Article type
Year
Abstract
Introduction: Systematic reviews using meta-analysis to combine the results of related randomised controlled trials (RCTs) are increasingly common. However, there is little empirical evidence of the advantages and disadvantages of the different techniques for these, which can include calculations based solely on information contained in a few published papers, more detailed analysis of aggregate data supplied by trialists, and time-to-event analysis of updated, and centrally checked, individual patient data (IPD).
Methods: The first and last of the above approaches have been compared using data from meta-analyses based on IPD from relevant RCTs assessing cisplatin-based combination chemotherapy versus single-agent non-platinum drugs in advanced ovarian cancer, and the use of ovarian ablation in operable breast cancer among pre-menopausal women.
Results: For the ovarian cancer meta-analysis, data extracted from publications gave an overall estimated survival benefit for combination chemotherapy of 7.5% at 30 months (p=0.03), whereas the result for the IPD meta-analysis was only 2.5% (p=0.30). Conversely, for the breast cancer example, a crude meta-analysis of the published results for each trial gave an overall estimate of a 1% (p=0.96) improvement in survival for ovarian ablation compared to control. The IPD meta-analysis showed a 21% (p=0.0004) improvement at 15 years. The differences were due to several factors (including publication bias, "ineligible" patients, time-to-event analysis, and longer follow-up), which had different impacts in the two examples.
Discussion: IPD meta-analyses are the only way to perform time-to-event analysis, and the most practical way of doing sub-group analyses by allowing common prognostic and outcome variables to be used. The detailed data checking possible also improves the accuracy of the data and allows the integrity of the randomisation and follow-up procedures to be assessed centrally. The effect of all these additional processes on the result of a particular meta-analysis will vary between health care questions, and the direction of the effect might not be predictable in advance. IPD meta-analyses may, therefore, represent a "gold-standard" for systematic reviews as they also allow for greater involvement of the trialists in the identification of trials, and the interpretation and dissemination of the results. They are, however, the most time and resource intensive of the approaches to meta-analysis. The Cochrane Collaboration Working Group on IPD meta-analyses will, therefore, help to address the issue of when and how such projects should be undertaken.
Methods: The first and last of the above approaches have been compared using data from meta-analyses based on IPD from relevant RCTs assessing cisplatin-based combination chemotherapy versus single-agent non-platinum drugs in advanced ovarian cancer, and the use of ovarian ablation in operable breast cancer among pre-menopausal women.
Results: For the ovarian cancer meta-analysis, data extracted from publications gave an overall estimated survival benefit for combination chemotherapy of 7.5% at 30 months (p=0.03), whereas the result for the IPD meta-analysis was only 2.5% (p=0.30). Conversely, for the breast cancer example, a crude meta-analysis of the published results for each trial gave an overall estimate of a 1% (p=0.96) improvement in survival for ovarian ablation compared to control. The IPD meta-analysis showed a 21% (p=0.0004) improvement at 15 years. The differences were due to several factors (including publication bias, "ineligible" patients, time-to-event analysis, and longer follow-up), which had different impacts in the two examples.
Discussion: IPD meta-analyses are the only way to perform time-to-event analysis, and the most practical way of doing sub-group analyses by allowing common prognostic and outcome variables to be used. The detailed data checking possible also improves the accuracy of the data and allows the integrity of the randomisation and follow-up procedures to be assessed centrally. The effect of all these additional processes on the result of a particular meta-analysis will vary between health care questions, and the direction of the effect might not be predictable in advance. IPD meta-analyses may, therefore, represent a "gold-standard" for systematic reviews as they also allow for greater involvement of the trialists in the identification of trials, and the interpretation and dissemination of the results. They are, however, the most time and resource intensive of the approaches to meta-analysis. The Cochrane Collaboration Working Group on IPD meta-analyses will, therefore, help to address the issue of when and how such projects should be undertaken.