Statistical models for diagnostic meta-analysis Seminar

Event details

Abstract: The talk first gives an introduction to diagnostic meta-analyses. Then, a linear mixed model is introduced to jointly analyze estimated pairs of sensitivity (Se) and specificity (Sp) from a meta-analysis of diagnostic tests Reitsma et.al. 2005. This bivariate statistical does not transform pairs of sensitivity and specificity of individual studies into a single indicator of diagnostic accuracy, but preserves the two-dimensional nature of the data, taking into account any correlation between the two. This model assumes a bivariate normal distribution for the corresponding random effects. The assumption of a normal distribution for the random effect may be too strong. Thus, a semiparametric mixture model is proposed. For sensitivity and specificity a bivariate normal distribution is assumed on a logit scale. Based on this bivariate normal distribution as mixed kernel a finite mixture model is fit to the data (?). Parameter estimates are obtained using the EM algorithm McLachlan and Krishnan 1997. The method is presented using data from a meta-analysis investigating the diagnostic accuracy of noninvasive coronary angiography using computer tomography. Schutz et.al. 2010; and the potential of Procalcitonin as a diagnostic marker for the sepsis. Walker et. al. 2013.