TY - JOUR
T1 - Multivariate random-effects approach: for meta-analysis of cancer staging studies
AU - Bipat, Shandra
AU - Zwinderman, Aeilko H.
AU - Bossuyt, Patrick M. M.
AU - Stoker, Jaap
PY - 2007
Y1 - 2007
N2 - RATIONALE AND OBJECTIVES: Meta-analyses of diagnostic accuracy studies produce summary estimates of sensitivity and specificity. Cancer staging relies on staging systems and meta-analysis is often performed after dichotomization of the staging results. For each dichotomization, summary estimates of sensitivity and specificity can be calculated by repeated bivariate random-effects analyses. In this process, staging information is lost and under- and overstaging can not be adequately expressed. MATERIALS AND METHODS: We propose a new multivariate random-effects approach, which is an extension of the bivariate random-effects approach. To illustrate the principles and outcomes of both approaches, we used data from a meta-analysisevaluating endoluminal ultrasonography in staging of rectal cancer. In the multivariate approach, results on correct staging and under- and overstaging were calculated. In addition, the results from this analysis were used to calculate sensitivity and specificity estimates for each dichotomization and these estimates were compared with the results of the repeated bivariate analyses. RESULTS: By the multivariate analysis, results on correct staging and under- and overstaging were obtained. The sensitivity and specificity estimates for the dichotomizations, calculated from the results of this multivariate approach, were also comparable with the sensitivity and specificity estimates obtained by the repeated bivariate analyses. CONCLUSIONS: The multivariate random-effects approach can be a useful meta-analytic method for summarizing cancer staging data presented in diagnostic accuracy studies
AB - RATIONALE AND OBJECTIVES: Meta-analyses of diagnostic accuracy studies produce summary estimates of sensitivity and specificity. Cancer staging relies on staging systems and meta-analysis is often performed after dichotomization of the staging results. For each dichotomization, summary estimates of sensitivity and specificity can be calculated by repeated bivariate random-effects analyses. In this process, staging information is lost and under- and overstaging can not be adequately expressed. MATERIALS AND METHODS: We propose a new multivariate random-effects approach, which is an extension of the bivariate random-effects approach. To illustrate the principles and outcomes of both approaches, we used data from a meta-analysisevaluating endoluminal ultrasonography in staging of rectal cancer. In the multivariate approach, results on correct staging and under- and overstaging were calculated. In addition, the results from this analysis were used to calculate sensitivity and specificity estimates for each dichotomization and these estimates were compared with the results of the repeated bivariate analyses. RESULTS: By the multivariate analysis, results on correct staging and under- and overstaging were obtained. The sensitivity and specificity estimates for the dichotomizations, calculated from the results of this multivariate approach, were also comparable with the sensitivity and specificity estimates obtained by the repeated bivariate analyses. CONCLUSIONS: The multivariate random-effects approach can be a useful meta-analytic method for summarizing cancer staging data presented in diagnostic accuracy studies
U2 - https://doi.org/10.1016/j.acra.2007.05.007
DO - https://doi.org/10.1016/j.acra.2007.05.007
M3 - Article
C2 - 17659244
SN - 1076-6332
VL - 14
SP - 974
EP - 984
JO - Academic Radiology
JF - Academic Radiology
IS - 8
ER -