Background: Relative survival is the most common method used for measuring survival from population-based registries. However relative survival relies on comparability of the reference population and independence in the cause of death in the cancer and general population. Both of these requirements are frequently violated. We propose Fine and Gray multivariable regression is a superior method.
Methods: We used whole of population, person-level linked Western Australian cancer registry data to evaluate changes in survival from cancer overall, female breast, colorectal, prostate, lung and pancreatic cancers, and grade IV glioma using Fine and Gray competing risks regression.
Results: We observed substantial decreases in the probability of death from cancer overall, female breast, prostate and colorectal cancers over the study period. In contrast, improvements in pancreatic and lung cancers, and grade IV glioma were less pronounced and the cumulative incidence of death from these cancer diagnoses remain high. Changes were consistent with known changes in diagnosis and management over the study period.
Conclusion: The ability to adjust for both competing events and confounding makes the Fine and Gray competing risks model useful for population-based assessment of the impact of cancer programs, removing bias and limitation inherent in measures such as relative survival.