Abstract
Objective
To evaluate the clinical performance of triage strategies for HPV-positive women in rural Yunnan, China, with the aim of optimizing HPV test-based cervical cancer screening.
Methods
We recruited 417 HPV-positive women, who additionally underwent mRNA testing, AI-assisted cytology, PAX1 and SOX1 DNA methylation testing. We compared the sensitivity, specificity, area under the curve (AUC), and number needed to colposcopy for one cervical lesion detection of several triage approaches, for detecting cervical intraepithelial neoplasia grade 2 or worse (CIN2+).
Results
Among single triage strategies, cytology demonstrated a sensitivity of 54.2% (95% CI: 33.2%–73.8%), a specificity of 75.1% (95% CI: 70.4%–79.2%), an AUC of 0.65 (95% CI: 0.53–0.77), and an number needed to colposcopy for one lesion detection of 8.5. Compared with cytology, PAX1 methylation exhibited superior sensitivity (87.5%, 95% CI: 66.5%-96.7%) and specificity (90.6%, 95% CI: 87.1%-93.2%), with an improved AUC of 0.89, and a reduced colposcopy referral of 2.8. Adding SOX1 to PAX1 maintained sensitivity (87.5%) but slightly reduced specificity (88.3%), resulting in an AUC of 0.88. mRNA testing showed higher sensitivity (75.0%) but lower specificity (44.5%) and a declined AUC (0.60). Among combined triage strategies, HPV-16/18 genotyping with reflex cytology for non-16/18 types yielded a sensitivity of 62.5%, a specificity of 54.5%, and an AUC of 0.59. HPV-16/18 genotyping with reflex PAX1/SOX1 methylation improved sensitivity to 91.7% and specificity to 64.1%, with an AUC of 0.78.
Conclusion
PAX1 methylation triage demonstrated better clinical performance and required fewer colposcopy referrals than cytology triage. It showed good potential as an triage strategy for HPV-positive women in rural Yunnan. The addition of SOX1 methylation or integration with HPV genotyping did not significantly enhance its clinical performance.
Keywords: HPV-positive, Cytology, PAX1/SOX1 DNA methylation, Triage, Clinical performance
Introduction
Cervical cancer is the fourth most common cancer among females globally, especially in low- and middle-income countries/regions [1]. In China, cervical cancer remains a significant public health concern, with 150,000 cases and 50,000 deaths been reported in 2022 [2], and an increasing trend in cervical cancer incidence has been observed over the past decade [2]. The World Health Organization (WHO) has launched a global call to eliminate cervical cancer through a comprehensive strategy that integrates human papillomavirus (HPV) vaccination, screening, and treatment [3]. China has actively embraced this global initiative, committing to the WHO’s strategy and releasing a national action plan to accelerate cervical cancer elimination [4]. This plan includes a target of achieving 70% screening coverage for women aged 35–64 by 2030.
The WHO and Chinese cervical cancer screening guidelines prioritize HPV testing as the primary screening strategy due to its high sensitivity, good reproducibility, and the long-term reassurance it provides after a negative test result [5]. Consequently, an increasing number of countries are transitioning from cytology-based to HPV test-based screening programs. However, compared to cytology, HPV testing yields more false positives, leading to a relatively lower specificity, which may prompt unnecessary colposcopy referrals and increased anxiety among screened women [6]. Therefore, effective triage strategies are essential for identifying individuals with high-risk cervical lesions among those who test positive for HPV.
Cytology is currently the most widely used triage method and is recommended by the WHO guideline [5]. However, its clinical accuracy is highly dependent on the experience of cytologists, leading to substantial variability in sensitivity and specificity across different settings. HPV-16/18 genotyping is another widely accepted molecular triage test [5], given its strong association with high-grade cervical lesions. Emerging evidence suggests that novel triage methods, such as DNA methylation, may achieve satisfactory clinical performance [7]. Promising candidates for DNA methylation include miR-124, FAM19A4, PAX1, and SOX1 [8, 9]. Specific gene methylations alone (PAX1, JAM3, EPB41L3, ZNF582, SOX1, FAM19A4, miR124), or a combined panel (FAM19A4/miR124, PAX1/JAM3, PAX1/EPB41L3, PAX1/ZNF582) have been evaluated and shown to be an effective triage method for HPV-positive individuals [8, 10–12]. Nevertheless, these methods are not yet widely implemented in clinical practice due to limited evidence, and the optimal methylation marker panel and their clinical efficacy requires further validation. Dual staining is another innovative approach [13], but it shares similar limitations with cytology, as it also requires high-quality cytologists.
China is currently in the process of transitioning from cytology to HPV testing in population-based cervical cancer screening. As a result, a significant number of HPV-positive individuals will require effective management. The Chinese guideline has recommended exploring novel triage methods, such as DNA methylation, which show potential for cervical cancer screening but require further population-based evidence. In light of this, we conducted this study to evaluate the clinical performance of traditional and novel triage strategies for HPV-positive individuals in rural Yunnan, China, to identify the most appropriate triage strategy for low-resource settings and provide evidence to tailor HPV test-based cervical cancer screening programs at the population level.
Materials and methods
Study population
The detailed methods used in this study have been described previously [14]. In brief, a population-based, cross-sectional study consecutively recruited 3,000 women from rural communities in Shuangjiang County, Yunnan province, China, from March 2022 to December 2022. Eligibility criteria were: Having initiated sexual activity, no cervical cancer screening history in the past three years, and not been pregnant at recruitment. Women diagnosed with severe vaginal inflammation at the time of screening were excluded.
Local healthcare personnel obtained informed consent from participants after explaining the study procedure. Then, every participant completed a confidential questionnaire covering demographic, gynecologic, obstetric, and sexual history.
Specimen collection and laboratory testing
A trained gynecologist obtained two exfoliated cervical cell samples from each participant after a pelvic examination. Sample 1 was transferred to Yunnan Cancer Hospital laboratory for “out-of-lab” HPV DNA testing using PCR-based assays. The residual from HPV-positive cases were subsequently tested for PAX1 and SOX1 DNA methylation. Sample 2 was shipped to the local hospital laboratory for cytology slides processing, and interpreted by cytologists assisted with the artificial intelligence system. The residuals were sent to a third-party laboratory for E6/E7 mRNA testing. Laboratory tests were performed blinded to other screening results.
