(4 of 4) Network Meta-Analysis Series: Model Implementation
Model Implementation with Network Meta-Analysis:
We will fit our model using WinBUGS and SAS version 9.4. For the Bayesian implementation we will employ the binomial likelihood for dichotomous outcomes and will use uninformative prior distributions for the treatment effects, and a minimally informative prior distribution for the common heterogeneity SD depending on the outcome. Also, we will assume uninformative priors for all meta-regression coefficients. We will check for convergence using appropriate MCMC diagnostics.
GRADE quality assessment of all comparisons in the network:
We will also assess the quality of evidence contributing to network estimates of the main outcomes with the GRADE framework, which characterizes the quality of a body of evidence on the basis of the study limitations, imprecision, inconsistency, indirectness and publication bias. The starting point for confidence in each network estimate is high, but will be downgraded according to the assessments of these five domains.
Summary of Findings Table
We will interpret the findings of Network Meta-Analysis through NMA ‘Summary of findings” table developed based on the principles of the GRADE approach to rating certainty of evidence from NMAs. Though there is no standard NMA summary of findings table format, at beginning of the table, the basic information of NMA, such as target population, treatments, comparator (reference ), outcomes, study population will be presented. In the table body, we will exhibit total studies and total participants for each treatment and comparator; relative effects with 95% Crl; anticipated absolute effect with 95% Crl which include absolute treatments, placebo, and difference between treatments and placebo; certainty of evidence; ranking with 95% Crl; and finally interpretation of findings which provide recommendation of superiority based on ranking for each treatment.
In summary, we will draw NMA certainty level in the evidence from direct estimate and indirect evidence, high, moderate, low and very low as first component. We then will obtain uncertainty in treatment ranking by reanalyzing NMA of randomized trials, such as first top three ranks, other ranks, as second component. Combining first component and second component together, we will give a rank to all treatments as definitely superior, probably superior, probably inferior and definitely inferior.