In order to work with comparable urban entities, we chose to use urban units to define cities. Indeed three delimitations can be used in France in comparative urban studies : the urban central municipality, the urban unitc and the functional urban aread. The urban central municipality, corresponding to a political grid, is not suited to epidemiological questions. Urban units and functional urban areas both offer relevant delimitations for socioeconomic comparison. The functional urban area, encompassing rural, suburban and urban types of space, covers very different types of physical environment and is therefore not suitable. The urban unit, based on morphological continuity (less than 200 meters separating buildings), covers exclusively urban built-up areas and is well suited to a comparative perspective. As air pollution monitoring is obligatory for urban units with more than 100,000 inhabitants, the 55 French urban units exceeding this threshold were included in the study.
In order to apprehend the intra-urban organization, we used the IRIS census subdivisions. These census tracts are statistical units defined by INSEE, and known as IRIS (Ilots Regroupés pour des Indicateurs Statistiques). They comprise between 1,800 and 5,000 inhabitants. It is common to use this subdivision in France in segregation and socio-spatial urban studies [30, 31].
Literature on the geography of respiratory health has been looking at respiratory mortality or hospitalization for causes such as lung cancer, COPD or asthma [32–39]. Authors investigating the link between physical characteristics, and more specifically pollution and health within cities either look at COPD or cardiovascular diseases [40, 41]. Some authors concentrate on respiratory health within cities and most often look at asthma and/or COPD [4, 6, 42–46]. We used the incidence of hospitalization of elderly males as a proxy for the respiratory health in a city. This data used to create proxies for the urban unit health status were derived from the PMSIe database which gathers information on all hospitalizations, comprising the patient’s place of residence (ZIP code), age, gender, admission and discharge dates and all diagnoses. COPD is a chronic limitation or obstruction of airflow [12, 47] and it has been documented as a respiratory disease that is exacerbated by atmospheric pollution [5–7, 10, 48]. It is often considered as an adult (over 25) disability  and is more frequently diagnosed in males than females. Prevalence studies usually consider adults over 20 [12, 33, 47]. Jeannin  states that the first COPD hospitalization generally occurs between ages 65 and 69. Halonen et al. 2013  linked COPD hospital emergency room visits by age group (children, adults and elderly > 64 years) with air pollution levels and showed a clear effect between COPD hospitalizations of elderly and particulate air pollution. Their findings also suggest that the mechanisms of respiratory effects of air pollution differ by age group. Therefore a study concentrating on only one age group can be interesting. According to the results of Tissot-Dupont  elderly people (> 60 years) are particularly vulnerable to respiratory infectious pathologies. For the present study 12 diagnostic codes describing COPD were chosen according to the Furham and Delmas  methodology. Data pertaining to COPD diagnoses for the year 2008 were extracted to identify and assess specific respiratory health patterns. In an initial stage we concentrated on all males over 65 (38,323). Several health indicators were tested: hospitalization rates by age group (65–70; 71–75; 75–80; 80–85; 86–90) and standardized hospitalization ratio (based on standardized mortality ratio methodology) for all the age groups pre-cited population. The 65–75 age group is considered to be one of the most vulnerable  for COPD. In the PMSI database, the 71–75 age group is more numerous than the 65–70 (8,340 versus 6,463), and the hospitalization ratio is higher (19 versus 15 per 1,000). Inter-urban patterns for elderly male COPD hospitalization rates are very similar to those provided by other COPD health indicators (age groups rates and standardized hospitalization ratio). Therefore elderly (age 71–75) males’ COPD hospitalization rates was chosen as indicator to characterize the respiratory health status of an urban unit. The same indicators were calculated for all hospitalizations whatever the diagnosis, and this overall hospitalization rate among elderly males served as a benchmark that enabled us to highlight the specific nature of respiratory health patterns (Additional file 1: Indicators and measurements).
A review of the literature on health and socioeconomic inequalities [35, 42, 43, 46, 50–55] pointed in the direction of a large number of indicators relating to various socioeconomic, physical and amenity aspects. For the socioeconomic dimension, classic indicators such as unemployment rates by age group [17, 56–59], the proportion of individuals with no qualifications or those having attended at least two years of higher education [56, 59–62], non-taxable household income were established for different scales of observation (Additional file 1: Indicators and measurements). INSEEf socioeconomic databases dating from 2006 were used to create this set of indicators. Family physicians and pulmonologists indicatorsg were also established to take into account the presence of one important type of health care amenity [63, 64].
In order to characterize the intra-urban organization of the urban units in terms of residential segregation, three complementary indicators [65, 66], were established: the coefficient of variation, which was used to represent the degree of heterogeneity across census tractsh in the urban unit; the Gini concentration ratio, assessing the evenness of the spatial distribution of a given characteristic, such as educational level, across census tracts; and Moran’s spatial autocorrelation coefficient, which significance level values were used to measure the degree of similarity across census tracts. Based on literature review [67–69], educational characteristics are more appropriate for the French case than ethnic variables.
