Computational protocol: Comparative Effectiveness of Different Forms of Telemedicine for Individuals with Heart Failure (HF): A Systematic Review and Network Meta-Analysis

Similar protocols

Protocol publication

[…] Bayesian network meta-analyses and direct frequentist pairwise meta-analyses were conducted for all outcomes. For the primary analysis, the frequency data from each trial were used in the network meta-analysis using WinBUGS (MRC Biostatistics Unit, Cambridge, UK)[,]. Bayesian network meta-analysis (NMA) using a binomial likelihood model, which allows for the use of multi-arm trials, were conducted. Random effects network meta-analyses with vague priors were assigned for basic parameters were conducted for the analyses. The WinBUGS code used for the Random Effects Model is available in []. Three chains were fit in WinBUGS for each analysis, with 40,000 iterations, and a burn-in of 40,000 iterations. Odds ratios and 95% credible intervals were modelled using Markov chain Monte Carlo methods. We constructed all evidence networks using NodeXL[].Assessment of model fit for NMA comprised of assessment of the deviance information criterion (DIC) and the residual deviance in comparison with the number of unconstrained datapoints []. Models with smaller DIC were preferred to models with larger DIC. Similarly, the total value for the residual deviance should be lower than the number of unconstrained data points. To ensure convergence was reached, Brooks-Gelman-Rubin plots were assessed []. Model convergence is evident when the Gelman-Rubin statistic approaches 1.A network meta-analysis also requires that studies are sufficiently similar in order to pool their results. We assessed available study and patient characteristics to ensure similarity and to investigate the potential effect of heterogeneity on effect estimates. Inconsistency was assessed by comparing statistics for the deviance and deviance information criterion in fitted consistency and inconsistency models. Additionally, fixed effects models with vague priors were conducted.When considered appropriate, pair-wise meta-analyses were conducted by combining studies that compared the same interventions using a random-effects model. Heterogeneity was investigated by examining both forest plots and the inconsistency index (I2). [] I2 values of less than 25% represented mild heterogeneity, between 25% and 50% represented moderate heterogeneity, and greater than 50% represented considerable heterogeneity. Results having a p-value of less than 0.05 and 95% confidence intervals (CIs) that excluded 1 were considered to be statistically significant. These analyses were carried out using Comprehensive Meta-Analysis and Review Manager of the Cochrane Collaboration. The results from our network meta-analysis were qualitatively compared with direct, frequentist, pairwise estimates. […]

Pipeline specifications

Software tools WinBUGS, NodeXL
Applications Drug design, Protein interaction analysis
Organisms Homo sapiens
Diseases Heart Failure