Performance of the S5 classifier to detect ≥HSIL+ cases
ROC analysis of the S5 methylation for detecting ≥HSIL+ cases is shown in Figure 5 . The cutoff value of 2.85 was selected for discrimination of “≥HSIL+” lesions from “≤LSIL” diagnoses in this present Chinese screening population, with the best AUC value as 0.80. Also, we validated the performance of S5 methylation with different cutoff value, in the present population (cutoff = 2.85) and in other specific populations (S5 cutoff 0.8 in a UK colposcopy referral population as previous reported31). The receiver operation characteristic curve (ROC) analysis demonstrated that the S5 methylation had an AUC (area under the curve) of 0.80 (95% CI 0.74–0.85) for detecting HSIL+ among abnormal results between HPV and cytology (Figure 5 ), which was similar to the performance in a China colposcopy referral population13 (ACU = 0.86) as previous reported. Also, we calculated the AUC of EPB41L3 (0.65, 95% CI 0.74–0.85), which was much lower than that of S5 methylation (0.80, 95% CI 0.74–0.85) (Figure 6 ). According to the area under the curve (AUC), when combining sensitivity and specificity, the accuracy of S5 at a cutoff 2.85 for detection of HSIL+ was significantly higher than the accuracy of the HPV test or combination HPV with cytology (Figure 5 ).
Cross tabulation of the number of cases or controls with negative or positive test results and the comparison of sensitivity and specificities of the tests to detect ≥HSIL+ are shown in Table 3 and 4, respectively. For detecting ≥HSIL+ cases, the sensitivity of HPV16/18 was 64.9% (95% CI 58.3-71.7) and of cytology was 48.9% (95% CI 42.8-53.2) (as most cases with LSIL or HSIL were HPV infected, here we analyzed the performance of HPV16/18 genotyping). S5 at the cut-point of 0.8 (predefined in a UK population) had a higher sensitivity (95.5%, 95% CI 92.6-97.1) but significantly lower specificity (19.4%, 95% CI 17.8-21.7) than HPV 16/18 (44.8%, 95% CI 40.1-49.4, P< 0.001) or cytology testing (54.6%, 95% CI 50.7-57.9,P < 0.001). At the cut-point of 2.85 (corresponds to setting the specificity of S5 for ≤LSIL at which is higher than cytology specificity), the S5 specificity was significantly higher than HPV16/18 genotyping (P < 0.001), while the sensitivity of S5 was 76.1% (95% CI 71.7-79.2), which remained significantly higher than sensitivity of HPV16/18 genotyping, cytology. When compared the performance of S5 methylation with the scenario “cytology plus HPV16/18” (scenario positive means that positive for any of the two tests), the sensitivity and specificity of S5 were also better than the scenario “cytology plus HPV16/18” (Table 4).
S5 classifier in predicting progression to(His)HSIL+
Figure 7 shows a significant difference (Likehood-ratio test,P = 0.02) between cumulative proportions of progression to ≥(His)HSIL+ distributed by time in women population: (“HPV positive except HPV16/18” and “abnormal cytology results except ≥ (TCT)HSIL”). In this specific population, comparison between women positive for the S5 methylation classifier (S5 methylation classifier ≥ 2.85) vs negative for the S5 methylation classifier (S5 methylation classifier < 2.85) were conducted. As depicted in Figure 7 , most of the cases (“HPV positive except HPV16/18” and “abnormal cytology results except ≥ (TCT)HSIL”) who progressed to ≥HSIL+ within 2 years, were calculated S5 methylation scores equal or greater than 2.85 (≥ 2.85) at the first visiting.