A system for evaluating the impact of noise pollution on the population’s health Sistema de apoio à avaliação de impactos da poluição sonora sobre a saúde pública Sistema de apoyo a la evaluación del impacto del ruido sobre la salud pública

The aim of this study was to develop a support system for the evaluation of noise pollution, applied to the central urban area of Rio Claro, São Paulo State, Brazil. Data were obtained from noise measurements and interviews with the population, generating the following indicators: equivalent sound level (Leq), traffic noise index (LTNI), and a participatory diagnosis (Dp), integrated through a fuzzy inference system (FIS). The proposed system allowed classifying the measurement points according to the degree of impact of noise pollution on the population’s health (IPS) in the study area. Impact was considered significant in 31.4% of the measurement points and very significant in 62.9%. The FIS can be adjusted to local conditions, allowing generalization and thus also supporting noise pollution evaluation and respective environmental noise management in other


Introduction
Studies on the effects of noise have identified it as a public health problem and one of the forms of pollution that most affects people 1,2,3,4,5 .However, due to its physical nature (propagating without causing permanent alterations), noise evaluation can be complex and subjective, especially concerning its impacts on health, often with psychosomatic and social manifestations 6,7,8,9 .
The integration of quantitative evaluation using field measurements and interviews with the population can thus serve as an alternative for the evaluation of environmental noise and its effects on health 10,11,12,13 .
Sound measurements form the basis for an objective analysis, providing quantitative indicators of noise levels 10,14,15,16 , while interviews allow a participatory diagnosis based on how the exposed population perceives the noise 17,18,19,20,21,22 .
Environmental noise evaluation is thus important for verifying noise levels, assessing compliance with maximum permitted levels, detecting evidence of health impacts, and providing orientation on prevention and control measures 12,15,19,20 .
This study aimed to develop a support system for the evaluation of impacts of noise pollution (SIPS) on the population's health, based on a case study in Rio Claro, São Paulo State, Brazil.

Materials and methods
The system's input variables were equivalent sound level (L eq ), traffic noise index (L TNI ), and participatory diagnosis (D p ).The output variable was the impact of noise pollution on health (I PS ).
To demonstrate the performance of the proposed system, an application was performed using data collected in 2007 in the central urban area of Rio Claro, a medium-sized city with 192,000d inhabitants, located 173km from the state capital (Prefeitura de Rio Claro.Síntese da leitura técnica do diagnóstico ambiental de Rio Claro,  2006.http://www.prefeiturarc.sp.gov.br/siterc2/iss/download.php,accessed on 16/Jan/2015; Fundação Sistema Estadual de Análise de Dados.Perfil municipal.http://produtos.seade.gov.br/produtos/perfil/perfilMunEstado.php,accessed on 16/Jan/2015).The downtown area of Rio Claro includes the headquarters of the municipal government, and although there are numerous residences, the area is occupied mainly by commercial activities and is thus classified as a mixed-use area.

