|F_{t-1})\propto \sum _{t=1}^n\left(Y_t\mathrm{l}\mathrm{n}{\mu }_t-{\mu }_t\right)$ (2)

in which the vector μ = (μ1,...,μ_n) depends on the parameters and the process {Y_t}. Let X_t = [X_1t,..., X_pt]^T be the vector of covariables in the dimension p in time t, in which T denotes the transpose, which can include past values of Y_t and other ancillary information such as pollutants and confounding variables (trend, seasonality, and meteorological variables, among others). In this study, the sequence X_1t,..., X_qt denotes the concentrations of pollutants PM₁₀, SO₂, NO₂, O₃, and CO, thus, q = 5, and X_(q+1)t,..., X_pt indicates the confounding variables in time t, (p > q).

The relationship between the vector is given by:

$\text{ln}\left({\mu }_t\right)=\sum \begin{array}{c}q\\ j=0\end{array}{\beta }_j{X}_{jt}+\sum \begin{array}{c}p\\ j=q+1\end{array}{f}_j\left(X_{jt}\right)$ (3)

in which (β₀,β), with β = (β1,..., β_q)^T is the vector of the coefficients to be estimated (β_j it is the j-th covariable coefficient) and fj is the smoothing function for the j-th confounding variable. In addition, β₀ indicates the curve’s intercept and is associated with X_0t= 1 for every t. The entire modeling process was performed in the R software (http://www.r-project.org) with the ARES package ²³.

The relative risk (RR) of a pollutant covariable Xi,j = 1,...,q is given as the relative variation in the expected count of respiratory disease events by the unit variation ξ in the covariable while maintaining the other covariables constant. According to Baxter et al. ²⁴, formula (8), the RR is given by:

${RR}_{X_j}\left(\xi \right)=\frac{E\left(Y\right|X_j=\xi ,X_i,i\ne j)}{E\left(Y\right|X_j=0,X_i,i\ne j)}$ (4)

For the Poisson regression, the RR does not depend on the x_i, i ≠ j values of the other covariables and can be expressed as:

${RR}_{X_1}\left(\xi \right)=\mathrm{e}\mathrm{x}\mathrm{p}\left(\widehat{{\beta }_j}\xi \right)$ (5)

For the GAM with Poisson marginal distribution, the RR and its approximate confidence interval (CI, or IC in Portuguese), at level of significance α of a covariable, X_j, j = 1...q, is estimated as follows:

${RR}_{X_j}{\left(\xi \right)}_{X_j}=\mathrm{e}\mathrm{x}\mathrm{p}\left(\widehat{{\beta }_j}\xi \right)$ (6)

$IC\left({RR}_{X_j}\right(\xi \left)\right)=\mathrm{e}\mathrm{x}\mathrm{p}({\widehat{\beta }}_j\xi \pm z_{\alpha /2}\left({\widehat{\beta }}_j\right)\xi )$ (7)

${\widehat{\beta }}_j$ is the estimated coefficient associated with pollutant Xj in a study with standard error $s\left({\widehat{\beta }}_j\right)$ and Z_α/2 is the quantile α/2 of the standard normal distribution. At level of significance α, the hypothesis to be tested is defined as H₀: RR_xj = 1 against H₀: RR_xj > 1, in which RR_xj = RR_xj(1), that is, RR of the unit variation in X_j. The rejection of H₀ implies statistically that the respective pollutant has a significant adverse effect on health.

In this study, calculations of the RR(xj) values correspond to an increase of 1,000μg/m³ in the CO levels and of 10μg/m³ for the other pollutants. The results are presented as percentage increases in the number of medical treatments and are calculated by the expression:

$\%RR\left(x\right)=\left(RR\left(x\right)-1\right)x100$ (8)

Lag

The biological manifestation of the effects of pollution on human health apparently display a behavior that shows a lag in relation to the individual’s exposure to the pollutants, that is, events that occurred on a given day may be associated with the pollution levels on that day and/or on previous days. Thus, according to previous studies ^25,26, we decided to investigate the association between the number of treatments for respiratory diseases and the pollution levels on the day of the emergency treatment (lag0) and on the previous days (lag1, lag2, lag3). The cumulative effect was assessed with the moving averages from two to eight days (MA01, MA02, MA03, MA04, MA05, MA06, MA07) ^27,28.

The project was approved by the Institutional Review Board of the Health Sciences Center of the Federal University of Espírito Santo, under number 04/11, on May 14, 2011.

Results

During the study period there were 46,421 emergency treatments recorded for respiratory diseases in children 0 to 6 years of age living in the area of influence of the eight RAMQAr monitoring stations. The average daily number of treatments in Greater Metropolitan Vitória was 21.19 (SD = 9.90), ranging from 1.72 to 4.84 per day, with higher monthly averages from March to June, corresponding to autumn and early winter. The highest number of treatments was in children living in the Enseada do Suá neighborhood.

Average temperature during the period varied from 20.85 to 29.36°C, and rainfall varied from 0mm to 117.80mm (average = 3.78mm) and relative humidity varied from 61.79% to 97.27% (average = 77.47%).

Concentrations of pollutants did not display a uniform behavior between the different RAMQAr stations. The highest average concentrations recorded in the period were: PM₁₀ (43.06μg/m³) in Cariacica, SO₂ (16.32μg/m³) in Enseada do Suá, O₃ (38.66μg/m³) in Ibes, and NO₂ (44.10μg/m³) and CO (1,730.91μg/m³) in Vitória-Centro. Table 1 shows the pollutants’ minimum, average, and maximum concentrations and 25%, 50%, and 75% quartiles and standard deviations (SD).

Table 1
Descriptive statistics of treatments for respiratory diseases, by daily averages of PM₁₀ (μg/m³), SO₂ (μg/m³), and NO₂ (μg/m³) and 8-hour moving averages of O₃ (μg/m³) and CO (μg/m³) in each Automatic Air Quality Monitoring Network (RAMQAr, in Portuguese), 2005-2010.

Municipalities	Mean	SD	Minimum	Maximum	p25	Median	p75
Carapina
ADR	2.31	2.10	0	17	1	2	3
PM10	23.02	7.96	5.75	88.25	18.08	21.67	26.50
SO2	-	-	-	-	-	-	-
Tmpmin	20.85	2.47	13.1	25.98	19.05	21.15	22.80
Tmpmed	24.43	2.45	17.00	30.80	22.63	24.41	26.36
Tmpmax	29.36	3.28	19.40	39.70	27.20	29.43	31.60
Humidity	77.47	6.21	61.79	97.27	73.11	77.10	81.44
Rainfall	3.78	11.31	0.00	117.80	0.00	0.00	1.80
Vila Velha-Centro
ADR	2.25	1.76	0	12	1	2	3
PM10	23.49	8.22	5.04	90.75	18.04	22.33	27.48
SO2	11.99	5.80	0.50	54.16	8.61	11.21	13.98
Tmpmin	20.85	2.47	13.10	25.98	19.05	21.15	22.80
Tmpmed	24.43	2.45	17.00	30.80	22.63	24.41	26.36
Tmpmax	29.36	3.28	19.40	39.70	27.20	29.43	31.60
Humidity	77.47	6.21	61.79	97.27	73.11	77.10	81.44
Rainfall	3.78	11.31	0.00	117.80	0.00	0.00	1.80
Cariacica
ADR	2.28	1.91	0	13	1	2	3
PM10	43.06	15.94	8.83	117.88	32.52	41.13	51.46
NO2	28.06	11.18	4.01	97.64	19.28	26.79	35.78
NO2 (maximum)	50.65	19.35	8.65	220.25	36.80	49.20	62.36
SO2	5.50	2.62	0.00	19.82	3.42	5.32	6.98
CO	609.79	266.86	87.28	3700.05	412.67	577.28	769.46
O3	24.78	9.32	4.50	73.35	18.33	23.76	30.45
O3 (maximum)	47.90	13.93	12.08	120.00	37.93	46.94	56.40
O3 (8h)	37.79	12.59	8.54	105.69	28.31	37.30	46.02
Tmpmin	20.85	2.47	13.10	25.98	19.05	21.15	22.80
Tmpmed	24.43	2.45	17.00	30.80	22.63	24.41	26.36
Tmpmax	29.36	3.28	19.40	39.70	27.2	29.43	31.60
Humidity	77.47	6.21	61.79	97.27	73.11	77.10	81.44
Rainfall	3.78	11.31	0.00	117.80	0.00	0.00	1.80
Enseada do Suá
ADR	4.84	2.86	0	19	3	4	6
PM10	29.39	9.18	7.46	83.58	23.21	28.13	34.46
NO2	23.69	8.00	5.23	81.40	18.06	22.75	28.35
NO2 (maximum)	44.10	14.29	9.04	198.80	35.51	42.90	51.40
SO2	16.32	7.91	1.92	49.81	10.49	15.02	21.99
CO	783.30	276.71	80.22	2250.68	580.37	740.89	944.23
O3	29.91	8.84	9.08	69.52	24.51	28.51	34.27
O3 (maximum)	49.05	13.12	10.18	115.03	40.30	46.81	56.08
O3 (8h)	38.66	10.99	9.40	94.62	31.45	36.72	44.79
Tmpmin	20.85	2.47	13.10	25.98	19.05	21.15	22.80
Tmpmed	24.43	2.45	17.00	30.80	22.63	24.41	26.36
Tmpmax	29.36	3.28	19.40	39.70	27.20	29.43	31.60
Humidity	77.47	6.21	61.79	97.27	73.11	77.10	81.44
Rainfall	3.78	11.31	0.00	117.8	0.00	0.00	1.80
Laranjeiras
ADR	2.05	1.83	0	12	1	2	3
PM10	32.9	11.39	7.45	106.88	25.21	31.6	38.86
NO2	22.07	7.17	3.36	59.38	17.12	21.16	25.73
NO2 (maximum)	41.23	14.85	10.00	115.27	31.19	38.82	47.7
SO2	12.61	5.79	2.07	52.61	8.59	11.61	15.93
CO	647.58	175.55	212.01	1750.02	529.60	619.98	724.33
O3	32.58	9.05	10.26	74.18	26.54	30.90	37.05
O3 (maximum)	52.46	16.34	18.15	139.43	41.26	48.89	60.26
