Table 3

Risk of bias assessment

Risk of bias criteriaBagul and Kirkham14Breau et al15Turner et al16Shukralla et al17Capili et al18Pitrou et al20Haidich et al21
Representativeness of sample of trials (low if trials were searched across unselected journals and across a reasonable time period)HighHighLowLowLowHighHigh
Blinding of reviewers during CONSORT-harms data extraction (low if reviewers blinded to study authors, institution, journal name and sponsors)HighLowLowHighUnclearHighHigh
Selective outcome reporting (low if all CONSORT-harms criteria assessed)Low*Low*Low*†Low*HighHighLow
Reliability of data extraction (low if more than one reviewer assessed the CONSORT harms criteria for each review that was undertaken, with a description of how agreement was achieved)HighLowLowLowLowLowLow
  • *Recommendation 9 was not included in these studies as subgroup analysis was either not reported in any of the included studies or considered to be irrelevant for the therapeutic area being investigated.

  • †Authors response: ‘Recommendation 8 has been captured elsewhere in data extraction, to report this item would be to duplicate information presented’.

  • ‘Recommendation 10 was considered too vague to assess with any objectivity so we decided to leave this item, especially given that some of our primary outcomes were already reasonably subjective’.