gms | German Medical Science

Background: The incidence density is an attractive measure in epidemiological applications, e.g., for infectious disease data. However, incidence densities may be misleading, if the constant hazard assumption fails [1]. Another difficulty arises, if competing events are present, e.g., if death of patients has to be taken into account when considering infection events. This difficulty appears to have attracted less attention in the epidemiological literature. However, there are situations where ignoring competing events obscures the analysis most, but not non-constant hazards.

Our motivating data example is on infectious complications in transplanted patients [2]. Here, a certain transplant type reduces the infection incidence density, but eventually increases the cumulative infection probability due to its effect on the competing event.

Methods: We investigate to which extent incidence densities allow for a reasonable analysis in the presence of competing events, e.g., for judging which transplant type is associated with a higher risk of infection. The aim is an easily manageable display of the competing event situation, which takes advantage of the computational simplicity of incidence densities.

We suggest a simple multistate-type graphic, which immediately displays the competing event situation, a potential effect of, e.g., transplant type and confidence intervals. We also suggest a more formal summary analysis in terms of a best approximating effect on the cumulative event probability scale for this type of graphic.

Results: The proposed methods are seen to work well in the infection data example, even in the situation that the underlying hazards have been found to be time-dependent, which is often the case in epidemiological applications.

Conclusion: Interpretational limitations of incidence densities may be due to non-constant hazards or to competing events. Competing events and even more complex event patterns may be adequately addressed with the suggested methodology.