Background
Neuroendocrine cervical carcinoma (NECC) is a rare entity of gynecologic malignancy that accounts for only 1 to 3% of all cervical cancer, which are classified as small-cell type, large-cell type, and non-neuroendocrine admixed with neuroendocrine carcinoma (NEC) type [
1‐
4]. Nevertheless, NECC is exceedingly aggressive, and cases of demise have been reported even in early stage due to locoregional recurrence or distant metastasis [
5‐
7]. Furthermore, it was reported that NECC affected a wide age range (21–87 years) with a median age at diagnosis from 37 to 49 years old [
1,
8,
9], which tends to be younger when compared with other common histology [
10]. Due to the contradiction between the need for more aggressive treatment and the strong desire for ovarian retention in young women with NECC, a debate has arisen concerning criteria and rationality for ovarian preservation (OP).
Several gynecologic oncologists have suggested that OP should not be recommended in NECC. It was reported only 65% of patients who accepted OP would maintain ovarian function after external beam pelvic radiation and/or brachytherapy [
11,
12], not to mention there are theoretical concerns regarding residual microscopic disease in the ovaries [
13]. However, there is an alternative point of view, though it would be better to preserve one or both ovaries at the time of radical surgery in a specific group of patients for their physiological and psychosexual well-being. While bilateral salpingo-oophorectomy (BSO) eliminated microscopic carcinoma in ovaries, this procedure causes climacteric symptoms due to the abrupt loss of estrogen, including hot flushes, neurasthenia, osteoporosis, and hypertension [
14‐
16]. Besides, increased risks of cardiovascular disease, venous thromboembolism, and breast cancer remain controversial in women undergoing oral hormone therapy [
17‐
19]. It is also difficult to determine the suitable dosage and frequency of the drug compared with the hormone produced by the body itself [
20]. Therefore, OP should be considered for partial patients whose overall health benefit would exceed the risk. However, it remains a challenge in clinical practice to discriminate a group of NECC patients that were relatively feasible for retaining ovaries.
To evaluate the safety of OP in women with NECC, the oncological outcomes were compared between women with OP and BSO in this study. The risk factors were evaluated based on machine learning, and a risk score was constructed to classify patients into subgroups, in which the impact of OP on prognosis was further discussed. Additionally, the risk factors of ovarian metastasis among women who accepted BSO were analyzed in this study.
Materials and methods
Patients and data collection
In approval of the Institutional Ethics Review Board of Obstetrics and Gynecology Hospital, Fudan University, a total of 10,225 patients with uterine cervical carcinoma who underwent surgical treatment from December 2013 to December 2021 were retrospectively reviewed. Of them, patients who met the following criteria were included in this study: (i) the diagnosis included pure NECC or NEC admixed with other carcinomas and (ii) underwent surgical treatment in our hospital. We excluded patients who met any one of the following conditions: (i) incomplete clinical dataset and patients who refused surgery, (ii) diagnosed as typical or atypical carcinoid tumor, (iii) distant metastasis at the first visit and diagnosed as stage 4B based on the new International Federation of Gynecology and Obstetrics (FIGO 2018) staging system, and (iv) combined with other primary cancers of non-cervical origin. All patients or their relatives signed informed consent. Of all 116 patients enrolled, 86 women underwent radical hysterectomy, pelvic lymphadenectomy, and BSO, while the remaining 30 women underwent radical hysterectomy, pelvic lymphadenectomy, and OP. To evaluate the safety of OP in a mimic clinic situation, the 116 women were randomly divided into training, and testing cohort (included 70 and 46 patients respectively) used “caret” R package.
Pathological diagnosis
Based on the criteria of central pathological review (CPR), the diagnoses of all patients with NECC were confirmed by histologic morphology and immunohistochemical staining of tissue samples that were read by two pathologists specialized in gynecological cancers. Specifically, the pathology committee had a consensus on the diagnosis of NECC according to WHO classifications. Small-cell type was composed of hyperchromatic nuclei and scanty cytoplasm; its nuclear molding and crushing artifact were also common. Large-cell type is recognized by its arrangement in well-demarcated nests, trabeculae, or cords with peripheral palisading, and tumor cells are large and polygonal, with vesicular or hyperchromatic nuclei and a prominent nucleolus [
3]. For cases with squamous differentiation or adenocarcinoma differentiation in the tumor, as long as the NEC component accounted for at least 20% of the tumor area, they were all designated as mixed histology types. Which kind of histology subtype NECC admixed with, and whether NECC was dominant in the whole tumor, is two indexes considered in the mixed histology subgroup. Furthermore, at least one of the biomarkers derived from the immunohistochemical staining, including neuron-specific enolase, chromogranin, synaptophysin, and neural cell adhesion molecule CD56, is positive. Nevertheless, positive neuroendocrine markers were not necessary for diagnosis.
