Better Evidence Synthesis via Innovative Methods

Organizers: Haitao Chu (UMN)
Chair: Rakhi Kilaru (PPD)
Vice Chair: Haitao Chu (UMN)

Speakers:
Yong Chen (UPenn)
Lifeng Lin (FSU)
Amy Shi (SAS)
Haitao Chu (UMN)

Abstracts:

Title: A Robust and Computational-efficient Method for Multiple-outcome Network Meta-analysis
Speaker: Yong Chen (UPenn)

In many biomedical settings, there is an increasing number of interventions available for a disease condition. It is critical for clinical decision-making to accurately evaluate and compare the relative efficacy and safety, as well as other patient centered outcomes of these interventions. In this talk, we propose a network meta-analysis model for multiple clinical outcomes. Inspired by the idea of composite likelihood, the proposed method only requires specification of the marginal distribution of each outcome, and a pseudolikelihood is then constructed under a working independence assumption. We also develop a novel inferential procedure with associated efficient computational algorithm, which is statistically robust (i.e., requires minimal distributional assumptions) and computational stable and fast. We will illustrate our method through multiple case studies including a network meta-analysis of comparing 12 labor induction methods.

Title: Evidence inconsistency degrees of freedom in Bayesian network meta-analysis
Speaker: Lifeng Lin (FSU)

Network meta-analysis (NMA) is a popular tool to synthesize direct and indirect evidence for multiple-treatment comparisons, while evidence inconsistency greatly threatens its validity. The inconsistency degrees of freedom (ICDF) assesses the potential that an NMA might suffer from inconsistency. Because multi-arm studies provide intrinsically consistent evidence and complicate the ICDF’s calculation, the existing measure of ICDF is applicable to NMAs with two-arm studies only. However, NMAs often contain multi-arm studies. Driven from the existing ICDF measure and the effective number of parameters of a Bayesian model, we generalize the method for calculating the ICDF. Under the fixed- or random-effects setting, the new ICDF measure is calculated as the difference between the effective numbers of parameters of the consistency and inconsistency NMA models. We used artificial NMAs created based on an illustrative example and 39 real-world NMAs to compare the existing and new measures and evaluate their performance. In NMAs with two-arm studies only, the proposed ICDF measure under the fixed-effects setting was nearly the same with the existing measure. Among the 39 real-world NMAs, 27 (69%) contained at least one multi-arm study; existing measure was not applicable to them. The proposed method produced interpretable ICDFs in all NMAs, regardless of the presence of multi-arm studies. In summary, the ICDF is an important characteristic of an NMA. The proposed method enables its calculation in generic NMAs, and it may be routinely reported in practice.

Title: Fitting Bayesian network meta-analysis models using SAS
Speaker: Amy Shi (SAS)

Network meta-analysis synthesizes direct and indirect evidence on multiple treatments from a collection of independent studies or randomized controlled trials. The most widely used method in this field is contrast-based, in which a baseline treatment needs to be specified in each study, and the analysis focuses on modeling treatment contrasts (relative treatment effects, typically log odds ratios). However, population averaged treatment-specific parameters, such as absolute risks, cannot be estimated by this method without an external data source or a separate model for a reference treatment. Therefore, the arm-based network meta-analysis method has been proposed and been used increasingly.  We show how to employ SAS Bayesian procedures, PROC MCMC and PROC BGLIMM, to fit various arm-based network meta-analysis models, to estimate both absolute and relative effects, and to handle binary, continuous, and count outcomes. Those SAS Bayesian procedures provide convenient access with high performance and efficient sampling. We present  several examples illustrating how to use them for estimation, inference, and prediction in Bayesian Network meta-analysis.

Title: Bayesian Hierarchical Models Estimating CACE Accounting for Noncompliance in Meta-analysis
Speaker: Haitao Chu (UMN)

Noncompliance to assigned treatments is a common challenge in the analysis and interpretation of a randomized clinical trial (RCT). One approach to handle noncompliance is to estimate the complier-average causal effect (CACE) using the principal stratification framework, where CACE measures the impact of an intervention in the subgroup of the population that complies with its assigned treatment. When noncompliance data are reported in each trial, intuitively one can implement a two-step approach (i.e. first estimating CACE for each study and then combining them using a fixed-effect or random effects model) to estimate the population-averaged CACE in a meta-analysis. However, it is common that some trials do not report noncompliance data.  The two-step approach can be less efficient and potentially biased as trials with incomplete noncompliance data are excluded. In this paper, we propose a flexible Bayesian hierarchical CACE framework to simultaneously account for heterogeneous and incomplete noncompliance data in a meta-analysis of RCTs. The performance of the proposed method is evaluated by extensive simulations, and an example of a meta-analysis estimating the CACE of epidural analgesia on cesarean section, in which only 10 out of 27 studies reported complete noncompliance data.