Aims  Time series of incidence counts often show secular trends and seasonal patterns. We present a model for incidence counts capable of handling a possible gradual change in growth rates and seasonal patterns, serial correlation and overdispersion.

Methods  The model resembles an ordinary time series regression model for Poisson counts. It differs in allowing the regression coefficients to vary gradually over time in a random fashion.

Data  In the period January 1980 to 1999, 17,989 incidents of acute myocardial infarction were recorded in the county of Northern Jutland, Denmark. Records were updated daily.

Results  The model with a seasonal pattern and an approximately linear trend was fitted to the data, and diagnostic plots indicate a good model fit. The analysis with the dynamic model revealed peaks coinciding with influenza epidemics. On average the peak-to-trough ratio is estimated higher using the dynamic model, and graduate changes in seasonal pattern is seen.


Conclusion  Analyses conducted with this model provide detailed insights not available from more traditional analyses. The post-hoc analysis gives ideas to identify possible causal factors and confounders.