Development of the Delay Model

Historically, statistical models that estimate "reporting delay" for cancer incidence rates were developed for both the SEER 9 registries and the SEER 13 registries. The delay-adjusted rates produced by these models have been reported by the NCI SEER Cancer Program since 2003.

Starting from the 2015 release of SEER data and statistics, for the first time, the delay-adjusted rates reported in Cancer Statistics Review were based on the data submitted to the North American Association of Central Cancer Registries (NAACCR). NAACCR is a registry member organization that includes registries in SEER and registries for the remainder of the U.S. funded by the Centers for Disease Control and Prevention's (CDC) National Program of Cancer Registries (NPCR) (some registries are co-funded by both NCI and CDC), and some Canadian registries. Annual cancer incidence and survival data are reported by U.S. registries to SEER and NPCR, while registries throughout the U.S. and Canada report annually to NAACCR. It was a coordinated effort by NCI, NPCR and NAACCR to expand delay modeling to registries throughout the U.S. and Canada.

Unified Effort to Estimate Delay Adjustment Factors for All NAACCR Registries

Instead of developing delay-adjusted rates separately, a coordinated effort by NCI, CDC, and NAACCR has led to a unified approach to estimate and report delay-adjusted rates. A joint NAACCR, SEER, NPCR delay model workgroup was formed. The goals of this joint effort are as follows:

Produce delay factors (and standard errors) for every (or almost every) U.S. and Canadian registry. Registries may be excluded if they do not meet criteria established for this analysis.
Ensure that the factors can be combined across registries to allow the calculation of delay-adjusted incidence rates and standard errors for any registry combination. This will be accomplished by making the delay-adjusted rates available in SEER*Stat.
Establish methodology that accounts for the varying first submission years of registries and potential holes in the registry data.

In NAACCR submission, there is a total of 70 U.S. and Canadian registries potentially eligible for delay modeling. Some of these 70 registries are excluded because they have too many combinations of reporting and diagnosis years either missing (not submitted) or did not meet the NAACCR’s "fit for use" criteria. With too many missing cells, the model parameters are not estimable or may be unstable. The 70 registries are comprised of SEER-funded, NPCR-funded, SEER and NPCR jointly funded, and Canadian registries. The registries of SEER-funded (including jointly funded by NPCR) are listed below:

Alaska Native Tumor Registry
Atlanta
California (without San Francisco-Oakland/San Jose-Monterey/Los Angeles)
Connecticut
Georgia (without Atlanta)
Hawaii
Idaho
Illinois
Iowa
Kentucky
Los Angeles
Louisiana
New Jersey
New Mexico
New York
San Francisco-Oakland/San Jose-Monterey
Seattle-Puget Sound
Texas
Utah

The 2025 Release of Delay Adjustment Factors and Rates

Delay adjustment factors have been produced from the December 2024 NAACCR submission. The NAACCR-based factors are stratified by:

Cancer Site
Registry

and are modeled as a function of the following variables (if they are statistically significant):

Age Group
Race
Ethnicity
Year of Diagnosis

The factors are linked to the appropriate cases (based on the stratifications above) in data submissions for each of the three partners in this joint effort (SEER, NAACCR, and NPCR). As of 2025, we produce delay-adjusted rates for each of these groups and make their results available to the public.

Delay-adjusted rates can be computed in SEER*Stat. To learn more about the delay databases that are available, refer to Rates in SEER*Stat.

Using Variables not Included in the Delay Model

In some situations, a user may be interested in estimating delay adjusted rates by variables not accounted for in the delay model (e.g., histologic subtypes of lung cancer). It is difficult to know if the delay factors would differ significantly by these variables, so in these cases SEER*Stat provides a warning dialog stating, "The following variable was not considered in the selected delay model and may cause misleading results."

When you get the warning message, it can be difficult to know whether to proceed and produce the delay adjusted rates or not. One general rule of thumb is to consider whether the variable categories might have very different types of encounters with the health care system, making it so one or more subtypes would be more difficult to find by registrars. For example, if one subtype would only be seen on an outpatient basis throughout their phases of care, there might be delays in finding these types of cases.

This warning appears for any factors other than cancer delay site*, registry, age, race, ethnicity, year of diagnosis, and sex (though sex is not included in the model, it was considered and found not to be important), and stage of diagnosis for prostate, female breast cancers, colorectal, and lung cancer (see section on Delay Databases by Stage).

* Delay site is a special site variable that includes the cancer site groupings used in the delay model. If you use other cancer site variables, you will get a warning and should use the same cautions as discussed for histologic subtype.