Assessing the impact of waste co-incineration at the Anhovo cement plant (Slovenia) on the regional cancer burden

An epidemiological study combining geographic and correlational methods was conducted, in which four key data sources were integrated using geographic information system software: (1) cancer cases from the population-based Slovenian Cancer Registry together with a socioeconomic deprivation index; (2) background population data from the Statistical Office of the Republic of Slovenia; and information on exposure to environmental pollutants from (3) the Slovenian Environment Agency (ARSO) and (4) the Biotechnical Faculty of the University of Ljubljana.

From the SCR⁹, we obtained data on cancer cases diagnosed in Slovenia in the 20-year period from 2001 to 2020, along with mesothelioma data up to 2022; the last ten years, 2011–2020, were used for the geographical analysis (for the previous decades, data is not available on the desired level for small-area geographical analyses). For each cancer case, the information on the patient’s sex, cancer type, age and place of residence at the time of diagnosis (at the level of geographical coordinates), as well as the calendar year of diagnosis and the socioeconomic index Slovenian European Deprivation Index (SI-EDI)^11,12 categorised in five classes (from the most affluent to the most deprived) were used. Based on a systematic review of the association between cancer risk and exposure to co-incineration pollutants¹⁰, we identified lung cancer (defined as C33–C34 according to ICD-10 classification¹³), non-Hodgkin’s lymphoma (C82–C85) and sarcoma (as defined in the RareCare study¹⁴ according to ICD-O-3.2 classification¹⁵) as cancers that should be studied in relation to exposure. Previous studies^5,8 have reported a high mesothelioma burden in the Anhovo region attributable to asbestos exposure. That’s why we provide an overview of the cancer burden in the study area for all cancers combined and also excluding mesothelioma cases. In the geographical analysis, where our focus is on non-asbestos environmental exposures, we only report site-specific results in all cases other than the point source analysis. Population data by settlements and grid obtained from Statistical Office of the Republic of Slovenia was further disaggregated by sex and 5-year age groups.

The ARSO provided data on the average daily particulate matter that do not exceed 10 μm in diameter (PM₁₀) concentrations in the air for the year 2021, mapped on a 1 km × 1 km grid covering the municipalities of Brda, Kanal, Nova Gorica and Tolmin. The data represents the total air pollution influenced by emissions from the co-incineration plant and other local sources (e.g., residential heating or traffic) and pollution transferred by the winds from the nearby Po Valley in Italy. ARSO calculations were based on a combination of direct measurements and atmospheric modelling.¹⁶ The modelled PM₁₀ values barely differ in the various parts of the study area and are below the official limit values. Nevertheless, we have divided the values (minimum value 12.2 μg/m³ and maximum 15.4 μg/m³ on a 1 km by 1 km grid level) into three (in this case arbitrary) categories for the purpose of statistical analysis: PM10 less than 14.0 μg/m³, PM10 between 14.0 μg/m³ and 14.5 μg/m³ and PM₁₀ more than 14.5 μg/m³.

The Biotechnical Faculty of the University of Ljubljana provided data on the chromium (Cr) content in the soil at 30 sampling locations near the waste co-incineration plant, sampled in 2023.¹⁷ For geographical units without measurements, the value of the Cr concentration was assigned based on the average value at the sampling locations closest to the centroid of the spatial unit, which was determined using Euclidean distance. Similar to the PM₁₀ analysis, the Cr measurements (minimum value 14 ppm and maximum value 76 ppm) were divided into three categories for statistical purposes: less than 50 ppm, between 50 ppm and 65 ppm, and more than 65 ppm.

In the absence of historical measurements, present-day concentrations of PM₁₀ and Cr were employed as approximations of past exposure. For this reason, the results should not be interpreted as a causal relationship between exposure and cancer risk. We have only selected environmental PM₁₀ and Cr as sample data sources for analysing the potential hazardous environmental exposure in air and soil to develop a methodological framework for future monitoring in the affected region, although some further possible cancerogenic substances were also available for analysis.

