Article type
Year
Abstract
Background: Individual Patient Data (IPD) meta-analyses are regarded as the 'gold standard' for systematic reviews. This conclusion also applies to systematic reviews of diagnostic test accuracy (DTA) studies. In recent years, an increasing number of DTA systematic reviews with IPD meta-analysis have been done, but so far there are no standard methods for data collection and statistical analysis in DTA IPD meta-analyses.
Methods: A systematic search was performed to identify published articles containing IPD meta-analyses of DTA studies. Embase and MEDLINE databases were searched from 2000 to 2014. Data was extracted with a data extraction form developed for this study including both multiple choices and open questions.
Results: Twenty-nine DTA IPD meta-analyses articles published between 2006 and 2014 were selected as the subjects for the final analysis. More than 80% (24/29) of these were based on datasets containing individual patient information provided by authors of primary studies; 60% (15/25) included data from fewer than 10 primary studies; 60% (15/25) had fewer than 2000 patients and only one meta-analysis included more than 10,000 patients. Sensitivity and specificity and AUROC (area under the receiver operating characteristic curve) are the most commonly used measures of test accuracy in DTA IPD meta-analysis. Regression models including fixed- and random-effects, multilevel and GEE (generalized estimating equation) logistic regression are the most commonly used statistical methods in DTA IPD meta-analysis. Additional analyses, which include subgroup analysis, covariate analysis and cut-off value analysis, added value to the DTA IPD meta-analysis.
Conclusions: Although the number of IPD meta-analyses in DTA systematic reviews is increasing, there is much variation in how these IPD MA were performed and what statistical methods were used. In this study, we showed the wide variation in methods used.
Methods: A systematic search was performed to identify published articles containing IPD meta-analyses of DTA studies. Embase and MEDLINE databases were searched from 2000 to 2014. Data was extracted with a data extraction form developed for this study including both multiple choices and open questions.
Results: Twenty-nine DTA IPD meta-analyses articles published between 2006 and 2014 were selected as the subjects for the final analysis. More than 80% (24/29) of these were based on datasets containing individual patient information provided by authors of primary studies; 60% (15/25) included data from fewer than 10 primary studies; 60% (15/25) had fewer than 2000 patients and only one meta-analysis included more than 10,000 patients. Sensitivity and specificity and AUROC (area under the receiver operating characteristic curve) are the most commonly used measures of test accuracy in DTA IPD meta-analysis. Regression models including fixed- and random-effects, multilevel and GEE (generalized estimating equation) logistic regression are the most commonly used statistical methods in DTA IPD meta-analysis. Additional analyses, which include subgroup analysis, covariate analysis and cut-off value analysis, added value to the DTA IPD meta-analysis.
Conclusions: Although the number of IPD meta-analyses in DTA systematic reviews is increasing, there is much variation in how these IPD MA were performed and what statistical methods were used. In this study, we showed the wide variation in methods used.