Background: For screenable cancers, geographic differences in stage-specific incidences may reflect both variation in underlying cancer risk and variation in the extent of early detection through screening or other diagnostic activity. A small-area resolution may provide an accurate differential between the sociodemographic characteristics of areas and aid localizing areas with deviant incidence patterns.
Objectives: To introduce a conceptual framework for identifying geographic areas in greatest need of screening efforts and a Bayesian bivariate spatial modeling approach for jointly analyzing early- and late-stage cancer incidence, allowing for both spatial correlation across areas and outcome correlation within areas to improve small-area estimation.
Methods: Using data aggregated at the small-area level, with numerator data from a high-coverage cancer register containing adequate information on stage at diagnosis and linked denominator data from a population register, we estimate age-standardized incidence rates of early- and late-stage cancers across a nation and propose an inferential approach to localize priority areas for screening efforts. Specifically, if ASIR_late,i and ASIR_early,i represent the age-standardized incidence rates of late-stage and early-stage (asymptomatic) cancer, respectively, for small area i = 1, 2,…, n, areas with both quantities ASIR_late,i and ASIR_late,i - ASIR_early,i higher than the national averages indicate a combination of elevated disease burden and limited diagnostic activity. We perform joint early- and late-stage disease mapping based on a bivariate spatial formulation estimated using integrated nested Laplace approximation. We demonstrate the approach using Swedish register data on low-risk (“early-stage”) and advanced (“late-stage”) prostate cancer, 2018-2019, linked to population data at the small-area level, viz. a small-area division of Sweden into 5,984 Demographic Statistical Areas (DESOs).
Results: First, we point out two general patterns that were observed: (i) When the DESOs were grouped according to national deprivation quintiles, a consistent gradient emerged – the most deprived areas showed both higher late-stage incidence and higher late-minus-early incidence. (ii) When comparing the 21 regions of Sweden (where the healthcare organization is regionalized), 3 regions showed statistical evidence of both ASIR_late and ASIR_late - ASIR_early exceeding the national averages, based on the region-specific 80% credible ellipses for the joint estimates. The small-area spatial model revealed 189 DESOs with a posterior probability >0.80 of simultaneously exceeding the national estimates of ASIR_late and ASIR_late - ASIR_early. This disease mapping model reduced isolated extreme areas and strengthened inference for specific clusters within 3 regions supported by both quantities.
Conclusions/Implications: Mapping late-stage incidence (as an indicator of disease burden) and late-minus-early difference (as an indicator of insufficient diagnostic activity relative to the underlying disease burden) jointly provides an empirically grounded way to localize areas where enhanced early detection may yield the greatest benefits.