The review addressed a well-defined question and had clear inclusion criteria. The search for published studies was thorough, but limiting the review to English language studies risked language bias. A diagnostic filter was used in the search, which may have resulted in relevant studies being missed. Unpublished studies were excluded, which raised the risk of publication bias.
The quality of included studies and overall strength of evidence were assessed with appropriate criteria and the results were used in the synthesis. However, the calculation of a composite quality score from the QUADAS results was inappropriate and meant that readers could not determine which studies were subject to which types of bias. It appeared that validity assessment was conducted by one person, which increased the risk of errors and bias. Measures were taken to minimise errors and bias in study selection and data extraction. Full details of included studies were available in the paper or the full report.
Data on diagnostic accuracy outcomes were synthesised by meta-analysis to produce pooled estimates of sensitivity and negative likelihood ratio. The methods used were chosen to compensate for statistical heterogeneity between studies and seem appropriate. However, clinical heterogeneity between studies was present and this could affect the reliability of the meta-analysis. The authors stated that statistical heterogeneity was assessed, but the results were not reported in the paper.
In view of these issues and the low quality of the evidence, the authors' conclusions should be treated with caution.