HPV DNA test: The Sansure HPV test targets 15 high-risk HPV types (HPV-16, 18, 31, 33, 35, 39, 45, 51, 52, 53, 56, 58, 59, 66, 68), provided specific genotyping for HPV-16 and HPV-18. The GenPlex HPV test detected 15 high-risk HPV types and 9 low-risk HPV types (HPV-6, 11, 42, 43, 44, 73, 81, 82, 83), and provided specific genotyping for the 24 types of HPV.
E6/E7 mRNA test: The mRNA test detect 14 high-risk HPV types (HPV-16, 18, 31, 33, 35, 39, 45, 51, 52, 53, 56, 58, 59, 68).
The PAX1 and SOX1 DNA methylation: The methylation was analyzed using the methylation-specific quantitative PCR for the SOX1 and PAX1 genes that conducted using the SOX1 and PAX1 gene methylation detection kit (HybriBio Pharmaceutical Technology Co. Ltd., Guangzhou, China). The β-actin (ACTB) gene served as an internal reference control for DNA quality assessment and normalization. The crossing point values of SOX1, PAX1, and ACTB were obtained for each sample, with the crossing point value of ACTB falling between 15 and 35 considered acceptable. Specimens with ACTB values outside this range were excluded from analysis and retested. The cut-off value was 38.0 for PAX1, and 38.6 for SOX1, a sample positive for either gene was considered PAX1/SOX1 methylation positive.
The artificial intelligence (AI)-assisted liquid-based cytology: Cytology results were interpreted with the assistance of the “Artificial Intelligence Cloud Diagnosis System”, and reported following the Bethesda 2014 classification system [15].
Women positive for any HPV DNA or mRNA, or with cytological ASC-US or worse (ASC-US+), or with unsatisfactory results were recalled for colposcopy examination, and biopsied if any suspicious lesion was identified. Biopsy tissues were transported to a third-party laboratory for processing and diagnosis by experienced pathologists who were blinded to other screening results. Pathology results were reported as normal, cervical intraepithelial neoplasia grade 1 (CIN1), grade 2 (CIN2), grade 3 (CIN3), microinvasive carcinoma (MIC), squamous cell carcinoma (SCC), adenocarcinoma in situ (AIS), and adenocarcinoma (ADC). Women with CIN2 or worse (CIN2+) were recommended for treatment. Women who were negative for all tests were considered to be negative for the outcome of CIN2+.
Statistical analysis
Clinical accuracy of triage strategies (Table 1) were evaluated with sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), area under the curve (AUC), with the 95% confidence interval (95% CI). The cytology at ASC-US+, HPV-16/18 genotyping with reflex cytology were defined as the reference triage strategy. The relative sensitivity (RSe), relative specificity (RSp), the number needed to colposcopy for one lesion detection of the triage strategies were compared with the reference methods, in detecting CIN2+ and CIN3+. Significant difference was considered if the relative sensitivity and relative specificity were entirely above or below one.
Table 1.
Evaluated triage strategies
| Index | Description of evaluated triage strategies |
|---|---|
| 1 | Cytology (ASC-US+) triage of HPV + women |
| 2 | Cytology (LSIL+) triage of HPV + women |
| 3 | PAX1 or SOX1 methylation triage of HPV + women |
| 4 | PAX1 methylation triage of HPV + women |
| 5 | SOX1 methylation triage of HPV + women |
| 6 | PAX1 and SOX1 methylation triage of HPV + women |
| 7 | E6/E7 mRNA triage of HPV + women |
| 8 | HPV-16/18 triage of HPV + women |
| 9 | HPV-16/18/52/58/33/68 triage of HPV + women |
| 10 | HPV-16/18 with reflex cytology (ASC-US+) for other 13 HPV + women |
| 11 | HPV-16/18 with reflex PAX1 or SOX1 methylation for other 13 HPV + women |
| 12 | HPV-16/18 with reflex E6/E7mRNA for other 13 HPV + women |
| 13 | Cytology (ASC-US+) with reflex HPV-16/18 for cytology < ASC-US women |
| 14 | Cytology (ASC-US+) with reflex PAX1 or SOX1 methylation for cytology < ASC-US women |
| 15 | Cytology (ASC-US+) with reflex E6/E7mRNA for cytology < ASC-US women |
HPV+: HPV positive
Data were analyzed on SPSS 20.0 and R software 3.6.2.
Results
Biomarker positivity among screening women
The demographic characteristics of the 3,000 enrolled women have been described previously [14]. The study population had a median age of 45 years, and minority ethnic groups constituted 83% of the participants. The biomarker positivity from highest to lowest were as follows: HPV testing (417 cases, 13.9%), mRNA testing (276 cases, 9.2%), AI-assisted cytology (at ASC-US+, 189 cases, 6.3%). Accordingly, the suspicious lesion rates under colposcopy were 12.9% (52/403) for HPV testing, 14.3% (39/272) for mRNA testing, and 12.8% (24/188) for cytology, respectively. Overall, 26 (0.9%) CIN2+ cases and 13 (0.4%) CIN3+ cases were detected. The detection rates for CIN2+ were highest by HPV testing (0.8%), followed by mRNA testing (0.6%), and AI-assisted cytology (0.5%). The data have been reported in our previously published article [14].
