Evaluation of menopausal endometrial lesions via mathematical modeling clinical indicators and ultrasonographic parameters

1.Introduction

Endometrial lesions (EL) is one of the most common diseases in female groups and its incidence rapidly rises with the increasing of age. With the changes of estrogen and progesterone, the related thickening and proliferation of endometrium would lead to endometrial hyperplasia or space occupying lesions in uterine cavity.

The EL can be classified into benign and malignant lesions. The former had three general subtypes of endometrial polyp, endometritis, endometrial hyperplasia (EH), and the latter mainly consisted of endometrial precancerous lesions (EPL), endometrial carcinoma (EC) and endometrial sarcoma. Accordingly, the EPL was divided into three kinds of mild, moderate and severe endometrial dysplasia. It was reported that the EC was the most frequently found in perimenopausal and postmenopausal women [1, 2]. According to the guideline of surveillance, epidemiology, and end results (SEER), the EC would have better survival rate at the cases of early detection, young patients, and low grade.

Generally, the diagnosis of EL was categorized as invasive and non-invasive approaches. The invasive methods, including diagnostic curettage (DC), hysteroscopy (HS), can directly observe the endometrial samples for biopsying and excising. However, it was easily influenced by the clinical conditions and the operators’ skills and sometimes would lead to uterine perforation, infection and other serious complications [3]. Among the non-invasive methods, the ultrasonic imaging is the most used technique, and the transabdominal ultrasonography (TAS) and TVS has the advantages of real-time dynamic, simple operation, radiation free and good repeatability, etc. [4]. Compared with the modality of TAS, TVS avoided the interference of abdominal fat and intestinal gas and displayed higher quality of image, and even did not need preparation before examination. Thus, the TVS has become popular for diagnosis of EL in clinical practice [5, 6, 7, 8].

For clinical applications, the diagnosis of MEL is often misled by complicated clinical indicators and ultrasonographic parameters. Therefore, a variety of mathematical models were constructed for the performance evaluation for clinical indicators and ultrasonographic parameters. In clinical practice, two clinical indicators of patient age (PA), abnormal vaginal bleeding (AVB) was regarded as the typical symptoms for the differentiation between benign and malignant MEL. Moreover, three ultrasonographic parameters of endometrial thickness (ET) and endometrial uneven echo (EUE), endometrial blood flow signal (EBFS) were closely correlated with the occurrence of MEL.

Some mathematical models with clinical indicators and ultrasonographic parameters were proposed in the literature. It was evident that the EC of older menopause patients had the higher risks of the malignant degree [9, 10, 11, 12]. By multi-variable analysis, the clinical indicator of PA was thought as an independent prognosticator and with a cut-off threshold of 80 years [13]. It was once reported that EC increased significantly after the age of 53 in patients with atypical endometrial hyperplasia [14]. A prospective cohort study for testing the emerging EC biomarkers proved that 90% cases had the symptom of abnormal uterine bleeding [15]. This might explain the significance of abnormal EUE for the EL discrimination. This suggested that the ultrasonographic parameter of ET above the cut off threshold of 3–5 mm was identified as critical indication for endometrial malignancy [16]. Combined with multiple indicators, the risk of EC scoring system was performed with TVS parameters. The overall predicted accuracy of endometrial malignancy was 95.7%. In 89 patients with ET ⩾ 4 mm, the sensitivity and specificity achieved 91.8% and 93.6% for the predication of malignancy. The scoring model presented robust predictive performance in differentiating EC and EH [17]. Of the studied ultrasound endometrial parameters, the highest areas under the curve (AUCs) were achieved for ET (AUC = 0.76; 95% CI: 0.71–0.81), and for the interrupted endomyometrial junction (AUC = 0.70, 95% CI: 0.65–0.75) [18]. With the calculated ET, 2D Doppler TVS presented valuable performance in differentiating benign and malignant EL with the sensitivity and specificity of 86.2% and 76.1%, respectively [19]. Notably, myometrial invasion probably happened when there was irregularity of the endometrium-myometrium border. That is, the disruption of inner layer of myometrium or the tumor extended asymmetrically into the myometrium [20].

In clinical applications, early EL generally lacked of specific clinical manifestations. Many woman menopause patients were easily misdiagnosed by persistent irregular vaginal bleeding or vaginal discharge, abnormal uterine echoes, abdominal pain, emaciation, etc. Therefore, the accuracy of early diagnosis and prognosis of treatment become more and more difficult only depending on the prior clinical symptoms or TVS. Therefore, the mathematical model combing clinical indicators and TVS parameters is worthy of being discussed for the potential capability in the diagnosis of MEL.

In our study, two mathematical models of improved scoring model and logistic linear regression were proposed to evaluate the performance of combined and individual indicators for the discrimination of benign and malignant MEL. The novelties of our method were summarized as follows: (1) five typical indicators of PA, AVB, ET, EUE and EBFS were implemented to construct mathematical models; (2) The scoring model and multivariate linear logistic regression were applied in the performance evaluation of combined and individual indicators. This would provide insights for the indicator interpretation and deliver valuable guidance for the early diagnosis of EL.

2.Materials and methods

3.Results

3.1Performance evaluation of individual indicators

For the diagnosis of benign and malignant MEL, the calculated thresholds of PA, ET were 57.5 years old and 11.5 mm, respectively. In Fig. 1, the AUCs of PA, AVB, ET, EUE, and EBFS for benign and malignant MEL were 0.68, 0.70, 0.64, 0.65, 0.58, respectively. Among the five indicators, the sensitivity of EUE was the highest of 96.4%, and the specificity of EBFS was the highest of 88.3% (Table 3).

