The assessment followed the steps involved in the "full chain" approach, illustrated in Figure 1.
Waste Generation and Management
Current waste management data were collected from country-specific environmental agencies. The Italian Institute for Environmental Protection and Research (ISPRA)  provided a database of the incinerators operating during the period 2001-2007. In addition, a detailed census of the 52 incinerators operating in 2005 was made available by a national research institute . Detailed data were also provided by the regional environmental authority of Emilia Romagna for all eight incinerators located in that region. From all these sources, we were able to identify the 40 incineration plants operating in 2001, obtain their geographical coordinates and get specific information on years of operation, number of lines, fumes capacity (Nmc/h), stack height (m), stack diameter (m), exit velocity (m/s), emission rate (m3/s), and exit temperature (°C). In a few cases, when the information on technical characteristics was missing, it was approximated using information from other plants with similar characteristics.
The ISPRA provided a database of landfills in Italy (a total of 619 in 2001) with information of the total capacity and wasteland filled per year. Geographical coordinates of the landfills were available for only five (of twenty) regions (Piedmont and Emilia Romagna (North), Tuscany and Abruzzi (Centre), and Campania (South)) for a total of 118 landfills. For the rest of the country, we assumed that the characteristics (sex, age and socioeconomic status) of people around the 501 missing landfills were similar to those of the entire sample of the 118 sites studied.
For Slovakia, information on the number of incinerators handling municipal waste in 2001 was obtained from the Slovak Environmental Agency (SEA). There were two incinerators for MSW in 2001, geo-coded and with detailed information on technical characteristics, obtained from managing companies. At the end of 2001, there were 165 active landfills for municipal wastes. The list of landfills, by region, was available from the website of the Slovak Ministry of Environment , with geographical coordinates, capacity and starting year obtained from the SEA. Out of 165 active landfills in 2001, 121 were geocoded. We assumed that the characteristics (sex, age and socioeconomic status) of people around the 45 missing landfills were similar to those of the 121 studied sites.
For England, we evaluated all 11 municipal waste incinerators operating during 2001. Data on emissions of toxic substances (Tonnes/annum) and location (x- and y-coordinates) were obtained from the Environment Agency (EA). Wherever possible, more specific data (e.g. stack height, stack diameter, emission rate, etc) were obtained from the waste companies' websites. Where no data could be found, we applied an average from the known incinerator data.
Data for all regulated landfill sites in England and Wales was obtained from the EA. No data for 2001 was available because of changes in the regulations. In 2001 landfill sites came under a different directive and were not required to report to the EA under the Pollution Prevention and Control (PPC) Regulations. The EA advised using the 2006 landfill data instead as a good indicator for the 2001 situation . The 2006 dataset contains information about 242 regulated landfill sites (e.g. geographic Cartesian coordinates, atmospheric releases).
Population data by gender, age and socioeconomic status
Population data at the smallest unit of aggregation for the 2001 census were available for the census blocks in Italy (about 100-200 (mean 162, sd 223) inhabitants per unit) and Slovakia (about 700-800 (mean 785, sd 1318) inhabitants). For England, census population data were disaggregated to postcode areas (mean 41, sd 37). For each census block in Italy, a deprivation index was available . It used census information that represents various aspects of deprivation: education, occupation, home ownership, family composition and nationality. An algebraic combination of these factors was used to create an index of socioeconomic position by census block, with the corresponding population distributed in quintiles, ranging from very well off (level 1) to very underprivileged (level 5). For Slovakia, an index of socioeconomic position was derived from the following census variables: education (proportion of population with university, secondary, basic or no education), proportion of families with children, proportion of employed among 16-64 year olds, house type (house or flat), and house ownership. Again an index per census block was distributed in quintiles, ranging from very well off (level 1) to very underprivileged (level 5). For England, the Carstairs score , which is based on four census variables (lack of car ownership, unemployed head of the household, low social class and overcrowding) was applied as the deprivation index. The Carstairs score was available at the smallest census area, the output areas (OAs). By means of a point-in-polygon analysis the Carstairs score was transferred from the OA to each postcode. As in Italy and Slovakia, the Carstairs score in England was divided into 5 quintiles, 1 being the most affluent to 5 being the most deprived.
We used the distance from the point source (landfill site and/or incinerator) to estimate the exposed population. We decided to use the 3 km surrounding incinerators  and 2 km surrounding landfill sites  as the likely limit of the dispersion of emissions. For both incinerators and landfills, we defined increasing radial distances (1, 2, and 3 km) from the centre (the formal address of the plant) and evaluated the census blocks (or the postcode districts) that matched these locations. In several cases, the distribution of census blocks did not precisely fit the circle and the borders were tailored in order to more precisely count the population. The validity of the method has been evaluated using individually geocoded data of the resident population in four areas of Emilia Romagna (Italy) and considering the border of the plant area rather than the formal address: the error range was between 1 and 10%.
