OC155: Normative data of the transverse diameter of the developing fetal thymus

Abstract

Oral communication abstracts a prevalence of less than 15%, between 15 and 30% and above 30%. To ascertain statistically significant differences in performance between the types of centers, the AUCs were compared using bootstrapping. The optimal cutoff level per center and type was chosen corresponding to a sensitivity level as high as possible (preferable above 90%) while still keeping a good specificity (80%) as this was considered to be very important in correctly identifying malignant cases. Results: Both LR1 and LR2 performed better, although not statistically significant, in centers with a lower prevalence of malignant cases. The AUC of centers with less than 15% of malignancy was 0.956 and 0.941, for LR1 and LR2 respectively; centers with prevalence between 15 and 30% had an AUC of 0.948 and 0.925, respectively and centers with more than 30% malignancies had an AUC of 0.933 and 0.914, respectively. The optimal cutoff per center varied between 0.05 and 0.20, but the performance in the centers with a higher percentage of malignant cases did not improve by choosing a different cutoff level. Conclusions: The performance of the logistic regression models increases with decreasing prevalence of malignancy. Because new cutoff levels per center would be based on 8 to 253 patients and the cutoff of 0.10 is optimal for all three types of center, it seems reasonable to use this cutoff in all centres.