Delay Databases with Individual Registry Identifiers
Databases that use delay adjustment factors and contain individual registry identifiers are released with the SEER Research Plus data. When using a database that contains individual registry identifiers, caution is necessary because some estimates for small populations may be unstable or biased.
Individual delay factors are estimated to strike a balance between two goals: (1) capture the unique pattern of delay for each registry; (2) provide stable estimates that are not too noisy. There is a trade off between stability and local relevance. This is achieved by estimating delay factors for sparse cancer site/registry combinations (defined as fewer than 50 cases per year) from groups of registries that have similar delay factors. Additionally, for each registry/cancer site combination, age, race, and ethnicity strata need to have an average of 25 cases per year or they are pooled across these groups before modeling. Thus, for some smaller cancer site/registry combinations, delay factors may have been estimated from groups of “similar” registries, rather than from single registries. Moreover, age/race/ethnicity-specific estimates for some of the smaller groups in any particular registry may have been estimated from pooled age, race, and/or ethnic group-specific strata. Delay factors for American Indians/Alaska Natives are estimated by the Purchased/Referred Care Delivery Areas (PRCDA) for all cancer sites combined, lung, female breast, and colorectal cancers, and at the US level only (from counties designated as PRCDA), for all other cancer sites.
The statistical modeling estimates individual delay factors by cancer site, registry, diagnosis year, age, race, and delay time. However, composite delay-adjusted rates for any desired grouping are computed in SEER*Stat by weighting each group appropriately. Some individual delay factors stored in SEER*Stat may be unstable or biased—perhaps because of some of the pooling that is performed before analysis to obtain sufficient sample sizes—and may mostly reflect the dominant groups in the pooled estimate. Composite delay-adjusted rates estimated across large groups of registries (e.g. SEER 9, SEER 13, SEER 18, and SEER 21) are generally very stable. Delay factors were developed with the idea that they could be used to estimate delay-adjusted trends for individual or small groups of registries. However, caution should be applied in computing registry-specific delay-adjusted rates and trends, especially for the small population racial/ethnic groups the rarer cancer sites. While the delay modeling team felt that reasonable trade-offs were made between stability and the local/subpopulation relevance of estimates, these estimates are by their nature trade-offs that are not perfect. Please report any estimates that appear spurious to email@example.com.