A Novel Model To Predict Endometriosis in Patients with Ovarian Cysts: A Retrospective Study.

Objective To develop a model that uses hematological indexes and clinical characteristics to help estimate the probability of endometriosis in patients with ovarian cysts. Methods A retrospective study was conducted on 2242 patients who underwent surgery for benign ovarian cysts from January 2008 to November 2016. Variables included in the model were serum tumor markers, blood routine test, age, BMI, reproductive history, history of hysteroscopy, menstrual episodes. Logistic regression was used to construct a predictive model for endometriosis, Receiver Operating Characteristic curves and the areas under the curve was used to verify the model’s validities. Ten-fold cross-validation was primarily used as an internal validation to evaluate the prediction accuracies of the model. Normalized mean square errors (NMSE) was obtained to compare the reliability of different models. symptoms, age, body mass index (BMI), reproductive history (mature delivery, premature delivery, abortion), menstrual history (menstrual cycle, duration of menstrual ow), surgical features (history of laparotomy, laparoscopy or hysteroscopy) and histopathology diagnosis following surgery were retrieved for each patient.


Introduction
Endometriosis is one of the most common benign gynecological disorders occurring in 6 to 10% of the general female population [1]. It is de ned as the development of endometrial tissue (gland and stroma) outside the uterus such as, ovaries, pelvic peritoneum, and rectovaginal space. Endometriosis is a recurring persistent disease that causes non-menstrual pelvic pain, dyspareunia infertility, dysmenorrhea, and menstrual irregularities [2]. It is reported as a disease of complex multifactorial etiology, among all the hypothesis, transplantation of endometrial tissues via retrograde menstruation is widely accepted [3].
The correlation between symptoms and lesions is quite incomprehensible because the symptoms are nonspeci c and not diagnostic. [4].
The diagnosis of endometriosis is based on clinical manifestations and imaging techniques [5] but con rm diagnosis of the disease can only be obtain by invasive procedures like direct visualization of peritoneal and ovarian implants by laparoscopy or laparotomy followed by histological analysis [6]. To increase the accuracy of the diagnosis of endometriosis, especially to avoid the use of invasive management, some investigators have begun to characterize the factors contributing to the detection of endometriosis. There is evidence that family history, immunological, menstrual and reproductive factors and are associated with endometriosis [7], different combinations of these biomarkers, are studied to increase the diagnostic accuracy of this disease. [8][9][10] The development of multiple factors to improve the accuracy of diagnosis of endometriosis is necessary.
The novel model was synthesized by correlating patient's hematological indexes and clinical characteristics in a multivariate regression model which could help us to recognize which ovarian cysts are more likely to be biopsied so as endometriosis can be diagnosed in early stage.

Patients
Retrospective data were collected from The Second A liated Hospital of Wenzhou Medical University. A total of 2242 premenopausal women who underwent either laparoscopic or laparotomic surgery in the Gynecology Department of our hospital from January 2008 to November 2016 were included in this study. Patients who underwent surgery for ovarian endometriosis, which was con rmed by surgical specimen histopathological examinations were eligible for our study. The exclusion criteria as follows: history of hormonal therapy for endometriosis, pregnant woman, abnormal hepatic and renal function tests, pelvic in ammatory disease, pathologically con rmed or clinically diagnosed with leiomyoma or adenomyosis, acute infection or history of chronic in ammatory disease, immune system diseases, or malignancy.

Data Collection
The data were obtained by reviewing the patients' medical records. All patients underwent routine preoperative laboratory studies, including CA-125, CA-19-9, carcinoembryonic antigen (CEA) and a complete blood count test was performed prior to surgery. Patient-related factors assessed included endometriosis-related symptoms, age, body mass index (BMI), reproductive history (mature delivery, premature delivery, abortion), menstrual history (menstrual cycle, duration of menstrual ow), surgical features (history of laparotomy, laparoscopy or hysteroscopy) and histopathology diagnosis following surgery were retrieved for each patient.
Most patients in the study had preoperative ultrasonographic evaluations. The reasons why we do not bring in the ultrasonographic diagnosis of cysts in the study were that sonographic evaluations were performed by different sonographers, transducers and types of ultrasounds.

