Introduction

Cervical cancer became the fourth cancer among females, with 600,000 cancer cases and 340,000 cancer deaths per year worldwide1. About 80% of new cancer cases and deaths occurred in low- and middle-income countries due to restricted access to HPV vaccination and high-quality screening1. In China, 110,000 cervical cancer cases and 34,000 deaths occurred in 2016, and it has been observed increasing trend in the past decade2. World Health Organization (WHO) has initiated the call for eliminating cervical cancer in 2018 by integrating HPV vaccine, screening, and treatment3. China has determined to fully support the global strategy of cervical cancer elimination, released the national action plan for accelerating cervical cancer elimination, and addressed the 70% screening coverage on females aged 35–64 by 20304. However, the current screening coverage in China is only 36.8% among females aged 35–64, and the coverage is lower in rural areas than that of in urban areas5. It’s a challenge to extend the screening coverage in China, especially in remote rural areas.

WHO priority recommended HPV test as primary screening strategy, with or without appropriate triage6. With the high sensitivity7, good reproducibility, and long-term reassurance after a negative test result for HPV test8, more and more countries are switching from cytology-based to HPV-based screening. WHO recommends that HPV self-sampling should be made available as an additional approach in cervical cancer screening services for women aged 30–60 years6, and self-sampling can help reach the global target of 70% coverage of screening by 2030. Population-based studies have indicated that self-sampling HPV testing showed similar accuracy with the provider-sampling HPV testing9. Furthermore, women may feel more comfortable taking their own samples rather than going to see a health worker for cervical cancer screening. However, HPV test is hard to popularize in low-resource areas, especially in remote rural areas where lack of the polymerase chain reaction (PCR) laboratory. The point-of-care (POC) testing is designed for resource limited areas, using a small device to finish HPV testing out of a professional laboratory, and the staff could be a non-expertise with basic training. The POC HPV testing therefore makes it possible to access high-precision screening technology in remote rural areas with limited economy.

China is switching from cytology-based to HPV-based primary cervical cancer screening. However, the clinical accuracy and feasibility of self-sampling HPV test is less evaluated in rural China. Yunnan located in southwest of China, adjacent to Myanmar, Laos, and Cambodia. It is an economic less-developed region with most areas are plateau mountain area. Cervical cancer screening coverage in Yunnan is estimated less than 40%5. Liquid-based cytology is using as the primary screening approach in Yunnan. The HPV testing promotion is hindered by lack of PCR laboratory and limited health service workers. Additionally, in some remote rural sites, the local minority rural women usually more rely on health givers and maybe not well accept and acknowledge new things such as self-provided health care. Thus, the conducted study was expected to evaluate the accuracy and acceptability of self-sampling "out-lab" HPV testing among remote rural females in Yunnan, to explore the appropriate HPV-based cervical cancer screening strategy in low resource areas, especially for the hard-to-reach population.

Materials and methods

Study population

Participants were recruited from rural minority communities in Shuangjiang County, Yunnan province of China, based on the National Cervical and Breast Cancer Screening Program, from March 2022 to December 2022. The inclusion criteria were: (1) women who reported initiating sexual activity; (2) had not undergone cervical cancer screening in the past three years; (3) were both physically and mentally capable of undergoing cervical sampling; (4) had an intact cervix, and were not currently pregnant. Women diagnosed with severe vaginal inflammation during the pelvic examination were excluded. All participating women provided signed informed consent.

Questionnaire

After providing informed consent, participants were individually interviewed by a trained health worker using a self-designed questionnaire (Supplementary material-Questionnaire) to gather demographic characteristics, and assess knowledge, attitudes, and practices regarding cervical cancer screening. Following the screening, the women were asked about their acceptance of the sampling model and their experiences with self-sampling.

