Challenging the assumption that experience- and training-related factors improve accuracy of cervical cancer screening image interpretation

Visual inspection of the cervix with acetic acid (VIA), although recognized as inaccurate, remains widely used for cervical screening in low-resource settings. Training and image review by more experienced providers − such as the smartphone-enhanced VIA (SEVIA) in Tanzania − are assumed to improve VIA performance, but remain untested against a reference standard of disease. Between June and December 2023, 486 VIA providers in Tanzania completed a questionnaire on experience- and training-related factors (years practicing VIA, patient volume, refresher training, SEVIA training) and independently reviewed a set of 50 high-quality cervical images (24,300 total reviews). Image reference diagnosis was defined using HPV status and histology: precancer if HPV-positive and histologically-confirmed CIN3, normal if HPV-negative and normal histology, otherwise indeterminate. For comparability, providers’ image reviews were recategorized into positive (VIA-positive for ablation, LEEP, or suspect cancer) or negative (VIA-negative). In an image-level analysis, we calculated proportion of providers labeling each image as positive and averaged those within subgroups of experience/training and reference diagnosis. Overall, providers (mean VIA experience 3.7 years, weekly VIA exams 12.2) showed moderate sensitivity: on average 60.0% (95% confidence interval 59.2 − 60.7) labeled precancers as positive and 54.4% (53.0 − 55.9) labeled normals as negative. More experienced providers tended to be more specific but less sensitive: for example, an average of 56.5% (55.2–57.9) of providers with ≥5 years of VIA experience labeled precancers as positive vs 61.4% (60.5–62.4) of providers with

Cervical cancer · Cervical screening · Cervix · Confidence interval · Medical diagnosis · Tanzania · AI in cancer detection · Cervical Cancer and HPV Research · Global Cancer Incidence and Screening

• Aviva

  Open Access•Olabanjo Okunlola Ogunsola, John O Olawepo et al.•BMJ Global Health•2023

  Añadir item