Article type
Year
Abstract
Background: The majority of diagnostic accuracy studies report pairs of sensitivity and specificity. The summary Receiver Operating Characteristic (sROC) approach is the method of choice for meta-analysing these studies. This method uses the diagnostic odds ratio as the main outcome measure. Recently, we described the bivariate approach to analyse such data, which preserves the two-dimensional nature (e.g. sensitivity and specificity) of the original data (J Clin Epidemiol accepted).
Objectives:
1. To discuss differences and similarities in the statistical assumptions and properties of the sROC and the bivariate model;
2. To compare directly the results from the sROC and bivariate model by (re)-analysing data from published reviews and simulated datasets;
3. To demonstrate the differences in outcome measures and in the way covariates are investigated between the two approaches.
Methods: We will re-analyse the data from a set recently published diagnostic reviews (n=31 reviews) by both methods, which involve a wide range of target conditions, index tests and design features. In addition, we will simulate diagnostic data, incorporating differences in the number of primary studies, differences in the number of diseased and non-diseased patients, differences in threshold, and effect of covariates.
Results: The sROC approach uses the diagnostic odds ratio as the main outcome measure, which removes the effect of a possible threshold but at the same time loses relevant clinical information about test performance. The bivariate approach preserves the two-dimensional nature of the original data, acknowledging any possible (negative) correlation between sensitivity and specificity as a result from difference in threshold or other causes. It also takes into account the differences in precision by which sensitivity and specificity have been measured within and across studies, and it incorporates and estimates the amount of between-study variability in both sensitivity and specificity (random effects model). The results of re-analysing data from published diagnostic reviews and simulated datasets will be available before the Cochrane Colloquium.
Conclusions: The bivariate model is a flexible approach for meta-analysing diagnostic accuracy studies. The bivariate approach can be seen as an extension and an improvement to the sROC approach.
Objectives:
1. To discuss differences and similarities in the statistical assumptions and properties of the sROC and the bivariate model;
2. To compare directly the results from the sROC and bivariate model by (re)-analysing data from published reviews and simulated datasets;
3. To demonstrate the differences in outcome measures and in the way covariates are investigated between the two approaches.
Methods: We will re-analyse the data from a set recently published diagnostic reviews (n=31 reviews) by both methods, which involve a wide range of target conditions, index tests and design features. In addition, we will simulate diagnostic data, incorporating differences in the number of primary studies, differences in the number of diseased and non-diseased patients, differences in threshold, and effect of covariates.
Results: The sROC approach uses the diagnostic odds ratio as the main outcome measure, which removes the effect of a possible threshold but at the same time loses relevant clinical information about test performance. The bivariate approach preserves the two-dimensional nature of the original data, acknowledging any possible (negative) correlation between sensitivity and specificity as a result from difference in threshold or other causes. It also takes into account the differences in precision by which sensitivity and specificity have been measured within and across studies, and it incorporates and estimates the amount of between-study variability in both sensitivity and specificity (random effects model). The results of re-analysing data from published diagnostic reviews and simulated datasets will be available before the Cochrane Colloquium.
Conclusions: The bivariate model is a flexible approach for meta-analysing diagnostic accuracy studies. The bivariate approach can be seen as an extension and an improvement to the sROC approach.