Among the 417 HPV-positive women, the methylation positivity (PAX1 or SOX1 positive) was 16.1% (67/417). Specifically, the positivity for PAX1 was 13.9% (58/417), and for SOX1 was 11.3% (47/417). The methylation positivity (PAX1 or SOX1 positive) were 9.0% in CIN1, 13.4% in CIN2, and 13.4% in CIN3. Of the HPV-positive individuals, 56.6% (236/417) were positive for mRNA, and 26.6% (111/417) were cytology abnormal. All samples were genotyped for HPV-16/18, of these, 192 samples were genotyped for 15 high risk HPV genotypes. The top six genotypes were: HPV-52 (22.9%), HPV-58 (16.7%), HPV-16 (14.1%), HPV-18 (13.5%), HPV-33 (9.9%), HPV-68 (5.2%). Among the HPV-positive individuals, 24 (5.8%) CIN2+ cases and 13 (3.1%) CIN3+ cases were detected (Table 2).
Table 2.
Positive proportions of screening methods among women with CIN
| Methods | %(n) | CIN1, % (n) | CIN2, % (n) | CIN3, % (n) | Cancer, % (n) | CIN2+, % (n) | CIN3+, % (n) |
|---|---|---|---|---|---|---|---|
| ASC-US+ | 26.6 (111) | 9.0 (10) | 3.6 (4) | 5.4 (6) | 2.7 (3) | 11.7 (13) | 8.1 (9) |
| NILM | 73.4 (306) | 5.9 (18) | 2.3 (7) | 1.3 (4) | 0.0 (0) | 3.6 (11) | 1.3 (4) |
| PAX1 | 13.9 (58) | 10.3 (6) | 15.5 (9) | 15.5 (9) | 5.2 (3) | 36.2 (21) | 20.7 (12) |
| SOX1 | 11.3 (47) | 10.6 (5) | 17.0 (8) | 17.0 (8) | 6.4 (3) | 40.4 (19) | 23.4 (11) |
| PAX1 or SOX1 | 16.1 (67) | 9.0 (6) | 13.4 (9) | 13.4 (9) | 4.5 (3) | 31.3 (21) | 17.9 (12) |
| PAX1 and SOX1 | 9.1 (38) | 10.5 (4) | 21.1 (8) | 21.1 (8) | 7.9 (3) | 50.0 (19) | 29.0 (11) |
| Negative methylation | 84.2 (351) | 6.3 (22) | 0.6 (2) | 0.3 (1) | 0.0 (0) | 0.9 (3) | 0.3 (1) |
| E6/E7 mRNA positive | 56.6 (236) | 8.5 (20) | 2.5 (6) | 3.8 (9) | 1.3 (3) | 7.6 (18) | 5.1 (12) |
| E6/E7mRNA negative | 43.4 (181) | 4.4 (8) | 2.8 (5) | 0.6 (1) | 0.0 (0) | 3.3 (6) | 0.6 (1) |
| HPV-16/18 | 26.4 (110) | 1.8 (2) | 0.9 (1) | 2.7 (3) | 1.8 (2) | 5.5 (6) | 4.6 (5) |
| HPV-16/18/52/58/33/68 | 81.8 (157) | 9.6 (15) | 1.9 (3) | 1.9 (3) | 1.3 (2) | 5.1 (8) | 3.2 (5) |
CIN cervical intraepithelial neoplasia, CIN2 cervical intraepithelial neoplasia grade 2, CIN3 cervical intraepithelial neoplasia grade 3, CIN2+ cervical intraepithelial neoplasia grade 2 or worse, CIN3+ cervical intraepithelial neoplasia grade 3 or worse, ASC-US+ atypical squamous cells of undetermined significance or worse, NILM negative for intraepithelial lesion or malignancy
Clinical performance of single triage strategies for HPV-positive women
The sensitivity was 92.3% and the specificity was 86.4%, in detecting CIN2+ for HPV testing without triage. For HPV-positive women triage, cytology (at ASC-US+) showed a sensitivity of 54.2% (95% CI: 33.2%−73.8%), a specificity of 75.1% (95% CI: 70.4%−79.2%), and an AUC of 0.65 (95% CI: 0.53–0.77). The number needed to colposcopy for one CIN2+ detection was 8.5, and the missed CIN2+ proportion was 50.0% (Table 3).
Table 3.
Clinical accuracy of single triage strategies for CIN2 + and CIN3 + detection
| Methods | CIN, n | Sensitivity (95%CI) | Specificity (95%CI) | PPV (95%CI) | NPV (95%CI) | RSe/RSp | AUC (95%CI) | Referral rate, %(n) | NNR | Missed CIN, %(n/N) |
|---|---|---|---|---|---|---|---|---|---|---|
| CIN2+ | ||||||||||
| Cytology (ASC-US+) | 13 |
54.2 (33.2–73.8) |
75.1 (70.4–79.2) |
11.7 (6.6–19.5) |
96.4 (93.5–98.1) |
1.0/1.0 |
0.65 (0.53–0.77) |
3.7 (111) |
8.5 (13/111) |
50.0 (13/26) |
| Cytology (LSIL+) | 8 |
33.3 (16.4–55.3) |
91.8 (88.6–94.2) |
19.5 (9.4–35.4) |
95.8 (93.2–97.5) |
0.6/1.2 |
0.63 (0.50–0.76) |
1.4 (41) |
5.1 (8/41) |
69.2 (18/26) |
| PAX1 or SOX1 methylation | 21 |
87.5 (66.5–96.7) |
88.3 (84.6–91.2) |
31.3 (20.9–44.0) |
99.1 (97.3–99.8) |
1.6/1.2 |
0.88 (0.80–0.96) |
2.2 (67) |
3.2 (21/67) |
19.2 (5/26) |
| PAX1 methylation | 21 |
87.5 (66.5–96.7) |
90.6 (87.1–93.2) |
36.2 (24.3–49.9) |
99.2 (97.4–99.8) |
1.6/1.2 |
0.89 (0.81–0.97) |
1.9 (58) |
2.8 (21/58) |
19.2 (5/26) |
| SOX1 methylation | 19 |
79.2 (57.3–92.1) |
92.9 (89.7–95.1) |
40.4 (26.7–55.7) |
98.7 (96.7–99.5) |
1.5/1.3 |
0.86 (0.76–0.96) |
1.6 (47) |
2.5 (19/47) |
26.9 (7/26) |
| PAX1 and SOX1 methylation | 19 |
79.2 (57.3–92.1) |
95.2 (92.4–97.0) |
50.0 (33.7–66.3) |
98.7 (96.8–99.5) |
1.5/1.3 |
0.87 (0.78–0.97) |
1.0 (30) |
1.6 (19/30) |
26.9 (7/26) |
| E6/E7 mRNA | 18 |
75.0 (52.9–89.4) |
44.5 (39.6–49.6) |
7.6 (4.7–12.0) |
96.7 (92.6–98.6) |
1.4/0.6 |
0.60 (0.49–0.71) |
7.9 (236) |
13.1 (18/236) |
30.8 (8/26) |
| HPV-16/18 | 6 |
12.5 (0.6–53.3) |
71.7 64.6–78.0 |
1.9 (0.1–11.4) |
95.0 (89.5–97.8) |
0.2/1.0 |
0.42 (0.24–0.61) |
3.7 (110) |
18.3 (6/110) |
76.9 (20/26) |
| HPV-16/18/52/58/33/68 | 8 |
100.0 (59.8–100.0) |
19.0 (13.8–25.6) |
5.1 (2.4–10.1) |
100.0 (87.7–100.0) |
1.9/0.3 |
0.60 (0.43–0.75) |
5.3(158) |
5.1 (8/158) |
69.2 (18/26) |
| CIN3+ | ||||||||||
| Cytology (ASC-US+) | 9 |
69.2 (38.9–89.6) |
74.8 (70.2–78.9) |
8.1 (4.0–15.2.0.2) |
91.9 (84.8–96.0) |
1.0/1.0 |
0.72 (0.57–0.87) |
-- | -- |
30.8 (4/13) |
| Cytology (LSIL+) | 5 |
38.5 (15.1–67.6) |
91.1 (87.8–93.6) |
12.2 (4.6–27.0) |
97.9 (95.7–99.0) |
0.6/1.2 |
0.65 (0.47–0.82) |
-- | -- |
61.5 (8/13) |
| PAX1 or SOX1 methylation | 12 |
92.3 (62.1–99.6) |
86.4 (82.6–89.5) |
17.9 (10.0–29.6.0.6) |
99.7 (98.2–99.9) |
1.3/1.2 |
0.89 (0.81–0.98) |
-- | -- |
7.7 (1/13) |
| PAX1 methylation | 12 |
92.3 (62.1–99.6) |
88.6 (85.0–91.5.0.5) |
20.7 (11.6–33.7) |
99.7 (98.2–99.9) |
1.3/1.2 |
0.91 (0.82–0.99) |
-- | -- |
7.7 (1/13) |
| SOX1 methylation | 11 |
84.6 (53.7–97.3) |
91.1 (87.8–93.6) |
23.4 (12.8–38.4) |
99.46 (97.8–99.9) |
1.2/1.2 |
0.88 (0.76–0.99) |
-- | -- |
15.4 (2/13) |
| PAX1 and SOX1 methylation | 11 |
84.6 (53.7–97.3) |
93.3 (90.3–95.5) |
29.0 (15.9–46.1) |
99.5 (97.9–99.9) |
1.2/1.3 |
0.89 (0.78–1.00.78.00) |
-- | -- |
15.4 (2/13) |
| E6/E7 mRNA | 12 |
92.3 (62.1–99.6) |
44.6 (39.7–49.6) |
5.1 (2.8–8.9) |
99.5 (96.5–99.9) |
1.3/0.6 |
0.68 (0.57–0.80) |
-- | -- |
7.7 (1/13) |
| HPV-16/18 | 4 |
20.0 (1.1–70.1) |
72.2 (65.1–78.4) |
1.9 (0.01–11.4) |
97.1 (92.3–99.1) |
0.3/1.0 |
0.46 (0.22–0.71) |
-- | -- |
69.2 (9/13) |
| HPV-16/18/52/58/33/68 | 5 |
100.0 (46.3–100.0) |
18.7 (13.5–25.2) |
3.2 (1.2–7.7) |
100.0 (87.7–100.0) |
1.4/0.3 |
0.59 (0.38–0.81) |
-- | -- |
61.5 (8/13) |
CIN cervical intraepithelial neoplasia, CI confidence interval, PPV positive predictive value, NPV negative predictive value, RSe relative sensitivity, RSp relative specificity, AUC area under the curve, NNR number needed to referral, CIN2+ cervical intraepithelial neoplasia grade 2 or worse, AI artificial intelligence, ASC-US+ atypical squamous cells of undetermined significance or worse, CIN3+ cervical intraepithelial neoplasia grade 3 or worse
Compared with cytology, triaging by PAX1 or SOX1 DNA methylation demonstrated superior sensitivity (87.5%, 95% CI: 66.5%−96.7%, relative sensitivity: 1.6), better specificity (88.3%, 95% CI: 84.6%−91.2%, relative specificity: 1.2), and higher AUC (0.88, 95% CI: 0.80–0.96). The number needed to colposcopy to detect one CIN2+ decreased to 3.2, and fewer CIN2+ cases were missed (19.2%). Triaging by PAX1 methylation alone exhibited higher sensitivity (87.5%, 95% CI: 66.5%−96.7%), superior specificity (90.6%, 95% CI: 87.1%−93.2%), and better AUC (0.89, 95% CI: 0.81–0.97). The number needed to colposcopy for one lesion detection decreased markedly to 2.8, and fewer CIN2+ cases were missed (19.2%). Triaging by SOX1 methylation alone also demonstrated better sensitivity (79.2%, 95% CI: 57.3%−92.1%) and specificity (92.9%, 95% CI: 89.7%−95.1%), as well as increased AUC (0.86, 95% CI: 0.76–0.96). The number needed to colposcopy (2.5) for one CIN2+ detection and the missed CIN2+ proportion (26.9%) were both reduced. PAX1 and SOX1 methylation triage showed higher sensitivity (79.2%, 95% CI: 57.3%−92.1%), superior specificity (95.2%, 95% CI: 92.4%−97.0%), and improved AUC (0.87, 95% CI: 0.78–0.97). The number needed to colposcopy in detecting one lesion decreased markedly to 1.6, and the proportion of missed CIN2+ cases also decreased (26.9%) (Table 3).
