(146) Considerations in estimating complete prevalence of cancers using Surveillance, Epidemiology, and End Results (SEER) databases using marginal zone lymphoma (MZL) and acute myeloid leukemia (AML) as case examples

Publication Number: 481

Background: SEER cancer registry data can be used to generate limited duration prevalence (LDP) or complete prevalence (CP) estimates for various cancer types in the US. While both LDP and CP are point prevalence estimates, LDP includes all patients diagnosed over the duration of cancer registries while CP also includes patients diagnosed before the start of cancer registries. CP can be derived using the LDP and the completeness index (CI) calculated from incidence and survival (CP = LDP/CI). Cancer-specific CIs are not available for all cancer types in SEER. While an all-site cancer CI or broader cancer type can be substituted for the disease-specific CI to derive CP, prevalence estimates could be over or underestimated depending on differences in age of onset and survival.

Objectives: To assess the variability in CP estimates derived from SEER databases using different CIs, using MZL and AML as case examples.

Methods: We estimated 19-year LDP and CP for MZL and AML using the SEER 21 registry databases. We selected a relatively indolent cancer (MZL) and a relatively aggressive cancer (AML). To calculate CP for AML, we used the AML-specific CI and all-cancer-site CI to assess the impact of using all-cancer sites CI on CP estimates. For MZL, we used the non-Hodgkin lymphoma (NHL)-CI as MZL-specific CI was not available. LDP and CP prevalence by age in years (y) ( < 20, 20-64, ≥65) and sex were estimated and standardized to 2022 US Census population.

Results: Using disease-specific CI, LDP and CP in the US in 2022 were 80,552 and 92,553 for MZL and 56,209 and 71,821 for AML, with an increase of 15% and 28% from LDP to CP, respectively. As expected, no difference was found between LDP and CP in patients < 20 years of age since 19y LDP covers the entire pediatric lifespan. Differences in prevalence estimates derived using LDP and CP were larger among patients with AML than MZL and the largest difference was among patients with AML aged 20-64y. For MZL, the age-stratified differences in LDP and CP ranged from 12-16%. The CP of AML calculated from all-site cancer CI vs AML-specific was 9% lower (65,606 vs. 71,821).

Conclusions: CI adjustment can have a substantial impact on CP estimates, especially for cancers or subgroups with high survival. Prevalence studies using SEER databases should preferentially use disease-specific CI and caution should be exercised when using all-site cancer CI as a proxy especially when factors such as survival differ from all-site cancers.