Date of Completion

8-10-2020

Embargo Period

8-9-2025

Keywords

Network Meta Analysis, Likelihood Ratio Test, Plausibility Index, Hypothesis Testing, Bucher's Test, Inconsistency Dectection

Major Advisor

Ming-Hui Chen

Associate Advisor

Lynn Kuo

Associate Advisor

Dipak Dey

Field of Study

Statistics

Degree

Doctor of Philosophy

Open Access

Open Access

Abstract

One of the long-standing methodological issues in network meta analysis (NMA) is that of assessing homogeneity and consistency in treatment comparisons. In this dissertation, we construct general linear hypotheses to investigate homogeneity and consistency under general fixed effects models, one-equation and two-equation models, within Frequentist and Bayesian framework, respectively. We introduce the concept of 'inconsistency testable loops', which is the fundamental key to inconsistency detection in NMA. An algorithm is developed to compute all the inconsistency testable loops in network meta-data, as well as the contrast matrices under homogeneity and consistency assumptions. Under the normal fixed effects model, we show the equivalence of the likelihood ratio test (LRT) under the proposed linear hypotheses and Bucher's method for testing inconsistency based on comparison of the weighted averages of direct and indirect treatment effects. A novel Plausibility Index (PI) is developed to assess homogeneity and consistency. Theoretical properties of the proposed methodology are examined in details. A road map of treatment comparisons %while adjusting for heterogeneity and inconsistency is given. We apply the proposed methodology to analyze the network meta data from 29 randomized clinical trials with 11 treatment arms on safety and efficacy evaluation of cholesterol lowering drugs.

Available for download on Saturday, August 09, 2025

Share

COinS