O3 (8h)	43.34	12.98	14.34	106.38	34.60	40.55	50.01
Tmpmin	20.85	2.47	13.1	25.98	19.05	21.15	22.80
Tmpmed	24.43	2.45	17.00	30.80	22.63	24.41	26.36
Tmpmax	29.36	3.28	19.40	39.70	27.20	29.43	31.60
Humidity	77.47	6.21	61.79	97.27	73.11	77.10	81.44
Rainfall	3.78	11.31	0.00	117.80	0.00	0.00	1.80
Ibes
ADR	1.72	1.48	0	9	1	1	3
PM10	29.24	9.67	5.00	88.13	23.19	28.13	34.46
NO2	20.32	5.98	4.80	58.03	16.40	19.69	23.89
NO2 (maximum)	38.39	11.15	9.60	100.50	31.00	37.70	45.99
SO2	10.73	6.20	0.25	41.38	5.97	9.79	14.15
CO	657.24	267.07	155.40	2551.97	475.72	608.19	773.41
O3	40.15	12.23	13.29	102.33	31.36	37.87	47.19
O3 (maximum)	65.78	17.09	22.20	148.00	53.88	64.01	75.60
O3 (8h)	54.48	16.38	17.89	140.11	42.34	52.37	64.51
Tmpmin	20.85	2.47	13.10	25.98	19.05	21.15	22.8
Tmpmed	24.43	2.45	17.00	30.80	22.63	24.41	26.36
Tmpmax	29.36	3.28	19.40	39.70	27.2	29.43	31.60
Humidity	77.47	6.21	61.79	97.27	73.11	77.10	81.44
Rainfall	3.78	11.31	0.00	117.80	0.00	0.00	1.80
Jardim Camburi
ADR	2.24	1.82	0	12	1	2	3
PM10	26.95	8.06	5.46	78.08	21.50	25.92	31.63
NO2	24.91	7.85	0.92	56.83	19.06	24.02	30.17
NO2 (maximum)	42.07	12.76	1.17	127.04	32.50	40.94	49.98
SO2	14.15	7.49	0.96	54.70	9.73	12.35	16.12
CO
Tmpmin	20.85	2.47	13.1	25.98	19.05	21.15	22.80
Tmpmed	24.43	2.45	17.00	30.8	22.63	24.41	26.36
Tmpmax	29.36	3.28	19.40	39.7	27.2	29.43	31.60
Humidity	77.47	6.21	61.79	97.27	73.11	77.10	81.44
Rainfall	3.78	11.31	0.00	117.8	0.00	0.00	1.80
Vitória-Centro
ADR	3.5	2.39	0	15	2	3	5
PM10	26.09	7.23	6.08	70.42	21.51	25.08	30.08
NO2	29.87	10.24	6.56	109.68	22.56	28.77	36.28
NO2 (maximum)	55.78	15.27	13.1	168.35	46.38	54.71	63.83
SO2	15.77	6.32	2.62	44.63	10.85	15.29	19.85
CO	1730.9	715.30	451.73	4,814.20	1,201.60	1,611.22	2,152.20
Tmpmin	20.85	2.47	13.10	25.98	19.05	21.15	22.80
Tmpmed	24.43	2.45	17.00	30.80	22.63	24.41	26.36
Tmpmax	29.36	3.28	19.40	39.70	27.20	29.43	31.60
Humidity	77.47	6.21	61.79	97.27	73.11	77.10	81.44
Rainfall	3.78	11.31	0.00	117.80	0.00	0.00	1.8
Greater Metropolitan Vitória
ADR	21.19	9.9	2	64	14	20	26
PM10	29.27	7.73	7.86	75.9	24.42	28.66	33.59
NO2 (maximum)	45.37	11.05	19.06	114.93	37.31	44.47	52.26
SO2	12.44	3.11	4.89	26.48	10.07	12.16	14.57
CO	885.76	231.24	295.17	2141.53	724.83	866.60	1031.10
O3 (8h)	43.57	10.83	16.93	93.63	35.54	42.21	49.91
Tmpmin	20.85	2.47	13.10	25.98	19.05	21.15	22.80
Tmpmed	24.43	2.45	17.00	30.80	22.63	24.41	26.36
Tmpmax	29.36	3.28	19.4	39.70	27.20	29.43	31.60
Humidity	77.47	6.21	61.79	97.27	73.11	77.10	81.44
Rainfall	3.78	11.31	0.00	117.80	0.00	0.00	1.80

• ADR: daily number of treatments for respiratory diseases; SD: standard deviation; Tmpmin: daily minimum temperature; Tmpmed: daily medium temperature; Tmpmax: daily maximum temperature.
• Note: “humidity” represents the daily relative air humidity index, “rainfall” is daily precipitation.

The series of daily counts of treatments in the overall region was smoothed with a “spline” with 12 degrees of freedom, defined by the Akaike modeling criterion and residual analysis. Non-parametric analysis evidenced seasonality and a downward trend over time, confounding factors that were included in the modeling process. The period from autumn (March) to winter (June) displayed a seasonality with an increase in the number of treatments for respiratory diseases.

