Meta-Analysis in DTA with Hierarchical Models Bivariate and HSROC: Simulation Study
In meta-analysis for diagnostic test accuracy (DTA), summary measures such as sensitivity, specificity, and odds ratio are used. However, these measures may not be adequate to integrate studies with low prevalence, which is why statistical modeling based on true positives and false positives is necessary. In this context, there are several statistical methods, the first of which is a bivariate random effects model, part of the assumption that the logit of sensitivity and specificity follow a bivariate normal distribution, the second, refers to the HSROC or hierarchical model, is similar to bivariate, with the particularity that it directly models the sensitivity and specificity relationship through cut points. Using simulations, we investigate the performance of hierarchical models, varying their parameters and hyperparameters and proposing a better management of variability within and between studies. The results of the simulated data are analyzed according to the criterion of adjustment of the models and estimates of their parameters.
KeywordsDiagnostic precision Meta-analysis HSROC model Bivariate model Low prevalence
- 7.Kotz, S., Balakrishnan, N., Johnson, N.L.: Bivariate and trivariate normal distributions. Contin. Multivar. Distrib. 1, 251–348 (2000)Google Scholar
- 14.Doebler, P., Holling, H.: Meta-analysis of diagnostic accuracy with mada. Reterieved at: https://cran.rproject.org/web/packages/mada/vignettes/mada.pdf (2015)