Patient, tumor, and treatment variables
The clinicopathological variables of each woman were obtained from medical records, including admission and discharge notes, as well as pathological slides. Tumor size was determined as the maximum diameter of gross tumors from pathology reports. Preoperative diagnosis was based on the pathological results of colposcopy biopsy or loop electrosurgical excision procedure (LEEP). Ovarian metastasis was defined as the occurrence of viable tumor cells in the ovarian tissues or vessels and imitated the particular cell arrangement and morphology of primary cervical neoplasm [
14].
The variables analyzed included age at diagnosis, chief complaint, preoperative diagnosis, preoperative human papillomavirus (HPV) infection status (Roche cobas 4800 HPV system [Roche Molecular Systems Inc., Pleasanton, CA, USA]), tumor size, FIGO stage, histological heterogeneity (pure NECC or mixed histology types), lymph node metastasis (LNM), depth of myometrial invasion (DIM), lymph-vascular space invasion (LVSI), parametrial invasion, vaginal invasion, incisal margin involvement, lower uterine segment involvement (LUSI), and postoperative radiotherapy (chemotherapy was routinely applied to all patients and was not needed to be compared. The role of radiotherapy was controversial, for distant metastasis is much more common than local recurrence, and several studies were against that radiotherapy would promote the prognosis of NECC [
21‐
24]). The para-aortic LNM and the LNM positive ratio were viewed as two indexes in LNM positive population.
Statistical analysis
The primary endpoint was any NECC-related death, and the secondary one was NECC recurrence. Overall survival (OS) was measured from the date of radical hysterectomy to death or censored at the last follow-up. Disease-free survival (DFS) was measured from the date of radical hysterectomy to cancer recurrence or censored at the last follow-up. Descriptive statistics for continuous covariates are classified into higher or lower groups according to the cutoff determined by maximally selected log-rank statistics (used R package “survminer”). Categorical variables were compared through the chi-squared test and Fisher’s exact tests in our baseline table using the R package “tableone.”
Kaplan–Meier (KM) method (log-rank tests) was used to evaluate the impact of OP on prognosis. The possible significant factors that affected final risk scores (p < 0.1) were screened out using KM and univariate Cox regression analyses (R package survival, survminer, and ggplot2 were used). Different ways were used to construct risk scores respectively, including least absolute shrinkage and selection operator (LASSO) regression, stepwise multivariate Cox analysis, optimum subsets logistic regression, and random survival forest analysis, of which the method owned the highest area under receiver operator curve (AUC of ROC) was selected to discriminate death and relapse risks comprehensively. The safety of OP was evaluated in low- or high-risk subgroups. The R package “glmnet,” “caret,” “randomForest,” “My.stepwise,” “forestploter,” “forestplot,””bestglm,” “leaps,” “genefilter,” “Hmisc,” “ISLR,” “rms,” “regplot,” and “ROCR” were used. All data processing was carried out using R 4.2.0 software.
Discussion
Considering the relatively poorer prognosis of NECC in previous studies, there are no data to support the consideration of fertility preservation, such as simple conization or radical trachelectomy, even in patients with early-stage disease. While fertility-sparing surgeries have also been reported in women with early-stage NECC [
25,
26], most gynecologic oncologists tend to apply more aggressive treatment, which contributed to the tendency of BSO in clinical practice. However, OP is thought to be particularly important for premenopausal women, who might be more common in NECC than other histology types [
1,
8‐
10]. But there was no consensus on the safety of OP in NECC. In this study, it was found that OP is safe in patients with NECC, especially in younger patients who owned better prognoses based on the machine learning model.
Low incidence and lack of prospective clinical trials made it difficult to draw conclusions on the management of NECC despite the urgent need of clinical practice [
21,
27,
28]. Therefore, the outcomes of women with NECC who chose OP were rarely reported. It was inferred from a study with 1965 patients that non-squamous histology should be a deterrent to OP due to the possibility of residual microscopic tumor [
29], yet whether the cases of NECC are included in this study was not pointed out. Zhang et al. found that BSO may improve the prognosis of patients through the comparison of KM curves, especially for OS (
p = 0.023). However, the impact was not significant for DFS (
p = 0.235); besides, this conclusion was not validated in univariate and multivariate cox regression analysis of their study [
30]. Furthermore, selection bias could have existed since it was obvious that several other significant risk factors, such as age, tumor size, and FIGO stage, would possibly affect patients’ choice of OP and their prognosis. In our study, the results compared KM curves of OP, and BSO in all enrolled patients showed OP had no significant effects on OS and even had a better prognosis for DFS. However, the adverse effects of BSO, such as cardiovascular disease or osteoporosis, would mostly occur in longer-term follow-up, and both death and recurrence were NECC related in our study [
14‐
16]. Thus, the existence of confounding factors was reminded. Additionally, for the possible risk factors that caused bias between the prognosis of OP and BSO groups, such as age, tumor size, and DIM, KM analysis in subgroups respectively showed OP still did not influence prognosis significantly.
