Cervical cancer screening using DNA methylation triage in a real-world population

The study addresses the need for improved triage methods for HPV-positive women in cervical cancer screening. While HPV-based primary screening is more sensitive than cytology and allows longer screening intervals, triaging HPV-positive results typically relies on cytology, which suffers from low reproducibility, variable sensitivity, and requires infrastructure not compatible with self-sampling. Given the higher risk associated with HPV16/18 and increasing emphasis on genotype information, the authors investigate whether a DNA methylation (DNAme)–based triage test (WID-qCIN, targeting methylation at DPP6, RALYL, and GSXI) alone or combined with HPV16/18 genotyping can better detect prevalent disease and predict incident CIN2+ and invasive cancers compared with cytology in a real-world, population-based cohort.

DNA methylation testing for cervical (pre)cancer detection was demonstrated two decades ago, and multiple methylation marker panels have been evaluated, often in case–control or small cohorts (<1,000 participants). Recent studies assessed markers such as FAM19A4/miR124-2 and S5 panels, with varying performance for triage of hrHPV-positive women. The WID-qCIN test was developed from an epigenome-wide screen (850,000 CpGs) in 170 CIN3+ cases and 202 controls and preliminarily validated in diagnostic and predictive settings (total n=761). Guidelines now recommend primary HPV screening (≥30 years) with triage, and some programs refer HPV16/18-positive women directly to colposcopy. Prior work indicates cytology’s variable sensitivity and decreasing performance over time, underscoring interest in molecular, automatable triage compatible with self-sampling.

Design and setting: Population-based cohort study including all women aged ≥30 years who attended the cervical cancer screening program in Greater Stockholm between 1 January and 31 March 2017 (KI-q1-2017 cohort). Primary screening used HPV testing (Cobas 4800) with genotyping information on HPV16/18 and other high-risk types. Cytology was assessed for HPV-positive women. Ethical approvals were obtained from the Swedish Ethical Review Authority and Medical University Innsbruck. Participants and data sources: Of 28,017 screened women, 2,377 were HPV-positive and included for triage analyses. Cytology-positive (ASC-US+) women were referred for colposcopy and biopsy; cytology-negative HPV-positive women were invited for follow-up at ~36 months. Histopathology outcomes (CIN2, CIN3, HSIL, AIS, CC) within 0–12 months were defined as prevalent; diagnoses after 12 months (13–72 months) were incident. Registry data came from NKCx (complete through 31 Dec 2022) and invasive CC verification from the Swedish National Quality Register for Gynecological Cancers (GCR). Discrepant invasive cancer diagnoses were adjudicated by blinded pathology review where possible. Intervention/Index tests: WID-qCIN is a quantitative real-time PCR DNA methylation assay on bisulfite-modified DNA targeting three human gene regions (DPP6, RALYL, GSXI). The assay was optimized and calibrated; percentage of fully methylated reference (PMR) values were computed. Samples with SUM-PMR > 0 were WID-qCIN positive; SUM-PMR = 0 were negative. HPV16/18 genotyping results were used alone and in combination with WID-qCIN (WID-qCIN/HPV16/18). Cytology was categorized as negative (NILM), positive (ASC-US+) or inadequate. Outcomes: Primary outcomes included sensitivity/specificity for detecting prevalent CIN2+ (and subcategories: CIN2, CIN3, AIS, CC); predictive performance for incident CIN2+ and CC over 13–72 months (hazard ratios, cumulative incidence); and efficiency estimated by number of colposcopy referrals required per CIN2+ detection over 72 months. Statistical analysis: Wilson method provided 95% CIs for proportions; comparisons via two-sided chi-squared tests without Yates’ correction. Kaplan–Meier estimators, Cox proportional hazards models (hazard ratios, 95% CIs, log-rank tests) were used for incident outcomes, with censoring defined by most recent negative test. A logistic Weibull mixture model (PIMixture R package) accounted for undiagnosed prevalent disease and interval-censored incident disease, yielding odds ratios, hazard ratios, and 6-year cumulative incidence estimates. Analyses performed in R 4.3.1.

