Introduction
Researching health at the ecological level is important although it is accompanied by numerous criticisms. These are mainly related to the transfer of results from the aggregated to the individual level and the associated ecological fallacy. Deprivation indices at the level of spatial units, despite certain limitations, are often used to measure and monitor inequalities in health, mainly due to the absence of socio-economic data of individuals in administrative databases. Even if individual data is available, access to them is usually problematic and time-consuming due to protection of privacy. The aim of the presented study was (1) to evaluate the degree of ecological fallacy in the use of aggregated deprivation indices on routinely collected population data, (2) to assess the agreement between the measurements of socio-economic deprivation in the Slovenian population with an individual or aggregated deprivation index, and (3) to determine by applying geographical analysis of the association of socio-economic deprivation and cancer incidence, whether it is possible to adequately substitute an individual deprivation index with an aggregated one.
Methods
The conducted research was based on routinely collected data. Information from three databases was used: European Union Statistics on Income and Living Conditions (EU-SILC) and the National Census (Census) data from the Statistical Office of the Republic of Slovenia for year 2011, and all first cancer cases diagnosed in Slovenia at aged 16 and older in the period 2011-2013 from population-based Slovenian Cancer Registry (SCR). As the background population data from Census in 2011 was used. For the purpose of this study the European Deprivation Index (EDI) was utilized, which is based on Townsend theorization of relative deprivation. The Slovenian version of EDI (SI-EDI) at the aggregated level (SI-EDI-A) was calculated for different geographical levels using EU-SILC and Census data. The SI-EDI was also calculated at the individual level (SI-EDI-I) by the approach that represents a methodological innovation. The degree of ecological fallacy was estimated with the ROC curves. By calculating AUC, the ecological fallacy was evaluated numerically. Agreement between measuring deprivation with SI-EDI-I and SI-EDI-A was analysed by two graphical methods (Bland-Altman method and mountain plot) and formal testing (Lin's concordance correlation coefficient–CCC). The association of the socio-economic status and cancer risk was analysed using SCR data. The relative risk of cancer (RR) was calculated for 17 cancer locations. Analysis was done for each level separately, for SI-EDI-I and for SI-EDI-A. The Poisson regression model was implemented for individual level and hierarchical convolutional Bayes model (BYM model) for aggregated level (at the level of polling station units). In order to compare the impact of the socio-economic status on cancer incidence at the same time at the individual and aggregated level, a generalized multilevel model for count outcomes was used with one explanatory variable at micro level (SI-EDI-I) and one explanatory variable at macro level (SI-EDI-A for polling station units).
Results
SI-EDI-A for year 2011 was calculated for five geographical levels: polling station units, settlements, municipalities, administrative units, and statistical regions. SI-EDI-I was developed and played a key role in the evaluation of ecological fallacy. The assessment of the ecological fallacy showed that none of the five aggregated deprivation indices was clearly superior, but the SI-EDI-A at the level of polling station units performed better than others. The degree of agreement between SI-EDI-I and SI-EDI-A using Bland-Altman and mountain plots reflected discrepancy between the methods. The same pattern was found for the CCC. To explore the association of socio-economic deprivation and cancer incidence, data of 1,739,865 individuals aged 16 years and older were included in the analysis together with 27,331 cancer cases. Certain types of cancer had higher incidence among socioeconomically deprived populations (e.g. head and neck, oesophageal, and lung cancers), while others were more common in affluent populations (e.g. skin melanoma, prostate cancer). These findings were consistent across both individual and aggregated levels of deprivation indices. Comparison of RR results at the individual (Poisson regression) and aggregated (BYM model) level showed that altogether fourteen out of seventeen cancer sites agreed. When comparing RR related to SI-EDI-I from the Poisson regression to RR from the multilevel analysis, fifteen out of seventeen comparisons agreed; and when comparing the results of RR related to SI-EDI-A from the BYM model and the multilevel analysis, all seventeen comparisons agreed.
Conclusion
The study clearly showed that ecological fallacy is unavoidable, but it is the lowest when the smallest available geographical units are used. When exploring the association of socio-economic status and cancer incidence in the Slovenian population on aggregated data, the ecological fallacy and the discrepancy in the degree of agreement did not significantly affect the final findings. Comparison of the RR related to SI-EDI-I and SI-EDI-A showed a sufficient agreement. Although the association of cancer incidence and socio-economic deprivation at the individual and aggregated levels was not the same for all cancer sites, the results were similar for the majority of investigated cancer sites. The study also confirmed previous research findings that there are certain types of cancers in Slovenia, which are more common in the socio-economically more deprived population, and others whose risk is greater in the socio-economically more affluent population. Regardless of the type of deprivation indicator (aggregated or individual) used, the findings showed a notable level of consistency across the different methods employed. It is possible, therefore, to adequately substitute SI-EDI-I with the SI-EDI-A at the level of polling station units when exploring the association of socio-economic status and cancer incidence in the Slovenian population. However, researchers should be aware of the potential for ecological fallacy and consider this when interpreting their results.
|
