Meta-analysis of joint longitudinal and event-time outcomes

Background: It is widely acknowledged that joint analysis of longitudinal and time-to-event data is preferable over separate longitudinal or time-to-event analyses as the joint approach can account for longitudinal measurement error, study dropout, and can produce less biased estimates [1]. However the literature focuses mainly on single study cases or doesn’t account for potential heterogeneity between studies or sites. In many situations, such as meta-analyses, it may be important to be able to account for this variability. Objectives: This investigation aims to assess the benefit of use of joint models compared to separate methods in meta-analyses. Methods: Joint longitudinal and time-to-event data were simulated under a range of associations, for high and low event rates, for 5 studies, under assumption of homogenous treatment effect or heterogeneous treatment effect across studies. A range of joint models, and separate longitudinal and survival models were fitted to each study within each simulation, and the results pooled using standard meta-analytic techniques. Estimates of the treatment affect, association parameters, bias, and coverage were output. Results: From the simulations, there appears to be a benefit for time-to-event coefficients to perform joint models during meta-analyses rather than separate analyses