The purpose of this study is to validate Automated Visual Evaluation (AVE), specifically the CINFinder version developed by DL Analytics, a point-of-care screening and triage diagnostic tool for cervical cancer based on the assessment of digital images through artificial intelligence. Several teams around the world have developed versions of AVE as a triage technology but none as a screening tool.

BACKGROUND:

Cervical cancer, caused by persistent infection of carcinogenic types of the human papillomavirus (HPV), is the second leading cause of female cancer in El Salvador. Effective screening and treatment of precancerous lesions have lowered cervical cancer rates in high-income countries, but the disease remains a leading cause of death in low-and middle-income countries (LMICs). Currently, up to 80% of the disease burden and 90% of global deaths occur in LMICs primarily due to lack of resources and poor infrastructure. A new test, Automated Visual Evaluation (AVE), is a point-of-care screening and triage diagnostic tool based on the assessment of digital images through artificial intelligence.

The purpose of this study is to validate Automated Visual Evaluation (AVE), specifically the CINFinder version developed by DL Analytics, a point-of-care screening and triage diagnostic tool based on the assessment of digital images through artificial intelligence. AVE has been in development since 2021. The current study will consist of a clinical trial to compare the sensitivity of AVE (CINFinder version) with traditional screening and triage tests.

STUDY DESIGN:

This is a prospective paired interventional study of 10,000 women in San Salvador, El Salvador, to test the difference in sensitivities between AVE (CINFinder) and with other screening and triage tests, including HPV tests and unaided visual inspection with acetic acid (VIA) to detect CIN2+ as a primary screening method. As a secondary aim, investigators will also validate the use of AVE as a triage test in patients with positive HPV results. Histopathology diagnosis will be used as the reference to determine true disease status.

STUDY PROCEDURES:

The design includes a screening visit, a colposcopy visit, and a result delivery and treatment visit. During the screening visit, participants will undergo routine HPV sample collection, VIA, AVE with the EVA System (digital colposcope), and an additional cervical image capture with an Android smartphone. Women with a positive screening result on any of the three main screening tests (HPV test, VIA, or AVE with the EVA System) will be referred to a second study visit to undergo HPV testing with a genotyping test, AVE with the EVA System for the second time, and colposcopy with biopsy. In addition, 5% of screen-negative women will undergo the same procedures. Results from VIA and AVE with the EVA System during the first and the second visit will be compared to determine AVE performance as a screening and triage test, respectively. Histopathology findings will serve as the reference to establish the true diagnosis of each case. Women will receive their histopathology results during the third visit and, if eligible, will be offered ablation treatment the same day. Women ineligible for ablation will be referred to LEEP or other care as appropriate.

DATA COLLECTION & MANAGEMENT:

This specific version of AVE (CINFinder) is designed to run on both EVA System and an Android phone, alongside additional tools for improving image quality and post-screening management.The AVE software (including CINFinder, CerVisibility, and CervManager modules), will be installed as user-friendly features that can be easily accessed by authorized

Data collection: All data will be recorded in real time on paper forms OR on an electronic tablet filled out by a nurse research assistant. All forms (digital or paper) will include the date and study ID to link information to specific patients and visits. Collected data will include eligibility criteria, a background questionnaire including relevant sociodemographic and medical history traits, clinical procedures, and laboratory results. In addition, providers conducting speculum exams during the first (screening) and second (colposcopy and biopsy) visits will use the CervManager application to create a profile for each patient to input clinical information. Thus, study data will be collected on both paper forms and CervManager for added back-up security. Providers will also capture cervical images using the CervManager application on both the EVA System and an Android mobile phone. Images belonging to the same patient will be synchronized under the same study ID. Other clinical data (e.g., HPV test results, histopathology) will also be inputted into the CervManager patient profile.

Data Management: For storage and management, data from paper forms will be transferred to REDCap, an electronic data management system widely used for clinical research. CervManager and REDCap can be downloaded and merged periodically to easily identify and correct any entry errors, duplications, or missing data on either database. All AVE algorithms implemented as part of this project and used for subsequent analyses will run on DL Analytics servers.

Non-participant documentation: Investigators will document the number of eligible women who decline to participate in the study (including main reasons why). If individuals agree, Investigators will collect general, de-identified information (i.e., age range, education, and previous screening). This will be done to identify any potential source of sampling bias.

STATISTICAL POWER CALCULATIONS AND SAMPLE SIZES

Aim 1: Validate AVE as a primary screening test compared to visual inspection with acetic acid (VIA) for the detection of high-grade cervical pre- cancer (CIN2+).

A study with 10,000 women will be able to detect a minimum difference of 10% between VIA and AVE with 80% power and a Type I error rate of 5%, assuming a minimum sensitivity of VIA of 65% and a correlation coefficient of the tests in the diseased population (Rho) of at least 50%. Based on the study design, all participants who test positive on any test will be referred to colposcopy with biopsy (i.e., proportion of verification = 1), as opposed to only 5% of those who are double negative being referred (i.e., proportion of verification = 0.05). Finally, based on the Investigators previous experience working in El Salvador, Investigators estimate the prevalence (lambda) of CIN2+ in the general screening population at 2%.

Objective 2. To compare AVE as a triage test compared to VIA for detecting CIN2+ among HPV-positive women.

Since all women who test positive with careHPV, VIA or AVE will undergo colposcopy with biopsy, the positive predictive value (PPV) for CIN2+ will be estimated directly using the women who test positive with a particular screening test. An unbiased estimate of NPV will be obtained by using only results for the 5% of women who are negative for the three screening tests who are randomly selected to receive colposcopy and a minimum of two biopsies. Methods that account for verification-biased sampling will be used to obtain unbiased estimates of sensitivity and specificity for CIN2+ detection.25 A method developed by Leisenring et.al. 26 will be used to test for significant differences in PPV and NPV.

Inclusion Criteria:

• Women between 30 and 59 years of age

Exclusion Criteria:

• Pregnancy at the time of colposcopy/biopsy
• Hysterectomy with surgically absent cervix
• HPV test in the last 5 years independently of negative or positive result
• Previous cervical cancer diagnosis or treatment in the last 5 years
• Lack of willingness or capacity to provide informed consent

Cisgender women