Some statistical problems in sequential meta-analysis

Dogo, Samson Henry (2016) Some statistical problems in sequential meta-analysis. Doctoral thesis, University of East Anglia.

Preview

PDF Download (77MB) | Preview

Abstract

The objective of meta-analysis is to combine results from several independent studies in
order to make evidence more generalisable and provide evidence base for decision making.
However, recent studies show that the magnitude of effect size estimates reported in many
areas of research have significantly changed over time. These temporal trends can be dramatic
and even lead to the loss or gain of the statistical significance of the cumulative treatment
effect (Kulinskaya and Koricheva, 2010). Standard sequential methods including cumulative
meta-analysis, sequential meta-analysis, the use of quality control charts and penalised z-test
have been proposed for monitoring the trends in meta-analysis. But these methods are only
effective when monitoring in fixed effect model (FEM) of meta-analysis. For random-effects
model (REM), the analysis incorporates the heterogeneity variance, t2 and its estimation
creates complications. This thesis proposes the use of a truncated CUSUM-type test (Gombay
method) for sequential monitoring in REM, and also examines the effect of accumulating
evidence in meta-analysis. Simulations show that the use of Gombay method with critical
values derived from asymptotic theory does not control the Type I error. However, the
test with bootstrap-based critical values (retrospective Gombay sequential bootstrap test
for REM) leads to a reduction of the difference between the true and nominal levels, and
thus constitutes a good approach for monitoring REM. Application of the proposed method
is illustrated using two meta-analytic examples from medicine. Two kinds of bias associated
with accumulating evidence, termed \sequential decision bias" and \sequential design bias" are
identified. It was demonstrated analytically and by simulations that both types of sequential
biases are non negligible. Simulations also show that sequential biases increase with increased
heterogeneity.

Item Type:	Thesis (Doctoral)
Faculty \ School:	Faculty of Science > School of Computing Sciences
Depositing User:	Jackie Webb
Date Deposited:	07 Jun 2016 08:34
Last Modified:	07 Jun 2016 08:34
URI:	https://ueaeprints.uea.ac.uk/id/eprint/59244
DOI:

Downloads

Actions (login required)

View Item