Compared with cytology, mRNA testing triage showed better sensitivity (75.0%, 95% CI: 52.9%−89.4%) but lower specificity (44.5%, 95% CI: 39.6%−49.6%), and slightly decreased AUC (0.60, 95% CI: 0.49–0.71). Although the proportion of missed CIN2+ cases decreased (30.8%), the number needed to colposcopy for one CIN2+ detection was increased to 13.1 (Table 3).
HPV-16/18 genotyping triage demonstrated inferior sensitivity (12.5%, 95% CI: 0.6%−53.3%), slightly lower specificity (71.7%, 95% CI: 64.6%−78.0%), and decreased AUC (0.42, 95% CI: 0.24–0.61), compared with cytology triage. The number needed to colposcopy was increased markedly to 18.3, and the proportion of missed CIN2+ cases remained high (76.9%). Extended HPV genotyping (HPV-16/18/52/58/33/68) triage exhibited superior sensitivity (100.0%, 95% CI: 59.8%−100.0%) but poor specificity (19.0%, 95% CI: 13.8%−25.6%), and lower AUC (0.60, 95% CI: 0.43–0.75). Although the number needed to colposcopy for one lesion detection decreased to 5.1, the proportion of missed CIN2+ cases increased significantly (69.2%) (Table 3).
Clinical performance of combined triage strategies for HPV-positive women
HPV-16/18 genotyping with reflex cytology (ASC-US+) for those without HPV-16/18 demonstrated a sensitivity of 62.5% (95% CI: 40.8%–80.4%), a specificity of 54.5% (95% CI: 49.4%–59.5%), and an AUC of 0.59 (95% CI: 0.47–0.70), for detecting CIN2+. The number needed to colposcopy for one lesion detection was 12.9, with the missed CIN2+ proportion of 42.3% (Table 4).
Table 4.
Clinical accuracy of combined triage strategies for CIN2 + and CIN3 + detection
| Methods | CIN, n | Sensitivity (95%CI) | Specificity (95%CI) | PPV (95%CI) | NPV (95%CI) | RSe/RSp | AUC (95%CI) | Referral rate, %(n) | NNR | Missed CIN2+, %(n/N) |
|---|---|---|---|---|---|---|---|---|---|---|
| CIN2+ | ||||||||||
| HPV-16/18 | ASC-US + for non-HPV-16/18 | 15 |
62.5 (40.8–80.4) |
54.5 (49.4–59.4) |
7.7 (4.5–12.7) |
96.0 (92.2–98.0) |
1.0/1.0 |
0.59 (0.47–0.70) |
6.5 (194) |
12.9 (15/194) |
42.3 (11/26) |
| HPV-16/18 | PAX1 or SOX1 for non-HPV-16/18 | 22 |
91.7 (71.5–98.5) |
64.1 (59.1–68.8) |
13.5 (8.8–19.9) |
99.2 (96.9–99.9) |
1.5/1.2 |
0.78 (0.70–0.86) |
5.4 (163) |
7.4 (22/163) |
15.4 (4/26) |
| HPV-16/18 | mRNA for non-HPV-16/18 | 19 |
79.2 (57.3–92.1) |
31.3 (26.8–36.2) |
6.6 (4.1–10.2) |
96.1 (90.7–98.6) |
1.3/0.6 |
0.55 (0.44–0.67) |
9.6 (289) |
15.2 (19/289) |
26.9 (7/26) |
| ASC-US+ | HPV-16/18 | 4 |
30.8 (10.4–61.1) |
76.5 (66.7–84.3) |
14.8 (4.9–34.6) |
89.3 (80.2–94.7) |
0.5/1.4 |
0.54 (0.37–0.71) |
0.9 (27) |
6.8 (4/27) |
84.6 (22/26) |
| ASC-US+ | PAX1 or SOX1 | 13 |
100.0 (71.7–100.0) |
72.5 (62.3–80.8) |
32.5 (19.1–49.2) |
100.0 (93.6–100.0) |
1.6/1.3 |
0.86 (0.79–0.93) |
1.3 (40) |
3.1 (13/40) |
50.0 (13/26) |
| ASC-US+ | mRNA | 13 |
100.0 (71.7–100.0) |
48.0 (37.8–58.2) |
20.3 (11.7–32.6) |
100.0 (90.6–100.0) |
1.6/0.9 |
0.74 (0.64–0.85) |
2.1 (64) |
4.9 (13/64) |
50.0 (13/26) |
| CIN3+ | ||||||||||
| HPV-16/18 | ASC-US + for non-HPV-16/18 | 11 |
84.6 (53.7–97.3) |
54.7 (49.7–59.6) |
5.7 (3.0–10.2.0.2) |
99.1 (96.5–99.8) |
1.0/1.0 |
0.70 (0.57–0.82) |
-- | -- |
15.4 (2/13) |
| HPV-16/18 | PAX1 or SOX1 for non-HPV-16/18 | 13 |
100.0 (71.7–100.0) |
62.9 (57.9–67.6) |
8.0 (4.5–13.5) |
100.0 (98.1–100.0) |
1.2/1.2 |
0.81 (0.75–0.88) |
-- | -- |
0.0 (0/13) |
| HPV-16/18 | mRNA for non-HPV-16/18 | 13 |
100.0 (71.7–100.0) |
31.7 (27.2–36.5) |
4.5 (2.5–7.8) |
100.0 (96.4–100.0) |
1.2/0.6 |
0.66 (0.55–0.77) |
-- | -- |
0.0 (0/13) |
| ASC-US+ | HPV-16/18 | 3 |
33.3 (9.0–69.1.0.1) |
76.5 (66.8–84.1) |
11.1 (2.9–30.3) |
92.9 (84.5–97.1) |
0.4/1.4 |
0.55 (0.35–0.75) |
-- | -- |
76.9 (10/13) |
| ASC-US+ | PAX1 or SOX1 | 9 |
100.0 (62.9–100.0) |
69.6 (59.6–78.1) |
22.5 (11.4–38.9) |
100.0 (93.6–100.0) |
1.2/1.3 |
0.85 (0.77–0.93) |
-- | -- |
30.8 (4/13) |
| ASC-US+ | mRNA | 9 |
100.0 (62.9–100.0) |
46.1 (36.3–56.2) |
14.1 (7.0–25.5.0.5) |
100.0 (90.6–100.0) |
1.2/0.8 |
0.73 (0.61–0.85) |
-- | -- |
30.8 (4/13) |
CIN cervical intraepithelial neoplasia, CI confidence interval, PPV positive predictive value, NPV negative predictive value, RSe relative sensitivity, RSp relative specificity, AUC area under the curve, NNR number needed to referral, CIN2+ cervical intraepithelial neoplasia grade 2 or worse, AI artificial intelligence, ASC-US+ atypical squamous cells of undetermined significance or worse, CIN3+ cervical intraepithelial neoplasia grade 3 or worse
