IARC 60th Anniversary - 19-21 May 2026
Session : Integrating AI and data science across the cancer screening continuum
Novel screening and triage approach to detect cervical precancer with HPV extended genotyping and AI-automated visual evaluation (PAVE)
MREMI A. 16, ALFARO K. 12, NORMAN J. 8, DONASTORG Y. 10, BOYD-MORIN J. 5, CLARK C. 20, KINDER S. 20, WENTZENSEN N. 4, KALPATHY-CRAMER J. 20, SCHIFFMAN M. 4, DE SANJOSE S. 1, INTURRISI F. 2, PERKINS R. 3, EGEMEN D. 4, RODRIGUEZ A. 2, BEFANO B. 5, GUILLEN D. 2, JEROMINO J. 4, CAMPOS N. 19, RIBEIRO A. 13, AJENIFUJA K. 14, YATES K. 15, CREMER M. 12, GOLDSTEIN A. 7, CHEN H. 18, FIGUEROA J. 11, MADELEINE M. 9, NORRIS T. 17, RAIOL T. 13, ADEPITI C. 14
1 ISGlobal, Barcelona, Spain; 2 Doctors withouth Borders Foundation, Paris, France; 3 Tufts University School of Medicine , Medford, MA, United States; 4 Clinical Epidemiology Unit, Clinical Genetics Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, , Rockville, MD, United States; 5 Information Management Services Inc,, Calverton, MD, United States; 6 University of Colorado Anschutz Medical Campus,, Aurora, CO, United States; 7 Gynecologic Cancers Research Foundation,, Arnold, AL, United States; 8 Mercy Medical Center,, Phnom Penh, Cambodia; 9 Fred Hutch Cancer Center,, Seattle, WA, United States; 10 Instituto Dermatológico y Cirugia de la Piel Dr. Huberto Bogaert Diaz,, Santo Domingo, Dominican Republic; 11 Secretaria de Salud , Tegucigalpa, Honduras; 12 Basic Health International,, Pittsburgh, PA , United States; 13 Oswaldo Cruz Foundation (Fiocruz), , Brasília, Federal District, B, Brazil; 14 Obafemi Awolowo University, , Ile-Ife, Nigeria; 15 Queen's University, , Kingston, Ontario, Canada; 16 KCMC University , Moshi, Kilimanjaro , Tanzania; 17 HPV Global Action , Montreal, QC, Canada; 18 Epicenter, MSF , Paris, France; 19 Center for Health Decision Science, Harvard T.H. Chan School of Public Health , Boston, MA,, United States; 20 University of Colorado Anschutz Medical Campus, Aurora, CO, United States
Limited access to screening demands accurate, scalable strategies to impact the target population effectively. The HPV-Automated Visual Evaluation (PAVE) consortium validated a cervical screening approach to detect CIN3 or worse (CIN3+) in resource-limited settings with varying HIV prevalence.
Methods
Participants aged 25–49 were enrolled before April 2025 from clinical sites in Brazil, the Dominican Republic, El Salvador, Honduras, Cambodia, Malawi, Nigeria, Tanzania, and Eswatini (N= 50,450). Self-collected vaginal samples (FLOQSwab, Copan) were tested using a low-cost HPV extended genotyping assay (ScreenFire, Atila Biosystems), which provides results in four risk-based channels: HPV16, HPV18/45, HPV31/33/35/52/58, and HPV39/51/56/59/68. HPV channels were analyzed hierarchically by carcinogenicity and evaluated by HIV status. All HPV-positive participants were referred for cervical imaging and biopsy (ecto and/or endo). Cervical images were captured using a low-cost mobile colposcope (IRIS, Liger Medical). An earlier deep learning algorithm was trained and validated to classify images into three categories of Automated Visual Evaluation (AVE): Normal, Indeterminate, and Severe. The algorithm is tested on IRIS images for external validation. Among HPV-positive women, we assessed the performance of HPV extended genotyping combined with AVE, to estimate absolute risk for confirmed CIN3+.
Results
The HPV prevalence was 27% among HIV negative/unknown status and 36.1% among women living with HIV (WLHIV) (range: 10.7% to 50.6%) . HPV-negative women (N= 34,042) were reassured. Among HPV-positive participants with complete data (N= 7,535), 6,113 were HIV-negative/unknown and 1,422 WLHIV. There were 359 (5.9%) CIN3+ cases among HIV negative/unknown and 174 (12.2%) CIN3+ cases among WLHIV.
Among HPV positives, visual triage with AVE provided the strongest risk stratification when restricting the analysis to CIN3+ cases with visible lesions irrespective of HIV status.
Across diverse settings, the PAVE strategy provided remarkable CIN3+ risk stratification. HPV16 was associated with the highest risk of CIN3+ and image severity added risk refinement to all HPV channels irrespective of HIV status. WLHIV have higher CIN3+ burden.
The combination of HPV extended genotyping and AVE enables targeted management of infections and lesions among HPV-positive women, regardless of HIV status. The project is now in the pre-implementation phase, evaluating the clinical use of HPV extended genotyping assisted by AVE in the triage phase. Additional work is required for the triage and management of women with no visible squamous columnar junction.