Materials and methods

This study evaluates the hypothesis that in the two provinces of Naples and Caserta overall mortality, cancer mortality and occurrence of congenital anomalies are spatially correlated with the intensity of legal and illegal landfills and waste dumping sites. The unit of analysis consists of the 196 municipalities comprised in the two provinces. Mortality data were retrieved from the Italian National Institute of Statistics for 1994–2001; specific death causes (table 1; supplementary online material) were selected on the basis of the published literature on the health effects of waste-related exposures.

Congenital anomalies data, regarding live births, stillbirths and termination of pregnancy, were obtained from the Campania region registry of congenital anomalies (a full member of the European network for surveillance of congenital anomalies (EUROCAT) and part of the International Clearinghouse for Birth Defects and Surveillance and Research (ICBDSR)) for 1996–2002, and were grouped by subtypes (table 1; supplementary online material), following standard classification. Yearly population data from the Italian National Institute of Statistics (ISTAT) were used as denominators: number of residents for mortality and number of resident total births for congenital anomalies. Using number of births from ISTAT as denominators, rather than from the congenital anomalies registry, was necessary because registry data on births do not include the residence municipality.

Data from the 1991 census were used to calculate a deprivation index at municipality level, applying an established methodology based on variables on education, unemployment, housing ownership, surface of the dwelling, and family structure.22 The 1991 census data were preferred to the 2001 data to allow for long latencies for cancer mortality, and used in all analyses also considering the high correlation between the two indices (correlation coefficient  =  0.85). Municipalities were subdivided in quintiles with respect to the value of the deprivation index (1  =  least deprived; 5  =  most deprived).

The degree of environmental pressure due to waste dumping activities was estimated through the creation of a synthetic index, used at municipality level.23 24

A database with over 300 waste landfilling or dumping sites, built by the regional Environmental Protection Agency, was used. This database included both authorised landfills and illegal waste dumping sites. Two hundred and twenty-seven sites were selected for the analysis (89 legal and 138 illegal) if: (1) hazardous wastes were present; or (2) total waste volume was more than 10 000 m³. A group of independent experts then evaluated each selected site in terms of its potential hazard, assessed on the basis of the likelihood of releases on water, soil and air. Sites were grouped into seven categories of decreasing hazard, as shown in table A1 in the Appendix.23 A 1 km-radius circle was drawn in a Geographic Information System around the centre of each site; the study area was thus partitioned into zones according to the presence of these circles (yes or no, number of overlapping circles) and their intersections. The choice of a 1 km radius, smaller than what has been used in some comparable studies,8 9 was motivated by the high density of the waste sites in the study area and the frequent overlap of the circles, and by the high population density, which guarantees high statistical power. Data on population counts by census tract (a geographical unit smaller than municipality, with 20 (median value) tracts per municipality) were combined with the ordinal classification of the zones, producing a value of population-weighted intensity of waste-related exposure. Each of the 196 municipalities was then given a summary index, resulting from summing the scores of all the zones comprised in the municipality. The ranking scale was designed in such a way that the presence of multiple low-hazard sites in one area, produces a lower score than one of the high-hazard sites. Finally, municipalities were categorised in five groups, using so-called natural breaks, that is, cut points that maximise homogeneity within groups and heterogeneity between groups.25 Distribution of the index by municipality and geographical distribution are shown in fig 1.

The correlation analysis between the health endpoints, the waste index and the deprivation index was carried out applying standard Poisson regression models, using the STATA software package. All mortality models included age (in 5-year groups except for the first group of 0–14 years of age); for both mortality and congenital anomalies models, the five groups defined by the deprivation index and the waste index were used as categorical factors in the analysis; linear trends were also estimated across the five groups of both indices, by coding the categories from 1 to 5, and taking these codes as quantitative figures. A variable for time period was also included, but subsequently discarded from the regression models given the lack of any effect. All mortality analyses were carried out separately for men and women, due to different patterns of occupational exposures and different mortality rates for some endpoints.

A hierarchical Bayesian conditional autoregressive normal model, including terms for spatially unstructured and spatially structured effects26 27 was also fitted, using the WinBUGS software, in order to evaluate the sensitivity of the risk estimates to extra-Poisson variability and spatial autocorrelation.

Results

The 196 municipalities had a population (average over 1994–2001) of 1 921 315 men and 2 023 954 women. There were, respectively, 123 627 and 116 872 deaths, in total in 1994–2001. A total of 4192 cases with congenital anomalies were registered out of 351 416 total births in 1996–2002. The classification of municipalities in groups by index of waste-related environmental pressure produced a skewed distribution with 8, 24, 25, 35 and 104 municipalities, respectively, in the five classes, from most exposed to least exposed, the latter being taken as reference group (see fig 1). Table 1 gives the details on population and mortality, including the causes of death studied, by the five levels of exposure (“exposure” henceforth refers to the summary index described above); table 2 gives the details on births and congenital anomalies.

Tables 3 and 4 present the results of the regression analysis. For mortality, statistically significant trends were found for all-cause mortality in men and women; statistically significant or borderline significant trends were also found for all cancers, liver cancer in both sexes, lung cancer in men, and stomach cancer in men. These trends reflect marked differences between the different groups of municipalities; using the group with lowest exposure as reference, excess relative risk estimates in the highest group are, for example, 9.2 and 12.4% for men and women, respectively, 4.1 and 6.6% for all cancer mortality, and 19.3 and 29.1% for liver cancer mortality. No association was detected between the waste index and mortality from the pool of all cancers other than those individually analysed.

For congenital anomalies, association is less marked, with only two significant trends observed: one positive for congenital anomalies of the internal urogenital system and one negative for cardiovascular anomalies. Comparisons between high exposure municipalities (group 5) with reference (group 1) were statistically significant for congenital anomalies of the internal urogenital system (82.7%) and for congenital anomalies of the central nervous system (83.5%).

These risks are adjusted for deprivation, which was associated with the waste index (correlation  =  0.30). As expected, the deprivation index was significantly associated with most mortality endpoints, for both sexes (see table 3), with significant trends of increasing overall mortality across deprivation quintiles; positive associations were obtained for all cancer mortality, lung cancer, liver cancer and kidney cancer in both sexes, bladder cancer and non-Hodgkin lymphoma in men only; associations were null or non-significant for stomach cancer and soft tissue sarcoma in both sexes and, for women, non-Hodgkin lymphoma; for all other cancers combined, significant associations were found for men and women.

Association between deprivation and risk of congenital anomalies was also found; the trend was statistically significant for all anomalies combined (p<0.001), cardiovascular anomalies (p<0.001), and limb anomalies (p = 0.013). For the other types of anomaly, associations were null or non-significant.

The analysis carried out with hierarchical Bayesian modelling is available in tables 2 and 3 in the supplementary online material. Results are robust with a comparable overall pattern, and little deviation suggesting limited extra-Poisson variability (spatially structured and unstructured). The one notable exception, however, was the risk estimate for all congenital anomalies combined: excess relative risks in all exposure levels, compared with the reference unexposed group were positive, although not statistically significant (11.2, 10.5, 10.5, 29.1% in groups 2 to 5, respectively).

Discussion

The intensity of waste-related exposure, as measured by the summary index based on density and degree of hazard of known waste disposal sites, correlates with the occurrence of several health endpoints, among those selected for this study. At municipality level, total mortality, all cancer mortality, and mortality from some specific cancer sites have increasing risks, for one or both genders, by increasing exposure, with statistically significant linear trends across five exposure groups. Trends are significant in five of nine causes of death in men, and three of nine in women, largely above what could be expected by chance due to multiple testing under the null hypothesis of no association (at the 5% significance level, 0.9 significant trends are expected by chance alone). Congenital anomalies are less markedly associated with exposure, with one significant increasing trend (for anomalies of the internal urogenital system) and one significant decreasing trend (for cardiovascular anomalies) out of 12 tested. For another type of anomaly, congenital anomalies of the central nervous system, the trend is not significant, but a significant excess risk, based on 46 cases, was observed in the most exposed group. All anomalies combined are not associated with exposure in the standard Poisson regression analysis, but are in the Bayesian analysis, where extra-Poisson variability and spatial autocorrelation are taken into account.

These results cannot be easily compared with those of other multi-site studies on landfills and health, because of the marked differences in the settings. For example, the UK studies on cancer incidence5 and the EUROHAZCON study on congenital anomalies9 address the residential risk due to living in the vicinity of known legal and controlled landfill sites, as opposed to the widespread occurrence of illegal sites of highly variable characteristics. This lack of comparability with previous studies makes causal inference all the more challenging. Several questions must be addressed when evaluating the nature of the associations. Data quality in the study area, for mortality, is in line with national figures, with proportions of ill-defined deaths ranging from 1.2 to 3% in the two provinces, for men and women. Some caution is needed, on the other hand, for congenital anomalies, given a coverage of approximately 75% of all deliveries, due to the fact that some maternity units did not provide delivery data to the regional registry in the study period.28

Exposure data are based on the available database of sites of legal and illegal waste dumping. While this database is quite detailed, the very nature of illegal waste dumping operations is almost certain to produce some degree of incompleteness of the data, and a non-systematic undercoverage of illegal dump sites. This in turn may result in some error in estimating the exposure index at municipality level, with uncertain consequences on risk estimates.

