Description

In the past decade, several portable thermal ablation (TA) devices have been developed to treat cervical precancer in low-resource settings. Current data shows that portable TA is safe and effective, but questions remain regarding the optimal treatment protocol. The purpose of this trial is to validate a portable TA device equipped with an endocervical probe that could improve the effectiveness of treatment, and to evaluate a Automated Visual Evaluation (AVE), a novel, AI-based imaging triage strategy designed to more accurately identify women at highest risk. This will be done using IRIS™ (Liger Medical, Lehi, UT), an FDA-approved (K202915) device that combines a digital colposcope and thermal ablator. Study aims will be 1) To clinically validate the modified IRISTM probes by estimating the cure rates of CIN2+ one year after treatment, and 2) To perform a retrospective validation of the AVE algorithm on IRISTM images using histopathologic disease status as ground truth.

The first aim will be a prospective, single-arm, interventional cohort study. The second aim will be a retrospective diagnostic performance and decision modeling study evaluating how AVE-based triage might have altered treatment decisions, using histopathologic diagnosis as the reference standard. The study will enroll 5,000 women who have tested HPV+ in the El Salvador national program but have not yet received treatment. Inclusion criteria will be women, aged 30-49, non-pregnant (determined by a urine pregnancy test), HPV+ positive per MOH records, and willingness to have colposcopy and biopsies. Exclusion criteria will be plans to become pregnant during the study, history of a LEEP or ablation procedure in the past 5 years, history of total hysterectomy (verified by medical record or pelvic evaluation), history of cervical cancer, and inability or unwillingness to provide a permanent and reliable address, and/or informed consent. Eligibility criteria will be same for both aims, except that women who are ineligible for ablation treatment will be eligible for Objective 2 if the SCJ is visible and other eligibility criteria are met.

An anticipated 6,250 HPV+ women will be invited to the study, and about 80% (n=5,000) are expected to agree to participate over the course of 30 months (167 per month). During the first visit, consented participants will undergo a provider-collected HPV genotyping test, visual assessment for treatment (VAT), capture of digital images with the IRISTM device for AVE, colposcopy, biopsy, and endocervical curettage (ECC). Women eligible for ablation during VAT (approximately 85%, or 4,250 participants) will receive thermal ablation using a two-probe protocol with IRIS™. Pain during treatment will be evaluated using the Baker-Wong scale. Women deemed ineligible will be referred for standard of care. Based on prior data, an estimated 10% of treated women (n = 425) will have biopsy-confirmed CIN2+ at baseline. These participants will be asked to return in one year for follow-up which will include HPV genotyping, colposcopy, biopsy, and endocervical curettage (ECC) to assess for persistent disease. Accounting for an expected 15% loss to follow-up, endpoint data will be available for approximately 362 women under Aim 1. All captured cervical images-including those from women ineligible for TA-may be included in the AVE analysis for Aim 2. Additionally, a subsample of 200 women will return at 6 weeks post-treatment to monitor for potential side effects. A separate subsample of 50 women diagnosed with CIN2+ at baseline will return at 4-6 months for colposcopy, biopsy, and ECC, to further evaluate the safety of the two-probe treatment protocol.

All biopsy and ECC samples from the initial, one-year, and safety evaluation visits will be sent to the same private laboratory in San Salvador for analysis. All cervical biopsy specimens will be reviewed by an expert in gynecologic pathology after local analysis is complete. If findings of the two pathologists agree, the diagnosis will be accepted as truth. If there is a discrepancy, a third expert in gynecologic pathology will review the specimen. Significant discrepancy occurs when pathologists disagree by more than one level: e.g., the diagnosis from pathologist 1 is normal and the diagnosis from pathologist 2 is CIN2+. Consensus will be reached when two of the three pathologists agree on a diagnosis. For cases in which there is no consensus, all three pathologists will hold an adjudication meeting and arrive at a consensus diagnosis (i.e., agreement of 2/3 pathologists). Results from the QA process will be considered ground truth in final analysis.

For Aim 1, cure rates of CIN2+ for IRISTM modified 2-probe treatment will be estimated after one year. Of the 425 women with CIN2+ treated with the IRISTM modified 2-probe approach, it is anticipated that 85% will return for the one-year follow-up visit (n=362) to have HPV testing, colposcopy/biopsy and ECC. The 1-year cure rate, defined as histopathological diagnosis of \<CIN2, one year after treatment with the IRISTM modified 2-probe approach will be estimated with a precision (half-width of 95% confidence interval) of 3.7% assuming a cure rate of85%. Women diagnosed with recurrent/untreated CIN2+ and those with suspected cancer on the 1-year biopsy will be referred to the local cancer hospital for standard of care treatment. For Aim 2, the Area Under the Curve - Receiver Operator Characteristic (AUC-ROC) displays a model's performance across various classification thresholds, plotting the true positive rate (sensitivity) against the false positive rate. This curve gives a complete picture of performance, not just for one cutoff point, but for all possible thresholds. It illustrates how the AVE algorithm changes in accuracy depending on where the cutoff point is - or what is considered "positive." With a sample size of 4,500 women, the AUC can be estimated with high precision. For an AUC in the range of 0.5 to 0.9, the expected half-width of the 95% confidence interval will be approximately 0.01 to 0.03, corresponding to an uncertainty of only 1%-3% around the point estimate. This level of precision will provide strong statistical confidence that the measured AUC is close to the true performance of the AVE algorithm in identifying cervical images most consistent with CIN2+.