Migration-adjusted lung cancer burden in China: a population data-based Bayesian spatial modeling approach

Cancer surveillance in Mainland China is based on household-registered residents (HRR) and therefore fails to cover migrant populations. This introduces selection bias and leads to a misestimation of the true cancer burden. Estimating lung cancer (LC) incidence and mortality among resident populations (RPs) provides a more accurate epidemiological and public health assessment.

Using 2016 data from 487 cancer registries and multidimensional covariates, we developed a Bayesian-integrated nested Laplace approximation with stochastic partial differential equation (INLA-SPDE) model to estimate LC incidence and mortality among the RP, with adjustments for interprovincial migration.

In 2016, the interprovincial migrant population in Mainland China reached 140.96 million, representing 10.1% of the HRR. The results indicate that the INLA-SPDE model outperformed the Bayesian hierarchical linear model in estimation accuracy, effectively captured spatial heterogeneity and achieved a Bayesian credible interval coverage exceeding 94%. Significant disparities in LC incident cases and deaths between RPs and HRR were observed in Henan (9159 cases and 7539 deaths), Guangdong (8851 cases and 7235 deaths), and Shanghai (5406 cases and 4332 deaths). The largest rate differences occurred in Shanghai (incidence, 20.4/100 000, 23.7%; mortality, 8.7/100 000, 15.1%).

Disparities in incidence and mortality vary with the direction and magnitude of interprovincial migration, indicating that household-registered residency-based registration overestimates LC burden in high-immigration regions and underestimates it in high-emigration regions. We recommend transitioning to RP-based registration to improve the accuracy of LC burden estimates of cancer surveillance, particularly in regions with substantial migrant populations.