Survival of all cancer sites combined: Partitioning of temporal changes into cancer, non-cancer and case-mix

Quantifying cancer survival is a crucial component of cancer surveillance and control. Survival for all cancers combined is an overall summary to explore differences between population groups and over time. Net survival is the usual measure for reporting survival for all cancers combined. Differences in the cancer site distribution between groups can be adjusted for using standardization. We propose using individual weights incorporated into the Pohar Perme estimator of net survival for standardized all cancers combined survival estimates, rather than a weighted average of stratum specific estimates. This removes sparse data problems, where estimates are unobtainable for some strata. Extending to reference adjusted all-cause survival gives an alternative, interpretable measure, enabling partitioning of all-cause survival differences into those due to cancer site/age/sex distribution differences, other cause mortality differences and cancer mortality differences. We illustrate the methods using data on 749 889 individuals diagnosed with cancer in Norway 1986-2021. Using individual weights gives very similar estimates to traditional and model-based standardization and avoids using ad-hoc sparse data methods. Reference adjusted all-cause survival provides measures with simpler interpretation. For example, between 1986 and 1990 and 2016-2021 there was a 25.9 %age point improvement in 5-year all-cause survival. This improvement was partitioned into changes in the site/age/sex distribution (2.0), changes in other cause mortality rates (4.4) with the majority (19.6) due to improvements in cancer survival. Survival of all cancers combined is easily analyzed non-parametrically using individual weights. Reference adjusted all-cause survival gives a more interpretable measure improving understanding of differences over time/between groups.