To assess serum VEGF-D levels as a biomarker both for LAM and for lymphatic involvement, we derived a model of prediction by plotting the Se (the proportion of subjects who have the predicted condition, LAM or lymphatic involvement) vs Sp (the proportion of subjects not having the predicted condition), with the curve of the plot (ROC) defining the optimal cutoff point (ie, serum VEGF-D levels above which the diagnosis of LAM disease or lymphatic involvement may be accepted). The area under the curve (AUC) was used as a measure of accuracy for the ROC. The closer the AUC is to 1, the better the diagnostic test result, which, for this study, was the ability to use serum levels of VEGF-D to classify patients with or without LAM or with or without lymphatic involvement. We based the model on Se and Sp, rather than on positive predictive value, as the positive predictive value depends on the prevalence of disease as well as the Sp of the test, and LAM is a rare disease. In evaluating lymphatic involvement among LAM patients, serum VEGF-D levels have a good discriminating ability (AUC, 0.813 [p < 0.0001]; Se, 0.792; Sp, 0.824; cutoff point, 949 pg/mL) [Fig 5, top left, A]. When the analysis group included healthy volunteers, the predictability increased (AUC, 0.845 [p < 0.0001]; Se, 0.753; Sp, 0.919; cutoff point, 1,317 pg/mL) [Fig 5, top right, B] The discriminating ability of serum VEGF-D level to predict LAM disease is not as robust (AUC, 0.751 [p < 0.0001]; Se, 0.577; Sp, 0.975; cutoff, 1,239 pg/mL) [Fig 5, bottom left, C] and is poor for predicting LAM disease for patients with only cystic lung disease (AUC, 0.66 [p = 0.046]; Se, 0.867; Sp, 0.529; cutoff, 1,202 pg/mL) [Fig 5, bottom right, D].