Table 3

Performance evaluation of individual indicators

Indicators	Sensitivity	Specificity	AUC	95% CI
PA	71.4%	64.8%	0.68	0.57 ∼ 0.79
AVB	78.6%	61.7%	0.70	0.60 ∼ 0.80
ET	57.1%	70.3%	0.64	0.52 ∼ 0.75
EUE	96.4%	32.8%	0.65	0.55 ∼ 0.74
EBFS	28.6%	88.3%	0.58	0.46 ∼ 0.71

Figure 1.

ROCs of individual indicators for MEL.

3.2Performance evaluation of combined indicators

According to the ROCs, the mean total scores of malignant and benign groups were 3.32 ± 1.09 and 1.82 ± 1.01, respectively. At the total cut-off value of 2.5, the sensitivity, specificity and AUC for the benign and malignant MEL were 78.6%, 75.8% and 0.83, respectively. And the 95% CI changed from 0.75 to 0.91 (P= 0.000), as shown in Fig. 2.

Figure 2.

ROC of combined indicators for MEL.

3.4Evaluation of optimized combined and individual indicators

With the establishment of the corresponding logistic regression equation, the results are shown in Table 6. It was evident that the three significant indicators of PA, ET and AVB showed statistical significance (P< 0.05). The AVB had the highest OR of 6.320 among the three indicators.

Table 6

Evaluation of optimized indicators via multivariate logistic regression

Variables	β	OR	χ2	P	95% CI
PA	0.147	1.150	17.123	< 0.001	(1.076, 1.244)
ET	0.085	1.068	7.272	0.046	(1.005, 1.143)
AVB	2.042	6.320	12.615	0.002	(2.127, 2.195)
Constant	-12.221	0.000	25.841	< 0.001

Table 7

Comparison of optimized indicators and individual indicators

Significant indicators	Sensitivity	Specificity	AUC	95% CI
PA	71.4%	64.8%	0.73	(0.638, 0.822)
AVB	78.6%	60.9%	0.70	(0.595, 0.800)
ET	57.1%	70.3%	0.66	(0.556, 0.770)
Multivariate logistic regression	75.0%	81.3%	0.85	(0.786, 0.925)

Figure 3.

ROCs of optimized combined and individual indicators.

The diagnostic efficacy of benign and malignant MEL obtained by regression equation is shown in Table 7 and Fig. 3. The sensitivity, specificity and AUC of the regression model in the diagnosis of MEL were 75.0%, 81.3% and 0.85, respectively.

4.Discussion

This study retrospectively analyzed the clinical indicators and ultrasonographic parameters in the diagnosis of MEL, and concluded that PA, AVB, ET, EUE and EBFS could be used to identify the benign and malignant MEL. With the evaluation of those indicators, the scoring model and logistic regression had the similar clinical significance in the diagnosis of MEL and would provide reliable basis for the classification of MEL.

Among the five indicators, three indicators of PA, AVB and ET showed higher sensitivity and specificity in the diagnosis of MEL.There was a significant difference in PA for discriminating the benign and malignant MEL. The PA of the malignant group was higher than those of benign group. The diagnostic efficacy indicated that PA was one of the valuable indicators of endometrial malignancy, and it was consistent with endometrial cancer diagnosis and treatment standards (2018 edition) as well as previous studies [23, 24, 25, 26]. It was learned from our experiments that AVB was statistically significant in differentiating benign and malignant MEL. Combining with the main clinical symptoms of EC patients, AVB was an important indicators for 90% of menopausal patients.

In this study, the ET of 11.5 mm was the cut off value for clinical diagnosis. For menopausal women without abnormal bleeding, the sensitivity and specificity for the diagnosis of EC were the best at the ET of 11 mm [27]. More studies showed that ET was associated with EC in postmenopausal women [28, 29, 30]. Meanwhile, the risk of EC increased with increased ET [31]. In our experiment, the EUE and EBFS of malignant group were higher than those of benign group. During the examination, the ultrasound images showed that the menopausal endometrium become thinner alike the line shape, and there was no obvious blood flow signal. According to the EH, it should be alert once the endometrial thicken or more detected chaotic blood flow signals [32]. Learned from our experiment, the combined indicators of the scoring model was better than the most of individual indicators except for the PA. Multivariate logistic regression analysis showed that AVB was the most correlated risk factors with the maximum OR value. This would bring reliable indications for the clinical applications in MEL. Therefore, the menopausal women with the two mentioned indicators of AVB, ET should be preserved the high risks of malignant lesions.

Referenced by the guidelines of 2010 college of obstetricians and gynecologists of Canada (SOGC), it recommend that only endometrial thickening of postmenopausal asymptomatic women don’t regularly need for endometrial clinical samples. Especially, if the clinical cases existed typical ultrasonic indicators, such as EUE, EBFS and ET above 11 mm, the further examination of diagnostic curettage or other follow-up treatments should be required. This is consistent with our results.

However, there is still no consensus on the ET cut-off value of menopausal women undergoing diagnostic curettage or hysteroscopic surgery. Usually, surgery was strongly suggested while the menopausal ET reached to 10–15 mm. The ET of 11 mm was also reported as the most common criterion [33]. Other studies proposed that the ET should not be used as a predictor of EC [34, 35, 36].

With the multiple indicators, the two mathematical models had better guiding significance for clinical diagnosis and treatment. For menopausal women with AVB, the TVS should be suggested in the screening of MEL. Together with the follow-up observations, diagnostic curettage or hysteroscopy examination, the early detection and treatment of MEL would greatly improve the recovery rate.

5.Conclusions

The combined indicators presented better performance in the diagnosis of benign and malignant MEL. The indicators of PA, AVB and ET were considered as independent risk factors, and the AVB demonstrated the most typical indication for clinical applications.