Article type
Year
Abstract
Background: Meta-analysis of individual patient data (IPD) is the 'gold-standard' for evidence synthesis. Yet for some studies the IPD may be lost, destroyed, or not provided. Simmonds et al1 found that seven (24%) of 29 IPD meta-analyses only obtained IPD for > 80% of all randomized patients. In such situations, the available IPD could be supplemented with aggregate data (AD), such as a treatment effect estimate and standard error, from those studies for whom IPD are not available. Methods for combining IPD and AD are then required, but this issue has received little attention.
Objectives: We performed a systematic review to: (i) assess if and how meta-analysts combine IPD and AD in practice, and (ii) identify statistical methods for combining IPD and AD in evidence synthesis as proposed in methodological papers.
Methods: We searched MEDLINE, The Cochrane Library and other relevant databases to identify, critically appraise, and then
summarise IPD meta-analyses and relevant statistical articles.
Results: Our preliminary results indicate that approximately 40 (of about 240) IPD meta-analyses have combined IPD and AD.
Additionally eight relevant statistical articles have been identified. Across this literature, four different methods have been proposed or applied, ranging from simply reducing the IPD to AD and employing standard AD meta-analysis techniques, to Bayesian simultaneous analysis of IPD-only and AD-only hierarchical related models. We will describe each of these methods and discuss their benefits and limitations; the detail of our final results will also be provided.
Conclusions: It is clear that the joint synthesis of IPD and AD is a relevant research issue. The four methods identified are thus potentially important, but their validation and development is needed. We will thus conclude with further research suggestions.
References
1. Simmonds MC, Higgins JP, Stewart LA, Tierney JF, Clarke MJ, Thompson SG. Meta-analysis of individual patient data from randomized trials: a review of methods used in practice. Clinical Trials 2005; 2:209-17.
Objectives: We performed a systematic review to: (i) assess if and how meta-analysts combine IPD and AD in practice, and (ii) identify statistical methods for combining IPD and AD in evidence synthesis as proposed in methodological papers.
Methods: We searched MEDLINE, The Cochrane Library and other relevant databases to identify, critically appraise, and then
summarise IPD meta-analyses and relevant statistical articles.
Results: Our preliminary results indicate that approximately 40 (of about 240) IPD meta-analyses have combined IPD and AD.
Additionally eight relevant statistical articles have been identified. Across this literature, four different methods have been proposed or applied, ranging from simply reducing the IPD to AD and employing standard AD meta-analysis techniques, to Bayesian simultaneous analysis of IPD-only and AD-only hierarchical related models. We will describe each of these methods and discuss their benefits and limitations; the detail of our final results will also be provided.
Conclusions: It is clear that the joint synthesis of IPD and AD is a relevant research issue. The four methods identified are thus potentially important, but their validation and development is needed. We will thus conclude with further research suggestions.
References
1. Simmonds MC, Higgins JP, Stewart LA, Tierney JF, Clarke MJ, Thompson SG. Meta-analysis of individual patient data from randomized trials: a review of methods used in practice. Clinical Trials 2005; 2:209-17.