Estimating the standardized incidence ratio (SIR) with incomplete follow-up data

Background: A standard parameter to compare the disease incidence of a cohort relative to the population is the standardized incidence ratio (SIR). For statistical inference is commonly assumed that the denominator, the expected number of cases, is fixed. If a disease registry is available, incident cases can sometimes be identified by linkage with the registry, however, registries may not contain information on migration or death from other causes. A complete follow-up with a population registry may not be possible. In that case, end-of-follow-up date and therefore, exact person-years of observation are unknown. Methods: We have developed a method to estimate the observation times and to derive the expected number of cases using population data on mortality and migration rates. We investigate the impact of the underlying assumptions with a sensitivity analysis. Results: The method provides a useful estimate of the SIR. We illustrate the method with a numerical example, a simulation study and with a study on standardized cancer incidence ratios in a cohort of migrants relative to the German population. We show that the additional variance induced by the estimation method is small, so that standard methods for inference can be applied. Conclusions: Estimation of the observation time is possible for cohort studies with incomplete follow-up.