Ovarian cancer is predominantly a cancer of postmenopausal women, and it is rare in women below the age of 40 years. Thus, it is classically described as a disease of older women. The median age for women with ovarian cancer ranges from 60 to 65 years in most developed countries. As life expectancy has increased in most countries worldwide, and because the incidence rate of ovarian cancer increases with age, more and more postmenopausal women will have ovarian cancer (11).
Regarding the sociodemographic data of the study participants, the overall median age of the study participants was 59 years. All of the included participants were postmenopausal and so, age didn’t affect the RMI value as all patients got a score of 3. However, age is a component of the ADNEX model and with increasing age, the probability of malignancy increase. Our study showed no significant difference between those who were older than 59 years and those who were 59 years or younger regarding histopathology results as no certain pathologic entity was significantly higher in either group. Our results were consistent with Huwidi et al., who assessed the diagnostic value of RMI among patients with adnexal mass; there was no significant difference between different age groups regarding either benign or malignant pathology however, the study included different age groups and was not restricted to postmenopausal women (12). Zhang et al., showed no significant age difference between those with benign pathology and those with borderline ovarian tumors in their retrospective study which tested the predictive ability of the RMI among the study patients (13). The performance of ADNEX model for prediction of ovarian cancer was assessed and there was no significant age difference between different pathology groups in the study conducted by Yang et al., however, the study conducted by Lam Huong et al., showed significantly higher median age among patients who were diagnosed with cancer which can be attributed to the much smaller number of patients with cancer (65 VS 396) (14, 15). As for parity, the median parity in our study was 3 and there was no remarkable difference at any histopathologic group between patients who were para 3 or less and those who were more than para 3.
The level of education of the study participants was assessed in our study. More than 75% of the study cohort reached university level. Alberg et al., evaluated the socioeconomic status of African-American women and their relation to the risk of ovarian cancer, the study revealed an inverse relationship between educational level and ovarian cancer risk after adjustment for ovarian cancer risk factors (16). Such relationship could be explained the cross sectional study which was conducted by Elshami et al., and showed higher level of awareness about risk factors and protective factors of ovarian cancer among those with post-secondary education (17).
Regarding past history of cancer among the study participants, only one patient in the study cohort had past history of cancer which was cancer breast and the histopathology revealed that the ovarian mass was already a metastatic deposit. Studies with much larger sample size which targets the relation between past history of gynecological or GIT cancer and present ovarian cancer as its primary outcome and the associated syndromes as BRCA1 or 2 mutation and Lynch syndrome can be of value for better assessment of the relationship between certain malignancy and ovarian cancer
Family history of ovarian or ovarian cancer is well known risk factor for development of OC. At our study, 16% of the study participants had family history of malignancy; only one case had past history of breast cancer but ultimately she had benign pathology and one case with borderline ovarian tumor had family history of ovarian cancer.
Studies that have thoroughly adjusted for the effects of factors like duration of oral contraceptive use and number of full-term pregnancies, have not noted a strong association between difficulty in conceiving and the risk of ovarian cancer among parous women. However, an increased risk among infertile women who remain childless despite long periods of unprotected intercourse has been reported in two large, pooled analyses. It remains to be understood whether such women are at risk due to the primary basis for their infertility, some correlate of infertility such as exposure to ovulation-inducing drugs, a shared genetic susceptibility to ovarian cancer and infertility, or some other reason (18). Previous studies have debated whether OI could increase the risk of invasive ovarian cancer (IOC) and borderline ovarian tumors (BOT). Although most studies have concluded that OI does not contribute to the risk of IOC and BOT, some scholars still proposed that OI may be associated with them (19). Infertility, its duration and the management which was adopted to deal with it was assessed in our study; 8 patients representing 16% of the study participants had history of infertility and out of those 8 patients, 3 received induction of ovulation for treatment of their fertility problems with only one patient developing ovarian cancer. Retrospective cohort studies with much larger sample size is more suitable for evaluating the relationship between infertility and ovarian cancer.
The RMI was evaluated in different histopathologic subtypes and it was shown to be significantly higher among those with malignant pathology stages Ⅱ - Ⅳ and in those with metastatic ovarian cancer. Our results were similar to those obtained by Lycke et al., who showed significantly higher mean RMI among patients with FIGO stage Ⅲ, Ⅳ ovarian cancer compared with those with benign or borderline ovarian tumors whether the patients were premenopausal or postmenopausal (20). Similar results were also achieved by Dora et al., who showed significantly higher RMI among patients with malignant ovarian masses (21). Only one patient among our study cohort was diagnosed with stage Ⅰ ovarian cancer and so a comparison with benign cases regarding the RMI value needs further studies with more cases ultimately diagnosed with stage Ⅰ ovarian cancer.
The components of the ADNEX model were evaluated in the five histopathological categories; the three components which were significantly different between the were the CA-125 level, the presence of ascites, and the number of locules. The first 2 were associated with malignant cases while from 1–10 locules were predominant in the benign cases. Lam Huong et al., showed that ascites was more prevalent in the cancer group however, there was significant difference regarding all other components; this difference can be attributed to the fact that the study wasn’t limited to postmenopausal women. Moreover, the histopathological results were either benign or malignant i.e the analysis was not based on the five histopathological groups which can be predicted by the ADNEX model (15). The results obtained by Yang et al., showed significant difference between benign and malignant cases regarding all ultrasound components of the ADNEX model however, the authors included the patients with borderline ovarian tumors into the malignant category despite being completely different entity and this could affect the reliability of the findings (14). In daily practice, the two most prevalent histological subtypes are the benign and the stage Ⅱ - Ⅳ OC and so, for prospective assessment, comparison between these two subtypes in particular would be more reliable in identifying which ultrasound feature correlate better with a given subtype. The other 3 histological subtypes are relatively rare and so, multicentric and retrospective studies would be more suitable for evaluation of the ultrasound features of these 3 subtypes.
The diagnostic performance of both RMI and ADNEX model for differentiating benign from malignant ovarian tumors was assessed; the ROC curve showed a bigger area under the curve (AUC) for the ADNEX model. Regarding the RMI, a cutoff of 115 was associated with 81.8% sensitivity, 60.7% specificity, positive likehood ratio of 2.06 and negative likehood ratio of 0.299 while the ADNEX model at a cutoff level of 10 was associated with 91.1% sensitivity, 65% specificity, 2.66 positive likehood ratio and 0.257 negative likehood ratio. By using the ADNEX model, Yoeli-Bik et al., achieved a sensitivity of 91%, specificity of 86%, LR + of 6.7, and LR- of 0.7 and these results were obtained at 10% cutoff; such higher specificity could be attributed to the fact that 33% of the study cohort didn’t undergo surgical intervention and were included in the study if they had adequate clinical or imaging follow-up which can point to a tendency towards operating on cases with high probability of malignancy which shall decrease the incidence of false positive results (22). Another muticenter cohort study by Van Calster et al., which included 4905 patients from 36 oncology centers assessed the predictive ability of the ADNEX model and the RMI for detecting ovarian cancer; regarding the ADNEX model, the overall sensitivity was 91% and the overall specificity was 85% and this was achieved with 10% risk threshold and 0.94 AUC. The higher AUC compared to our study can be attributed to the much larger sample size and the fact that 2151 patients (44%) of the study cohort were postmenopausal women which would increase the probability of ovarian malignancy in the cohort. Regarding the RMI, at a cutoff of 200, the overall sensitivity was 60% and the specificity was 95%, such lower sensitivity and higher specificity compared to our study is attributed to the lower cutoff value which was set (115 vs 200) (23). Another study by Pelayo et al., which assessed the predictive accuracy of the ADNEX model yielded 94% sensitivity and 82% specificity with 0.92 AUC; the lower false positive cases compared to our study can be attributed to the fact that 39% of the study participants suffer from digestive symptoms which should increase the probability of malignancy in contrast to our study participants who were asymptomatic besides the fact that 16% of the study participants underwent sonography by non-expert sonographers and 8% came from the emergency room; such diversity could affect the reliability the interpretation of the ultrasound findings (24). Results obtained by Poonyakanok et al., showed 98% sensitivity and 87% specificity when using 10% threshold of malignancy probability; the higher sensitivity and specificity compared to our study can be attributed to the fact that the authors excluded 13 patients from those who were recruited as they were ultimately diagnosed with uterine or abscess lesions which would indirectly raise the accuracy of the ultrasound which is integral part of the ADNEX model (25). Results were also similar to those obtained by Peng et al., who achieved with the same cutoff value a 94% sensitivity, 74% specificity, 3.06 LR + and 0.08 LR-; both studies were conducted at a tertiary oncology centers and the incidence of benign and malignant pathology were also similar (26). Yang et al., achieved 93% sensitivity, 73% specificity, 3.39 LR + and 0.1 LR- using 10% malignant probability; the authors excluded masses which didn’t originate from the ovary on the histopathology specimen; this exclusion will raise the predictive ability of the ultrasound and will decrease the false positive results (14).
