Ranking spatial areas by risk of cancer: modelling in epidemiological surveillance

Background: The representation and analysis of maps of events within a fixed time frame/period has been established as a basic tool for disease monitoring. In addition to having methods that can address the study of certain problems, the existence of criteria to discriminate relevant results is equally important. In chronic diseases such as cancer, monitoring the spatial distribution of mortality/morbidity in small areas through relative risk (RR) estimators is used frequently, but there is no clear strategy to discriminate which regions are important. Moreover, it usually requires substantial time for an effective surveillance or an advanced technical knowledge. The objectives of this study are to first establish a data analysis pipeline that allows users to make an initial screening for exploratory purposes so they can identify regions of interest in the context of chronic diseases monitoring and second to develop an R-Shiny application to implement this strategy in a straightforward way without requiring strong technical knowledge.

Methods: First, a pipeline of seven steps for ranking risk of disease spatial areas was developed taking into account relative and absolute risk estimators, using observed and expected cases in spatial units of a study region. Second, an R-Shiny application (RANKSPA, Ranking Spatial Areas) was developed to perform the pipeline. Third, we applied the pipeline using RANKSPA to simulated and real data of lung cancer municipal mortality 2005–2009 in Galicia (North-East of Spain), a region with 314 spatial units.

Results: There was a clear excess of mortality in the middle-east of the studied region using simulated data where a spatial mortality cluster is also located, existing 5 spatial units outside this cluster that occupy the top positions in the ranking generated by the application. From the total spatial units of the study region [314], only 14 had an excess of mortality whose posterior probabilities are greater than or equal to 80%. In addition, all the spatial tests implemented, with the exception of Moran’s I test, were statistically significant. In the study of real data, a clear excess of mortality was observed in the east part of the study region where several of the spatial mortality clusters are also located. Moreover, there are three spatial units located outside these clusters that occupy the top positions in the ranking generated by the application. Eleven spatial units have an excess of mortality, with their posterior probability (PP) greater than or equal to 80%. All the spatial tests implemented were statistically significant. Both in simulated and real data, there was a positive correlation between absolute and relative measures. However, a greater dispersion was observed when these measures take the highest values.

Conclusions: The work presented shows a strategy of exploratory analyses to provide an initial assessment of geographical patterns in disease risk, focused primarily on chronic diseases such as cancer. Furthermore, an R-Shiny application has been created to ease the implementation of this strategy without requiring substantial technical knowledge.