Roxanne Bouchard-Bellavance1, Florence Perrault1, Gilles Soulez1,2, Miguel Chagnon3, Gregory A Kline4, Isabelle Bourdeau2,5, André Lacroix2,5, Benny So6, Eric Therasse1,2. 1. Department of Radiology, Centre Hospitalier de l'Université de Montréal (CHUM), Montreal, QC, Canada. 2. Centre de Recherche du CHUM (CRCHUM), Montreal, QC, Canada. 3. Department of Mathematics and Statistics, Pavillon André-Aisenstadt, University of Montreal, Montreal, QC, Canada. 4. Division of Endocrinology, Department of Medicine, Foothills Medical Center (FMC), Calgary, AB, Canada. 5. Division of Endocrinology, Department of Medicine, Centre Hospitalier de l'Université de Montréal (CHUM), Montreal, QC, Canada. 6. Department of Radiology, Foothills Medical Center (FMC), Calgary, AB, Canada.
Abstract
BACKGROUND: Adrenal vein sampling (AVS) failure is mainly due to right adrenal vein unavailability. Multinomial regression modelling (MRM) and left adrenal vein-to-peripheral vein ratio (LAV/PV) were proposed to predict the lateralization index without the right AVS. OBJECTIVE: To assess external validity of MRM and LAV/PV to predict lateralization index when right adrenal vein sampling is missing. DESIGN: Diagnostic retrospective study. PATIENTS: Development and validation cohorts included AVS of 174 and 122 patients, respectively, from 2 different centres. MEASUREMENTS: Development and validation cohort data were used, respectively, for calibration and for validation of MRM and LAV/PV to predict the lateralization index without the right adrenal vein sampling. Sensitivity and specificity of MRM and LAV/PV were compared between both centres at different pre-established specificity thresholds based on receiver operating characteristic curves generated from the development cohort data. RESULTS: At a specificity threshold of 95% set in the development cohort, specificity values exceeded 90% (range, 90.6%-98.8%) for all verified MRM and LAV/PV models in the validation cohort. Corresponding sensitivities for MRM and LAV/PV, respectively, range from 54.1% to 83.7% and 32.8% to 88.4% for the development cohort compared to 33.3%-87.5% and 2.8%-79.2% for the validation cohort. Overall, diagnostic accuracy of both methods was higher to detect right (82.8%-93.5%) than left (70.2%-80.6%) lateralization index status in both centres. CONCLUSIONS: Minimal changes in specificity from development to validation cohorts validate the use of MRM and LAV/PV to predict the lateralization index when the right AVS is missing. Both methods had better accuracy for right than left lateralization detection.
BACKGROUND: Adrenal vein sampling (AVS) failure is mainly due to right adrenal vein unavailability. Multinomial regression modelling (MRM) and left adrenal vein-to-peripheral vein ratio (LAV/PV) were proposed to predict the lateralization index without the right AVS. OBJECTIVE: To assess external validity of MRM and LAV/PV to predict lateralization index when right adrenal vein sampling is missing. DESIGN: Diagnostic retrospective study. PATIENTS: Development and validation cohorts included AVS of 174 and 122 patients, respectively, from 2 different centres. MEASUREMENTS: Development and validation cohort data were used, respectively, for calibration and for validation of MRM and LAV/PV to predict the lateralization index without the right adrenal vein sampling. Sensitivity and specificity of MRM and LAV/PV were compared between both centres at different pre-established specificity thresholds based on receiver operating characteristic curves generated from the development cohort data. RESULTS: At a specificity threshold of 95% set in the development cohort, specificity values exceeded 90% (range, 90.6%-98.8%) for all verified MRM and LAV/PV models in the validation cohort. Corresponding sensitivities for MRM and LAV/PV, respectively, range from 54.1% to 83.7% and 32.8% to 88.4% for the development cohort compared to 33.3%-87.5% and 2.8%-79.2% for the validation cohort. Overall, diagnostic accuracy of both methods was higher to detect right (82.8%-93.5%) than left (70.2%-80.6%) lateralization index status in both centres. CONCLUSIONS: Minimal changes in specificity from development to validation cohorts validate the use of MRM and LAV/PV to predict the lateralization index when the right AVS is missing. Both methods had better accuracy for right than left lateralization detection.