Development of the model
To deal with the imprecision associated with reading noise levels (L eq and L TNI ) and the inherent uncertainty and subjectivity of interviews with the population (D p ), a fuzzy inference system (FIS) was constructed with the MatLab software (MathWorks, Natick, USA).
Constructed on the basis of expert consultation and expertise, the FIS simulates human reasoning to support decisions based on a given condition 23,24 , such as diagnoses and monitoring in the health area 25,26,27,28,29,30,31 .Construction of the FIS thus involves four principal stages: fuzzification, construction of the rules set, inference, and defuzzification 32 .
Fuzzification involved modeling the input and output variables using fuzzy sets, developed from reference values in the literature.In this sense, the linguistic classification of the variable L eq was based on Guedes et al. 33 , as well as the Standards of Reference (Nível de Critério de Avaliação -NCA) of the Brazilian Association of Technical Standards (ABNT) 34 and Langdon & Scholes 35 for L TNI .
The point of reference in the study area was 60 dB(A) 34 .According to Guedes et al. 33 , places can be considered slightly noisy when L eq is less than or equal to 65 dB(A), noisy between 65 and 75 dB(A), and very noisy when greater than 75 dB(A).
According to Langdon & Scholes 35 , the expected degree of annoyance is low L TNI for less than 65 dB(A), medium from 75 to 65 dB(A), high between 90 and 75 dB(A), and very high when greater than 90 dB(A).
In fuzzification of the D p , the percentage of interviewees who felt that noise pollution had affected their health was proposed as the criterion for defining the low, medium, and high degrees of perceived noise, and gradual transition between classes was based on the margin of error estimated by the Krejcie & Morgan method 36 , described below.Thus, for a margin of error of ±d, the region of uncertainty was defined around the limit (L) in the range [L-d, L+d].
Finally, the output variable was proposed by means of linguistic classes pertaining to the significance of the I PS , namely the following degrees of impact: insignificant (I), scarcely significant (SS), significant (S), very significant (VS), and extremely significant (ES).
Next, expert consultation was used to develop the rules set and included an urban planning architect with expertise in environmental comfort research, a physicist, and two engineers with expertise in studies on the diagnosis of noise pollution.Inference was developed with the Mam-dani method, which establishes a fuzzy relationship R(v,u) that maps the degree of association between the input (v) and output parameters (u).In this study, inference occurred according to rules that relate the evaluation criteria (L eq , L TNI , and D p ) to antecedents (A j ) and the significance of the impact as consequences (C j ), as expressed in equation ( 1): (1) Meanwhile, defuzzification was performed by the centroid method, which establishes compatibility φB(u) between output (u) and the concept modeled by fuzzy set B, as shown in equation ( 2): (2) Next, the defuzzified values were linearly normalized for the range [0, 10] by means of equation (3): (3) Where: x, generated value in the defuzzification; x min , lowest generated value in the defuzzification; x max , highest generated value in the defuzzification.

Noise level measurements
Data on sound measurements were collected in the study by Mochizuki 5 , using a digital sound pressure meter, Instrutherm brand, model DEC-470, class 2 (Instrutherm, São Paulo, Brazil), calibrated and adjusted to operating in the weighting circuit dB(A), best adjusted to the human ear's sensitivity.
The sampling perimeter was demarcated by streets 1 and 7 and avenues 9 and 10, totaling 54 blocks, in which the measurement points were located at the intersections of streets and avenues, distributed evenly in a grid consisting of 35 points (Figure 1).
Measurements under the influence of atypical sources were avoided, i.e., noisy events that were uncharacteristic of the location, such as tempo-rary or occasional construction work and natural phenomena like thunder and heavy rainfall.In the absence of these sources, measurements were taken at points 1.2 meters aboveground and at least 2 meters from rebounding surfaces, through 30 sound pressure level readings at 10-second intervals at each of the points, always preventing the effect of wind on the microphone by using a shield.
This procedure was performed twice nonconsecutively at each sampling point to attenuate atypical readings.Based on these measurements, the target indicators were L eq (A) and L TNI (A), shown by equations ( 4) and ( 5), respectively 35,38,39 : (4)   Where: L eq (A), noise level with energy equivalent A from the target period; f i , frequency of readings with intensity L i ; L i , instantaneous noise level read at each time interval, adopted as 10 seconds. ( Where: L TNI (A), level evaluated by the traffic noise index; L 10 , noise level exceeded 10% of the time; L 90 , noise level exceeded 90% of the time.
The results of these indices were analyzed and interpreted according to NBR 10,151 standard  (2000) for noise evaluation in inhabited areas, aimed at community comfort 36 and taking parameters from the literature into account.