Specimen collection

Women collected one vaginal self-sample using a vaginal brush (Qiagen, Jiangsu, China) according to sampling standard under the supervision of healthcare providers, and placed it in PreservCyt solution (ThinPrep, Hologic, Marlborough, MA, USA). Subsequently, a trained gynecologist collected two exfoliated cervical cell samples using a paired special sampling brush and placed each in PreservCyt solution (ThinPrep, Hologic, Marlborough, MA, USA).

Laboratory testing

All collected samples were stored in a 4 °C refrigerator. The self-collected samples, along with residuals from one physician-collected sample, were transferred to Yunnan Cancer Hospital laboratory for HPV DNA testing using the Sansure HPV test® (Sansure Biotech, Changsha, China) and GenPlex HPV test® (Bohui Biotech, Beijing, China). The HPV testing process, including nucleic acid extraction, purification, amplification, and detection, is automated using the GenPlex microfluidic fully automatic nucleic acid testing system developed by Bohui Innovations. This compact device is placed outside the laboratory, and in this study, we refer to it as the “out-lab” HPV testing system. Additionally, one physician-collected sample was sent to KingMed Diagnostic Laboratory (Yunnan, China) for HPV E6/E7 mRNA testing (Hologic, Marlborough, MA, USA). Another physician-collected sample was shipped to the local hospital (Shuangjiang maternal and child health hospital) laboratory for cytology slides processing, and interpreted by cytologists from Wuhan Landing Intelligence Medical CO. LTD (Wuhan, China), assisted with the artificial intelligence system.

The Sansure HPV test is a PCR-based diagnostic tool that utilizes One-step Fast Release technology and real-time fluorescent quantitative PCR to target 15 high-risk HPV types (HPV-16, 18, 31, 33, 35, 39, 45, 51, 52, 53, 56, 58, 59, 66, 68), with specific genotyping for HPV-16 and HPV-18. This test has received European Union Certificate, indicating compliance with European health and safety standards. HPV status is determined by the cycle threshold (Ct) values observed; a Ct ≤ 39 is considered HPV positive, and a Ct > 39 is considered negative10.

GenPlex HPV Test utilizes PCR for the amplification of HPV DNA and low-density microarray to target-specifically hybridize the amplified DNA for detection. GenPlex Test detected 15 high-risk HPV types (HPV-16, 18, 31, 33, 35, 39, 45, 51, 52, 53, 56, 58, 59, 66, 68) and 9 low-risk HPV types6,11, and provided specific genotyping for the 24 types of HPV. The cut-off value of the probe at each point is calculated by receiver operating characteristics (ROC) method. When the detection signal value of the probe is greater than or equal to the cut-off value of the probe, the corresponding subtype of the probe is judged to be positive, otherwise, judged to be negative.

HPV E6/E7 mRNA test was performed by using the branched strand DNA hybridization capture principle to quantitatively detect 14 high-risk HPV types16,18 in accordance with the experimental instructions. Detection results HPV E6/E7 mRNA ≥ 1.0 copy indicates positive, while < 1.0 copy indicates negative.

The operators for HPV DNA test were the non-experts from Yunnan Cancer Hospital, who have been accepted the simple training and were qualified to operate the testing according to Standard Operating Procedure (SOP) provided by the biotech enterprise. The equipment were calibrated by enterprise engineer before the testing process, and strictly internal quality control was implemented during the detection. Factors such as sample degradation (especially RNA degradation), poor collection, and test limitations could impact the results. Therefore, all collected samples were preserved in a specialized solution and promptly shipped to the laboratory for DNA or RNA testing.

AI Cytology is the artificial intelligence (AI)-assisted liquid-based cytology systems that used pathological image data modeling to form an automatic and intelligent cytopathological diagnosis system ‘The Artificial Intelligence Cloud Diagnosis System’ (Patent number: ZL 2019 1 0,964,425.7). It based on machine-learning-derived thresholds that has been trained on more than 1 million cytological cases, including around 400,000 NILM, 200,000 LSIL, 100,000 HSIL, 200,000 ASC-US, 100,000 ASC-H, and the number of cell markers exceeds 100 million. The collected samples were shipped to the local hospital for cytology slides processing, then the full cytology slides were scanned and were automatically uploaded as image data to the cloud platform using the remote scanning and transmission equipment of Wuhan Landing Intelligence Medical CO. LTD, where the cytology results were first automatic classified as normal and cancerous cells, by The Artificial Intelligence Cloud Diagnosis Platform, while the results were finally been confirmed by cytologists form Landing Intelligence Medical CO. LTD, and the final cytology interpretation results were read and printed by local hospital workers. The cytology results were interpreted and diagnosed in accordance with the International Society of Cytology’s Bethesda System guidelines for cervical cytology reporting11. Diagnostic categories included: negative for intraepithelial lesion or malignancy (NILM), atypical squamous cells of undetermined significance (ASC-US), atypical squamous cells-cannot exclude HSIL (ASC-H), atypical glandular cells (AGC), low-grade squamous intraepithelial lesion (LSIL), high-grade squamous intraepithelial lesion (HSIL), squamous cell carcinoma (SCC), adenocarcinoma in situ (AIS), and adenocarcinoma (ADC).

Women positive for any HPV DNA or mRNA, or with cytological ASC-US or worse (ASC-US +) or unsatisfactory were recalled for colposcopy examination, and biopsied if any suspicious lesion was identified. Endocervical curettage was performed if the squamocolumnar junction was invisible. Biopsy tissues were immediately immersed in 10%-buffered formalin and transported to KingMed diagnostics for processing and diagnosis by experienced pathologists who were blinded to other screening results.

Pathology results were the golden standard and 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 according to clinical guidelines6. Women who were negative for all tests (no HPV detected and cytology < ASC-US) were considered to be negative for the outcome of CIN2 + .

All methods were performed in accordance with the national cervical cancer screening and treatment guidelines and regulations.

Statistical analysis

Clinical accuracy of screening approaches 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 relative sensitivity and specificity (reference for provider-sampling model) were calculated with 95% CI. Significant difference was considered if the 95% CI of the values were entirely above or below one.

The knowledge, attitude, and practice (KAP) for cervical cancer screening were evaluated by answer scores. The KAP score were defined as: know about HPV (2 points), never know about HPV (1 points); know about HPV causes cervical cancer (3 points), not sure about HPV causes cervical cancer (2 points), not know about HPV causes cervical cancer (1 points); had cervical cancer screening before (2 points), never had cervical cancer screening before (1 points). If the score more than 5 points were defined as good, while if the score less than 5 points were defined as inferior. Self-sampling model acceptance rate and the experience were described, and the influencing factors were analyzed using logistic regression.

Data were analyzed on SPSS 20.0 and R software 3.6.2.

Results

Study population and screening positive rate

Among the 3000 recruited women, the median age was 45 years old (95% CI 36–52). Minority ethnic females constituted 82.5% (n = 2475) of the study population, with breakdowns as follows: Dai minority at 53.4% (n = 1602), Lahu minority at 9.1% (n = 273), Wa minority at 7.3% (n = 219), Bulang minority at 7.2% (n = 216), Yi minority at 3.4% (n = 102), and Bai minority at 2.1% (n = 63). The Han ethnic group (non-minority) comprised 17.5% (n = 525) of the participants. Additionally, 94.8% (n = 2843) of the women were married, 85.5% (n = 2565) had an education level of junior high school or below, and 51.9% (n = 1557) had never participated in cervical cancer screening before. These data are presented in Table 1.

Table 1 Demographic characteristic of the screened women.
Full size table

Flowchart of the study procedure is shown in Supplementary Fig. 1. Among the 3000 women, the positive rate was 16.8% (n = 504) for self-sampling HPV DNA testing, 13.9% (n = 417) for provider-sampling HPV DNA testing, 9.2% (n = 276) for E6/E7 mRNA testing, and 6.3% (n = 189) for AI-assisted cytology at ASC-US + (data are not shown in table). A total of 610 women tested positive on any of the tests were recommended referred to colposcopy, and 546 of these (89.5%) have accepted the colposcopy. The detection rates for CIN1, CIN2, CIN3, and cervical cancer were 1.06% (n = 32), 0.43% (n = 13), 0.33% (n = 10), and 0.10% (n = 3) respectively, with a CIN2 + detection rate of 0.86% (n = 26), detailed in Table 2 and Supplementary Fig. 1. Additionally, of the 1400 HPV samples from both self-sampling and provider-sampling that were selected for head-to-head genotyping, 238 (17.0%) tested positive for high-risk HPV. The most common genotypes were HPV-52, HPV-58, HPV-16, and HPV-18 detected by both self- and provider-collected samples (data are not shown).

Table 2 Clinical accuracy of primary screening approaches for CIN2 + and CIN3 + detection.
Full size table

Accuracy of self- and provider-sampling HPV DNA testing

The positive agreement rate was 96.9%, negative agreement rate was 96.1%, and overall agreement rate was 96.2% respectively, for self- and provider-sampling HPV DNA testing, and the Kappa value was 0.86 (95% CI 0.82–0.90). Data are not shown.

The CIN2 + detection rate was 0.83% (n = 25) for self-sampling and 0.80% (n = 24) for provider-sampling HPV DNA testing respectively. The sensitivity was 96.2% (95% CI 90.8%–8.9%) and the specificity was 83.9% (95% CI 82.5–84.0%) for self-sampling HPV testing for detecting CIN2 + . The sensitivity and specificity were 92.3% (95% CI 81.4–98.5%) and 86.4% (95% CI 85.8–88.4%) respectively for provider-sampling HPV testing for CIN2 + detection. Data are shown in Table 2. The relative sensitivity and specificity of self-sampling testing was 1.03 (95% CI 0.76–1.62) and 0.96 (95% CI 0.65–1.46) compared to provider-sampling testing but insignificant (data are not shown). The PPV was 5.0% (95% CI 3.8–6.1%), NPV was 100.0% (95% CI 99.8–100%) for self-sampling HPV testing, while the PPV was 5.6% (95% CI 3.5–8.3%), NPV was 99.9% (95% CI 99.7-100%) for provider-sampling HPV testing for CIN2 + detection. The AUC was 0.90 for both self- and provider-sampling HPV testing for CIN2 + detection and showed no statistical differences. Data are shown in Table 2.

The CIN3 + detection rate was 0.43% (n = 13) and 0.40% (n = 12) for self- and provider-sampling HPV DNA testing respectively. The sensitivity was 100% (95% CI 99.8–100.0%) for self-sampling HPV testing and 92.3% (95% CI 90.1–94.5%) for provider-sampling HPV testing respectively for detecting CIN3 + . The specificity was 83.5% (95% CI 82.6–84.3%) and 86.4% (95% CI 85.3–87.4%) for self- and provider-sampling HPV testing respectively for CIN3 + detection. The AUC was 0.92 for self-sampling HPV testing and 0.89 for provider-sampling HPV testing for CIN3 + detection and have no statistical differences. Data are shown in Table 2.

Accuracy of HPV DNA testing, E6/E7 mRNA testing and AI-assisted cytology

The CIN2 + detection rate was 0.83% (n = 25) for HPV DNA testing, 0.63% (n = 19) for E6/E7 mRNA testing, and 0.47% (n = 14) for AI-assisted cytology respectively. The sensitivity, specificity, PPV, and NPV was 96.2% (95% CI 90.8%–98.9%), 83.9% (95% CI 82.5–84.0%), 5.0% (95% CI 3.8–6.1%), and 100.0% (95% CI 99.8–100%) respectively for HPV DNA testing for CIN2 + detection. While for E6/E7 mRNA testing, the sensitivity was 73.10% (95% CI 60.3–95.3%), specificity was 91.4% (95% CI 71.1–97.3%), PPV was 6.9% (95% CI 4.7–9.7%), and NPV was 99.74% (95% CI 99.6–99.9%) in detecting CIN2 + . For AI-assisted cytology, the sensitivity, specificity, PPV, and NPV was 53.9% (95% CI 32.7–67.9%), 94.2% (95% CI 83.6–96.7%), 7.4% (95% CI 5.3–10.7%), and 99.6% (95% CI 99.2–99.5%) respectively for CIN2 + detection. The AUC was 0.90 for HPV DNA testing, 0.82 for E6/E7 mRNA testing, and 0.74 for AI-assisted cytology respectively in detecting CIN2 + . Data are shown in Table 2.

