Introduction/Background The incidence of adenocarcinomas of the cervix (AC) is on the rise, particularly among young women. The safety of ovarian preservation in patients with early-stage AC remains controversial. This study aimed to develop a machine learning-based predictive model to assess the risk of adnexal metastasis and guide personalized surgery strategies for patients with early-stage AC.

Methodology Patients with early-stage AC were from five major hospitals in China between September 2013 and December 2022. Univariate and multivariate logistic regression analyses were used to identify the independent risk factors associated with adnexal metastasis. Multiple machine learning algorithms were applied to construct several prediction models. Subsequently, a 5-fold cross-internal validation was conducted to identify the optimal model based on the efficacy, which was evaluated by the area under the receiver operating characteristic curve (AUC), relative sensitivity, specificity, positive predictive value, negative predictive value and overall accuracy.

Results A total of 987 patients with early-stage AC were included, with 43 (4.4%) patients having pathologically-confirmed adnexal metastasis. Predictive variables including human papillomavirus infection status, depth of myometrial invasion, corpus uteri invasion, preoperative carbohydrate antigen 125 and Ki-67 status were significantly associated with the adnexal metastasis. The AUC for the logistic regression (LR) model was the highest (0.816) among the six machine learning prediction models. The LR prediction model exhibited an accuracy of 74.37%, a sensitivity of 68.33% and a specificity of 74.69%. Further evaluation demonstrated that the model was well-calibrated and exhibited satisfactory clinical utility. A nomogram and web calculator based on the LR model were developed for clinical interpretability.

Conclusion We have developed the first machine learning-based prediction model to determine the risk of adnexal metastasis in patients with early-stage AC. The model can provide individualized risk assessment and tailor the decision-making for the surgical management of adnexa.

Disclosures All authors declare no competing interests.