Compared with the reference triage, HPV-16/18 genotyping with reflex to PAX1 or SOX1 methylation for those without HPV-16/18 showed higher sensitivity (91.7%, 95% CI: 71.5%–98.5%, relative sensitivity: 1.5), improved specificity (64.1%, 95% CI: 59.1%–68.8%, relative specificity: 1.2), and increased AUC (0.78, 95% CI: 0.70–0.86). The number needed to colposcopy for one lesion detection decreased markedly to 7.4, and the proportion of missed CIN2+ cases declined significantly to 15.4%. Triaging by HPV-16/18 genotyping with reflex to mRNA for those without HPV-16/18 demonstrated improved sensitivity (79.2%, 95% CI: 57.3%–92.1%) but inferior specificity (31.3%, 95% CI: 26.8%–36.2%), and similar AUC (0.55, 95% CI: 0.44–0.67). Although the proportion of missed CIN2+ cases declined (26.9%), the number needed to colposcopy for detecting one CIN2+ increased to 15.2 (Table 4).
Triaging by cytology (ASC-US+) with reflex PAX1 or SOX1 methylation exhibited superior sensitivity (100.0%, 95% CI: 71.7%–100.0%), higher specificity (72.5%, 95% CI: 62.3%–80.8%), as well as better AUC (0.86, 95% CI: 0.79–0.93). The number needed to colposcopy for one lesion detection decreased markedly to 3.1, but the proportion of missed CIN2+ cases slightly increased (50.0%). Cytology (ASC-US+) with reflex mRNA triage demonstrated significantly higher sensitivity (100.0%, 95% CI: 71.7%–100.0%), increased AUC (0.74, 95% CI: 0.64–0.85), but inferior specificity (48.0%, 95% CI: 37.8%–58.2%), compared with the reference triage approach. The number needed to colposcopy to detect one CIN2+ decreased to 4.9, but the proportion of missed CIN2+ cases remained unchanged (50.0%). Cytology (ASC-US+) with reflex HPV-16/18 genotyping triage showed better specificity (76.5%, 95% CI: 66.7%–84.3%), but inferior sensitivity (30.8%, 95% CI: 10.4%–61.1%), and decreased AUC (0.54, 95% CI: 0.37–0.71). Although the number needed to colposcopy for one lesion detection decreased significantly to 6.8, the proportion of missed CIN2+ cases increased markedly (84.6%), compared with the reference triage (Table 4).
Discussion
For HPV-positive women, PAX1 methylation triage demonstrated superior clinical performance compared to the frequently used triage strategy (cytology at ASC-US+). It significantly reduced the number needed to colposcopy to detect one cervical lesion and missed fewer CIN2+ cases. Adding SOX1 methylation to PAX1 did not improve the clinical performance over PAX1 alone. Neither mRNA triage nor extended HPV genotyping triage exhibited significantly improved clinical performance compared with cytology. HPV-16/18 genotyping with reflex PAX1 or SOX1 methylation triage showed better clinical performance compared to HPV-16/18 genotyping combined with cytology triage. Nevertheless, these combined triage strategies did not outperform PAX1 methylation triage alone.
The current clinical recommendation for HPV-positive triage is cytology [5]. Individuals with high-risk HPV positivity and abnormal cytology, or those with non-HPV-16/18 positivity and abnormal cytology, are referred to colposcopy. With its high specificity, cytology has been identified as a strong triage strategy for HPV primary screening. However, as a morphology-based diagnosis, cytology’s clinical accuracy largely depends on cytologists’ experience. In resource-limited regions such as rural Yunnan, cytology triage may not be effectively used due to a shortage of experienced cytologists. A study conducted in rural Xinjiang, China reported that the sensitivity was 63.8% and the specificity was 97.9% for cytology triage in detecting CIN2+ [16]. A previous study in China indicated that, the AI-assisted cytology for triaging HPV-positive individuals demonstrated sensitivity of 86.5% and specificity of 51.3% for detecting CIN2+ [17]. Differently, AI-assisted cytology triage in our study showed lower sensitivity (54.2%) and higher specificity (75.1%) for CIN2+ detection. These finding indicated that the clinical accuracy of cytology triage varied across different settings. Therefore, it may not be an ideal triage approach for the HPV-positive population especially in resource-limited areas that lack high-quality cytologists.
