Results
Simulated probabilities of being classified as LTFU are shown in Figure 1A for patients enrolled during different time periods. Although the model assumes rates of ART interruption and mortality to be constant over time, the simulations show a trend towards increasing LTFU in more recently enrolled cohorts.
(Enlarge Image)
Figure 1.
Association between period of enrollment in antiretroviral treatment (ART) and probability of being classified as lost to follow-up (LTFU). In part A, the probability of LTFU is calculated using standard assumptions (persons who return to care after an interruption are treated as if they have remained in care for the duration of the interruption). In part B, the probability of LTFU is calculated on the basis of the assumption of no return to care after an interruption. In part C, patients are classified as LTFU when they first interrupt ART for more than 6 months (prospective LTFU definition).
Differences between cohorts enrolled during different periods remained significant in sensitivity analyses that considered alternative definitions of LTFU (3 months with no visit instead of 6 months), alternative assumptions about rates of ART interruption and ART resumption (including constant rates of interruption), and alternative assumptions about mortality, as well as in competing-risks analyses (Web Appendix 3 http://aje.oxfordjournals.org/content/180/12/1208/suppl/DC1). However, differences between cohorts enrolled in different periods became less pronounced when lower rates of ART resumption were assumed, and in the extreme case in which rates of restarting ART were set to zero, differences between cohorts ceased to be significant (Figure 1B). Similar results were obtained when using the prospective LTFU definition (Figure 1C). In Web Appendix 4 http://aje.oxfordjournals.org/content/180/12/1208/suppl/DC1, we provide a mathematical proof of the dependence of the retrospective LTFU definition on the time to analysis closure and demonstrate that the prospective LTFU definition is independent of the time to analysis closure.