Research Synthesis Methods and Their Application to Evidence-Based Medicine

Meta-analysis is a valuable tool to synthesize evidence and pool results from multiple sources. It plays an integral role in evidence-based medicine and may have a direct impact in the clinical setting. This dissertation explores several topics related to research synthesis methods and their application to evidence-based medicine. Chapter 2 methodologically assesses the current systematic reviews and meta-analyses on COVID-19. COVID-19 has caused an ongoing public health crisis. Many systematic reviews and meta-analyses have been performed to synthesize evidence for better understanding this new disease. However, some concerns have been raised about rapid COVID-19 research. The current systematic reviews and meta-analyses on COVID-19 might suffer from low transparency, high heterogeneity, and suboptimal statistical methods. In chapter 3, we investigate the performance of several methods commonly used for meta-analyses of diagnostic studies, including the summary ROC (SROC) approach, bivariate linear mixed model (LMM), and generalized linear mixed model (GLMM). The variation of estimates calls into question the appropriateness of the normality assumption required by the SROC method and LMM. In cases of notable differences presented in these methods' results, the GLMM is preferred. Network meta-analysis (NMA), an extension of pairwise meta-analysis, is a statistical method often used to draw conclusions about multiple treatment comparisons. It simultaneously synthesizes both direct and indirect evidence, where the direct evidence comes from head-to-head trials, and the indirect evidence comes from indirect comparisons with common comparators. The prior choice of between-study heterogeneity is critical in Bayesian NMAs. Chapter 4 assesses recent NMAs published in high-impact medical journals to evaluate how conclusions differ based on the application of various prior distributions. We re-analyze the NMAs using 3 commonly-used non-informative priors and empirical informative log-normal priors. The informative priors produce substantially more precise estimates than non-informative priors, especially for NMAs with few studies. The surface under the cumulative ranking curve (SUCRA) and the P-score are increasingly used to quantify treatment ranking under the Bayesian and frequentist NMA frameworks, respectively. They provide summary scores of treatments among the existing studies in an NMA. Clinicians are frequently interested in applying such evidence from the NMA to future decision-making. The heterogeneity between existing studies in the NMA and the future study needs to be considered in this prediction process. Chapter 5 extends the frequentist treatment ranking measure, P-score, to the Bayesian framework and the future study setting, focusing on NMAs with binary outcomes. Two empirical examples are used to illustrate the proposed measure. Chapter 6 considers an extension of the approach presented in chapter 5 beyond binary data. Using an extensive database of NMAs identified in a recent survey, we re-perform all eligible NMAs and calculate the differences between treatment ranking measures and those between predictive treatment ranking measures. We investigate the empirical distribution of these differences and assessed the changes of treatment ranking measures when making predictions. The predictive P-score can help inform medical decision-making in future studies. Clinicians may use these findings to interpret the magnitudes of differences between treatment ranking measures and quantitively assess treatment rankings in future NMAs.