The CIN3 + detection rate was 0.43% (n = 13) for HPV DNA testing, 0.30% (n = 9) for E6/E7 mRNA testing, and 0.20% (n = 6) for AI-assisted cytology respectively. For detecting CIN3 + , the sensitivity was 100% (95% CI 99.8–100.0%) for HPV DNA testing, 69.2% (95% CI 57.3–78.3%) for E6/E7 mRNA testing, and 46.2% (95% CI 33.5–58.7%) for AI-assisted cytology respectively. The specificity was 83.5% (95% CI 82.6–84.3%) for HPV DNA testing, 91.1% (95% CI 77.8–95.4%) for E6/E7 mRNA testing, and 93.9% (95% CI 87.6–95.4%) for AI-assisted cytology respectively for CIN3 + detection. The corresponding AUC was 0.92 for HPV DNA testing, 0.80 for E6/E7 mRNA testing, and 0.70 for AI-assisted cytology respectively. Data are shown in Table 2.

Acceptability of self-sampling screening model

Among the screened women, 37.0% (n = 1110) preferred the self-sampling model, while 63.0% (n = 1890) initially chose the provider-sampling model. Acceptance varied by age: 53.2% of women under 40 years, 28.3% of women aged 40–59 years, and 15.4% of women 60 years and older preferred self-sampling (χ2 = 215.9, p < 0.001). By ethnicity, 42.7% of Han ethnic women, 40.6% of Dai minority women, and 27.0% of women from other minorities showed a preference for self-sampling (χ2 = 53.2, p < 0.001). Among women with a senior high school education or higher, 81.3% tended toward self-sampling, compared to only 34.8% of women with a junior high school education or less (χ2 = 96.7, P < 0.001). These data are presented in Table 3.

Table 3 Acceptability of the self- and provider-sampling model among screened women.
Full size table

For all screened women, 33.1% (n = 993) demonstrated good KAP towards cervical cancer screening (with an answer score ≥ 5 points), while 66.9% (n = 2007) showed inferior KAP (answer score < 5 points). Among those, 45.7% with good KAP and 32.7% with inferior KAP tended to prefer the self-sampling model (χ2 = 48.0, P < 0.01). Additionally, 51.9% (n = 1557) of the women had never undergone cervical cancer screening before, of whom only 27.2% preferred self-sampling. In contrast, 48.1% (n = 1443) of women who had previously been screened were more inclined towards self-sampling (χ2 = 134.3, P < 0.001). Data are presented in Table 3.

Age, ethnicity, education level, marriage status, KAP for cervical cancer screening, and history of cervical cancer screening were firstly included into univariate logistic regression, and into multivariate logistic regression analyses adjusted for the potential confounders. The results showed that several factors negatively influenced the selection of the self-sampling model: older age (≥ 40 years old vs. < 40 years old, aOR = 2.75), belonging to a rare minority ethnic group (rare minority ethnics vs. Han and Dai Ethnic, aOR = 2.55), lower education level (junior high school and below vs senior high school and above, aOR = 2.13), and inferior KAP (KAP score < 5 points vs. ≥ 5 points, aOR = 2.87). Data are presented in Table 4.