Figure 2 shows the goodness-of-fit analysis for the model described previously via the results from Poisson regression for estimation of the effect of PM₁₀ on the 6-day moving average for Greater Metropolitan Vitória.

Figure 2
Diagnosis of the central model for Greater Metropolitan Vitória, Espírito Santo State, Brazil.

The estimated coefficient for PM10 was 0.0032, with a standard error of 0.0015 and p-value of 0.0301, which we represent as [MA6, Greater Metropolitan Vitória] (0.0032; 0.00145; 0.0301).

The model’s fit to the data for the areas in loco resulted in [MA6, Vila Velha-Centro] (0.0086; 0.0036; 0.0173), [MA6, Jardim Camburi] (0.0114; 0.0045; 0.0112), [MA6, Cariacica] (0.0044, 0.0021, 0.0331), and [MA6, Carapina] (0.0116; 0.0043; 0.0074). The results for the other sites were not statistically significant.

With the values estimated by the models, we calculated the relative risks corresponding to the estimates and proceeded to the analyses as presented.

We observed no empirical evidence of the model’s poor fit, that is, the residuals are not correlated and are approximately normal. The periodogram also proved that the residuals presented characteristics of white noise. These residual graphic analyses provide the necessary support for the model’s good fit and thus for performing inferences. Thus, the quality of the fitted model was guaranteed by the empirical properties shown by the residuals.

We calculated the relative risks (RR) and confidence intervals (CI) for each model fitted to the station’s neighborhood and for the overall region for each pollutant. Table 2 shows the estimated relative risks for an increase of 10 in the concentration of each of the air pollutants for the overall region. When we analyzed the patterns shown in this table, the gradients clearly evidenced the significant effects of PM₁₀ on the treatments, for cumulative lags. For SO₂, in the simple lag, the RR values were significant for lag0 and lag2. NO₂, CO, and O₃ did not show statistical significance in the RR for Greater Metropolitan Vitória.

Table 2
Relative risk (RR) of treatments for respiratory diseases in children under 6 years of age for 10µg/m³ increases in PM₁₀ (μg/m³), SO₂ (μg/m³), NO₂ (μg/m³), O₃ (μg/m³), and CO (μg/m³) in Greater Metropolitan Vitória, Espírito Santo State, Brazil, 2005 to 2010.

Exposure	%RR	Lower	Upper	p-value
PM10
Current day	0.99	-0.50	2.50	0.19
Lag, 1 day	0.04	-1.35	1.46	0.95
Lag, 2 days	0.81	-0.58	2.23	0.25
Lag, 3 days	1.10	-0.29	2.51	0.12
Cumulative, 2 days	0.69	-1.02	2.42	0.43
Cumulative, 3 days	1.13	-0.82	3.12	0.26
Cumulative, 4 days	1.77	-0.44	4.03	0.12
Cumulative, 5 days	2.43	-0.05	4.97	0.05 *
Cumulative, 6 days	2.73	0.00	5.53	0.05 *
Cumulative, 7 days	3.29	0.31	6.36	0.03 *
Cumulative, 8 days	2.50	-0.67	5.77	0.12
SO2
Current day	4.47	-0.01	9.14	0.05 *
Lag, 1 day	4.20	-0.20	8.79	0.06
Lag, 2 days	5.26	0.81	9.90	0.02 *
Lag, 3 days	1.43	-2.89	5.94	0.52
Cumulative, 2 days	6.47	0.99	12.26	0.02 *
Cumulative, 3 days	8.80	2.55	15.44	0.01 *
Cumulative, 4 days	8.76	1.93	16.05	0.01 *
Cumulative, 5 days	7.09	-0.13	14.84	0.05 *
Cumulative, 6 days	3.70	-3.71	11.69	0.34
Cumulative, 7 days	2.23	-5.45	10.53	0.58
Cumulative, 8 days	0.00	-7.85	8.52	1.00
NO2
Current day	0.25	-0.89	1.40	0.67
Lag, 1 day	-0.90	-2.04	0.25	0.12
Lag, 2 days	-0.45	-1.56	0.67	0.43
Lag, 3 days	-0.16	-1.25	0.94	0.78
Cumulative, 2 days	-0.50	-1.90	0.93	0.49
Cumulative, 3 days	-0.77	-2.39	0.88	0.36
Cumulative, 4 days	-0.82	-2.63	1.01	0.38
Cumulative, 5 days	-0.91	-2.86	1.09	0.37
Cumulative, 6 days	-0.91	-3.02	1.25	0.41
Cumulative, 7 days	-0.37	-2.63	1.95	0.75
Cumulative, 8 days	-0.31	-2.72	2.16	0.81
O3
Current day	0.54	-0.54	1.62	0.33
Lag, 1 day	0.46	-0.59	1.52	0.39
Lag, 2 days	0.07	-0.97	1.13	0.89
Lag, 3 days	-0.32	-1.36	0.73	0.55
Cumulative, 2 days	0.74	-0.55	2.05	0.26
Cumulative, 3 days	0.69	-0.78	2.19	0.36
Cumulative, 4 days	0.45	-1.20	2.12	0.60
Cumulative, 5 days	0.34	-1.46	2.17	0.71
Cumulative, 6 days	0.92	-1.04	2.91	0.36
Cumulative, 7 days	0.66	-1.42	2.78	0.54
Cumulative, 8 days	0.60	-1.60	2.86	0.60
CO
Current day	1.83	-3.25	7.17	0.49
Lag, 1 day	-3.61	-8.21	1.23	0.14
Lag, 2 days	0.39	-4.37	5.39	0.87
Lag, 3 days	4.60	-0.26	9.70	0.06
Cumulative, 2 days	-1.69	-7.76	4.76	0.60
Cumulative, 3 days	-1.24	-8.30	6.37	0.74
Cumulative, 4 days	2.15	-5.97	10.98	0.61
Cumulative, 5 days	3.23	-5.74	13.05	0.49
Cumulative, 6 days	2.53	-7.11	13.17	0.62
Cumulative, 7 days	4.89	-5.72	16.70	0.38
Cumulative, 8 days	5.83	-5.60	18.65	0.33