In order to reduce heterogeneity, the smallest possible geographical areas were selected and two levels of geographical subdivision were applied: settlements (the smallest formally defined Slovenian administrative unit) and the 1 km × 1 km grid. The potential excess cancer risk was estimated separately for (a) settlements in four municipalities – Brda, Kanal, Nova Gorica and Tolmin (located in the western part of Slovenia as shown in Figure 1); (b) all the settlements in the municipality of Kanal; and (c) a smaller area within the Kanal municipality (where Cr was sampled). The definitions of the different geographical areas are described in detail elsewhere.¹⁸

FIGURE 1.

Map of the settlements categorized into three zones constructed for point source analysis and the location of the study area in four municipalities in Western Slovenia. The radius r indicates the distance from the three chimneys of the Anhovo cement plant.

For the point source analysis, a radius map was created with settlements classified into three zones based on the proximity to the three chimneys of the Anhovo cement plant. The first zone extends up to a distance of 2 km from the chimneys, the second zone extends from 2 to 5 km and the third zone from 5 to 10 km from the chimneys (Figure 1). The radii of the zones were chosen to be as small as possible (to detect possible increased cancer risks in the immediate vicinity of the Anhovo cement plant), but at the same time large enough (in terms of population size) to allow relevant estimates to be made. If the pollutants from these point sources (chimneys) have increased the cancer risk, the highest risk would be expected in the first zone, which is closest to the sources, and the lowest in the third zone, which is furthest from the Anhovo cement factory.

The absolute numbers of new cancer cases are strongly influenced by the population size and age distribution in the background population, both of which vary over time and across geographical areas. To account for these differences, we applied a widely used approach to handle unreliable observations in the spatial analyses, namely Bayesian hierarchical modelling. This is the same approach used in our previous studies to estimate the potential excess cancer risk in small geographic areas.^2–4 An indirect standardisation method was used to calculate the standardised incidence ratios (SIR) by dividing the observed number of cancer cases by the expected number of cases.¹⁹ The expected number of cancer cases was estimated using Slovenia’s national age-specific incidence rates. SIRs represent an approximation of the relative risk of an individual geographical unit compared to the reference population. A SIR value of 1.0 indicates that the number of cases in the observed population is equal to the expected number of cases in the reference population. SIR values greater than 1.0 indicate more cancer cases than expected and SIR values lower than 1.0 indicate fewer cancer cases than expected.

As cancer is a relatively rare disease, the underlying analyses may be unreliable due to low or no case counts in some small geographical areas. To address this, we used the Besag-York-Mollié (BYM) Bayesian hierarchical spatial model,^20–22 implemented via integrated nested Laplace approximation (INLA), to smooth the observed values and account for spatial correlation and sampling variability. Spatial clustering was assessed using the ratio of precision parameters (τ_s/τ_h). A ratio of less than 1 indicates that the spatial structure accounts for more of the variability than random heterogeneity. To account for potential risk factors (PM₁₀, Cr and SI-EDI), we included them as explanatory variables in the BYM model and effectively adjusted the smoothed SIR value for the risk factor. We ran four models: Model 1: without additional explanatory variables; Model 2: SI-EDI variable only; Model 3: one explanatory variable (PM₁₀ or Cr); Model 4: two explanatory variables, SI-EDI and PM₁₀ or Cr.

The proportion of cancer cases attributable to the risk factor (Population Attributable Fraction; PAF) was calculated taking into account the proportion of the population in each exposure category and SIR same categories.^23,24

All the analyses were performed using CanMapTool (v1.1)²² for cancer incidence mapping, and RStudio (v4.0.2) with the dplyr package (v1.0.2) for data processing. Shapefiles for municipalities and settlements were provided by the Slovenian Surveying and Mapping Authority.

Approximately 16,000 people were diagnosed with cancer annually in Slovenia during the five-year period from 2016 to 2020, including 950 (6%) in the Goriška region and just over 50 (0.3%) in the municipality of Kanal. The temporal trends for the ten most common types of cancer in the Goriška region and in the municipality of Kanal do not differ significantly from the trends in Slovenia, with the exception of mesothelioma. Due to the small absolute number of cases in the municipality of Kanal, the 95% confidence interval for the age-standardised cancer incidence rate (ASR) of the individual cancer types is very wide and a comparison of the values for the municipality of Kanal with Slovenia and the Goriška region is not reliable.

