The ability of aggregate data meta-analysis in predicting individual patient data meta-analysis




Poster session 3


Tuesday 25 October 2016 - 10:30 to 11:00


All authors in correct order:

Mao C1, Tang J1, Huang Y2
1 Hong Kong Cochrane Branch, Hong Kong
2 School of General Practice and Continuing Education, Capital Medical University, Beijing, China
Presenting author and contact person

Presenting author:

Chen Mao

Contact person:

Abstract text
Background: Aggregate data meta-analyses (ADMAs) are easier and less resource-consuming to conduct than individual-patient data meta-analyses (IPDMAs). The latter, however, is generally considered to have scientific advantages over the former, particularly in controlling for confounding and assessing interactions.

Objectives: We compared the overall results of the IPDMAs with those of their prior corresponding ADMAs to see how often the former were predicted by the later. We also explored factors that may make a difference between their results.

Methods: IPDMA articles were identified with a comprehensive search of PubMed, Embase and the Cochrane Database of Systematic Reviews. The ADMA articles published immediately prior to the IPDMA and matched in the research topic according to the patient, intervention, comparator, outcome and setting (PICOS) were then identified from PubMed and references of each IPDMA identified. We considered that the matched meta-analyses agreed with each other if the direction of the summary effect was the same in both the ADMA and its matched IPDMA. Sensitivity analyses were conducted by changing the definition of agreement slightly. Factors that might influence the agreement were investigated.

Results: We identified 829 IPDMA articles published and indexed before 9 August 2012. We identified a matched ADMA article for 129 (15.6%) of these 829 IPDMA articles, and this resulted in a total of 204 pairs of the ADMA and IPDMA matched to the same topic. Agreement in the direction of effect was observed in 187 (91.7%) of the 204 paired meta-analyses. The ADMA was more likely to agree with its corresponding IPDMA (P ≤ 0.05) when grey literature was searched, data were requested from authors, intention-to-treat analysis was used, and the overall result in ADMA was statistically significant.

Conclusions: Most ADMAs can provide a valid result on the direction of effect by summarizing grouped data from published primary studies, but should make greater efforts to search for grey literature, request necessary data from original authors, and use intention-to-treat analysis to increase its validity further.