Διερεύνηση στατιστικών μεθόδων για την ανίχνευση του συστηματικού σφάλματος δημοσίευσης και της επίδρασης μικρών μελετών στα μοντέλα μετά-ανάλυσης (Master thesis)
Meta-analysis has been an established evidence synthesis method in the field of medicine and is increasingly employed for assessing the effectiveness of interventions. It ranks at the top of the evidence-based hierarchy and is considered by many, including the World Health Organization (WHO) to be the most appropriate method for improving clinical practice. However, there are some methodological weaknesses, which may compromise the results form meta-analysis. In the context of this postgraduate dissertation, we will focus on the effect of publication bias and on the possible influence of small size studies on estimating a meta-analytic treatment effect (small-study effect). Initially, Chapter 1 describes the study designs of evidence-based science. More specifically, the different types of research with their characteristics and differences are described. Additionally, it states the purpose of postgraduate thesis and the main contents of the chapters. Chapter 2 refers to the systematic review methodology, describing its basic principles and stages its implementation in detail. In Chapter 3 arises the need to quantify and compare the effectiveness of the interventions satisfying the eligibility criteria in systematic review. We describe effect sizes for dichotomous and continuous outcomes. Chapter 4 describes the statistical method of meta-analysis, which synthesizes quantitatively the effect sizes from the included studies to get a pooled result. We present the two most popular meta-analysis models, the fixed effect model and the random effect model. We also define the heterogeneity, the variance of the underlying true effects, and present methods to identify it. We also present meta-regression models that are commonly applied to explore heterogeneity. In Chapter 5, we present the problem of publication bias and the possible effect of small size studies on the results of meta-analysis (small-study effect). We describe the funnel plot and its more advanced version, contour enhanced funnel plot. We then concentrate on interpreting the possible asymmetry of the funnel plot. For this reason, reference is made to the various statistical tests and especially to the linear regression tests developed for asymmetry control. Finally, for the identification of the publication bias the Fail-safe N, Trim and Fill methods are described as well as a statistical method based on the Copas selection model. As reported, a large number of statistical tests have been developed to control the asymmetry of the funnel plot. The most prominent ones are Egger's, Begg's rank correlation and Thompson & Sharp's tests. We carry out a simulation study to evaluate these tests for detecting small-study effects and publication bias in Chapter 6. The dissertation ends with an appendix describing the R code used to carry out the simulation study.
|Institution and School/Department of submitter:||Πανεπιστήμιο Ιωάννινων. Σχολή Θετικών Επιστημών. Τμήμα Μαθηματικών|
|Keywords:||Μετα-ανάλυση,Συστηματικό σφάλμα δημοσίευσης,Επίδραση μικρών μελετών,Meta-analysis,Small study effect,Publication bias,Systematic review|
|Appears in Collections:||Διατριβές Μεταπτυχιακής Έρευνας (Masters)|
Please use this identifier to cite or link to this item:This item is a favorite for 0 people.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.