Prediction of Occult Invasive Cervical Cancer After Conization for High-Grade Squamous Intraepithelial Lesions: Model Development, Temporal Validation, and a Web-Based Tool
Objective This study aimed to develop and temporally validate a prediction model for occult invasive carcinoma in patients with cervical high-grade squamous intraepithelial lesion (HSIL) undergoing conization and to develop a web-based prediction tool for individualized risk assessment. Methods We retrospectively included patients with cervical HSIL who underwent conization and were subsequently classified as having invasive carcinoma or non-upgraded disease according to the final histopathological findings. Least absolute shrinkage and selection operator (LASSO) regression were used for candidate predictor selection. Multivariable logistic regression and several machine-learning algorithms were developed and evaluated using internal cross-validation, followed by temporal validation in an independent cohort.The selected model was subsequently implemented as an R Shiny–based web prediction tool. Results Invasive carcinoma was detected after conization in approximately 6% of patients with cervical HSIL. LASSO selection identified age, high-grade cytology on TCT, HPV-16/18 positivity, involvement of 3–4 cervical quadrants, CIN2 or CIN3 on endocervical curettage (ECC), and CIN3 on biopsy as candidate predictors. In multivariable logistic regression, high-grade cytology, HPV-16/18 positivity, involvement of 3–4 quadrants, and CIN2 or CIN3 on ECC remained independently associated with histopathological upgrade. Among the evaluated algorithms, AdaBoost achieved the highest AUC in internal cross-validation and it was implemented as a web-based tool that provides individualized predicted probabilities and threshold-based risk classification using six routinely available preoperative variables. Conclusion Approximately 6% of patients with cervical HSIL in this cohort had occult invasive carcinoma identified after conization. Prediction models incorporating clinical and pathological variables showed good discriminative performance in the derivation cohort and retained satisfactory performance in temporal validation. A web-based prediction tool was subsequently developed to facilitate individualized risk assessment and may support clinical risk stratification for patients with cervical HSIL before conization.
Authors
- Jiangping Chen (ORCID: https://orcid.org/0000-0002-7016-4583)
- Jianmiao Chang
- Xuntao Liu
- Liu Xiaojuan (ORCID: https://orcid.org/0009-0003-0014-8451)
- Chunxing Ma
- Wang Jing
Institutions
- Hebei North University (CN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-18
- DOI
- https://doi.org/10.21203/rs.3.rs-11374123/v1
- Primary Topic
- Cervical Cancer and HPV Research
- Type
- preprint