A consensual approach to domain-partitioning of a cancer data sample space : lessons from Tanzania

Interest in the patterns and predictability of various types of cancer has been steadily growing over the years (e.g. Steliarova-Foucher et al. 2004). However, comparative studies on geo-ethnic grounds (particularly Africa-specific) have been quite limited. Further, the influence of commonly known predisposing factors on various types of cancer has mainly been studied using conventional statistical analysis approaches which, typically, make assumptions about the underlying data distributions. We critically examine the common methods used in studying cancer predisposing factors and propose a novel triangular approach to detecting patterns in data samples. Three samples of cervix cancer, conjunctival cancer and Kaposi's sarcoma, collected from various regions in the United Republic of Tanzania, are subjected to techniques for graphical visualisation (GDV), association rules (AR) and dimensional reduction (DR). We uncover interesting patterns in the distribution of the three cancer types hinging on Tanzania's basic cultural attributes and geographical diversity. The results are analysed in a multi-disciplinary context - presenting an unprecedented approach to addressing cancer-related research issues. Our findings show that the triangular approach is capable of uncovering subtle cultural-driven relationships among data attributes. The findings present a rigorous basis for assessing and evaluating the impact of predisposing factors on the three types of cancer hence providing some useful guidelines into informed intervention, prevention and treatment of the diseases. On the basis of the results from the analysis, the paper makes recommendations as to how to structure future cancer studies in areas of similar geo-ethnic features. The triangular approach to extracting knowledge from data provides good insights into avoiding knowledge masking effects and highlights the need for formulating and designing a prototype model for updatable cancer-related data sources and storages aimed at connecting cancer registries across the African continent and beyond.