Table 6 Data extraction
Step | Sub-step | Methods/approaches | Sources ▪ Examples |
|---|---|---|---|
1.0 Plan the data elements to extract | |||
1.1 Determine the data to extract on the characteristics of SRsa | Becker 2008 [1]; Caird 2015 [31]; JBI 2015 [40, 41]; Li 2012 [44]; Ryan 2009 [53, 54] | ||
1.2 Determine the data required to assess which SRs address the overview question and allow assessment of the overlap across SRsa | Smith 2011 [57] | ||
1.3 Determine data to extract about the results from the SRs for each relevant primary outcome | |||
1.3.1 Extract M-A results | Becker 2008 [1]; Caird 2015 [31]; Hartling 2012 [35]; Smith 2011 [57] | ||
1.3.2 Extract numeric trial results | Thomson 2013 [59] | ||
1.3.3 Extract narrative results | Bolland 2014 [30]; JBI 2015 [40, 41]; Li 2012 [44]; Ryan 2009 [53, 54] | ||
1.3.4 Extract a combination of 1.3.1–1.3.3 | |||
1.3.5 Extract risk of bias assessment (overall assessment, or domain/item level data, or both) and certainty of the evidence | Becker 2008 [1]; Hartling 2012 [35]; JBI 2015 [40, 41]; Li 2012 [44]; Ryan 2009 [53, 54] | ||
1.4 Determine the data to extract from primary studiesa | |||
1.4.1 Extract numerical trial results | Caird 2015 [31] | ||
1.4.2 Extract data required to assess risk of bias for each domain or item | Hartling 2012 [35] | ||
1.5 Develop a data extraction forma | Becker 2008 [1]; Cooper 2012 [32]; Hartling 2012 [35]; JBI 2015 [40, 41]; Singh 2012 [56] | ||
2.0 Plan the data extraction process | |||
2.1 Determine the sources where data will be obtained from | |||
2.1.1 SRs | Becker 2008 [1]; Bolland 2014 [30]; Caird 2015 [31]; CMIMG 2012 [4]; Hartling 2014 [37]; JBI 2015 [40, 41]; Pieper 2012 [6, 45] | ||
2.1.2 Primary studies | Caird 2015 [31]; Salanti 2011 [3]; Thomson 2013 [59]; Whitlock 2008 [48,49,50,51,52] | ||
2.1.3 Registry entries (for SRs and/or trials) | Inferred method | ||
2.1.4 A combination of the above | Caird 2015 [31]; Salanti 2011 [3]; Thomson 2013 [59]; Whitlock 2008 [48,49,50,51,52] | ||
2.2 Determine how overlapping information across SRs will be handled | |||
2.2.1 Extract information from all SRs | Bolland 2014 [30]; Caird 2015 [31]; CMIMG 2012 [4]; Cooper 2012 [32]; Hartling 2014 [37]; JBI 2015 [40, 41]; Pieper 2014 [46]; White 2009 [48,49,50,51,52] | ||
2.2.2 Extract information from only one SR based on a priori eligibility criteria | Cooper 2012 [32]; CMIMG 2012 [4]; Foisy 2011 [34]; Hartling 2014 [37]; Pieper 2012 [6, 45]; Pieper 2014 [47]; Thomson 2013 [59] ▪ SR with the greatest number of trials (Cooper 2012 [32]) | ||
2.3 Determine how discrepant data across SRs will be handled in data extraction | |||
2.3.1 Extract all data, recording discrepancies | Becker 2008 [1]; Bolland 2014 [30]; Caird 2015 [31]; Kovacs 2014 [42]; Pieper 2012 [6, 45]; Pieper 2014 [46]; Smith 2011 [57]; Thomson 2010 [58] | ||
2.3.2 Extract data from only one SR based on a priori eligibility criteria | Cooper 2012 [32]; Pieper 2014 [47] ▪ Most recent SR and SR of the highest quality (Pieper 2014 [47]) ▪ Highest quality SR (Cooper 2012 [32]) | ||
2.3.3 Extract data element (e.g. effect estimates, quality assessments) from the SR which meets decision rule criteria | Bolland 2014 [30]; Cooper 2012 [32] ▪ SR that reports the most complete information on effect estimates (Bolland 2014 [30]) | ||
2.3.4 Reconcile discrepancies through approaches outlined in 2.4 | Bolland 2014 [30]; Caird 2015 [31]; Flodgren 2011 [33]; JBI 2015 [40, 41]; Salanti 2011 [3]; Thomson 2010 [58]; Whitlock 2008 [48,49,50,51,52] | ||
2.4 Determine additional steps to deal with missing data from SRs, or when there is variation in information reported across SRs | |||
2.4.1 Retrieve reports of the primary studies | Bolland 2014 [30]; Caird 2015 [31]; CMIMG 2012 [4]; Flodgren 2011 [33]; Pieper 2012 [6, 45]; Pieper 2014 [47]; Salanti 2011 [3]; Thomson 2010 [58]; White 2009 [49,50,51] | ||
2.4.2 Contact SR or trial authors, or both, for missing info and/or clarification | Bolland 2014 [30]; Flodgren 2011 [33]; JBI 2015 [40, 41]; Whitlock 2008 [49,50,51] | ||
2.4.3 Search SR or trial registry entries for information | Inferred method | ||
2.4.4 A combination of the above approaches | Bolland 2014 [30]; Caird 2015 [31]; Salanti 2011 [3]; Thomson 2010 [58]; Whitlock 2008 [48,49,50,51,52] | ||
2.4.5 Do not take additional steps to deal with missing data or discrepancies | Becker 2008 [1]; Caird 2015 [31]; Foisy 2011 [34]; JBI 2015 [40, 41] | ||
2.5 Pilot the data extraction forma | |||
2.6 Determine the number of overview authors required to extract dataa | |||
2.6.1 Single, double, or more | Becker 2008 [1]; Bolland 2014 [30]; Hartling 2012 [35]; JBI 2015 [40, 41]; Li 2012 [44]; White 2009 [48,49,50,51,52] | ||
2.6.2 Data extraction versus data checking | Becker 2008 [1]; CMIMG 2012 [4]; Singh 2012 [56]; Whitlock 2008 [48,49,50,51,52] ▪ Evaluate a random sample of primary studies to ensure that data abstraction is accurate and reproducible (Whitlock 2008 [48,49,50,51,52]) | ||
2.7 Determine if authors (co-)authored one or several of the reviews included in the overview, and if yes, plan safeguards to avoid bias in data extraction | Buchter 2015 [60] ▪ Overview authors do not extract data from their co-authored SRs | ||