A graphical assessment of the potential impact of losses to follow-up on the validity of study results

Losses to follow-up in clinical trials-patients for whom we do not know if the outcome of interest has occurred-can bias study results. If we investigate extreme case scenarios and find the study results do not change much, impact is negligible. If not, we may need to interpret the study's results with caution. At issue is how much caution do we need? We describe a graphical approach to assess the potential impact of losses to follow-up on the validity of study results. One can create the graphs using design estimates and interim or final data. We give two examples using design parameters and another example modelled after observed data from clinical trials conducted by the Terry Beirn Community Programs for Clinical Research on AIDS. The examples illustrate that tolerable levels of losses to follow-up change depending on the overall outcome and direction of differential losses.