Long-term net survival among women diagnosed with cancer: accuracy of its estimation and evaluation of its determinants.

Breast cancer is a major public health challenge. It affects a very large numbers of women across the globe. Although improvements in its management have dramatically transformed its prognosis at diagnosis, breast cancer remains associated with an increased long-term risk of death, persisting even decades after diagnosis. A comprehensive understanding of this underlying pattern of death from breast cancer in the long-term is currently lacking but increasingly important as the number of long-term survivors rises. The reliability of the cause of death is of particular interest in this context. In this thesis, I use data from the Geneva Cancer Registry to first, determine the best methodology for examining long-term net survival, and second, to evaluate its determinants. Two data settings are available for the estimation of net survival: the cause-specific setting, where the cause of death is required, and the relative-survival setting, where it is not. I first evaluated the accuracy of routinely collected cause of death information and the impact of inaccuracies upon survival estimates. I observed small but non-negligible advantages in using a reviewed cause of death when estimating survival. I then compared the cause-specific to the relative survival setting for the estimation of long-term net survival and demonstrated that the relative-survival setting was less sensitive to violations of the assumptions both for breast cancer patients as well as for patients diagnosed with cancer at three other localisations. I further investigated the long-term effects of key prognostic factors and treatment for women with breast cancer in the relative survival setting using an appropriate strategy for model selection. Although I demonstrated insightful non-linear and time-dependent effects for some prognostic variables, the analyses were limited by issues of convergence and misspecification of the model. High quality population-based data and additional statistical tools are required to understand with greater certainty the determinants of breast cancer long-term excess mortality.