Statistical Analysis
Patients characteristics were compared by using variance analysis (t-test, for continuous variables), chisquare test (for dichotomous variables) or Kruskal-Wallis test (for continuous variables in skewed distribution) [11]. Logistic regression was used to perform multivariate analysis, and forward method was used to select variables [12]. 10-fold cross validation was used to predict the accuracy of the internal validation model. We randomly divided the data set into 10 copies, with 9 of them as a training set, used to establish the forecast model, the remaining 1 data set as a validation set, as a validation set. Training set to build the model, with the validation set to predict, this process was continued for 10 times and ultimately obtained a complete set of predictions. The predicted value was used to establish a crossvalidated ROC curve. At the same time, 10 models were established, and 10 normalized mean square errors (NMSE) were obtained for the veri cation set and the average NMSE of the model was obtained to compare the reliability of the model. The smaller the NMSE, the more reliable the model is. Delong method was used to compare the signi cant difference between ROC curves, in which P < 0.05 indicated the difference was statistically signi cant. The Hosmer-Lemeshow goodness-of-t test was used to test the predictive ability (calibration) of the model [13]. The model was evaluated with different indicators, including sensitivity, speci city, area under curve (AUC), and Youden Index, where the signature = sensitivity − (1-speci city). SPSS 13.0 for statistical analysis, R 3.2.3 software pROC package to do the Receiver Operating Curve (ROC) and ROC curve comparison. Bilateral test P < 0.05 showed statistically signi cant.

Patients Characteristics:
A total of 2242 women were enrolled in the study and the age ranges from 18 to 46 years, with a mean age of 31.37 years. 978 patients had endometriosis while the rest were 860 patients with mature teratoma,145 patients with serous cystadenomas, 183 patients with mucinous cystadenomas, and 76 patients with other benign conditions. The characteristics included baseline demographic, clinical, surgical, and laboratory features of patients with and without endometriosis (Table 1).

Clinicopathological Features With Endometriosis:
After analysis of endometriosis with clinicopathological features, we found that there was no statistical signi cance with premature delivery (P = 0.  (Table 1).
Univariate And Multivariate Analysis: We selected 16 variables to establish the multivariate logistic regression analysis. As a result, 11 variables were retained in the nal logistic regression model (Table 2), it revealed that Log(CA-125), dysmenorrhea, history of hysteroscopy, age and mature delivery remained as signi cant variables associated with endometriosis. The Hosmer-Lemeshow test results revealed an adequate goodness-oft for the regression model (P > 0.05). The association between characteristics and diagnosis of endometriosis was explored and shown in (Table 3), all the models (model 1 ~ model 5) were signi cantly associated with endometriosis. A ROC curve for the model with CA-125 alone was constructed (Fig. 1), the area under the curve was 0.888 (P < 0.001) for the model, with a sensitivity of 0.816% and speci city of 0.835%, These results indicate a moderate predictive performance of the model,after adjustment for other multiple covariates, the result presented a little rise of accuracy in diagnosis of endometriosis. The modle5 of included other values (combined clinical characteristics and haematological indexes) in the CA-125 based probabilistic model showed an AUC of 0.916 (P 001), a sensitivity of 0.849 and speci city of 0.864 (Fig. 2), The Hosmer-Lemeshow test (p = 0.060) in the combined model5 indicate a good tness of the model characteristic (dysmenorrhea, Irregular menstruation,), laboratory characteristics (Platelet count, monocyte count, CA-19-9, CA-125, CEA).

