We simulated under three censoring distributions with maximum follow-up of 20 years. The first assumed that all the non-cases had complete follow-up of 20 years ('complete follow-up model'). The second assumed each individual was equally likely to be censored at one of four surveys: 5, 10, 15 and 20 years ('survey follow-up model'). The third randomly censored individuals across the 20 years of follow-up using a uniform distribution with minimum 0 years and maximum 20 years ('random follow-up model'). We would expect the censoring distribution of most studies to lie somewhere in-between these distributions. Within each simulation, an individual was considered to have experienced the disease outcome if their event time was less than their censoring time, and to be disease free otherwise. Their time in the study was set to be the minimum of their event time and censoring time.
