Substantial overestimation of standard errors of relative survival rates of cancer patients

Relative survival rates are among the most commonly reported outcome measures of cancer patients. They are calculated as ratios of observed survival rates and the expected survival rates in the absence of cancer. Standard errors of relative survival rates are commonly calculated by dividing the standard error for absolute survival rates by the expected survival, without taking possible random variation of the latter into account. The aim of this study was to empirically assess the validity of these commonly reported standard errors. Using data from the nationwide Finnish Cancer Registry, the authors calculated 5- and 10-year absolute, expected, and relative survival rates for patients with 25 common forms of cancer in Finland in 1989. The authors used bootstrap analysis to empirically assess the random error of absolute and relative survival rates and then compared the results with conventionally derived estimates of standard errors. The conventional and bootstrap standard errors were closely similar for all estimates of absolute survival. By contrast, the conventional estimates of standard errors of 5- and 10-year relative survival exceeded the bootstrap estimates by up to 17% and 32%, respectively. The authors conclude that conventional derivation may substantially overestimate standard errors for relative survival.