Completeness of ascertainment by cancer registries: putting bounds on the number of missing cases.

Background When comparing cancer incidence or mortality rates between different regions, it is important to know how complete the registration data are on which these figures are based. A number of ways of estimating completeness have been proposed, but it is often difficult to say how precise these estimates are. We describe a computer program developed to produce measures of precision for estimates of completeness obtained by one such method, the flow method. Methods The program works by resampling the required data sets, and repeatedly calculating completeness estimates until convergence of the standard errors occurs. It was tested on colorectal tumours from a single health district, and empirical confidence limits for 1 and 5 year completeness were compared with those obtained by applying various normalizing transformations and a beta distribution. The method was then applied to tumours of the head and neck, breast and lung and the results compared with those from a capture-mark-recapture exercise carried out 4 years previously. Results The sampling distribution was close to normal for 1 year completeness, but much less so for 5 year completeness, as assessed by quantile plots. Approximation by a beta distribution was better than by normalizing transformation. Although there were differences between the results produced by the flow method and capture-recapture, the flow method is more reproducible and easier to apply. Conclusion It is now possible to estimate confidence limits for the results of the flow method, and thus determine whether comparative results between registries are likely to be affected by sampling error.