Table 3 presents a summary of the RR values for the effects of each pollutant whose calculations showed statistical significance (the results in bold print are those with the highest magnitudes). For the pollutants PM₁₀ and SO₂, the relative risks were higher when estimated for each station separately, compared to the RR for the entire Greater Metropolitan Vitória, that is, the adjusted model in which the covariables correspond to the average concentrations. As seen previously, the effects of NO₂, CO, and O₃ for the aggregate area (Greater Metropolitan Vitória) did not show statistical significance in the RR, which can be considered a spurious result, since the effects of these pollutants for the disaggregated sites led to highly significant relative risks.

Table 3
Percentage increase and 95% confidence interval (95%CI) in emergency pediatric treatments for respiratory problems. Greater Metropolitan Vitória, Espírito Santo State, Brazil, 2005-2010.

Exposure	%RR	95%CI	p-value	RAMQAr
PM10
Cumulative, 5 days	2.43	-0.05; 4.97	0.05	Greater Metropolitan Vitória
Cumulative, 6 days	2.73	-0.00; 5.53	0.05	Greater Metropolitan Vitória
Cumulative, 7 days	3.29	0.31; 6.36	0.03	Greater Metropolitan Vitória
Lag, 1 day	4.49	1.45; 7.62	0.00	Laranjeiras
Cumulative, 2 days	4.5	1.04; 8.08	0.01	Laranjeiras
Cumulative, 3 days	5.17	1.35; 9.13	0.01	Laranjeiras
Cumulative, 4 days	4.66	0.60; 8.89	0.02	Laranjeiras
Lag, 1 day	4.66	0.48; 9.02	0.03	Carapina
Cumulative, 6 days	8.36	0.16; 17.23	0.05	Carapina
Cumulative, 7 days	12.27	3.15; 22.19	0.01	Carapina
Cumulative, 8 days	11.5	1.85; 22.06	0.02	Carapina
Cumulative, 6 days	10.59	1.79; 20.16	0.02	Jardim Camburi
Cumulative, 7 days	12.08	2.63; 22.40	0.01	Jardim Camburi
Cumulative, 8 days	11.82	1.87; 22.73	0.02	Jardim Camburi
Lag, 3 days	2.41	0.17; 4.71	0.03	Enseada do Suá
Current day	4.08	0.54; 7.76	0.02	Vitória-Centro
Cumulative, 6 days	6.58	-0.03; 13.64	0.05	Vitória-Centro
Cumulative, 7 days	7.13	0.01; 14.75	0.05	Vitória-Centro
Cumulative, 7 days	9.05	1.54; 17.12	0.02	Vila Velha-Centro
Cumulative, 8 days	9.01	1.19; 17.42	0.02	Vila Velha-Centro
Lag, 3 days	2.62	0.58; 4.71	0.01	Cariacica
Cumulative, 7 days	4.53	0.36; 8.88	0.03	Cariacica
Cumulative, 8 days	4.77	0.40; 9.34	0.03	Cariacica
SO2
Current day	4.47	-0.01; 9.14	0.05	Greater Metropolitan Vitória
Lag, 2 days	5.26	0.81; 9.90	0.02	Greater Metropolitan Vitória
Cumulative, 2 days	6.47	0.99; 12.26	0.02	Greater Metropolitan Vitória
Cumulative, 3 days	8.8	2.55; 15.44	0.01	Greater Metropolitan Vitória
Cumulative, 4 days	8.76	1.93; 16.05	0.01	Greater Metropolitan Vitória
Cumulative, 5 days	7.09	0.13; 14.84	0.05	Greater Metropolitan Vitória
Current day	10.68	1.09; 21.17	0.03	Laranjeiras
Lag, 2 days	9.71	1.68; 18.37	0.02	Jardim Camburi
Cumulative, 7 days	13.52	-0.24; 29.18	0.05	Jardim Camburi
Current day	8.7	1.12; 16.85	0.02	Vila Velha-Centro
Cumulative, 2 days	11.31	1.23; 22.40	0.03	Vila Velha-Centro
Lag, 3 days	11.9	0.04; 25.17	0.05	Vila Velha-Centro
Lag, 2 days	7.57	1.17; 14.38	0.02	Cariacica
NO2
Lag, 2 days	3.85	0.99; 6.80	0.01	Jardim Camburi
Cumulative, 6 days	3.56	0.35; 6.88	0.03	Vitória-Centro
Cumulative, 7 days	3.88	0.46; 7.42	0.03	Vitória-Centro
Cumulative, 8 days	4.32	0.70; 8.06	0.02	Vitória-Centro
CO
Lag, 3 days	1.78	-0.02; 3.62	0.05	Laranjeiras
Lag, 2 days	2.24	0.72; 3.79	0.00	Ibes
Lag, 3 days	1.97	0.46; 3.51	0.01	Ibes
Cumulative, 4 days	3.01	0.60; 5.48	0.01	Ibes
Cumulative, 5 days	2.96	0.34; 5.65	0.03	Ibes
Cumulative, 6 days	2.99	0.20; 5.86	0.04	Ibes
O3
Current day	3.23	0.17; 6.37	0.04	Laranjeiras
Cumulative, 4 days	4.13	-0.08; 8.51	0.05	Laranjeiras

