Introduction Cancer-specific and relative survival analyses are the two main methods of estimating net cancer survival. Bias through misclassification of cause of death is well recognised for cancer-specific survival, but to date there has been no systematic examination of the potential bias from using an external comparison group for relative survival. This latter bias may be particularly important for smoking-related cancers where the expected survival is lower than the general population because of the high incidence of non-cancer smoking-related mortality.
Methods We use unique data from the New Zealand census which provides information on individual smoking status, allowing us to produce smoking-adjusted life tables. We apply these to relative survival estimates for lung and bladder cancers, known to be strongly associated with smoking. We also compare these with simulations to estimate the effect of misclassification bias on cancer-specific estimates.
Results Five-year relative survival estimates were similar regardless of which life tables were used for lung cancer. For bladder cancer, estimates varied more markedly; for example, 5-year survival estimates ranged between 0.70 and 0.84 for male smokers. Simulations suggested that for cancer-specific analyses, bias of up to 20% misclassification of cause of death resulted in estimates of 5-year survival that were 1.3–19% different from “true” estimates, with largest error for those cancers with poorest survival.
Conclusions Both cancer-specific and relative survival methods are potentially valid for population-based cancer survival studies. Both methods are susceptible to bias which is sensitive to the survival probability of the cancer under study.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.