Setting
Rome is the largest Italian city with a population of about 2.6 millions inhabitants on a surface of 1290 km2. It is a radiant city, and the most important roads are still the ancient roman roads that starting from the centre, the Roman Forum, connect the city with the rest of the country in all the directions. During the last century, the urban development in Rome took place gradually from the centre to the suburbs, with a higher population density in the centre compared to the periphery [28].
Study Population
The cohort was defined from Rome Municipal Register's data. We enrolled all residents of Rome on the 21st October 2001; data were available on gender, age, and residential history. Using a variety of record-linkages procedures, under strict control to protect individual privacy, we collected additional information for each study member. In particular, individual data from the 2001 Census were used for indices of socioeconomic position at the baseline. The 2001 residential addresses were used to estimate environmental exposures and traffic-related air pollution indices. Individual hospital admissions from public and private hospitals in Italy, during the period 1996 to 2001, were available to provide the morbidity history of the subjects.
Area-based and individual information on the socioeconomic position
A composite area-based index of socioeconomic position (SEP) by census block was built using the 2001 Census of Rome. Briefly, we used 4,888 census blocks with at least 50 inhabitants (average population: 500 subjects) as the units of observation. We considered census information that represented various socioeconomic parameters (occupation, education, housing tenure, family composition, and foreign status (yes or no)) and each census block was characterized. We performed a factor analysis to create a composite indicator, and we used the quintiles of its distribution in census blocks to obtain a 5-level area-based index [29]. To obtain the index for all census blocks of Rome, we assigned a SEP level to census blocks with fewer than 50 inhabitants (0.4% of the population) according to the levels of contiguous blocks. The area based SEP has been validated with individual census data [29] (for example in the highest category of area based SEP there was 29% of people with a university degree vs. 5% in the lowest category of SEP), the index is highly correlated with a small area income index based on 1998 Tax Register data [30], and it has been associated with overall and cause-specific mortality and incidence of specific diseases such as stroke [31, 32]. Figure 1 shows the map of the city by SEP.
From the 2001 Census we obtained individual data on educational level (grouped into four categories: University, High school, Secondary school, Primary school), employment status (Employed, Looking for first employment, Unemployed, Student, Housewife, Retired, Military or civil service, Unable to work, Other), occupation (Non-manual: Managers, Highly-skilled, Medium-skilled, Unskilled; Manual: Services, Farmer, Highly-skilled, Medium-skilled, Unskilled; Military forces).
Environmental characteristics at the residential address
Geographical information system (GIS) indices were developed for each individual. We geocoded each subject's residence as of 21st October 2001 using the interpolation method within road segments. To locate the address on the map we used the Italian road network (Tele Atlas, Italy). The City Council of Rome provided the traffic data for all major roads in Rome as of 2005, i.e. 6,585 road segments which represented the 26% of all roads, and included the totality of roads with more than 10,000 vehicles per day (2,228 segments).
We defined as high traffic (HTR) roads all road segments where at least 10,000 vehicles travelled per day. Figure 2 shows the map of Rome with the HTRs. We defined different GIS indicators for each residential address: the distance from the residence to the nearest HTR, the total length of the HTR segments within a 150 m buffer zone, the daily average traffic counts from the closest HTR within 150 m, and the traffic density within 150 m. The latter was defined as the sum of the products of each HTR segment length by the estimated annual average daily traffic count of the HTR segment (within the 150 meter buffer zone around the residence address) [10], divided by the area of the buffer:
We also defined a categorical variable of traffic density within 150 m as the quartiles of the distribution of the continuous variable.
Similar to the SAPALDIA study [8], we applied buffers of different radii (50, 100, 150, and 250 m) to the residences and intersected the buffers with the list of high traffic roads to create a categorical variable indicating distance to HTR (high traffic road more than 250 m, between 150 and 250 m, between 100 and 150 m, between 50 and 100 m, and less than 50 m away). We calculated a five category variable of total length of high traffic road segments within a 150 m buffer zone as the quartiles of the sum of segments' lengths within the 150 meters buffer (none, low <166 m, medium 166-266 m, high 266-323 m, very high 323-1445 m).
We collected and stored all geographical variables using ArcGis 9.1 (ESRI, Redlands, California, USA). We used the Word Geodetic System of 1984 with the Universal Transverse Mercator 33N as the coordinate system and map projection.
Baseline health status: individual morbidity history
In Italy there is a National Health Service that provides medical care to all the population. The morbidity history of the study population was based on data from the Health Information System of the Lazio region, where Rome is located. The regional Health Information System collects individual discharge records from all hospitals, both public and private. All the records are linkable using a unique identifier, but privacy protection is assured from strict management rules. Discharge records are routinely collected and contain: patient demographic data, admission and discharge dates, up to six discharge diagnoses (International Classification of Disease, 9th revision, Clinical Modification [ICD-9-CM]), medical procedures or surgical interventions (up to six), and status at discharge (alive, dead, transferred to other hospital). In order to describe the baseline health characteristics of the cohort, we used hospital discharges from 1996 to 2001 to identify those individuals who had at least one hospitalisation. We considered hospitalisations for all causes excluding accidents, those with a principal diagnosis of cardiovascular diseases (ICD-9-CM: 390-459), with principal or secondary diagnoses of hypertension (ICD-9-CM: 401-405), with principal diagnosis of ischemic heart disease (ICD-9-CM: 410-414), congestive heart failure (ICD-9-CM: 398.91, 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93, 425.4-425.9, 428), peripheral vascular disorders (ICD-9-CM: 093.0, 437.3, 440, 441, 443, 447.1, 557.1, 557.9, V43.4), arrhythmia (ICD-9-CM: 426.0, 426.13, 426.7, 426.9, 426.10, 426.12, 427.0-427.4, 427.6-427.9, 785.0, 996.01, 996.04, V45.0, V53.3), cerebrovascular disease (ICD-9-CM: 430-438); for cancer (ICD-9-CM: 140-239); diabetes (ICD-9-CM: 250), and chronic obstructive pulmonary disease (ICD-9-CM: 490-492, 494, 496).
Statistical analysis
We studied the association of traffic variables with age and socioeconomic position. We considered two SEP indices (area-based SEP and individual level of education) and three traffic measures (a binary variable to identify who lives within 50 meters of an HTR, the distance from HTR, and the traffic density within 150 m).
We used logistic regression to evaluate the associations with living close (50 m) to an HTR (Odds ratios, OR, with 95 percent confidence intervals were calculated). We log-transformed both the distance from an HTR and the traffic density to obtain two normally distributed variables, and we used them as dependent variables in a multivariate linear regression analysis. For both logistic and linear regression models, the independent variables were age (0-17, 18-34, 35-64, 65-74, 75+ years), education, and area-based SEP. In the presentation of the results of the linear regression models, we calculated the exponential function of the regression coefficients in order to estimate the ratio of the dependent variable in the specific subgroup compared to the reference group (geometric mean ratio, GMR). We calculated 95 percent confidence intervals of the GMR. To take into account the clustering of the subjects within census blocks, we performed all multiple regression analyses with robust variance estimate.
As a final step, we performed a stratified analysis dividing the entire population by area of residence (inside and outside the central railway ring) to better understand the relationship between age, small area SEP, and education with traffic exposure.