• RAMQAr: Automatic Air Quality Monitoring Network; RR: relative risk.

The entire analysis was based on the comparison of RR for the different pollutants and spatial points and compared to an average

Discussion

This study analyzed the effect of air pollution on children’s health in Greater Metropolitan Vitória by estimating relative risk in a GAM. Concentrations of PM₁₀, SO₂, NO₂, and O₃, even complying with Brazil’s prevailing legislation and WHO guidelines, showed a significant association with increases in emergency treatments for respiratory diseases in children 0 to 6 years of age, proving that there is no limit on these pollutants that is safe for human health ¹.

The number treatments increased from March to June, corresponding to autumn to early winter. This expected increase is due to different factors such as low temperatures that predispose to aggravation of preexisting respiratory diseases, higher incidence of viral respiratory diseases, and increased concentration of primary pollutants determined by scarce rainfall and thermal inversion.

The neighborhood with the most children treated at the emergency departments was Enseada do Suá. This neighborhood is heavily impacted by mobile polluting winds, since it is close to roadways with intense traffic, including a bridge connecting the city of Vitória to Vila Velha, with up to 70,000 motor vehicles circulating per day, as well as fixed sources, mainly an industrial complex inside the urban grid. A recent study showed that the resident population in this neighborhood of Vitória reported the most discomfort and aggravation from sedimented dust ²⁹.

During the study period, the concentrations of pollutants varied between the different RAMQAr stations, which can be explained by each neighborhood’s specific polluting activities. The average pollutant levels in these stations are based on on-site data and are usually quite close and significantly different from the overall average. This empirical evidence shows that the local averages, E(μ^L _t), (L = 1,...,8) differ from the Greater Metropolitan Vitória average E(μ^R _t). This result is expected, since the process is not stationary in time, that is, E(μ^L _t) ≠ E(μ^R _t), for L=1,…,8, and justifies the proposed study in the sense of comparing the health outcome to the pollutants in spatially discriminated areas.

PM₁₀ and SO₂ were the pollutants that showed the most consistent association with the increase in treatments for respiratory problems. PM₁₀ showed a substantial effect on treatments in all the lags and in nearly all the neighborhoods, with the highest magnitude in Jardim Camburi and Carapina. This effect pattern is consistent with other studies. Samoli et al. ¹⁰, in Greece, found that an increment of 10μg/m³ in PM₁₀ and SO₂ was associated with increases of 2.54% and 5.98% in the number of hospitalizations from respiratory diseases, respectively. In Itabira, Minas Gerais, Brazil, an increase of 10μg/m³ in the concentration of PM₁₀ was associated with a 4% increase in emergency department treatments for respiratory diseases in children for lag0 and lag18. In Greater Metropolitan Vitória, Souza et al. ¹³ observed a significant relationship between the concentrations of all the pollutants monitored by the RAMQAr (PM₁₀, SO₂, NO₂, O₃, and CO) and the number of emergency treatments for respiratory problems in children up to 6 years of age, through hybrid modeling, vector autoregression (VAR), principal components analysis (PCA), and GAM.

We observed high concentrations of PM₁₀ in the Cariacica station, which can be explained by the station’s location inside the Espírito Santo Farm Produce Clearinghouse (CEASA/ES), which has intense internal truck traffic and is located close to two federal highways with heavy truck traffic (BR-262 and BR-101).

The higher effect of PM₁₀ and SO₂ in the Jardim Camburi, Carapina, and Laranjeiras neighborhoods can be explained by their proximity to heavily travelled roadways and the largest industrial complex in the state of Espírito Santo, both of which are potential sources of PM₁₀ and SO₂.

We found the highest RR for SO₂ around the Jardim Camburi and Laranjeiras stations. In Enseada do Suá, the pollutant with the highest average was SO₂, but due to the lack of available data, the RR was only calculated for PM₁₀ in this neighborhood, representing a shortcoming in the study.