DNA methylation is a crucial epigenetic modification that plays a significant role in the occurrence and development of cervical cancer, and has been reported to be associated with the progression from low-grade to high-grade premalignant cervical lesions [18]. In this study, the positivity of PAX1 and SOX1 methylation were significantly increased in higher-grade cervical lesions compared to low-grade lesions. In our study, PAX1 methylation triage enhanced both sensitivity (87.5% vs. 54.2%) and specificity (90.6% vs. 75.1%) compared with cytology triage. The sensitivity and specificity were both higher than those reported in a cohort study from Hong Kong, China (sensitivity: 73.5%, specificity: 70.3%) [9]. Both that study and ours found that adding SOX1 methylation did not significantly improve the triage efficacy compared with using PAX1 methylation alone. Furthermore, the PAX1 methylation triage detected more CIN2+ cases while requiring fewer colposcopy referrals than cytology triage. Although the PAX1 methylation combined with HPV-16/18 genotyping triage in this study showed improved clinical accuracy than other combined triage strategies, the combined triage strategy did not significantly outperform the single triage strategy by DNA methylation. Therefore, PAX1 methylation alone demonstrated the potential to be an triage recommendation for HPV-positive women in rural Yunnan. Moreover, PAX1 methylation is an objective molecular method that can be performed as a reflex test on self-collected samples. As HPV self-sampling is highly recommended to be integrated into cervical cancer screening programs [19], a reflex molecular test that can be performed on self-samples would be ideal.
HPV mRNA has been identified as having equivalent sensitivity and superior specificity for CIN2+ detection compared with HPV DNA as a primary screening approach [20], but it has been less evaluated as a triage strategy for HPV-positive individuals. Previous studies have reported that mRNA testing showed satisfied sensitivity (96.0%) but poor specificity (34.7%), and high colposcopy referral rate on triage for HPV-positive population [21]. Similarly, in our study, the E6/E7 mRNA triage did not improve the specificity (44.5%) and required more colposcopy referrals for cervical lesion detection than cytology triage. Consequently, it had not shown potential as an optimal triage method in rural Yunnan.
HPV genotypes carry varying carcinogenic risks for cervical cancer. HPV-16/18 account for 70% of cervical cancers [22], and primary screening with HPV-16/18 positivity have been recommended for colposcopy due to their high risk for cervical lesions [5]. In this study, HPV-16/18 triage was observed to markedly decrease sensitivity and missed more cervical lesions compared to cytology triage. We subsequently evaluated an extended HPV genotyping triage strategy based on six most prevalent genotypes (HPV-52/58/16/18/33/68). This strategy demonstrated poor specificity (19%), despite reduced unnecessary colposcopy referrals. In contrast, a pooled study in China reported that an extended six HPV genotyping (HPV-16/18/58/52/33/31) triage achieved both satisfactory sensitivity (87.5%) and specificity (93.1%), along with a reduction in colposcopy referrals [23]. Identifying the optimal extended HPV genotyping panel for triaging HPV-positive women in rural Yunnan requires further evaluation with a larger sample size.
Yunnan and similar low-resource areas are transitioning from cytology-based to HPV test-based cervical cancer screening. Assessing the accuracy and feasibility of triage approaches is crucial for population-wide screening programs. While these triage strategies have been evaluated in various Chinese settings, they have been less evaluated in real-world setting in rural Yunnan, a frontier region home to many ethnic minorities and with limited health resources. Evidence from such settings is essential to inform government-led, HPV-based cervical cancer screening program. However, this study has several limitations. Firstly, it is a cross-sectional study with limited number of CIN2+ cases (n = 24), which may result in unstable estimates of sensitivity and specificity. Secondly, the HPV DNA test used was only an “out-of-lab” test which was not been verified by a PCR-based laboratory, adapting to a real-world situation of lacking PCR laboratory in the study site. Thirdly, not all samples were tested for 15 high-risk HPV genotypes, limiting the evaluation of HPV genotyping triage efficacy. Additionally, several potential biases, such as selection and geographic biases, may have influenced the sample selection in this study.
In conclusion, PAX1 methylation demonstrated better clinical performance and required fewer colposcopy referrals for cervical lesion detection compared with cytology triage. It showed good potential as an triage strategy for HPV-positive population in cervical cancer screening in rural Yunnan, China. Future studies are recommended to evaluate the long-term clinical performance of PAX1 methylation triage, its efficacy on self-collected samples, and its cost-effectiveness to generate more valuable evidence for cervical cancer screening programs.
Acknowledgements
Jie He and Chen Lu from the Third Affiliated Hospital of Kunming Medical University were appreciated for laboratory testing, all the local collaborative investigators were appreciated for their efforts in conducting the study, and all women who participated in this study were appreciated.
Authors’ contributions
RF Duan and HP Zhang initiated and designed the study. XW Zhai, T Kong, X Yang, J Yu, Y Dao, and RF Duan participated in sample collection, experiment, and data management. XW Zhai, T Kong, and RF Duan participated in manuscript writing, data analysis, and interpretation. HY Yang and HP Zhang provided constructive comments and revisions on the manuscript. All authors have reviewed and approved the final article.
Funding
This study was financially supported by Yunnan Province Medical Discipline Reserve Talent Training Program [H-2024010], Basic Research Project of Yunnan Provincial Department of Science and Technology [202301AT070237], “Xingdian Talent Support Plan” for Outstanding Doctors in Yunnan Province [XDYC-MY-2022-0056], The Joint Special Funds for the Yunnan Provincial Department of Science and Technology and Kunming Medical University [202301AY070001-245].
Data availability
Original data are available and can be accessed by contacting Drs. Rufei Duan and Hongping Zhang. All the patient records were de-identified before been obtained for analysis. The analysis outcomes of all available data were reported in the article.
Declarations
Ethics approval and consent to participant
This study adhered to the Declaration of Helsinki in the ‘Ethics approval and consent to participate’ section of the Declarations. The study protocol was approved by the Ethical Review Committees of Yunnan Cancer Hospital The Third Affiliated Hospital of Kunming Medical University. Women who participated in this study gave written informed consent before specimen collection.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Xiuwei Zhai, Tong Kong, Hongping Zhang and Rufei Duan contributed equally to this work.
Contributor Information
Hongping Zhang, Email: kmzhp@126.com.
Rufei Duan, Email: rufeiduan11@163.com.
