The concordance and correlation of Global Burden of Disease stroke incidence rates with stroke incidence rates from population-based stroke registers

Background The Global Burden of Disease (GBD) 2021 study and population-based stroke registers are principal sources of stroke incidence estimates. This study aims to assess the concordance and correlation between GBD stroke incidence rates and stroke incidence rates from population-based stroke registers. Methods Crude and age-standardized stroke incidence rates were sourced from high quality population-based stroke registers and compared to GBD estimates matched by year and location, using GBD subnational data where available. Studies were categorized by country income group status using the World Bank country classifications; High income countries (HICs); Upper middle-income countries (UMICs) and Lower middle-income countries. Studies were categorized as to whether they were reported as informing the GBD 2021 model, using the online GBD 2021 sources tool. Concordance and correlation were assessed using Lin’s concordance correlation coefficient and Pearson’s correlation coefficient respectively. Bland-Altman plots were created to display 95% limits of agreement. Findings 50 crude matched incidence rates and 31 matched age-standardized rates were compared. Concordance and correlation for crude stroke incidence were 0.67 and 0.68 overall, 0.66 and 0.68 for HICs, 0.59 and 0.77 for UMICs and 0.03 and 0.97 for LMICs respectively. Overall, 11 (22.0%) GBD estimates, accounting for UIs, matched population-based stroke register crude incidence rates. 95% limits of agreement were -110.2/100,00 to 134.1/100,000 overall. Concordance and correlation for age-standardized incidence rates were 0.56 and 0.59 overall, 0.59 and 0.63 for HICs, 0.12 and 0.17 for UMICs and 0.25 and 0.42 for LMICs. 95% limits of agreement were from -94.6 to 84.1/100,000. Subgroup analysis including only studies where more specific subnational geographical GBD estimates were available marginally improved crude incidence (n=18) concordance (0.67 to 0.71) but not age-standardized incidence (n=13) concordance (0.53 to 0.49). Subgroup analysis limited to population-based stroke registers included as GBD 2021 sources, did not significantly improve correlation or concordance. Interpretation Our findings demonstrate limited concordance and corelation in crude and age-standardized stroke incidence rates between population-based stroke registers and the GBD 2021 model, with lower concordance for UMICs and LMICs, compared to HICs. The wide 95% limits of agreement demonstrated should provide caution in the use of GBD stroke incidence estimates to guide policy or assess progress in the primary prevention of stroke.