We evaluated mortality, cancer incidence and hospitalization for cardiovascular and respiratory diseases among people living close to incinerators using dispersion modelling to assess exposure. The internal analyses showed no association with non-cancer related mortality and morbidity. However, predicted heavy metals concentrations, as indicator of air pollution from the incinerators, were somehow related to cancer mortality in women, in particular for stomach, colon, liver and breast cancer. In addition, a combined analysis of men and women suggested an increase in soft-tissue sarcoma mortality related to exposure to incinerators. The results were adjusted for socioeconomic status whereas there was no important confounding effect from pollution due to other sources.
The excesses detected in the areas with higher exposure levels were observed mainly among females. Of course, a chance finding could be an explanation given multiple testing but it should also considered that women are a more stable population than men and misclassification of exposure is less likely to have occurred. On the other hand, most of the associations that were found for mortality cancer outcomes were not confirmed by incidence data although the time window of follow-up was the same (1990-2003). A possible explanation of these findings is that the effect of the exposure on cancer incidence precedes the time window of our study so that only mortality is affected.
From the results of the present study it is difficult to determine the causality of the associations and which specific agent emitted from the plants could have an etiological role in the excess risk that we have found. Of course, we considered heavy metals as a surrogate marker for exposure to a complex mixture of pollutants. As for the results of other studies on incinerators, the role of exposure to dioxins could be of importance in this context. Dioxin refers to 210 congeners/isomers of structurally and chemically related polychlorinated dibenzo-para-dioxins (PCDDs) and polychlorinated dibenzofurans (PCDFs), and the 2,3,7,8-tetra-CDD (TCDD) is considered the most toxic dioxin congener in this group. Dioxins are persistent in the environment and resistant to biodegradation and are considered human carcinogens . The MSW incinerator in Forlì did emit dioxins and the values were considered relatively high until 1996 . Like in our female population, increases for all cancers and in particular for cancer of the digestive system (stomach and colon rectum) have been observed among occupational cohorts exposed to dioxin  and in the Seveso population, among residents in the more contaminated areas . In addition, among women we observed an increase in mortality for Hodgkin's disease and myeloma (based on few cases and not statistically significant) as it has been reported among Seveso women  and in a French study  where the increase of blood cancer was related to dioxin exposure from incinerators. We observed a clear excess for breast cancer but the literature on the risk of breast cancer in the proximity of incinerators is rather poor [1, 2]. No breast cancer excesses were observed in Seveso's longitudinal study, following the 1976 accident ; however, the Seveso Women's Health Study reported a two-fold risk for breast cancer among pre-menopausal women with highest serum levels of TCDD . There are several other studies that have found increased breast cancer incidence [27–29] in females occupationally exposed to dioxins. Finally, as in our observation, several studies have related residency in proximity of incinerators with liver cancer  and soft tissue sarcoma [10, 2, 14, 30, 31] although negative results also exist .
The strength of this work is the longitudinal study design adopted, in which individuals were followed for various health outcomes, exposure was assessed with advanced modelling techniques, socioeconomic status and other environmental exposures were also considered as potential confounders. To our knowledge, there are no other studies on incinerators conducted at the individual level with the details that we took into account. Despite that, the methodological aspects of the study and the main limitations should be considered.
Exposure assessment is a critical component of the study. We define the study population as people living up to 3.5 km far from the plants. This choice is based on previous studies [8, 10] and information by model simulations on the profile of the distribution of pollutant emitted by plants. In fact, this choice provided a good contrast of exposure conditions and a better comparability of the contrasted population groups. We geo-coded all the residential addresses and exposure was assessed using the results of a model of dispersion of pollutants into the atmosphere. A French study has validated Gaussian dispersion model for dioxins from an incinerator with a campaign of measurements on the ground in 75 sampling points . The results confirmed the validity of the model in defining the different gradients of exposure, and identified inconsistencies between measured levels and those estimated by the model only in the presence of complex topographical situations (e.g. hills), a condition that does not apply to the Po Valley. A recent British study compared the use of distance as a proxy of exposure from a source of pollution by means of estimates derived from dispersion models and concluded that the use of the models significantly reduces the risk of misclassification embedded in the use of the distance from a point source .
The approach we used for exposure assessment has several assumptions and limitations. First of all, we considered only exposure to air pollution whereas other exposure routes, such as soil contamination or food and water consumption, could have importance. We considered only the individual residences at the beginning of the follow-up and this choice was supported by the observation that the exposure category never changed during the study for over 90% of the subjects (Table 2). Also, exposure was assessed at the beginning of follow-up to account for diseases with a long period of induction-latency (such as cancer) where the relevant exposure does not necessarily correspond to when it was diagnosed but to exposure levels in the previous years or decades. We used authorized emission values of pollutants to simulate dispersion from incinerators. This could have overestimated concentration values, but the shape of the fallout and the gradients of exposure are not sensitive to this choice. Finally, we assumed that heavy metals are better tracer for incinerator pollution than other pollutants since there is vast literature that indicates different heavy metals (such as Cd, Ni, As, Pb, Zn, Cu, Mn) as possible tracers of incinerators [16, 35, 36], and this choice was supported by monitoring campaigns conducted during the study period.
The role of the potential confounders, in particular other occupational and/or environmental exposures, should be considered. We observed a cluster of incidence and mortality for pleural cancer among men in the second exposure category; the absolute number of cases is low, but the relative risk is high. This cluster is due to occupational exposure to asbestos in the small industrial area in the second exposure category. There are other environmental factors in the study area due to the proximity of the highway and the urban area. We took into account the effect of exposure to vehicular traffic using predicted NO2 levels; the dispersion model for NO2 showed no overlap between the areas of high NO2 and the areas of high heavy metal levels. The traffic-related pollutants in this case would act, at least in theory, as negative confounders. However, when NO2 levels where considered in the analysis no important changes in the relative risk estimates were noted, even for respiratory mortality.
An additional limit, like many other epidemiological investigations, is the lack of individual data about potential confounding factors such as individual socioeconomic conditions, occupational exposure, and personal lifestyle factors such as smoking habits. Data on socioeconomic status, available at the aggregate level (census tract), allowed us to indirectly take into account other factors related to mortality and/or cancer incidence (i.e. smoking habits and occupational exposure are strongly linked to socioeconomic conditions). In fact, rate ratios estimates were attenuated after adjusting for socioeconomic status (all causes, cardiovascular diseases and some cancer types; data before adjustment are not reported). On the basis of these findings, we cannot exclude the possibility of a residual confounding by socioeconomic status.
It is notable that the lack of information about residential history before 1990 has limited the possibility of evaluating the effects of duration of exposure and latency since first exposure, two common useful measures in cohort analysis. Finally, as already indicated, power limitations given the size of the population studied may have limited the possibility to provide stable results. In fact, the study was able to detect a relative risk of 1.3 in the last category of exposure for all cancer combined (with α = 0.95 and β = 0.80).