Table 4 Influencing factors for self-sampling acceptability among screened women.
Full size table

Experience of self-sampling screening model

Totally 1110 women prone to self-sampling model. Among them, 90.3% thinking it protects self-privacy, 88.8% considering it was convenient that they needn’t go to hospital, and 88.2% thinking this model reducing the pain and feeling more comfortable during sampling. For all women who preferred self-sampling model, 69.1% of the them had no difficulty for sample collection, 83.2% of them feeling no pain during the sampling process. 49.5% of them hope the sampling sites in nearest village clinics, 25.6% of them wish the sampling sites in near hospitals, and 19.1% of them hope self-sampling at home. However, 73.2% of women who prone to self-sampling model thinking the screening result unreliable. While for the 1,890 women choosing provider-sampling model, 91.2% of them thinking the results will be more accurate than self-sampling, and 76.7% of them believing more disease could be identified during the pelvic examination. Data are shown in Supplementary Table 1 and Supplementary Table 2.

Discussion

Self-sampling HPV DNA “out-lab” testing demonstrated excellent sensitivity and good specificity for detecting cervical precancerous and cancerous lesions in the rural minority areas of Yunnan, China. The clinical accuracy of this method was comparable to that of conventional provider-sampling for cervical cancer screening. For primary screening, the HPV DNA test showed a higher CIN2 + detection rate and higher sensitivity, but slightly lower specificity, compared to E6/E7 mRNA testing and AI-assisted cytology. Despite its efficacy, only 37% of the screened women preferred the self-sampling model, and 73% of them questioned the reliability of the screening results. Women of older age, belonging to rare minority groups, with lower education levels, or having inferior knowledge, attitudes, and practices regarding cervical cancer screening were less likely to choose self-sampling.

WHO screening guidelines highly recommend HPV DNA tests for cervical cancer screening through either self- or provider-sampling6. These guidelines also highlight that self-sampling HPV testing is crucial for achieving the target of 70% screening coverage by 2030, a goal set for the elimination of cervical cancer6. Self-sampling is particularly beneficial in increasing coverage among underscreened populations, especially in low-resource settings where medical infrastructure and specialized personnel are scarce. Studies have shown that self-sampling HPV testing is both accurate and feasible in low- and middle-income countries12 as well as in some regions of China13. Furthermore, PCR-based self-sampling HPV tests have demonstrated superior clinical accuracy compared to signal amplification-based HPV tests9. In this study, the self-sampling HPV test showed comparable accuracy to conventional provider-sampling, achieving higher sensitivity (96.2% vs. 92.3%) but slightly lower specificity (83.9% vs. 86.4%) for detecting CIN2 + . These figures exceed those reported in the meta-analysis conducted by Tatara et al., where sensitivity and specificity ranged from 74 to 86% and 80% respectively14.

HPV-based cervical cancer screening using a point-of-care (POC) model is recommended6. This innovative approach enables HPV testing to be conducted without a specialized laboratory, and it can be operated by non-specialists following simple training. The POC model operates at or near the patient’s treatment location, providing rapid results to support timely clinical decision-making. Self-sampling HPV POC testing has been proven to be effective, acceptable, and safe when scaled up in primary healthcare facilities in low-resource settings15. In rural Yunnan, China, self-sampling “out-lab” HPV testing offers a promising strategy to overcome multiple barriers to cervical cancer screening and increase participation among women who have not been screened previously. Although the testing method in this study did not fully meet the criteria of a traditional point-of-care test in real-world settings, we employed an “out-lab equipment” approach for PCR-based HPV testing. This method is characterized by non-PCR lab-based operations, non-expert operation, and faster results. It offers an affordable and high-quality screening alternative for rural areas, enabling quicker access compared to traditional cytology screening. This approach can be considered a form of “point-of-care” testing tailored for hard-to-reach populations. Implementing this approach will be crucial for achieving the WHO’s target of 70% cervical cancer screening coverage by 2030 in China.

