Implications of Age and Parity on Squamocolumnar Junction (SCJ) Visibility for Implementing Artificial Intelligence-Assisted Cervical Cancer Screening: Multi-Country Evidence From the PAVE Study
Abigail Ukwuani, Olusegun Ajenifuja, Imran Morhason-Bello, Jose Jeronimo, Ana Cecilia Rodriguez, Rebecca Perkins, Brian Befano, Didem Egemen, Federica Inturrisi, Diego Guillen, Nicole Campos, Kanan Desai, Jayashree Kalpathy-Cramer, Nicolas Wentzensen, Jennifer Boyd-Morin, Christopher Clark, Scott Kinder, Ignacio Prieto-Egido, Ana Ribeiro, Karen Yeates, Margaret Madeleine, Elizabeth Torres, Jacqueline Figueroa, Huiwu Chen, Susie Lau, Te Vantha, Clement Adepiti, Bariki Mchom, Tiphaine de Foucher, Juan Carlos Rauda, Yeycy Donastorg, Manuel Sandoval, Sovannara Thay, Zeev Rosberger, Flavia Correa, Taina Raiol, Silvia de Sanjosé, Mark Schiffman, Harrison Ndetan,PURPOSE
Extended HPV-based screening and automated visual evaluation (AVE) may improve cervical cancer screening in resource-limited settings (RLS), but these tools should work well across different groups of women. The HPV–Automated Visual Evaluation (PAVE) approach is promising. Regardless, its performance may be limited by whether the SCJ (a marker of the transformation zone) can be seen on the cervix. We examined how age and parity affect SCJ visibility and how SCJ visibility influences AVE performance, with a focus on implementation challenges in RLS.
METHODS
We analyzed 5,754 cervical images from HPV-positive women in the nine-country PAVE study (Brazil, Cambodia, Dominican Republic, El Salvador, Eswatini, Honduras, Malawi, Nigeria, Tanzania). PAVE scientific experts independently classified blinded participant characteristics images taken after acetic acid application as SCJ “not visible,” “partially visible,” and “fully visible.” We assessed associations between SCJ visibility, patient characteristics (age, parity), and AVE's performance in predicting high-grade cervical lesions (CIN3+).
RESULTS
SCJ visibility declined with age. The proportion of women with a non-visible SCJ increased from 7.2% among women aged 25-29 to 21.3% among women aged 45-49. Full SCJ visibility decreased with age. Parity also mattered. Non-visible SCJ was more common among women who had never given vaginal birth (increased from 11.4% in women who had ≥5 children to 22.4% in nulliparous women); this increased further at older ages. Considered together, the effect became stronger; 42.3% of nulliparous women aged 45-49 had a non-visible SCJ, compared with 15.0% of parous women. AVE performance was better when the SCJ was visible (OR 4.3), associating severe visual classification and case status, and weaker when it was not visible (OR 2.4).
CONCLUSION
SCJ visibility is an important factor for equitable and effective implementation of AVE-based cervical screening. Older and nulliparous women are more likely to have non-visible SCJ, limiting AI-assisted visual triage performance in RLS. HPV positivity without visible lesions in these populations poses a critical management challenge. To achieve equitable global cervical cancer control as AI-enabled cervical screening expands globally, settings lacking robust surveillance should develop tailored management strategies for these specific subgroups.