Article type
Abstract
Background: Previous research shows that many authors of Cochrane overviews were also involved in some of the included systematic reviews (SRs). This type of dual (co-)authorship (DCA) may be considered to be a conflict of interest and a potential source of bias. No research has been conducted to investigate this in non-Cochrane overviews. Whether DCA constitutes a potential source of bias has also not been examined empirically.
Objectives: To estimate and compare the prevalence of DCA in overviews of reviews and investigate potential bias arising from DCA regarding quality assessments of included SRs.
Methods: We selected a sample of Cochrane (n=20) and non-Cochrane (n=78) overviews for analysis. We extracted data on the number of reviews affected by DCA and whether quality assessment of included reviews was conducted independently. We also extracted data on the quality assessments of the included SRs and compared mean quality scores of SRs affected versus not affected by DCA (for example, the number of items fulfilled in AMSTAR assessments). We calculated standardised-mean differences (SMD) to account for different assessment tools.
Results: Forty out of 78 non-Cochrane overviews (51%) and 18 out of 20 Cochrane overviews (90%) had included at least one SR with DCA. For Cochrane overviews, a median of 5 [interquartile range (IQR): 2.5 to 7] SRs were affected by dual (co-)authorship (median of included reviews 10). For non-Cochrane overviews a median of 1 [IQR: 0 to 2] of the included SRs were affected (median of included reviews 14). SRs affected by DCA scored significantly better in methodological quality assessments than SRs not affected by DCA (SMD: 0.58 [95%-CI: 0.27 to 0.90]).
Conclusions: Many authors of overviews often have an authorship on one or more of the underlying reviews. Our analysis shows that, on average, authors of overviews give higher-quality ratings to SRs in which they were involved than to other SRs. Conflict of interest is one explanation, but there are several others such as reviewer expertise. Independent and blinded reassessments of the reviews would provide more robust evidence on potential bias arising from DCA.
Objectives: To estimate and compare the prevalence of DCA in overviews of reviews and investigate potential bias arising from DCA regarding quality assessments of included SRs.
Methods: We selected a sample of Cochrane (n=20) and non-Cochrane (n=78) overviews for analysis. We extracted data on the number of reviews affected by DCA and whether quality assessment of included reviews was conducted independently. We also extracted data on the quality assessments of the included SRs and compared mean quality scores of SRs affected versus not affected by DCA (for example, the number of items fulfilled in AMSTAR assessments). We calculated standardised-mean differences (SMD) to account for different assessment tools.
Results: Forty out of 78 non-Cochrane overviews (51%) and 18 out of 20 Cochrane overviews (90%) had included at least one SR with DCA. For Cochrane overviews, a median of 5 [interquartile range (IQR): 2.5 to 7] SRs were affected by dual (co-)authorship (median of included reviews 10). For non-Cochrane overviews a median of 1 [IQR: 0 to 2] of the included SRs were affected (median of included reviews 14). SRs affected by DCA scored significantly better in methodological quality assessments than SRs not affected by DCA (SMD: 0.58 [95%-CI: 0.27 to 0.90]).
Conclusions: Many authors of overviews often have an authorship on one or more of the underlying reviews. Our analysis shows that, on average, authors of overviews give higher-quality ratings to SRs in which they were involved than to other SRs. Conflict of interest is one explanation, but there are several others such as reviewer expertise. Independent and blinded reassessments of the reviews would provide more robust evidence on potential bias arising from DCA.