Similar to the rest of Slovenia, the number of new cancer cases (incidence) in men and women in the Goriška region has been increasing steadily since 1961. The ASR has risen by 16% in Slovenia (an average annual change of 1.0%) and by 23% in the Goriška region (an average annual change of 1.4% in the observed 20-year period) (Figure 2A). In Kanal, the ASR is higher than in Slovenia and Goriška, but remained stable (the average annual change is not statistically significantly different from zero) (Figure 2A). Mesothelioma, which is particularly high in the Kanal municipality, accounts for the largest share of the excess incidence; however, it has not increased in the last two decades. After excluding mesothelioma cases, the ASR for Kanal corresponds to the Slovenian average for the 2016-2020 period and remains stable over the last 20-year period (Figure 2B).

FIGURE 2.

Age-standardised cancer incidence rates (Slovenian standard) with the average annual change (expressed as a percentage) in the Kanal municipality, Goriška statistical region and Slovenia in the 2001-2020 period for (A) all cancers and (B) all cancers excluding mesothelioma. Statistically significant (at the 5% level) average annual changes are marked with an asterisk.

The peak of mesothelioma incidence in Slovenia was already reached in 2004, followed by a steady ASR in the 2004-2014 period and a decline from 2014 to 2022 with an average annual change in the ASR of mesothelioma of -3.1% per year (95% CI is -3.4% to -2.8%). The incidence of mesothelioma in Slovenia is around 40 newly diagnosed cases in the last decade (2013-2022). As expected given the small population in Kanal, the ASR of mesothelioma fluctuates greatly from year to year, but since 1999 the trend of the ASR of mesothelioma in Kanal has remained stable (an average of 6.5 cases per year in the last observed period of 2013-2022). Since 1988, the number of mesothelioma cases in the municipality of Kanal accounts for around 15% of all cases detected in Slovenia and has not changed significantly during this period. However, compared to the incidence of other cancers in Kanal, mesothelioma has decreased from the most common cancer in the 2001–2005 period to the fourth most common cancer in the last ten years. Even today, almost half of all mesothelioma cases in Slovenia are diagnosed in residents of the Goriška region.

A key strength of the study is the synthesis of four high-resolution, high-quality data sources within a unified methodological framework. Because of its completeness, longstanding operation, rigorous data validation and national representativeness, the population-based Slovenian Cancer Registry is a gold standard source for population-level cancer research that is particularly suitable for studying spatial, temporal and environmental patterns in cancer incidence and outcomes.³⁰ Population-level data is crucial to avoid the selection and participation biases that occur in studies based on cases and selected controls.¹⁹

For this study, data on cancer cases and the background population were available at the level of residential x- and y-coordinates, enabling the flexible and spatially precise definition of the areas to be analysed, which is a major strength of this study. To assess the sensitivity of our spatial unit definitions, separate analyses were conducted at three levels of spatial resolution: settlements (the smallest available administrative units), a 1 km × 1 km grid, and three distance-based zones relative to the three chimneys of the Anhovo cement plant. The results indicate that the choice of spatial aggregation does not have a meaningful impact on the study outcomes, as the results obtained with different aggregations are consistent. One exception is non-Hodgkin’s lymphoma for the highest category of PM₁₀ concentration, calculated using a 1 km × 1 km grid (Table 1), which could also be a consequence of the multiple testing that such an approach entails. In spatial epidemiology, analyses are frequently constrained to the use of administrative units (e.g. regions or municipalities) as the spatial resolution of analysis, due to the limited availability of geocoded data on the individual level. The main problem with administrative areas is that they are arbitrary when mapping the cancer burden, as environmental or behavioural risk factors are usually not aligned with such administrative boundaries.