Population: 28,017 screened; 2,377 HPV-positive. Mean age of HPV-positive women: 40.8 years (range 30–64). At baseline, 27.5% were HPV16/18-positive and 28.9% WID-qCIN-positive. Prevalent disease (0–12 months): - Sensitivity for CIN2+: Cytology 98.4% (300/305; biased by referral protocol), HPV16/18 53.3% (163/306), WID-qCIN 77.0% (224/291), WID-qCIN/HPV16/18 85.9% (250/291). - Sensitivity for CIN3: HPV16/18 58.9% (56/95), WID-qCIN 85.7% (78/91), WID-qCIN/HPV16/18 93.4% (85/91). All CCs (11/11) detected by HPV16/18 and by WID-qCIN/HPV16/18; WID-qCIN alone detected 90.9% (10/11). - Specificity (≤CIN1): Cytology 80.1% (1,656/2,067), HPV16/18 76.3% (1,577/2,068), WID-qCIN 76.9% (1,534/1,996), WID-qCIN/HPV16/18 60.7% (1,210/1,993). - Predictive values for CIN2+: PPV—Cytology 42.2% (300/711), HPV16/18 24.9% (163/654), WID-qCIN 32.7% (224/686), WID-qCIN/HPV16/18 24.2% (250/1,033). NPV—Cytology 99.7% (1,656/1,661), HPV16/18 91.7% (1,577/1,720), WID-qCIN 95.8% (1,534/1,601), WID-qCIN/HPV16/18 96.7% (1,210/1,251). Incident prediction (13–72 months): - CIN2+: Cytology-positive predicted 18.2% (49/269) of incident cases; HPV16/18-positive 45.6% (123/270), HR 2.72 (95% CI 2.14–3.45); WID-qCIN-positive 46.3% (120/259), HR 3.01 (2.36–3.85); WID-qCIN/HPV16/18-positive 69.4% (179/258), HR 3.55 (2.73–4.63). - Incident CC: Cytology-positive 20.0% (2/10), HR 0.93 (0.20–4.40); HPV16/18-positive 54.5% (6/11), HR 3.87 (1.18–12.68); WID-qCIN-positive 40.0% (4/10), HR 2.46 (0.69–8.72); WID-qCIN/HPV16/18-positive 80.0% (8/10), HR 6.44 (1.37–30.35). - Mixture model HRs for incident CIN2+: WID-qCIN 2.31 (1.31–4.08), HPV16/18 2.47 (1.40–4.37), WID-qCIN/HPV16/18 2.83 (1.55–5.16). Program-level efficiency over 72 months (HPV-positive cohort n=2,377): - Cytology triage detected 60.8% of CIN2+ (349/574) and 63.6% of CCs (14/22) with 1,432 colposcopy referrals; 4.10 referrals per CIN2+ detected. - HPV16/18 triage detected 49.7% CIN2+ (286/576) and 75.0% CCs (18/24) with 654 referrals; 2.29 referrals per CIN2+. - WID-qCIN triage detected 62.5% CIN2+ (344/550) and 69.6% CCs (16/23) with 686 referrals; 1.99 referrals per CIN2+. - WID-qCIN/HPV16/18 triage detected 78.1% CIN2+ (429/549) and 91.3% CCs (21/23) with 1,033 referrals; 2.41 referrals per CIN2+. Notably, 7/8 incident CCs in cytology-negative women were identified by the combined triage.

The findings demonstrate that combining a DNA methylation-based triage (WID-qCIN) with HPV16/18 genotyping substantially improves detection of prevalent CIN3 and prediction of incident CIN2+ and invasive CC compared with cytology-based triage. This addresses limitations of cytology, including variability, dependence on morphology, and incompatibility with self-sampling. The combined WID-qCIN/HPV16/18 approach identified nearly all invasive cancers developing up to 72 months, including most cancers arising in cytology-negative women, highlighting its clinical relevance for preventing progression by enabling earlier treatment. While specificity of the combined approach is lower than single tests, overall program efficiency is improved, requiring fewer colposcopy referrals per CIN2+ detection than cytology. The molecular, automatable nature of WID-qCIN suggests feasibility for integration into screening pathways, particularly those incorporating self-sampling, potentially increasing uptake and reducing logistical barriers.

In a large, real-world screening cohort, WID-qCIN combined with HPV16/18 genotyping outperformed cytology for triaging HPV-positive women, detecting a higher proportion of prevalent CIN3 and predicting a larger share of incident CIN2+ and CC cases, while reducing colposcopy referrals needed per CIN2+ detected compared with cytology. The results support implementing WID-qCIN/HPV16/18 as an improved triage strategy, especially compatible with self-sampling workflows. Future research should include prospective randomized clinical trials to quantify the impact on invasive cancer prevention and evaluate optimal screening intervals; data suggest intervals up to 5 years may be safe for HPV-positive women testing negative by WID-qCIN/HPV16/18.

Key limitations include potential ascertainment bias for prevalent disease, as only cytology-positive women were referred for colposcopy and biopsy, inflating cytology’s apparent sensitivity in the prevalent setting. Swedish pathology reporting transitioned to HSIL (combining CIN2 and CIN3), limiting separate incident CIN3 analyses. Nonproportional hazards were observed when stratifying incident CIN2+ by cytology status, complicating model fitting; a mixture model was used to address undiagnosed prevalent and interval-censored incident disease. Definitive assessment of invasive cancer prevention impact requires a prospective randomized clinical trial.