Large variations in cancer survival have been recorded between populations, e.g., between countries or between regions in a country. To understand the determinants of cancer survival differentials between populations, researchers have often applied regression analysis. We here propose the use of a non-parametric decomposition method to quantify the exact contribution of specific components to the absolute difference in cancer survival between two populations. Survival differences are here decomposed into the contributions of differences in stage at diagnosis, population age structure, and stage-and-age-specific survival. We demonstrate the method with the example of differences in one-year and five-year breast cancer survival between Denmark's five regions. Differences in stage at diagnosis explained 45% and 27%, respectively, of the one- and five-year survival differences between Zealand and Central Denmark for patients diagnosed between 2008 and 2010. We find that the introduced decomposition method provides a powerful complementary analysis and has several advantages compared with regression models: No structural or distributional assumptions are required; aggregated data can be used; and the use of absolute differences allows quantification of the survival that could be gained by improving, for example, stage at diagnosis relative to a reference population, thus feeding directly into health policy evaluation.
Keywords: Denmark; breast cancer; decomposition; regions; stage at diagnosis; survival.