Purpose: We aim to identify risk factors associated with pathological upgrading/downgrading after conization in patients with cervical high-grade squamous intraepithelial lesion and provide risk stratification management based on machine learning predictive model. Methods: 1. This retrospective study included patients who visited Obstetrics and Gynecology Hospital of Fudan University from January 1 to December 31, 2019, and were diagnosed with cervical high-grade squamous intraepithelial lesion (HSIL) by colposcopy directed biopsy (CDB) and subsequently underwent conization. A wide variety of data were collected from medical records, including demographic data, laboratory findings, colposcopy descriptions, and pathological results. Patients were categorized into three groups according to their post-conization pathological results: LSIL or below (downgrading group), HSIL (HSIL group), and cervical cancer (upgrading group). 2. Univariate and multivariate analysis were conducted to obtain the independent risk factors for pathological changes in cervical HSIL patients. 3. Machine learning prediction models were established and evaluated, subsequently verified in external testing data. Results: 1. A total of 1585 patients were included, with 65 cases (4.1%) being upgraded to cervical cancer after conization, 1147 cases (72.4%) remaining HSIL, and 373 cases (23.5%) being downgraded to LSIL or below. 2. Multivariate analysis results showed that a 2% decrease in the incidence of pathologic downgrade was found for each additional year of age and each 1% increase in lesion area. Patients with cytology >LSIL (OR=0.33, 95%CI: 0.21-0.52), HPV infection (OR=0.33, 95%CI: 0.14-0.81), HPV 33 infection (OR=0.37, 95%CI: 0.18-0.78), coarse punctate vessels on colposcopy examination (OR=0.14, 95%CI: 0.06-0.32), HSIL lesions in endocervical canal(OR=0.48, 95%CI: 0.30-0.76) , and HSIL impression (OR=0.02, 95%CI: 0.01-0.03) were less likely to experience pathologic downgrading after conization. 3. The independent risk factors of pathological upgrading to cervical cancer after conization including: age (OR=1.08, 95%CI: 1.04-1.12), HPV16 infection (OR=4.07, 95%CI: 1.70-9.78), the presence of coarse punctate vessels during colposcopy examination (OR=2.21, 95%CI: 1.08-4.50), atypical vessels (OR=6.87, 95%CI: 2.81-16.83), and HSIL lesions in endocervical canal (OR=2.91, 95%CI: 1.46-5.77). 4. In 6 machine learning prediction models, the BP neural network model demonstrated highest and most uniform predictive performance in the downgrading group, HSIL group, and upgrading group, with AUCs of 0.90, 0.84, and 0.69, respectively, sensitivities of 0.74, 0.84 and 0.42, specificities of 0.90, 0.71 and 0.95, accuracies of 0.74, 0.84 and 0.95. In the external testing set, the BP neural network model showed a higher predictive performance than the logistic regression model, with the overall AUC of 0.91. Thus, a web-based prediction tool (http://115.29.79.17:8082/) was developed. Conclusion：BP neural network prediction model has excellent predictive performance and can be used for risk stratification of CDB diagnosed HSIL patients.

Persistent high-risk human papilloma virus (HR-HPV) infection is the main risk factor for cervical cancer, threatening women’s health. Despite growing prophylactic vaccination, annual cervical cancer cases are still increasing and show a trend of younger onset age. However, therapeutic approaches towards HPV infection are still limited. 25-hydrocholesterol (25HC) has a wide-spectrum inhibitory effect on a variety of viruses. To explore efficient interventions to restrict HPV infection at an early time, we applied different pseudoviruses (PsV) to evaluate anti-HPV efficacy of 25HC. We tested PsV inhibition by 25HC in cervical epithelial-derived HeLa and C-33A cells, using high-risk (HPV16, HPV18, HPV59), possibly carcinogenic (HPV73), and low-risk (HPV6) HPV PsVs. Then we established murine genital HPV PsV infection models and applied IVIS to evaluate anti-HPV efficacy of 25HC in vivo. Next, with the help of confocal imaging, we targeted 25HC activity at filopodia upon HPV exposure. After that, we used RNA-seq and Western blotting to investigate 1) how 25HC disturbs actin cytoskeleton remodeling during HPV infection and 2) how prenylation regulates the cytoskeletal remodeling signaling pathway. Our findings suggest that 25HC perturbs F-actin rearrangement by reducing small GTPase prenylation. In this way, the phenomenon of HPV virion surfing was restricted, leading to failed infection.