Discussion
Endometriosis can only be diagnosed by invasive procedures such as laparoscopic or laparotomy exploration. We constructed a non-invasive predictive model based on medical history and hematological indexes (blood routine, serum tumor markers examination) that can diagnose endometriosis in ovarian cyst patients. We found association between the CA-125, CA-19-9, age, partus matures, menstrual episodes, history of hysteroscopy, dysmenorrhea and blood routine test with endometriosis, but no single characteristic predicted endometriosis with a high accuracy. Our study supported the retrograde menstruation theory because the history of hysteroscopy is shown to be associated with an increase in risk of developing endometriosis.
Our study con rmed the belief that an increased frequency of and duration of menstruations is associated with endometriosis [14,15]. Dysmenorrhea was the main symptom of endometriosis infertile women (46.92%) with endometriosis and the mechanism of dysmenorrhea in endometriosis lie in increased production of prostaglandins (PGs) [16]. Moreover, BMI showed a negative correlation with the presence of endometriosis, as was reported previously [17]. Obesity is often associated with long menstrual cycles, a factor that reduce the risk of endometriosis. It is considered that the reduction of the frequency of menstrual episodes counterbalances the relative hyperestrogenism of women [18].
Endometriosis is rare before the menarche and tends to decrease after the menopause. Studies conducted in women under age of 45 years suggested that the frequency of endometriosis increases with age until menopause [19]. While Fuldeore, M. J, et al [2] reported that the average age of women with endometriosis in their study was 37.8 years compared to 33.8 years women without endometriosis (p < 0.0001), it is possible that incidence of endometriosis increases as women age increases which can be because of the hormonal changes that occur during peri-menopause [20].
Screening for the diagnosis of patients with clinical suspicion of endometriosis is based on serum CA-125 which have been con rmed in many studies. Shen, A et al [21] reported that endometriosis is signi cantly associated with elevated serum CA-125 concentrations, con rmed CA-125 as an auxiliary biological marker in endometriosis diagnosis. Some studies [22,23] did not agree with this nding and showed that the diagnosis of endometriosis on CA-125 alone is not accurate, mainly in relation to their sensitivity, Hirsch, M, et al [23] reported that CA-125 with a cut-off of ≥ 30 u/ml has a sensitivity of 0.57, which did not meet the criteria for a triage test, and international guidelines do not recommend CA-125 testing in women with suspected endometriosis [24]. However, in the study we nd model 1 which consisted of CA-125 alone predicted endometriosis with high sensitivity (81.6%) and predicted the absence of endometriosis with a speci city of 83.5%. Nevertheless the timing of blood collection for CA-125 is uncontrolled because it's a retrospective design, the relationship with the menstrual cycle is known to affect this test [25].
The study shows an inverse association between the number of mature delivery and endometriosis, but no association between the number of abortions and endometriosis has been found. This has also been observed in many studies of endometriosis [14,26]. Parazzini,F,et al [26]reported that the risk of endometriosis decreased with increasing number of births, compared with nulliparous women, the OR of endometriosis at stage 1 was 0.1 (95% CI 0.1, 0.2) in women reporting two or more births was respectively 0.1 (95% CI 0.1, 0.3), 0.2 (95% CI 0.1, 0.4).
It has been reported that [27] reproductive history may in uence hormonal milieu, Estradiol levels is higher among nulliparous women than among parous women, whereas androgen levels have an opposite effect, and reproductive history may in uence the volume of endometrial cells released into the peritoneal cavity. The other studies revealed that CA-19-9 can be used to discriminate between patients with or without endometriosis, and it is correlation to severity of the disease, their results showed that CA-19-9 was signi cantly associated with advanced stage (stage III and IV) endometriosis [10]. Our results is in concordance with a former study that the mean levels of CA-19-9 are signi cantly elevated compared with the control group.
Endometriosis is associated with increased in ammatory activity which is an important stimulant for platelets [28], suggest platelet indices is an important and effortless hematological parameter that can be useful in evaluation of endometriosis [29,30]. Evsen, M. S et al reported that platelets count in patients with peritoneal endometriosis were found to be higher from the control group (p = 0.038)[30], particularly more apparent in advanced stage peritoneal endometriosis. Monocytes also was also implicated as prognostic factor of in ammatory response but there is no evidence supporting that monocyte count is associated with endometriosis, our study shows that it is a protective factor for endometriosis, further study should be done to con rm this nding. This model provides guidance about con rmation of endometriosis. CA-125 can be useful in directing the diagnosis of the disease, and clinical history, tumor marker and routine blood tests increase the diagnosis of endometriosis more accurately. For instance, a peri-menopause woman with multiple reproductive history and irregular menstruation, has a higher chance of containing endometriosis, and if CA-125 is quite high, ovarian cysts would be appropriate to con rm the presence of the disease.

Conclusion
In this study, we found that CA-125, clinical history, tumor marker and routine blood term testing are predictors of endometriosis. Our model can contribute in diagnosis as a predictor for endometriosis in patients with ovarian cysts.