The adjustments to the model for each of the neighborhoods around the RAMQAr monitoring stations allowed better observation of the effect of all the pollutants on the number of emergency treatments for respiratory diseases. Evidence has shown that in time-series studies, measurements from a central station, compared to statistical or dispersion models or satellite-derived estimates, can be an adequate exposure metric for pollutants distributed homogeneously over the area ³⁰, which does not apply to the current study.

Comparing the estimates when we simulated for the entire Greater Metropolitan Vitória to the analyses around the RAMQAr stations, some effects were only perceptible when the analyses were done in the sites disaggregated by neighborhood, signaling a greater effect compared to the estimate for the entire Greater Metropolitan Vitória. The explanation is that the analysis by the average for all the stations tends to smooth the data and thus decrease their variability, disguising some effects and underestimating air pollution’s effects on health. Wilson et al. ³¹ also concluded that consideration of intraurban variations in concentrations in epidemiological studies allows minimizing exposure errors and uncertainties in the relative risk, but they found that this only applies to long-term studies. However, using this approach we observed short-term effects of NO₂, CO, and O₃ on the number of treatments for respiratory diseases in young children, especially in the Jardim Camburi, Vila Velha-Centro, and Laranjeiras neighborhoods.

Despite the association between the various air pollutants and the risk of emergency treatment for respiratory diseases in children in Greater Metropolitan Vitória, we observed a downward trend in the levels of pollutants over time, which can be explained by better local air pollution control.

For future studies, other susceptible groups could be investigated in the same neighborhood in order to allow elaborating a complete picture of the acute effects of air pollution on the population’s health. As an alternative methodology, bootstrap techniques could be used to obtain intervals with the same precision but with a narrower sampling range. Another potential methodology is estimation of variance with heteroscedastic models. The GLARMA ³² and PINAR ³³ models, with greater structural complexity, are statistical tools that could be addressed in these studies.

The consistency of associations and the magnitude of the observed effects in the neighborhoods, even in an environment with pollution levels that comply with limits set by regulatory agencies, are extremely relevant to public health. The findings provide backing for measures to minimize health risks, besides contributing to environmental and urban health planning and improvement of public policies.

Acknowledgments

The authors wish to acknowledge the support of various research funding agencies: Brazilian National Research Council (CNPq), Graduate Studies Coordinating Board (Capes), Espírito Santo State Research Foundation (FAPES), Vitoria Science and Technology Support Fund (FACITEC), Espírito Santo State Institute for the Environment (IEMA), Nossa Senhora da Glória Children’s Hospital (HINSG), and the Unimed Vitória Integrated Healthcare Center (CIAS).

