Use this URL to cite or link to this record in EThOS:
Title: Synthesis of clustered time-to-event outcomes
Author: Nisar, Wirda
ISNI:       0000 0005 0287 7604
Awarding Body: University of Liverpool
Current Institution: University of Liverpool
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Thesis embargoed until 01 Jan 2026
Access from Institution:
Background: Previous studies have proposed a random effects hierarchical Cox Proportional Hazard (PH) model for the Meta-analysis (MA) of Individual Participant Data (IPD) with time- to-event (TTE) outcomes. In these models the random effects either capture the variation in the treatment effect across different trials or the between-trials variation caused by baseline risk, and are suitable for the MA of trials comparing two treatments. However, if trials compare more than two treatments that require a simultaneous analysis, such as in the case of network meta-analysis (NMA), there is a need to account for the correlation between treatment effects. The aim of this thesis is to explore, demonstrate and extend the use of MA methodologies to account for correlated random-effects in the analysis of treatment effect in clinical trials. In particular, an emphasis is done on statistical inference to ensure appropriate implementation of such analyses. Objectives: (1) To perform a review of methods for the synthesis of clustered time-to-event data; (2) To undertake a simulation study to compare alternative methods for the estimation of random effects Cox PH models under different conditions; (3) To explore methods through application of data from epilepsy clinical trials; (4) To extend a current method to incorporate multiple correlated random-effects; (5) To explore the adequacy of the model and the inferential procedure through simulation study; (6) To apply the random effects Cox PH model with correlated random effects to data from epilepsy clinical trials. Results: The methodology review identified 55 articles, among which 36 articles described semi-parametric methods, 10 described parametric methods, and 9 described both, 2 parametric and semi-parametric methods. The simulation study identified that, in general, one-stage and two-stage approaches produced identical estimates of the pooled log Hazard Ratio (HR). For one-stage approaches described by Vaida and Xu 2000 and Ripatti and Palmgren 2000, there exist a moderate underestimation of between-trials heterogeneity variance. On the contrary, the results revealed that easy to compute DerSimonian and Laird methods of moments (MOM) is not suitable for the estimation of between-trials heterogeneity variance. From the analysis of epilepsy data set h-likelihood approach appeared to be computationally intensive for all the three one-stage random-effects IPD models. The simulation results further showed that proposed IPD NMA models for accounting for correlated random-effects performed well, achieving nominal coverage of 95% for the overall multiple treatment effects under correct settings. For the epilepsy data, results of the proposed models showed good agreement across all comparisons. Conclusion: We highlighted pragmatic approaches to analysis of IPD TTE data based on a methodology review and existing empirical comparisons of one and two-stage IPD MA methods through simulation studies and application, which could be helpful in the choice of methods. The NMA models presented in this thesis are novel in terms of modelling framework and estimation procedure. The proposed novel one-stage IPD NMA models can be further modified to allow for the complicated covariance structure between random effects, likely to be considered as a future project.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral