PNS147 VISUALIZATION AND COMMUNICATION OF RESULTS FROM BAYESIAN NETWORK META-ANALYSIS: PAST, PRESENT, AND FUTURE

Abstract

Network meta-analysis (NMA) is commonly conducted within a Bayesian framework, with presentation and communication of results informed by previously published guidance documents. Recently, authors have proposed a multiplicity of approaches to visualize and report findings. This has caused issues in comparability of analyses across publications and complication in interpretation of results by non-statistical stakeholders. We provide recommendations for the use and interpretation of Bayesian NMA results and propose a standard battery of visualizations and statistics to improve understanding of this methodology. We compared communication and visualization of results of Bayesian NMAs from recent publications. Resulting approaches were classified according to their purpose and were assessed as to whether they: (a) support the intended use; (b) introduce difficulty in interpretation; and (c) adhere to a Bayesian decision-making framework. Core visualizations of Bayesian NMAs have remained relatively unchanged since initial publications. The choice of summary differs according to whether the goal is to compare a single agent to comparators (i.e., typically sponsored publications) or to estimate comparative effectiveness more broadly for academic or health technology assessment purposes. Heatmaps have been introduced to provide rapid summaries of various output statistics. Attempts to communicate balance of benefit and harm has focused on bivariate SUCRA plots that may be difficult to interpret when absolute rates differ. Composite forest plots augmented by additional summary characteristics such as probability better offer an attractive option to display interval estimates alongside Bayesian summaries, although the conventional focus on 95% intervals may inadvertently encourage dichotomization of evidence. We introduce two new visualizations: the ridge forest plot and minimally important benefit curves to support Bayesian presentation and interpretation of results. Communication of Bayesian NMAs for non-statistical audiences can be improved through clear recommendations for standard comprehensive reporting. We provide initial recommendations for visualization and communication of Bayesian NMA results.