Participatory diagnosis of impacts on health
The D p variable took into account the data collected in studies by Bressane et al. 7 using interviews with residents of the central urban area of Rio Claro, with previously prepared forms consisting of multiple-choice questions.In that study 7 , the sample needed for a ±0.05 margin of error was calculated with the statistical method proposed by Krejcie & Morgan 36 , according to the following equation ( 6): (6)   Where: s, sample size; X 2 , table value of the chi-square for 1 degree of freedom, assumed as 3.841; N, size of the study population; P, proportion of the population (equal to 0.50 for greatest variance); d, tolerated margin of error.
However, since the area in which the interviews were conducted is larger than that in which the noise measurements were taken, for the study's purposes only a proportion of the sample from the area common to both analyses (interviews and noise measurements) was considered, totaling 109 interviewees.Considering an estimated resident population of 1,800 in the area, the resulting margin of error was ±0.09.
To reach the number of interviewees, sampling was conducted such that residents from one out of five households were approached on one side of the street or avenue in the preselected blocks.When nobody answered the door, the adjacent households were selected, and when necessary the procedure was repeated on the opposite side of the street or avenue.
The questionnaire consisted of two questions on the participant's profile (age and gender) and a question with 5 items on noise perception, as follows: "Do you feel that the noise pollution has hurt your: (1) nighttime sleep or rest periods; (2) disposition or physical performance; (3) emotional condition or well-being; (4) concentration or mental performance; and/or (5) hearing or communication?".

Results
The expert consultation produced a base with 36 rules, shown in Table 1.
The support system for the evaluation of the impacts of noise pollution on health resulted in the architecture shown in Figure 2.
The modeling of variables with fuzzy sets aimed to establish a gradual transition between noise sets or conditions.For example, in the variable L eq (A), the NCR (60 dB) was defined as the upper limit for the "slightly noisy" state.From this level upward, the noise was classified as belonging to the "noisy" state, defined around 70 dB in the ]60-80 dB[ range.Next, the "very noisy" state began at the upper limit of the previous class (70 dB) and became certain for noise levels of 80dB or higher.Thus, the fact that sampling points belong to more than one class is the consequence of the gradual transition, evaluated in the fuzzy inference process.
Meanwhile, Table 2 shows the results of the interviews with the resident population in the study area, with the proportion of answers for each of the five target items: 66.4% of the interviewees felt that noise pollution had some impact on their health.
Noise measurements at the 35 sampling points totaled 5 hours and 50 minutes of environmental noise evaluation.Figure 3 shows the values and spatial distribution of L eq , L TNI and the result of the Table 1 Rules set in the fuzzy inference system (FIS).

Input variables
Output variable L eq (A) L TNI (A) D p I PS integrated analysis with D p by means of the FIS, as a map of the impact of noise pollution on health in that area.

