Abstract
Quantifying cancer patient survival is an important, but challenging, task. Difficulties attributing death to the cancer of interest mean that relative survival analysis is now widely used. This method determines the survival of cancer patients relative to the expected survival of that group based on population life-tables. In epidemiological terms, therefore, cancer diagnosis is the exposure of interest, survival the outcome, and the relative survival rate (RSR) is the measure of association for those with the specified cancer compared to the population at risk. Further, it is often desirable to compare RSRs between population sub-groups, for example, ethnic groups. However, a framework for understanding, identifying and minimising potential sources of systematic error familiar to epidemiologists (i.e., confounding, selection and information biases) is absent. This presentation details an epidemiological framework to apply to relative survival analyses, both in terms of initial calculation of RSRs and to the comparison of RSRs between sub-populations. Using examples from New Zealand, we will illustrate how this framework allows us to assess and address these sources of error, and clearly allows us to identify the strengths and limitations of relative survival analyses.