Methodological issues in meta‐analysis of diagnostic accuracy studies

To the Editor: I wish to comment on several methodological issues in a recent systematic review and meta-analysis on the diagnostic accuracy of ultrasound-guided core-needle biopsy for head and neck lesions. This study suffers from several problems that are commonly seen in meta-analyses of diagnostic accuracy studies. First, meta-analysis requires specialized statistical methods to synthesize evidence from several studies to form a summary receiver operating characteristic (ROC) curve. The statistical methods used for evidence synthesis were not reported in the study by Novoa et al; however, it does not seem that ROC methods were used. Simple aggregation of studies can lead to errors in the estimates of sensitivity and specificity. Second, comparisons of accuracy should be made using the area under the ROC curve. Novoa et al compared the accuracy of cutting needle biopsy to the accuracy of fine-needle aspiration (FNA) biopsy using the data from Tandon et al, but neither study used ROC curves to assess accuracy. In addition, the Tandon et al study used simple aggregation rather than meta-analytic methods to estimate sensitivity and specificity and, for that reason, the estimates may not be reliable. Third, neither study used adequate methods of quality appraisal. Quality appraisal of included articles should be done using an instrument such as the quality assessment of diagnostic accuracy studies (QUADAS). It is well known that diagnostic studies often suffer from design deficiencies. Recent reviews have shown that level 2b diagnostic accuracy studies of FNA for salivary gland lesions (most FNA diagnostic accuracy studies are grade 2b) commonly suffer from partial verification, which leads to biased estimates of sensitivity and specificity. Therefore, a meta-analysis based on grade 2b studies is likely to produce biased results. Statistical methods are available to correct for verification bias, but such corrections were not performed in the Tandon et al study. For these reasons, the accuracy estimates obtained in both the Novoa et al and Tandon et al studies may be unreliable. Researchers should be aware of these methodological issues when reporting and interpreting the results of systematic reviews and meta-analyses.