OBJECTIVES: In cohort-based approaches to cost-effectiveness analyses, the way the impact of demographics characteristics on outcomes is modeled has a large impact on the results over long-time horizons. Methods based on both means and full distributions of demographic characteristics have been presented. However, both omit to re-adjust such parameters or distributions over time to account for individuals experiencing an event (e.g. death or discontinuation) and will thus not contribute anymore to the demographic characteristics of the rest of the cohort for estimating quantities of interest.

METHODS: We present an approach that accounts for change in population heterogeneity over time. At discrete time points in economic models, we re-set key population demographics to a new baseline accounting for subjects that are no longer part of the model cohort, e.g. subjects who experienced a death event are excluded from the calculation of background mortality.

RESULTS: A series of simulations shows how differences in the change of age and weight distributions (among other factors considered) over time can significantly affect long-term clinical and cost outcomes in the case of identical Kaplan-Meier curves (or survival patterns). For diseases that can affect people in all age groups, accurate estimates of long-term life-year gains will depend on whether deaths are predominantly experienced by younger or older individuals within the cohort, as this will have a profound impact on background mortality risk over time which will in turn lead to significantly different long-term results. Similar considerations can apply to drug costs and weight distributions.

CONCLUSIONS: The implementation of changing population heterogeneity has been shown to impact the prediction of long-term outcomes in health-economic models. Here we suggest re-setting the demographic characteristics after events to further improve model accuracy. This approach has the potential to be extended to analyses of specific external populations of interest by means of propensity score methods.