A large number of studies have confirmed that respiratory diseases are related to the physical characteristics of the living area: COPD has multi-factorial etiology, including exogenous factors like air pollution, cold  and topography [71–73]. Little is however known about the long term effects of climate, but there is evidence of statistical associations between the prevalence of respiratory diseases in different locations and various climatic factors or climate zones [11, 74–77].
Recent scientific papers have discussed the link between meteorological and climatic factors and COPD prevalence . Across France, because of its geographical situation and its rather large surface area (674,800 square kilometers), there is significant spatial variability in climate. Various climatic zones occur across the country (oceanic, semi-continental, mountain, mediterranean). It is therefore reasonable to hypothesize that climate can affect inter-urban spatial patterns of COPD prevalence. The effect could be direct, for example, an effect of air temperature or humidity on the reactivity of the respiratory tract. The effect could also be indirect through a differential exposure to air pollutants, infections or aeroallergens. If this is the case, spatial patterns of hospital admissions for COPD might be related to climatic zones and variability in climatic parameters like temperature or humidity. The climatic parameters considered in the literature vary: annual mean or seasonal mean temperature [11, 74–77, 79], mean temperature for the coldest and hottest months , annual variation of mean temperature [76, 77, 79], relative outdoor humidity [74, 76, 80, 81] frequency of fog , and wind force or direction [11, 80]. This review of the literature on climatic factors and their potential effects on respiratory health (most papers focus on asthma) led us to choose a wide array of indicators describing the urban units. Firstly, each urban unit was assigned to a climate zone which constitutes its regional climatic context. On urban unit scale, the following parameters were chosen: minimum and maximum temperatures in January and July, annual mean outdoor relative humidity minima, average number of foggy days per year (visibility less than one kilometer), average number of days with strong winds per year (wind > 57 km/h) and average number of hot days per year (> 25°C) were selected. The last parameter was chosen because there is clear evidence in the literature reporting temporal studies that high temperature events (heat waves) increase morbidity in general, and in particular for elderly people [82, 83]. All these variables were derived from Météo-France meteorological data for the period of 1981–2010.
Two indicators defining the altitude of each urban unit were also included: firstly the altitude of the urban unit center. There is evidence that COPD prevalence is linked to altitude [71–73, 84], but there are diverging results. Some studies suggest that higher altitude is associated with higher COPD prevalence [71, 72, 84], whereas others suggest the reverse relationship . Secondly, in addition to the altitude of the urban unit centre, a measure of intra-urban variation in altitude was also introduced. There is no direct effect of altitude variability on COPD prevalence related in the scientific literature. This indicator was included because it might have an indirect link with COPD prevalence via effects on climatic parameters.
Some studies have investigated associations between daily variations in airborne pollen concentrations and respiratory morbidity or mortality [6, 85–87]. These results indicate a positive relationship between daily rates of COPD mortality and airborne pollen concentrations. Their findings suggest that other small particles of biological origin can potentially have inflammatory effects and exacerbate COPD symptoms. In order to take account of the aerobiological characteristics of the urban units, RNSAi pollen risk indexes were included. For each pollen type, RNSA defines a level of allergic risk and its geographical extension. The mapping of the situation in 2008 was used to construct a global index, derived from aeroallergen speciesj risk levels. This index was established for each urban unit. Only pollens classified as medium, high or very high allergen risk by RNSA in 2008 were considered.
According to the literature there is strong evidence that pollutants exacerbate COPD symptoms [4, 6, 7, 39, 88–92]. Indicators of air pollution were created on the scale of the urban unit. Two ambient air pollutants, NO2, and PM10 were used. The dataset used for assessing the air pollutant levels for each pollutant in each urban unit is derived from ADEMEk and the Geovariances interpolation model. The ADEME-Geovariances model was developed from concentrations measured in urban background stations and estimated emission data . The model estimates a range of pollutant levels: NO2 (annual mean of daily concentrations; annual mean of daily 95th percentile concentrations; mean of daily winter concentrations) and PM10 (annual mean of daily concentrations; annual mean of the 95th percentile of daily concentrations) within cells of a 4x4 kilometer grid. For each urban unit we calculated a range of concentration parameters for the two pollutants, averaging all cells for each urban unit. As there is a considerable temporal variability in pollutant levels throughout the day and the seasons, in addition to the annual mean we used parameters that illustrate this variety: annual mean of daily 95th percentile and mean of daily winter (NO2). The intra-urban level of variability of air pollutant was assessed by calculating a coefficient of variation for each urban unit and each pollutant. These variation coefficients were calculated from estimated air pollution concentrations within cells in a grid of 4 km x 4 km. These heterogeneity indicators contribute to the description of cities’s physical “profile”.