PhD Tat Thang Vo: Causal inference in meta-analysis and mediation analysis.

03-09-2020 from 16:00 to 18:00
Het Pand, Onderbergen 1, 9000 Gent.
Tat Thang Vo
Add to my calendar

PhD defense Tat Thang Vo

In this thesis, we aim to contribute to the two fast-growing fields of meta-analysis and mediation analysis. In the first part, we develop a novel framework for individual patient data meta-analysis of randomized controlled trials that allows one to (i) control for differences the case-mix across studies and reduce heterogeneity, and to (ii) infer the treatment effect for a population that is well-defined in terms of case mix. This is achieved by standardizing the results from the different trials to the same patient population, e.g. the patient population observed in one of the trials or any other population of interest, before meta-analyzing them as in a classical twostage IPD meta-analysis. We the extend the conventional I2 statistic and the prediction interval proposed in standard meta-analysis methodology, using them to support the casemix and beyond-case-mix heterogeneity decomposition in the new framework.

In the second part, we shift the focus on mediation analysis. We first conduct a systematic review to describe the methodological characteristics of mediation analyses (MAs) reported in recent randomized controlled trials (RCTs) and propose recommendations on the planning, conduct and reporting of MAs in practice. We then develop a novel mediation approach which allows one to investigate the effect of a treatment mediated through a mediator that is repeatedly measured over time. The proposed approach accounts for the feedback relationship that may exist between the mediators and the confounders over time, while addressing the complications when the primary outcome is time-to-event and subject to competing risks. This approach is illustrated on the data of the English Longitudinal Study of Aging, to identify whether loneliness mediated the impact of hearing loss on dementia, accounting for mortality as a competing event.