Permanent residence at the time of cancer diagnosis served as a surrogate measure of exposure status in geographical analyses. People do not necessarily live where they have a registered permanent residence, which also holds for Slovenia.¹² Further, using a single time-point for exposure determination does not allow for analysing the actual duration and level of exposure in an individual, especially in the adult population, which is generally mobile and may work and spend many hours away from the permanent place of residence, being exposed to other pollutants (including occupational) that are not connected to the area observed in our study. If we take into account the patient’s permanent residence at the time of diagnosis in the analyses, we assume that the patient spent the entire observation period in the same environment, i.e. at the same exposure level (the Slovenian Cancer Registry and Statistical Office of the Republic of Slovenia cannot provide data on residential history). If the patient relocated prior to diagnosis, the impact of risk factors could undoubtedly vary.³¹ Precise data on population migration is not routinely documented in cancer registers, rendering such an analysis unfeasible with routinely collected data.

Important inputs in our study also include recent measurements of air and soil pollutants provided by the ARSO and Biotechnical Faculty of the University of Ljubljana, which are recognised as quality and internationally validated institution and scientific partners. The selection of the two pollutants (PM₁₀ particles in the air and Cr in the soil) was partially based on a review of potential carcinogenic substances associated with industrial sources¹⁰ and partially on a practical need to test the complex methodological framework, while considering the type of available measurements and the time frame of actual exposure measurements. The actual measured concentrations were below the official limit levels for air and soil pollution. Additionally, we did not have at our disposal any information on the form of the Cr. The predominant source of the measured Cr concentrations in the soil in the observed area (i.e. the data we used in the analyses) was assessed to be geogenic¹⁴, while only hexavalent Cr is carcinogenic.³² The range of pollutant concentrations across spatial units was also limited, indicating low variability in exposure levels. Still, our study extrapolates the current PM₁₀ and soil Cr exposure levels to represent past conditions, although actual historical concentrations may have varied.

Data on potential confounding factors were limited in this study. Among them, information on smoking habits is particularly important when assessing lung cancer risk. National surveys conducted in the Goriška statistical region³³ indicated that the smoking prevalence was comparable to or lower than in other regions of Slovenia; however, more granular data at the local level is not available. In our analysis, the Slovenian socioeconomic deprivation index (SI-EDI) was used as a proxy for smoking, based on established evidence that tobacco use is more common in socioeconomically disadvantaged populations.²⁷ Other potential confounders include non-industrial sources of environmental pollution, such as traffic emissions and residential heating (e.g. wood-burning stoves). Adjustments for non-industrial environmental sources of pollution may be warranted in future studies, as it was shown that the main sources of PM₁₀ air pollution in Kanal and the Goriška region are domestic wood-burning and traffic.¹⁶

The latency period must be considered in studies on cancer incidence, which has already been discussed previously to some extent. For many cancer types, long-term exposure to risk factors is required for their development, usually spanning between 15 and 20 years for solid tumours³⁴ and up to 40 years for mesothelioma.³⁵ The population diagnosed with cancer during the observation period (2011–2020) was likely exposed to relevant risk factors around the year 2000 or earlier, including the 1990s – a period when the co-incineration of waste had not yet commenced in the Anhovo cement plant.³⁶ Thus, the potential impact of waste co-incineration on cancer incidence could not have been properly assessed yet. For considering the latency period of the carcinogenicity of environmental pollutants related to novel cement production and waste co-incineration, it is necessary to repeat the present study 10–15 years from now.

Small spatial units contain small populations and, consequently, a small number of cancer cases. In particular, the low incidence rates of these events may lead to sparse data in some populations, which means that estimates derived from these models may be unstable and less reliable. A low number of rare events reduces the statistical power of the analysis, making it more difficult to detect significant associations or differences, which can lead to wider confidence intervals and less precise estimates. Aggregating data to larger spatial units was not appropriate for this analysis, as localised exposures could be diluted or masked when results are averaged across heterogeneous regions. For this reason, modern approaches to estimating the relative risk in small spatial areas often rely on smoothing methods³⁷, which we also applied in this study. The basic idea behind spatial smoothing is to use information from neighbouring regions to obtain a more stable and less noisy estimate, thus separating the spatial pattern from the noise.³⁷ We used the scientifically established and internationally recognised statistical approach of Bayesian hierarchical models, which is robust for small numbers, as it provides more stable estimates and helps to mitigate the impact of rare events.³⁸ This approach has been used in our previous studies examining the impact of environmental exposures on cancer incidence in Slovenia.^2–5