All models used in the analysis included a municipality-level deprivation index, used for controlling the possible confounding effect of socioeconomic variables. The index has been used in correlation studies in Italy and its effectiveness in removing confounding has been documented.29 Results from this study suggest an important role of socioeconomic factors, and the need to control for them in this kind of analysis. However, it is possible, also in consideration of the ecological study design, that residual confounding, not captured by the deprivation index, may be present and explain part of the associations found for waste exposure. Other risk factors that may partially explain the associations include determinants such as quality of primary care, lifestyle, tobacco consumption, hepatitis infections and other kinds of environmental degradation, for which systematic data are not available in the study area. Hepatitis infection, in particular, shows an elevated prevalence in the study area.30 31 This might produce possible confounding on mortality from liver cancer; however, the possibility of interaction between hepatitis virus and hepatotoxic carcinogenic chemicals should be considered, if these chemicals have a fibrogenic action, as in the case of vinyl chloride.32 In more general terms, while some of the risk factors mentioned above are likely to be more powerful health determinants than waste-related exposures, there are no obvious reasons (perhaps other than socioeconomic factors) why they should correlate, at the municipality level, with intensity of waste dumping and thus produce a positive confounding effect.

Conversely, some factors are supportive of a causal role of waste-related exposures on the health outcomes under study. The different analyses have a certain internal consistency, with regards to: a relatively large number of positive and significant risk estimates, and only one negative (that may be due to multiple comparison); comparability of risks in men and women; coherent behaviour of estimates when adjusting for deprivation index (the estimates were reduced); absence of association between the waste index and all cancers other than those selected a priori for the analysis, combined; and robustness of models when allowing for extra-Poisson variability (spatially structured and unstructured). Positive associations were found for outcomes with long latency times, such as lung cancer mortality, and for one outcome with latency of less than 1 year (internal urogenital congenital anomalies), consistently with the fact that waste mismanagement has been a problem in the area for at least two decades, and that waste dumping sites established and used in the past have not yet been cleaned up.16

Thus, on balance, factors suggestive of a role of waste-related exposures as a contributory causal component of the association are somewhat stronger than those suggestive of an artefact. The overall picture, though not clear, is in fact compatible with the occurrence of adverse health effects of intense environmental pressure due to waste disposal practices in Campania. Validating this observation may be difficult, as no systematic environmental monitoring has so far taken place in the area, and only sporadic ad hoc measurements have been performed, aimed at documenting highly polluted spots, of particular interest for local environmental remediation or for criminal prosecution. In these measurements, a range of chemical agents, namely heavy metals, polychlorinated biphenyls, dioxins and polycyclic aromatic hydrocarbons have been detected in several locations in soil samples and both surface and underground water.33 34 35 Extensive measurements, including biomonitoring of dioxin levels in human tissues, are currently in progress.36

The association between the waste index and several health endpoints is difficult to interpret also in consideration of the potential ecological bias; on the other hand the index may capture a variety of pathways through which harmful exposures occur in the local population. If so, risks are likely to be underestimated because causal agents are only approximated, with a likely non-differential exposure misclassification. In addition the health outcomes selected for this study represent only a part of the possible health impacts: preliminary analyses indicate associations similar to those here described also with non-cancer mortality (data not shown). The adverse effect on non-cancer endpoints is not inconsistent with effects on cancer mortality, as living in an area with intensive uncontrolled waste disposal has visual, olfactory and safety implications which impinge on perceived health, quality of life and stress.

More research is needed, and more complete data on congenital anomalies are desirable, to confirm and describe the existence of multiple risks due to the complex mix of exposures associated with poorly managed and uncontrolled waste dumping and burning, to clarify which agents and which pathways are most important, to explore the effects with regard to more health outcomes and to evaluate the health impact of these exposures. It is, however, more urgent that action is undertaken in the region for removing the most acute cases of waste-related exposures and for implementing an efficient cycle of waste management. Insights from Campania may be valuable elsewhere: recent reports from international agencies indicate that waste mismanagement occurs in Europe and beyond, with substantial illegal shipment of hazardous waste, especially from the Organisation for Economic Co-operation and Development countries to new EU member states, Balkan countries and the Commonwealth of Independent States37 and substantial quantities of waste dumped in illegal sites.38 Illegal transport and disposal of hazardous waste is also a reason for concern in Africa, due to the economic profitability of uncontrolled dumping in countries where regulations are absent or not adequately enforced.39