Statistical modeling and verification for the synthesis of median survival time in multilevel meta-analysis of survival data

Abstract

Objective: Past meta-analyses of survival data have been over simplistic because of restricting to proportional hazard model, lackof intuitive results, and potential omitting information. These had the potential to recommend sub-optimal policies. Here wedevelop multilevel methods for combining median survival times (MSTs) for meta-analysis of survival data.Methods: We used simulated data to test and verify the synthesis model we developed. We generated the study-level data to fit multilevel model and individual patient data to calculate gold standard. We then used the Bland-Altman method and the relativechange from the gold standard to evaluate the fit of the models. Examples were presented in a meta-analysis to illustrate the feasibility of the models.Results: We generated eight sets of simulated datasets of different number of studies and sample size. We established themulti-level fixed and random effect models to pool the MSTs. The test of the fitness of the model showed that the means ofdifference (d) for all simulated datasets between the calculated values and the gold standards are no more than -0.230 and -0.329days and the largest 95% CIs of d are -3.823 3.364 and -3.936 3.278 days respectively. At least 91.9% and 92.3% of the difference between the estimated values and the gold standards are small. The real examples of a meta-analysis were provided with combined MSTs along with pooled HR.Conclusions: The multilevel models of synthesizing MSTs in survival data AD meta-analysis were verified with good fitting effects and provide more intuitive information.