References
- 1.Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229–63. [DOI] [PubMed] [Google Scholar]
- 2.Zheng RS, Chen R, Han BF, Wang SM, Li L, Sun KX, et al. [Cancer incidence and mortality in China, 2022]. Zhonghua Zhong Liu Za Zhi. 2024;46(3):221–31. [DOI] [PubMed] [Google Scholar]
- 3.Basu P, Hutubessy R, Broutet N. Cervical cancer: WHO called for elimination, not eradication. BMJ. 2019;366:l5668. [DOI] [PubMed] [Google Scholar]
- 4.Zhao FH, Ren WH. [Accelerating the elimination of cervical cancer in China and building a paradigm for “Healthy China” cancer prevention]. Zhonghua Yi Xue Za Zhi. 2021;101:1831–4. [DOI] [PubMed] [Google Scholar]
- 5.WHO guideline for screening and treatment of cervical pre-cancer lesions for cervical cancer prevention [Internet]. 2nd ed. Geneva: World Health Organization; 2021. [PubMed]
- 6.Bhatla N, Singhal S. Primary HPV screening for cervical cancer. Best Pract Res Clin Obstet Gynaecol. 2020;65:98–108. [DOI] [PubMed] [Google Scholar]
- 7.Cuschieri K, Ronco G, Lorincz A, Smith L, Ogilvie G, Mirabello L, et al. Eurogin roadmap 2017: triage strategies for the management of HPV-positive women in cervical screening programs. Int J Cancer. 2018;143(4):735–45. [DOI] [PubMed] [Google Scholar]
- 8.Dick S, Vink FJ, Heideman DAM, Lissenberg-Witte BI, Meijer C, Berkhof J. Risk-stratification of HPV-positive women with low-grade cytology by FAM19A4/miR124-2 methylation and HPV genotyping. Br J Cancer. 2022;126(2):259–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Chan KKL, Liu SS, Lau LSK, Ngu SF, Chu MMY, Tse KY, et al. PAX1/SOX1 DNA methylation versus cytology and HPV16/18 genotyping for the triage of high-risk HPV-positive women in cervical cancer screening: retrospective analysis of archival samples. BJOG. 2025;132(2):197–204. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Chen X, Jiang H, Xu H, Wang L, Liu P, Ma D, et al. Cervical cancer screening: efficacy of PAX1 and JAM3 methylation assay in the triage of atypical squamous cell of undetermined significance (ASC-US). BMC Cancer. 2024;24(1):1385. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Rezhake R, Wang Y, Zhao X, Arbyn M, Shen G, Pan Q, et al. Performance of human gene EPB41L3 and HPV 16/18 viral DNA methylation to triage hrHPV-positive women. Vaccines. 2023. 10.3390/vaccines12010046. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Su Y, Tang Y, Zhang T, Xu Y, Zhang Y, Shen Y, et al. Methylated ZNF582 as a triage marker for occult cervical cancer and advanced cervical intraepithelial neoplasia. Future Oncol. 2022;18(23):2583–92. [DOI] [PubMed] [Google Scholar]
- 13.Mazurec K, Trzeszcz M, Mazurec M, Streb J, Halon A, Jach R. Triage strategies for non-16/non-18 HPV-positive women in primary HPV-based cervical cancer screening: p16/Ki67 dual stain vs. cytology. Cancers (Basel). 2023. 10.3390/cancers15205095. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Duan R, Yang H, Zhai X, Yu J, Lu C, Yang W, et al. Accuracy and acceptability of self-sampling HPV testing in cervical cancer screening: a population-based study in rural Yunnan, China. Sci Rep. 2025. 10.1038/s41598-025-09292-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Nayar R, Wilbur DC. The pap test and Bethesda 2014. Cancer Cytopathol. 2015;123(5):271–81. [DOI] [PubMed] [Google Scholar]
- 16.Rezhake R, Abuduxikuer G, Abudurexiti G, Zhuo Q, Muhetaer K, Abulimiti T, et al. Evaluation of the multiple HPV-based “screen and triage” algorithms in real-world settings of rural China. Cancer Biol Med. 2025. 10.20892/j.issn.2095-3941.2025.0384. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Xue P, Xu HM, Tang HP, Wu WQ, Seery S, Han X, et al. Assessing artificial intelligence enabled liquid-based cytology for triaging HPV-positive women: a population-based cross-sectional study. Acta Obstet Gynecol Scand. 2023;102(8):1026–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Pulliero A, Cassatella G, Astuni P, Khalid Z, Fiordoro S, Izzotti A. The role of microRNA expression and DNA methylation in HPV-related cervical cancer: a systematic review. Int J Mol Sci. 2024. 10.3390/ijms252312714. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Serrano B, Ibáñez R, Robles C, Peremiquel-Trillas P, de Sanjosé S, Bruni L. Worldwide use of HPV self-sampling for cervical cancer screening. Prev Med. 2022;154:106900. [DOI] [PubMed] [Google Scholar]
- 20.Arbyn M, Simon M, de Sanjosé S, Clarke MA, Poljak M, Rezhake R, et al. Accuracy and effectiveness of HPV mRNA testing in cervical cancer screening: a systematic review and meta-analysis. Lancet Oncol. 2022;23(7):950–60. [DOI] [PubMed] [Google Scholar]
- 21.Giorgi Rossi P, Carozzi F, Ronco G, Allia E, Bisanzi S, Gillio-Tos A, et al. P16/ki67 and E6/E7 mRNA accuracy and prognostic value in triaging HPV DNA-positive women. J Natl Cancer Inst. 2021;113(3):292–300. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Yu YQ, Hao JQ, Mendez MJG, Mohamed SB, Fu SL, Zhao FH, et al. The prevalence of cervical HPV infection and genotype distribution in 856,535 Chinese women with normal and abnormal cervical lesions: a systemic review. J Cytol. 2022;39(4):137–47. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Dun C, Yuan M, Zhao X, Hu S, Arbyn M, Zhao F. Clinical evaluation of primary human papillomavirus (HPV) testing with extended HPV genotyping triage for cervical cancer screening: a pooled analysis of individual patient data from nine population-based cervical cancer screening studies from China. Cancer Med. 2024;13(11):e7316. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Original data are available and can be accessed by contacting Drs. Rufei Duan and Hongping Zhang. All the patient records were de-identified before been obtained for analysis. The analysis outcomes of all available data were reported in the article.