Impacts of ovarian preservation on the prognosis of neuroendocrine cervical carcinoma: a retrospective analysis based on machine learning

Abstract Background Neuroendocrine cervical carcinoma (NECC) is a rare but aggressive malignancy with younger patients compared to other common histology types. This study aimed to evaluate the impacts of ovarian preservation (OP) on the prognosis of NECC through machine learning. Methods Between 2013 and 2021, 116 NECC patients with a median age of 46 years received OP or bilateral salpingo-oophorectomy (BSO) and were enrolled in a retrospective analysis with a median follow-up of 41 months. The prognosis was estimated using Kaplan–Meier analysis. Random forest, LASSO, stepwise, and optimum subset prognostic models were constructed in training cohort (randomly selected 70 patients) and tested in 46 patients through receiver operator curves. Risk factors for ovarian metastasis were identified through univariate and multivariate regression analyses. All data processing was carried out in R 4.2.0 software. Results Among 116 patients, 30 (25.9%) received OP and showed no significantly different OS compared with BSO group (p = 0.072) and got better DFS (p = 0.038). After construction of machine learning models, the safety of OP was validated in lower prognostic risk group (p > 0.05). In patients ≤ 46 years, no impacts of OP were shown for DFS (p = 0.58) or OS (p = 0.67), and OP had no impact on DFS in different relapse risk population (p > 0.05). In BSO group, regression analyses showed that later stage, para-aortic LNM, and parametrial involvement were associated with ovarian metastasis (p