Referências

• ¹
  World Health Organization. Inheriting a sustainable world? Atlas on children's health and the environment. Geneva: World Health Organization; 2017.
• ²
  Holgate ST. Every breath we take: the lifelong impact of air pollution - a call for action. Clin Med (Lond) 2017; 17:8-12.
• ³
  Cifuentes LA, Vega J, Köpfer K, Lave LB. Effect of the fine fraction of particulate matter versus the coarse mass and other pollutants on daily mortality in Santiago, Chile. J Air Waste Manag Assoc 2000; 50:1287-98.
• ⁴
  Gouveia N, Mendonça GAS, Ponce de Leon A, Correia JEM, Junger WL, Freitas CU, et al. Poluição do ar e efeitos na saúde nas populações de duas grandes metrópoles brasileiras. Epidemiol Serv Saúde 2003; 12:29-40.
• ⁵
  Daumas RP, Mendonça GAS, Ponce de León A. Poluição do ar e mortalidade em idosos no município do Rio de Janeiro: análise de série temporal. Cad Saúde Pública 2004; 20:311-9.
• ⁶
  Bakonyi SM, Danni-Oliveira IM, Martins LC, Braga AL. Poluição atmosférica e doenças respiratórias em crianças na Cidade de Curitiba - PR. Rev Saúde Pública 2004; 38:695-700.
• ⁷
  Nascimento LF, Pereira LA, Braga AL, Módolo MCC, Carvalho Jr. JA. Effects of air pollution on children's health in a city in southeastern Brazil. Rev Saúde Pública 2006; 40:77-82.
• ⁸
  Braga ALF, Pereira LAA, Procópio M, André PA, Saldiva PHN. Associação entre poluição atmosférica e doenças respiratórias e cardiovasculares na cidade de Itabira, Minas Gerais, Brasil. Cad Saúde Pública 2007; 23 Suppl 4:S570-8.
• ⁹
  Chen R, Chu C, Tan J, Cao J, Song W, Xu X, et al. Ambient air pollution and hospital admission in Shanghai, China. J Hazard Mater 2010; 181:234-40.
• ¹⁰
  Samoli E, Nastos PT, Paliatsos AG, Katsouyanni K, Priftis KN. Acute effects of air pollution on pediatric asthma exacerbation: evidence of association and effect modification. Environ Res 2011; 111:418-24.
• ¹¹
  Almeida SP, Casimiro E, Calheiros J. Short-term association between exposure to ozone and mortality in Oporto, Portugal. Environ Res 2011; 111:406-10.
• ¹²
  Souza JB, Reisen VA, Santos JM, Franco GC. Principal components and generalized linear modeling in the correlation between hospital admissions and air pollution. Rev Saúde Pública 2014; 48:451-8.
• ¹³
  Souza JB, Reisen VA, Franco GC, Ispány M, Bondon P, Santos JM. Generalized additive models with principal component analysis: an application to time series of respiratory disease and air pollution data. J R Stat Soc Series B Stat Methodol 2018; 67:453-80.
• ¹⁴
  Nascimento AP, Santos JM, Mill JG, Souza JB, Reis Júnior NC, Reisen VA. Associação entre concentração de partículas finas na atmosfera e doenças respiratórias agudas em crianças. Rev Saúde Pública 2017; 51:3.
• ¹⁵
  Froes Asmus CI, Camara VM, Landrigan PJ, Claudio L. A systematic review of children's environmental health in Brazil. Ann Glob Health 2016; 82:132-48.
• ¹⁶
  Castro HA, Hacon S, Argento R, Junger WL, Mello CF, Castiglioni Júnior N. Air pollution and respiratory diseases in the Municipality of Vitória, Espírito Santo State, Brazil. Cad Saúde Pública 2007; 23 Suppl 4:S630-42.
• ¹⁷
  Instituto Estadual de Meio Ambiente e Recursos Hídricos. Relatório da qualidade do ar: Grande Vitória 2013. Vitória: Instituto Estadual de Meio Ambiente e Recursos Hídricos; 2013.
• ¹⁸
  Namdeo A, Tiwary A, Farrow E. Estimation of age-related vulnerability to air pollution: assessment of respiratory health at local scale. Environ Int 2011; 37:829-37.
• ¹⁹
  Zeger SL, Thomas D, Dominici F, Samet JM, Schwartz J, Dockery D, et al. Exposure measurement error in time-series studies of air pollution: concepts and consequences. Environ Health Perspect 2000; 108:419-26.
• ²⁰
  Santolim L. Rede otimizada para o monitoramento da qualidade do ar da Grande Vitória, Espírito Santo [Dissertação de Mestrado]. Vitória: Universidade Federal do Espírito Santo; 1991.
• ²¹
  Junger WL. Análise, imputação de dados e interfaces computacionais em estudos de séries temporais epidemiológicas [Tese de Doutorado]. Rio de Janeiro: Universidade do Estado do Rio de Janeiro; 2008.
• ²²
  Hastie TJ, Tibshirani RJ. Monographs on statistics and applied probability 43: generalizes additive models. New York: Chapman & Hall; 1990.
• ²³
  Junger W, Leon AP. Ares: environment air pollution epidemiology. A library for time series analysis. Vienna: The Comprehensive R Archive Network; 2011. (CRAN Archives).
• ²⁴
  Baxter LA, Finch SJ, Lipfert FW, Yu Q. Comparing estimates of the effects of air pollution on human mortality obtained using different regression methodologies. Risk Anal 1997; 17:273-8.
• ²⁵
  Zhang H, Niu Y, Yao Y, Chen R, Zhou X, Kan H. The impact of ambient air pollution on daily hospital visits for various respiratory diseases and the relevant medical expenditures in Shanghai, China. Int J Environ Res Public Health 2018; 15:E425.
• ²⁶
  Chen R, Yin P, Meng X, Liu C, Wang L, Xu X. Fine particulate air pollution and daily mortality. a nationwide analysis in 272 Chinese cities. Am J Respir Crit Care Med 2017; 196:73-81.
• ²⁷
  Katsouyanni K, Schwartz J, Spix C, Touloumi G, Zmirou D, Zanobetti A, et al. Short term effects of air pollution on health: a european approach using epidemiologic time series data: the APHEA protocol. J Epidemiol Community Health 1996; 50 Suppl 1:S12-8.
• ²⁸
  Qu Y, Pan Y, Niu H, He Y, Li M, Li L, et al. Short-term effects of fine particulate matter on non-accidental and circulatory diseases mortality: a time series study among the elder in Changchun. PLoS One 2018; 13:e0209793.
• ²⁹
  Melo MM, Santos JM, Frere S, Reisen VA, Reis Jr. NC, Leite MDFS. Annoyance caused by air pollution: a comparative study of two industrialized regions. International Journal of Environmental and Ecological Engineering 2015; 9:182-7.
• ³⁰
  Dionisio KL, Baxter LK, Burke LK, Özkaynak H. The importance of the exposure metric in air pollution epidemiology studies: when does it matter, and why? Air Qual Atmos Health 2016; 9:495-502.
• ³¹
  Wilson JG, Kingham S, Pearce J, Sturman AP. A review of intraurban variations in particulate air pollution: implications for epidemiological research. Atmos Environ 2005; 39:6444-62.
• ³²
  Davis RA, Dunsmuir WTM, Streett SB. Maximum likelihood estimation for an observation driven model for poisson counts. Methodol Comput Appl Probab 2005; 7:149-59.
• ³³
  Monteiro M, Scotto MG, Pereira I. Integer-valued autoregressive processes with periodic structure. J Stat Plan Inference 2010; 140:1529-41.

Publication Dates

• Publication in this collection
  07 Oct 2019
• Date of issue
  2019

History

• Received
  24 July 2018
• Reviewed
  20 Jan 2019
• Accepted
  05 Apr 2019