However, in this study, only 37% of women preferred the self-sampling model. This acceptance rate was lower than those observed in more developed areas of China16 and in some high-income countries17. Factors such as older age, belonging to rare minority groups, lower education levels, and inferior KAP (Knowledge, Attitude, and Practice) regarding cervical cancer screening contributed to the lower preference for self-sampling. The most commonly cited reason for not preferring self-sampling was a lack of self-confidence in collecting a reliable sample. Among those who preferred self-sampling, 73.2% still expressed uncertainty about the accuracy of the results. In high-income countries, home-based self-sampling is often facilitated by efficient postal systems that enable quick and reliable sample processing17. In contrast, most women in this study preferred to collect samples at nearby village clinics rather than at home, maybe due to the generally lower educational level and elder age of this recruited population, they usually more rely on local health care providers’ instructions on sampling. Organizing health education sessions by community health workers, followed by self-sampling in village clinics, would likely be more feasible in rural Yunnan. Therefore, targeted health education is especially needed for older women, rare minorities, and those with lower education levels. Additionally, it is highly recommended to provide health education in local minority languages in rural Yunnan. During this study, some members of the rare ethnic minorities struggled to fully understand Mandarin-based health education. Offering education in their native languages would significantly enhance comprehension and screening engagement.

HPV DNA testing is increasingly adopted globally, but in China, cytology has remained the primary screening method for cervical cancer since 200918. Due to the generally lower quality of samples and the shortage of qualified cytologists, screening accuracy is often unsatisfactory, particularly in low-resource rural areas18. In this study, the AI-assisted cytology (ASC-US +) showed inferior sensitivity (53.9% vs. 92.3%) and superior specificity (94.2% vs. 86.4%) compared to HPV DNA in detecting CIN2 + . However, the sensitivity was lower than that of other AI-assisted cytology interpretation19. While AI-assisted cytology could reduce the workload for cytologists, its sub-optimal sensitivity suggests it may not be suitable for primary screening unless HPV testing is unavailable in Yunnan. Additionally, the E6/E7 mRNA test showed higher specificity (91.4% vs 86.4%) but lower sensitivity (73.1% vs 92.3%) compared to the HPV DNA test in this study. Contrarily, mRNA testing has been reported to offer similar sensitivity for CIN2 + and CIN3 + and slightly higher specificity than DNA tests20. Given its relatively lower sensitivity and higher specificity, HPV E6/E7 mRNA testing is not recommended for primary cervical screening but could be considered as a triage method for HPV DNA-positive women if it is affordable.

To the best of our knowledge, this study is the first to evaluate the clinical accuracy and acceptability of self-sampling HPV DNA “out-lab” testing in a real-world setting in rural Yunnan, a frontier minority area in China with limited health resources. Our goal was to determine if this novel screening approach could be the optimal strategy for cervical cancer screening in Yunnan, particularly combined with applicable health education to expand HPV-based screening coverage, especially for under-screened women. The HPV DNA testing with E6/E7 mRNA testing and AI-assisted cytology were compared to provide evidence for specific cervical screening guidance. Our study demonstrates the significant potential of novel self-sampling HPV testing as a faster and more cost-effective alternative for future cervical cancer screening algorithm. This approach is particularly well-suited for females in rural China and similar low-resource settings where traditional screening models may be limited by accessibility and resource constraints. However, the study has limitations. Firstly, 10.5% of women with positive primary screening results were lost to follow-up and were defined as normal, which could introduce information bias to the study results, such as the CIN2 + prevalence might be slightly underestimated. Secondly, both the self- and provider-sampling HPV DNA testing were not been validated in a conventional PCR laboratory in this study, and the validation testing is recommended in the near future using specimen residuals.

Conclulsions

Self-sampling “out-lab” HPV testing could be the most feasible and optimal primary cervical cancer screening method in rural Yunnan, China, as well as in similar low-resource areas where PCR laboratory-based HPV testing is not affordable. There is an urgent need for targeted health education, particularly for elder women, those with lower education levels, and women from rare minority groups, in Yunnan, to enhance the popularization and acceptance of the self-sampling cervical cancer screening model. A large-scale prospective study is recommended to further validate the clinical accuracy and cost-effectiveness of one-time self-sampling HPV-based screening, with or without triage strategies, for the future.