The investigation of the most significant prognostic factors could help us discriminate the risk of patients comprehensively, thus evaluating the safety of OP in different risk subgroups. From previous studies, significant prognostic variables are varied, which may include age, FIGO stage, tumor size, LNM, LVSI, DIM, histology heterogeneity, and the use of adjuvant therapies [
4,
27,
31‐
33]. And whether the different risk population classified by these variables was safe to accept OP was uncertain. Our patients were divided into training and testing cohorts, and LASSO, stepwise, optimum subsets, and random forest models were constructed and validated through tenfold cross validation in training cohort. It was demonstrated that random forest models owned highest AUC in testing cohort whether for DFS or OS. Therefore, tumor size, age, vaginal invasion, DIM, stage (1, 2, and ≥ 3), LNM, and LUSI were calculated to predict DFS, while age, tumor size, stage (whether in early stage), LNM, LUSI, DIM, parametrial involvement, and vaginal invasion were used to estimate OS. Then, prognoses between OP and BSO were compared through KM analysis, and it was validated that OP should be considered if the patients wished in the population of lower risk. However, no patients with higher risk have chosen OP, which might be related to the age of the high-risk prognosis group (82.8% patients of high DFS risk group and 100% patients of high OS risk group were > 46 years old).
After that, OP was evaluated in the cohort of younger women (≤ 46 years old) independently. Vaginal invasion, tumor size, and para-aortic LNM were used to construct recurrence risk scores and divided patients into different groups, in which the safety of OP was confirmed. On the other hand, only accepting radiotherapy was found to be possibly associated with death risk in patients ≤ 46 years, and none of the patients in this study had pelvic recurrence. It might remind us that the risks and benefits of radiotherapy need further estimation, though theoretically radiotherapy was recommended in patients with higher risks. Besides, the impact of radiotherapy on ovarian function should be considered for women who require OP [
11,
12]. In our present study, several strategies were applied to preserve ovarian function, for instance, all of the preserved ovaries were suspended outside the radiation field, and GnRH-α was used for young patients 14 days before chemotherapy. In total, for those patients ≤ 46 years, there was no significant effect of OP on prognosis even in higher risk population. Therefore, the need to preserve ovaries in these patients should be considered in treatment, with necessary strategies for protecting ovarian function.
Notably, one main concern for the safety of OP in NECC would be ovarian metastasis. The ratio of ovarian metastasis was seldomly mentioned in published studies of NECC and was higher than other common histology. A study of 133 NECC patients found 2 (1.5%) cases of ovarian metastasis at diagnosis [
30], which was higher than 0.9% reported in whole cervical carcinoma [
29]. Ngamcherttakul et al. even concluded that non-neuroendocrine would be the prerequisite of OP since one of two enrolled NECC patients occurred ovarian metastasis in their study [
13]. In our study, 3 (3.5%) ovarian metastases were found in 86 women who underwent BSO based on the pathological reports. While the conclusion was affected by rarity, the univariate analysis showed that the incidence of ovarian metastasis was increased in patients with later FIGO stage, para-aortic LNM, and parametrial involvement (
p < 0.05). Besides, none of the patients in subgroups of these factors (stage 4, para-aortic LNM, and parametrial involvement) accepted OP, though para-aortic LNM has been considered in our DFS risk model for patients ≤ 46 years old. Therefore, for patients who found these risk factors pre- and intraoperatively, OP should be cautiously considered, and the possibility of ovarian metastasis should be ruled out.
The main limitation of this study is that it is a retrospective study with limited sample size. Though the rarity of NECC might restrict the implementation of prospective randomized studies, large sample sized retrospective study with longer follow-up period is warranted to evaluate the safety of OP. Another limitation is that the patients enrolled were required to be eligible for surgery and had diseases in earlier stages than other studies of NECC, which is also reflected by lower rates of nodal and distant metastasis and higher 5-year overall survival rate [
5,
10,
31,
33]. However, in general, our risk model would be more applicable in patients eligible for surgery to consider OP, rather than risk prediction in all populations.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.