Local air dispersion modelling
Local air dispersion modelling was used to calculate increased pollutant concentrations (particulate matter, PM10, and Nitrogen dioxide, NO2) within 3 km of the waste incinerators. Dispersion modelling for incinerators was based on the national information on incineration census, actual waste throughput data and meteorological data. We have used the Atmospheric Dispersion Modelling System (ADMS-Urban)  for modelling dispersion at the local scale for 40 incinerators in Italy, 2 in Slovakia and 11 in England. Meteorological data requirements include temperature (°C), wind speed (m/s), wind direction (°), precipitation (mm), cloud cover (oktas), relative humidity (%), boundary layer height (m), and surface sensible heat flux (W/m2). We have used official meteorological data available from the nearest meteorological station. Usually 2001 meteorological data were used.
For PM10 and NO2 we have used emission rates based on national limits derived from EU legislation in 2001, namely daily emission rates of 10 and 200 mg/Nm3, respectively. However, since actual emissions could be estimated from Italy and England, we conducted an additional analysis using this measurement data [17, 18].
A number of incinerators in both Italy and England were located on hilly terrain. ADMS-Urban contains a hill module that takes into account the surrounding terrain when modelling the dispersion. Terrain data was therefore obtained for both these countries. For England the Ordnance Survey PANORAMA ™ Digital Terrain Model was used to obtain surface heights for 50 × 50 m cells up to 10 km away from 8 of the 17 incinerators. For Italy the terrain data was collected from the Italian Environmental Protection Agency for 35 of the 40 incinerators.
ADMS air pollution dispersion model provided "contours" of additional concentrations of PM10 and NO2 for the incinerators. These output files (one per country) have been transferred into the GIS system. The population database at the smallest available unit (i.e. census block or postcode district) for the given radius of 3 km has been added to the GIS as another data layer. Using an overlay function in GIS, the population data was combined with the air pollution concentration data with a grid of 200 meters. In this way, different statistics regarding population-weighted exposure levels have been estimated according to gender, age and socioeconomic status.
Following a systematic review of the literature , we chose to use the excess risk values reported by Elliott et al.  of cancer for incinerators, and of congenital malformations and low birth weights  for landfills. Cancer incidence between 1974 and 1987 among over 14 million people living near 72 solid waste incinerator plants in Great Britain were studied . The excess risk estimate for living within 3 km of an incinerator for all cancers combined was 3.5% (95%CI = 3-4%). However, Elliott et al. point out that there was an indication of residual confounding from socioeconomic status and a concern of misdiagnosis among registrations and death certificates for liver cancer. These aspects lowered our overall confidence in the results and we rated the level of confidence of the risk estimates for cancer as "moderate".
In the national study conducted by Elliott et al.  on 9,565 landfill sites in Great Britain, operational at some time between 1982 and 1997, statistically significant increased risks were found for all congenital malformations, neural tube defects, abdominal wall defects, surgical correction of gastroschisis and exomphalos, and low and very low birth weight in residents within 2 km of the sites. Although several alternative explanations, including ascertainment bias, and residual confounding could not be excluded in the study, it provides quantitative effect estimates: the relative risk for congenital anomalies was 1.02 (99% CI = 1.01-1.03) and 1.06 (99% CI = 1.052-1.062) for low birth weight. Again, on the basis of the systematic review  our level of confidence in these relative risks was "moderate".
Linear and no-threshold exposure-response functions related to the long-term effects on mortality from PM10
have been derived from the extensive existing reviews of epidemiological and toxicological data [19
]. We assumed a linear relationship between the air pollutants and associated health effects as most epidemiological studies on large populations have been unable to identify a threshold concentration below which ambient air pollutants has no effect on morbidity and mortality. The following values were used:
Background health statistics
Background gender-age specific cancer incidence data for the three countries were retrieved [20–22] together with national mortality statistics [22–24]. Prevalence of congenital malformations at birth was derived from the Annual Report (data for 2000) of the International Clearinghouse for birth defects monitoring system  for Italy and England, and from The Statistical Office of the Slovak Republic  for Slovakia.
Time frame of the assessment
Health impacts were estimated for the period 2001-2050, assuming that the incinerators operating in 2001 will be operating until 2020 - a realistic assumption given that these plants are usually in operation for a long time. The choice of 2050 ensures that the time period under consideration is long enough to account for chronic effects. For incinerators, cancer incidence "attributable" to exposure before 2001 ("past exposure") was estimated (burden of disease non-modifiable by policy) as it is likely that it will continue to appear until 2050. In addition, cancer incidence "attributable" to exposure during 2001-2020 was estimated ("current exposure") as these effects could be, at least in part, prevented. In addition, Years of Life Lost (YoLL) were estimated as attributable to current exposure (2001-2020) to PM10 and NO2 in the cohort of 2001 residents followed up to 2050.
