Spatial estimation of incidence rates (IR) using individual-level nation-wide population data: Acute myocardial infarction (AMI) in Denmark 2005-11

Background Acute myocardial infarction has severe individual and public health consequences. Spatial analysis of public health data is often performed using data aggregated in administrative areas. Reliable individual-level data on the population and disease enable examination of spatial variation of AMI in Denmark using a high geographical resolution of data independently of administrative boundaries. However, these analyses are very computer intensive limiting the general use of the method.

 

Objective To estimate the geographical distribution of incidence rate of AMI. To compare methods for estimating IR.

 

Methods Data were obtained from nation-wide registers. Three methods were used and compared. Initially, aggregated data at municipality level using hierarchical Bayesian analysis was used. Secondly, an iterative spatial generalized linear model (isGLM) was used to estimate IR accounting for the spatial distribution by modelling residuals using semivariogram models. Finally, a generalized linear model accounting for the IR in a 3km neighborhood (smoothGLM) was used.

 

Results The study population consisted of 3,501,621 residents (≥30 years) of which 74,037 (2.1%) experienced an incident AMI. IR was 318 per 100.000 person-years. The isGLM converged after 3 iterations. The isGLM and the smoothGLM resulted in very detailed IR maps identifying variation that were not seen using aggregated data.

Conclusions Spatial analysis of individual-level geocoded population data results in less smoothed maps of IR. Although methods are computer intensive, estimation is possible. Knowledge of the geographical distribution of AMI may contribute to more appropriate allocation of treatment resources.