Assessment of six markers for cervical (pre)cancer detection 49 2 of controls and CIN3, with Spearman coefficients between 0.64 and 0.83 (Supplementary Table 2.1) and variance inflation factors between 3.67 and 6.97. The individual ability of the six markers to distinguish controls from CIN3 was evaluated by logistic regression and resulted in single marker models with AUCs between 0.774 and 0.886 (Table 2.1). Validation by LOOCV produced a similarly good performance of all markers with AUCs greater than 0.800 for ASCL1, LHX8 and ZIC1 (Figure 2.2; Table 2.1). Corresponding sensitivities and specificities at the best threshold can be found in Table 2.1. Sensitivities at a threshold corresponding to a predefined specificity of 70 and 80% varied from 70.3 to 89.1% and 65.7 to 84.6%, respectively (Table 2.2). Adding age of the patient as a covariate to the models did not improve performance of the models (Supplementary Table 2.2). COMPLEMENTARITY BETWEEN METHYLATION MARKERS WITH A MARKER PANEL LASSO logistic regression was used to evaluate whether a marker panel involving two or more markers could further improve CIN3 detection. This resulted in a bi-marker panel consisting of ASCL1 and LHX8 with an AUC of 0.890 (Table 2.1). LOOCV yielded a near identical AUC of 0.882, which corresponded to a CIN3 sensitivity of 83.4% and a specificity of 82.4% at the best threshold (Figure 2.3 & Table 2.1). At a predefined specificity of 70% and 80%, corresponding CIN3 sensitivities were 89.1 and 85.7%, respectively (Table 2.2). Methylation marker non-CV AUC Sensitivity* Specificity* LOOCV AUC Sensitivity* Specificity* ASCL1 0.886 84.0% 81.8% 0.876 80.0% 85.2% LHX8 0.852 72.0% 84.4% 0.845 73.1% 82.7% ST6GALNAC5 0.788 67.4% 81.2% 0.784 66.3% 82.1% GHSR 0.801 65.7% 85.2% 0.795 65.7% 85.2% SST 0.774 65.7% 80.1% 0.768 65.7% 79.8% ZIC1 0.843 70.9% 85.2% 0.836 70.3% 85.5% Bi-marker panel (ASCL1/LHX8) 0.890 84.0% 83.2% 0.882 83.4% 82.4% * Sensitivities and specificities are calculated for CIN3 detection and based on Youden’s J-index. Abbreviations: AUC, area under the curve; CIN, cervical intraepithelial neoplasia; LOOCV, leave-oneout cross-validation; non-CV, non-cross-validated Table 2.1 AUC received from logistic regression and leave-one-out cross-validation analysis for CIN3 detection of six DNA methylation markers (ASCL1, LHX8, ST6GALNAC5, GHSR, SST and ZIC1) and the bi-marker panel (ASCL1/LHX8).
RkJQdWJsaXNoZXIy MjY0ODMw