Cause-specific or relative survival setting to estimate population-based net survival from cancer? An empirical evaluation using women diagnosed with breast cancer in Geneva between 1981 and 1991 and followed for 20 years after diagnosis

Abstract Background : Both cause-specific and relative survival settings can be used to estimate net survival, the survival that would be observed if the only possible underlying cause of death was the disease under study. Both resulting net survival estimators are biased by informative censoring and prone to biases related to the data settings within which each is derived. We took into account informative censoring to derive theoretically unbiased estimators and examine which of the two data settings was the most robust against incorrect assumptions in the data. Patients and methods : We identified 2489 women in the Geneva Cancer Registry, diagnosed with breast cancer between 1981 and 1991, and estimated net survival up to 20-years using both cause-specific and relative survival settings, by tackling the informative censoring with weights. To understand the possible origins of differences between the survival estimates, we performed sensitivity analyses within each setting. We evaluated the impact of misclassification of cause of death and of using inappropriate life tables on survival estimates. Results : Net survival was highest using the cause-specific setting, by 1% at one year and by up to around 11% twenty years after diagnosis. Differences between both sets of net survival estimates were eliminated after recoding between 15% and 20% of the non-specific deaths as breast cancer deaths. By contrast, a dramatic increase in the general population mortality rates was needed to see the survival estimates based on relative survival setting become closer to those derived from cause-specific setting. Conclusion : Net survival estimates derived using the cause-specific setting are very sensitive to misclassification of cause of death. Net survival estimates derived using the relative-survival setting were robust to large changes in expected mortality. The relative survival setting is recommended for estimation of long-term net survival among patients with breast cancer.