For landfills operating in 2001, we assumed that the emissions will last up to 2030 (an assumption supported by the available knowledge that landfilled biodegradable waste starts to emit biogas a few years after deposit and continues to do so for several decades) and the health effects, in terms of congenital anomalies and low birth weight, are constant throughout this period. It was assumed that there would be no improvement in the technology of gas recovery. Of course, the same could be applied to contamination of groundwater and soil.
Estimating cancer incidence near incinerators
The basic formula to compute the number of cancer cases attributable to an incinerator is:
where AC = the attributable cancer incidence
= background incidence rate in the general population
ER = excess risk in the exposed population (relative risk - 1)
= number of exposed people
We assumed that the excess risk is not constant over time, but varies for a specific individual of the population at a given age and specific time as a function of various characteristics: level of cumulative exposure, latency since first exposure and latency since cessation of exposure (if any). We therefore assumed a theoretical model of cancer occurrence and imputed the varying excess risk around different incinerators, as a function of the different characteristics of the plant and of the nearby population. The methods are fully described in the additional file 1 (Appendix 1). Briefly, we modified the excess risk for overall cancer incidence estimated by Elliott et al.  (i.e. 3.5% for people exposed at incinerators operating before 1980, assuming 20 years of exposure) as a function of cumulative exposure (with exposure coefficients varying with time), latency since first exposure and latency since cessation of exposure. This algorithm was applied to the estimated 2001 population (by gender and age) living within 3 km of each specific incinerator to estimate the number of excess cancer cases in 2001-2050 attributable to exposures before 2001 and during 2001-2020. In Appendix 1, we illustrate the basic assumptions and we show how the excess risk during the 2001-2050 evaluation period varies in relation to time since the start of the operation of the plant and the time since cessation. The key assumption we made is that after 1980, due to technological improvements and as a result of national and European laws, the emissions from incinerators were reduced. For instance, measured particulate matter emissions from one incinerator in Italy (Modena) were 0.19 g/s in 1980-1989 (two lines), 0.0347 and 0.0376 g/s in 1995-1996 (two lines), 0.0196, 0.0273 and 0.104 g/s (three lines) in 1997-2002, and 0.0081, 0.0101, and 0.013 g/s (three lines) in 2003-2006. On the other hand, emission limits in the UK were reduced through legislation from 460 mg/m3 (1968) to 200 mg/m3 (1983) to 30 mg/m3 (1989/1990) and finally to 10 mg/m3 in 2000. Based on these data, we assumed that if the exposure level was 1 before 1980, it was 0.8 in 1980-1989, 0.2 in 1990-2000, and 0.05 after 2000. In other words, we are assuming that the exposure levels during the eighties were somewhat lower (0.8) than during the seventies, during the nineties were fourfold lower, and in more recently they were twentyfold lower than in the seventies. We are highly confident about the scores we gave the exposure levels, as they are confirmed from measured data and are reflected in the legislation. Overall, we have a moderate level of confidence in the estimates of cancer cases, mainly due to the uncertainty characterizing the excess risk used.
Estimating years of life lost (YoLL)
Assuming that chronic effects may continue to manifest themselves until 2050 in the entire population living close to incinerators in the three countries in 2001, and that their mortality rate was similar to that of the national population in 2001, we estimated Years of Life Lost attributable to PM10 and NO2 exposure as derived from the air dispersion model. In particular, we assumed that the impact of PM10 and NO2 was that which occurs during 2001-2020. As was done by Miller , we assumed constant future birth rates, constant hazard rates over time, immediate mortality effects after change in population-weighted exposure (no lag). The effects are calculated up to the year 2050 and all over a whole life span (105 years).
Overall, we have a high level of confidence in the estimates of YoLL, mainly due to the rather stable and well-established coefficients for NO2 and PM10.
Estimating congenital anomalies and low birth weight near landfills
With a moderate level of confidence, we assumed that the only health impacts on populations living near landfills are congenital malformations and low birth weight. Other possible health effects were not considered, as our review of the literature did not reveal any other significant health effects. As already indicated, the time of the evaluation was 2001-2030 on the assumption that gas emissions (methane, carbon dioxide, non-methane volatile organic compounds, hydrogen sulfide and ammonia) from landfills last several years even after a landfill closure (we assumed 30 years). We cannot exclude that the effect on reproductive health occurs via groundwater or soil contamination. The formula to calculate the cases of malformation and babies of low birth weight attributable to residence near landfills is the same as for cancer incidence, where incidence should be changed with prevalence at birth and the number exposed are newborns.
Overall, we have a moderate level of confidence in the estimates of attributable cases of congenital anomalies and low birth weight, mainly due to the uncertainty in characterizing the excess risk ratios.