Article type
Year
Abstract
Background: A number of 'meta-epidemiological' studies have provided empirical evidence on sources of bias in randomized controlled trials (RCTs). However, findings vary between studies. The extent of overlap between the meta-analyses and trials included in these studies is unknown.
Objectives: We aimed to combine data from six meta-epidemiological studies into a single data set in which there was little or no duplication of trials or systematic reviews.
Methods: Data were combined into a single Access database. Unique identifiers (PMID numbers) were assigned for each metaanalysis and each trial using the identification number assigned by cataloguing databases. The hierarchy of searching was MEDLINE followed by EMBASE. References indexed in neither database were hand-checked for duplicates and unique studies were assigned a new identification number. Interventions and outcomes were extracted for each meta-analysis. The PMID numbers of trials included in each meta-analysis were compared with those in every other meta-analysis. Meta-analyses containing overlapping RCTs were grouped together. Within these groups, we coded trials in each pair of meta-analyses according to their similarity in terms of participant numbers and event rates in each arm. Finally, a single meta-analysis was chosen from a pair or set of meta-analyses containing important numbers of identical RCTs, using a well-defined hierarchy.
Results: Compared with the original database of 3468 RCTs in 296 meta-analyses, the final deduplicated database comprised 2792 trial results from 2714 trial bibliographic references in 248 meta-analyses (245 systematic reviews). Examples of overlapping sets of metaanalyses and trials will be presented.
Conclusions: Deduplication of the data set was complicated by a number of issues including the use of data from different trial arms of the same trial. There was surprisingly little duplication across the included meta-epidemiological studies. The final database will be used to examine combined evidence of bias in RCTs and meta-analyses.
Objectives: We aimed to combine data from six meta-epidemiological studies into a single data set in which there was little or no duplication of trials or systematic reviews.
Methods: Data were combined into a single Access database. Unique identifiers (PMID numbers) were assigned for each metaanalysis and each trial using the identification number assigned by cataloguing databases. The hierarchy of searching was MEDLINE followed by EMBASE. References indexed in neither database were hand-checked for duplicates and unique studies were assigned a new identification number. Interventions and outcomes were extracted for each meta-analysis. The PMID numbers of trials included in each meta-analysis were compared with those in every other meta-analysis. Meta-analyses containing overlapping RCTs were grouped together. Within these groups, we coded trials in each pair of meta-analyses according to their similarity in terms of participant numbers and event rates in each arm. Finally, a single meta-analysis was chosen from a pair or set of meta-analyses containing important numbers of identical RCTs, using a well-defined hierarchy.
Results: Compared with the original database of 3468 RCTs in 296 meta-analyses, the final deduplicated database comprised 2792 trial results from 2714 trial bibliographic references in 248 meta-analyses (245 systematic reviews). Examples of overlapping sets of metaanalyses and trials will be presented.
Conclusions: Deduplication of the data set was complicated by a number of issues including the use of data from different trial arms of the same trial. There was surprisingly little duplication across the included meta-epidemiological studies. The final database will be used to examine combined evidence of bias in RCTs and meta-analyses.