Discussion
Based on L eq , all the measurement points exceeded the limit of 60 dB(A) established by Brazil's prevailing legislation for areas with mixed Cad. Saúde Pública, Rio de Janeiro, 32(5):e00021215, mai, 2016 Modeling of input data: equivalent sound level (L eq ), traffic noise index (L TNI ) and participatory diagnosis (D p ) -and; output data: impact of noise pollution on health (I PS ) using fuzzy sets.
ES: extremely significant; I: insignificant; S: significant; SS: scarcely significant; VS: very significant.commercial and administrative use.Still, the community's response can vary according to the amount exceeded.According to Gerges 40 , limits exceeded by about 10 dB(A) commonly lead to complaints in the community, and differences greater than 15 dB(A) can spark more energetic responses such as class action.In the area evaluated here, 52.8% of the measurement points exceeded the noise limit by 5 to 10 dB(A) and 36.1% exceeded the limit by 10 to 15 dB(A).
Considering L eq (A), 21 points (60%) had noise levels in the [60-70 dB[ range, which defines partial membership of the slightly noisy state, although lower than presence of the noisy state, starting at 65 dB, which occurred in 18 of the measurement points in this range.In the [70-80 dB[ range, the remaining 14 points (40%) also showed partial membership in the "noisy" state, but among these, 1 showed mainly the very noisy state.
Although L eq is the standard applied to noise evaluation aimed at community comfort 35 , in areas with lower average noise, the annoyance can be even greater due to wider sound range, which occurs when the difference between L 10 and L 90 (levels exceeded 10% and 90% of the time) is more significant 41 .An important source of such variation is traffic noise, which was evaluated as L TNI .
Based on L TNI , 7 of the points (20%) belong totally to the definition for high annoyance and another 5 (14%) for very high annoyance, while the rest fell into ranges defined as transitional.In this sense, 10 measurement points had greater membership for medium annoyance (29%), 12 points Distribution of equivalent sound levels (L eq ), traffic noise index (L TNI ), and mapping of the impact of noise pollution on health (I PS ).Rio Claro, São Paulo State, Brazil, 2007.ES: extremely significant; I: insignificant; S: significant; SS: scarcely significant; VS: very significant.
Thus, a large proportion of the points met the definition for high annoyance (54%) or very high annoyance (17%), and L TNI values greater than 100 dB(A) were observed at three intersections (avenue 4 x street 2, avenue 7 x street 5, and avenue 7 x street 7).The points with the lowest degrees of annoyance were mainly located along avenue 2 and street 1.At certain intersections evaluated as slightly or moderately noisy, the expected degree of annoyance was evaluated as high, or even very high, since in these cases the wide variation in noise levels and very high instantaneous peaks had a heavy impact on the population.
As observed during the measurements, the concentration of commercial and service establishments produced intense circulation, making vehicle traffic the main source of noise in the study area.
When asked about noise pollution and the effect on their sleep and rest, 70.6% of the interviewees reported a negative impact.Environmental sound is one of the most important synchronizers of sleep, which can become shallow or even interrupted, causing the loss of restorative sleep stages, with psychological and intellectual harm and negative impact on mood and creativity 42,43,44,45 .
Although at a lower proportion, 39.4% of subjects reported negative effects on their disposition or physical performance.Noise pollution can alter rhythmic breathing movements, heart rate, and blood flow and viscosity and lead to hypertension and decreased tissue oxygenation 3,46,47,48 .
In relation to emotional well-being, 73.4% of subjects reported excessive noise as a stress factor.Studies indicate that high noise levels can induce emotional instability and a tendency to hostility, intolerance, and aggressiveness 3,49,50,51,52 .Some 71.5% of the research subjects reported negative effects of noise pollution on their con-centration or mental performance.Studies have shown that excessive environmental noise alters the brain's electric conductivity, leading to more rapid physical and intellectual fatigue and compromising motor activity, concentration, and task performance 53,54,55,56 .
Negative effects on hearing and communication were the most widely perceived, reported by 77% of subjects.Environmental noise can comprise communication, increasing the likelihood of mistakes and accidents 57,58 .Higher noise levels can cause auditory stress and temporary changes or even permanent loss of the auditory threshold 59,60,61 .
According to the integrated analysis using the system proposed here, the impact of noise pollution on the exposed population's health could only be considered slightly significant at a single measurement point (intersection of street 4 and avenue 1), while it was significant at 11 points (31.4%), very significant at 22 points (62.9%), and extremely significant at 1 point, located at the intersection of avenue 4 and street 2.
Thus, the SIPS using fuzzy inference to allow classifying the points according to the degree of impact and thus prioritize locations for urgent identification of noise sources and implementation of appropriate control measures.
The possibility of adjusting the fuzzy sets to the local conditions (NCA) according to the land use and occupation, as well as the margin of error in the D p , allows generalization of the system to evaluate other geographic areas.
Based on the above, the support system for evaluation of the impact of noise pollution on health proved to be an appropriate tool for environmental noise evaluation and its respective management in order to prevent and control the impacts of noise pollution on the population's health.

Contributors
A. Bressane participated in the collection, and analysis of data from interviews with the population, writing, revision, approval of the final version, and responsibility for the article accepted for publication.P. S. Mochizuki collaborated in the collection, treatment, and analysis of data from the acoustic survey, writing, revision, approval of the final version, and responsibility for the article accepted for publication.R. M. Ca-ram contributed in the project conception for the data collection, analysis, and interpretation, critical revision of the manuscript, approval of the final version, and responsibility for the article accepted for publication.J. A. F. Roveda collaborated in the conception of the fuzzy inference model for integrated evaluation, critical revision of the manuscript, approval of the final version, and responsibility for the article accepted for publication.

Figure 1 Location
Figure 1 Location of the study area and distribution of noise level measurement points.Rio Claro, São Paulo State,Brazil, 2007.

Table 2
Distribution of answers from the participatory diagnosis on the impacts of noise pollution on the population's health.Rio Claro, São Paulo State, Brazil, 2007.