The ADNEX model discriminates ovarian tumors into five subtypes; benign, borderline, malignant stage Ⅰ, malignant stage Ⅱ-Ⅳ, and metastatic ovarian deposits. At our study, the model performed best at discriminating between benign and malignant ovarian tumors with AUC of 0.864 while the least performance was observed with differentiating between benign and borderline and stage I OC with AUC of 0.722 and 0.724 respectively. Results achieved by Sayasneh et al., showed the highest performance when discriminating between benign and stage Ⅱ - Ⅳ OC with AUC of 0.99; the higher AUC compared to our study in particular when the model discriminated between benign and stage Ⅰ OC is attributed to the much higher sample size in the that multicenter study which led to higher percentage of patients with stage Ⅰ OC compared to our study (8% vs 2%) and the fact that the percentage of cases diagnosed with stage Ⅱ - Ⅳ OC in our study was the double (22% vs 11%) had led to the lesser AUC in our study when ADNEX model was used to discriminate between benign and malignant cases (0.864 vs 0.99) (27). Meys et al., assessed the performance of the ADNEX model for the five tumor subtypes and the highest AUC was obtained when the model was used to discriminate between benign and stage Ⅱ - Ⅳ OC (0.97) which is higher compared to our study (0.823) which could be related to the fact that the percentage of benign cases was lower in our study (56% vs 64%) (28). These results are in agreement with the results obtained by Van Calster et al., which also showed in their multicentric study that the highest AUC obtained when benign lesions are compared against stage Ⅱ - Ⅳ OC; the comparison of power of discrimination between benign and all other four histological subtypes showed excellent performance by ADNEX model with all AUCs higher than 0.9; the high percentage of benign lesions among the studied cohort (67%) in comparison with other histological subtypes can contribute to this performance (29).
The ADNEX model can change the future management of ovarian cancer by prediction of the staging of ovarian cancer which is particularly important as it largely improves the prognosis when the cancer is detected at stage Ⅰ. It has better sensitivity and specificity for differentiating between benign and malignant tumors when compared with RMI; however, it needs more experience in the ultrasound evaluation of adnexal masses so it can be implemented in screening programs on a large scale. Moreover, the discrimination between the five histopathologic subtypes is of a great value as it can lead to proper triaging of the patients; when the model predicts that the mass is benign, the patient can be managed by in a general gynecology hospital while if it predicts a malignant nature of the mass, the patient must be referred to a gynecological oncology for multidisciplinary team (MDT) consultation as this will largely influence the management and prognosis. Such MDT consultation will guide the management to achieve the best possible results; when the model predicts borderline or stage Ⅰ OC, the patient should have optimum surgical staging by an experienced gyne-oncologist while patients with high probability of stage Ⅱ - Ⅳ can have further imaging as Computed Tomography (CT) scan abdomen and pelvis for better detection of advanced disease and then receive neoadjuvant chemotherapy first followed by interval debulking. Finally, if the model predicts that the mass is metastatic, other investigations can be ordered to find out the primary origin such as mammogram, upper and lower GIT endoscopy or even Positron Emission Tomography (PET)-Scan and the patient can avoid surgery and its potential complications.
The major drawback of the ADNEX model is the fact that it needs significant experience in the field of ultrasound so as to be fruitful and of value. Training programs must be adopted in order to upgrade the skills needed for precise evaluation of adnexal masses by ultrasound which is extremely important before using the ADNEX model for screening purposes.
Our study was not without limitations; data were collected from single tertiary center and this can negatively impact the representation of different regions of the country in this study. Further studies from different centers are needed so as to produce a larger sample size which will be a better representative of the predictive ability of the ADNEX model particularly for rare findings such as borderline and stage Ⅰ OC