Taking Account of Inobservable Periods in Longitudinal Drug Treatment Follow-up

The question of the inclusion of periods of interruptions and periods unobservable has been studied as part of medico-administrative databases. However, the specificity of the health insurance databases justify the realization of dedicated methodological research. A thorough knowledge of the purpose of these data, how they collect and restitution is an essential prerequisite to carry out this work.

One type of scheme cohort exposed/unexposed led to the formulation of a finding on the basis of the comparison between the exposure groups. However, this traditional approach has the major drawback of neglecting the changing nature of drug exposure, and therefore fails to take into account any changes or interruptions doses. Approaches to take into account the time dependent variables were developed. These allow to take into account any interruptions. However, as part of study on health databases, sometimes we do not have the status in relation to the exhibition on the entire monitoring period. This is for example the case during hospitalization. The bias generated by the failure to take account of these periods ("non measurable time bias") can be problematic when a long exposure and chronic disorders require hospitalization.