Novel approaches to evidence synthesis

Authors

  • C. Vögele
  • M. Johnston
  • M. de Bruin
  • R. Powell
  • X. Li
  • D. French
  • M. Byrne

Abstract

Aims: This symposium aims to highlight new approaches to tackling challenges encountered in evidence synthesis. Specifically, it considers 1) how an ontology of key elements of research reports is necessary to enable machine learning (ML) to support extraction; 2) how variability in ‘control’ groups can be taken into account when synthesizing evidence; 3) the role of advanced statistical procedures to a) tease apart intervention effects and b) estimate the influence of interactions between moderators; 4) methods for synthesising the evidence of multiple systematic reviews. Rationale: Given the increasing quantity of studies addressing issues of interest to health psychology and behaviour change, systematic, high quality synthesis of evidence is vital to drive forward the scientific process and to inform health policy, enhancing population health. This symposium discusses important challenges to the effective synthesis of such evidence and presents novel solutions. Summary: Marie Johnston considers how ML could be harnessed to support researchers with the time-consuming process of data extraction, and describes the development of an ontology of elements which may allow machines to facilitate data extraction. Using data from several systematic reviews, Marijn de Bruin addresses the problem of assessing behavioural support provided to control groups, especially those receiving ‘treatment as usual’, accounting for variability in control groups in meta-analyses, and estimating intervention effect sizes adjusted for variability in control groups. Two presentations provide examples of how advanced statistical methods can allow investigation into how interventions take effect: Rachael Powell presents findings illustrating a component-based network meta-analysis approach to teasing apart the impact of intervention components. Xinru Li demonstrates how Meta-CART (Classification-And-Regression-Trees) allows interactions between moderators to be identified, and their significance tested. Finally, David French demonstrates how systematic review methods can be applied to the synthesis of systematic reviews before Molly Byrne provides an overview discussion of the symposium themes.

Published

2017-12-31

Issue

Section

Symposia