70 Chapter 3 pathologist, blinded to the original diagnosis. For this, the original haematoxylin-eosin (H&E)-stained slides and formalin-fixed paraffin-embedded tissue blocks were retrieved from the local pathology laboratories. Ki-67 and p16INK4A immunohistochemistry was performed as described before 15. In case no tissue was left in the blocks or lesions had an uninterpretable immunohistochemical staining, the revision diagnosis was based solely on expert review of the original H&E-stained slide. HOST-CELL DNA METHYLATION ANALYSIS DNA from cervical screening samples was isolated using the NucleoMag® 96 (MachereyNagel GmbH&Co. KG, Düren, Germany) and a Microlab Star robotic system (Hamilton, Gräfelfing, Germany) according to the recommendations of the manufacturer. DNA concentrations were quantified using a Qubit fluorometer (ThermoFisher Scientific, Waltham, MA, USA). Sodium bisulphite treatment of DNA and multiplex qMSPs were performed as described previously for markers GHSR/SST/ZIC1 8 and ASCL1/LHX8/ ST6GALNAC5 9. Samples with a quantification cycle (Cq) value for ACTB above 30 in one or both qMSP assays were considered to have an inadequate sample quality and excluded from further analyses. Methylation levels were normalised to the reference gene ACTB using the Cq values (2-ΔCq ×100) to obtain ΔCq ratios 16. All methylation testing was performed blinded for cyto- and histopathology outcomes. DATA AND STATISTICAL ANALYSIS All histology until two years after baseline was included. To calculate the agreement between original and revision diagnosis, the intraclass correlation coefficient (ICC) with 95% confidence intervals (95% CI) was used. To assess differences in DNA methylation levels across disease categories, the KruskalWallis omnibus test was performed on square-root transformed ΔCq ratios. Following a significant result, post hoc testing was performed using the Wilcoxon rank-sum test. Square-root transformed ΔCq ratios were visualised in boxplots. In an earlier study, single-marker classifiers using univariable logistic regression and a multi-marker classifier using a LASSO regression model, which yielded a bi-marker panel consisting of ASCL1 and LHX8, were developed for triage of hrHPV-positive women on clinician-collected cervical samples 10. All classifiers were trained to discriminate between CIN3 and controls. Predicted probabilities (value range 0 to 1), representing the risk of underlying CIN3, were calculated and thresholds calibrated at 70% clinical specificity among hrHPV-positive women, were defined for all classifiers 10. In the current study,
RkJQdWJsaXNoZXIy MjY0ODMw