A comparative study of the risk of malignancy index and the ovarian crescent sign for the diagnosis of invasive ovarian cancer

Authors


Abstract

Objective

To compare the value of the risk of malignancy index (RMI) and the ovarian crescent sign (OCS) in the diagnosis of ovarian malignancy.

Methods

This was a prospective observational study of women with ultrasonographic diagnosis of an ovarian cyst. The RMI was calculated in all cases using a previously published formula (RMI = U (ultrasound score) × M (menopausal status) × serum CA125 (kU/L)). A value > 200 was considered to be diagnostic of ovarian cancer. The OCS was defined as a rim of visible healthy ovarian tissue in the ipsilateral ovary. Its absence was taken as being diagnostic of invasive cancer.

Results

A total of 106 consecutive women were included in the study, of whom 92 (86.8%) had a benign ovarian tumor, five (4.7%) had borderline lesions and nine (8.5%) had an invasive ovarian cancer. The absence of an OCS diagnosed invasive ovarian cancer with a sensitivity of 100% (95% CI, 70–100%), specificity of 93% (95% CI, 86–96%), positive predictive value (PPV) of 56%, negative predictive value (NPV) of 100% and positive likelihood ratio (LR+) of 13.86 (95% CI, 6.79–28.29). This compared favorably with a sensitivity of 89% (95% CI, 57–98%), specificity of 92% (95% CI, 85–96%), PPV of 50%, NPV of 99% and LR+ of 10.78 (95% CI, 5.34–21.77), which were achieved using RMI > 200 (P < 0.01).

Conclusions

The RMI and the OCS are useful tests for discriminating between invasive and non-invasive ovarian tumors. The application of these tests in a sequential manner might improve the overall accuracy of ovarian cancer diagnosis. Copyright © 2006 ISUOG. Published by John Wiley & Sons, Ltd.

Introduction

In recent years ultrasonography has been increasingly used for the assessment of women presenting with a wide range of gynecological complaints. During a pelvic scan an attempt is routinely made to identify the ovaries and to examine their morphological appearance. This has led to a substantial increase in the number of asymptomatic women diagnosed with ovarian abnormalities1. Although the quality of ultrasound equipment has significantly improved in recent years, there is still no consensus on the most effective diagnostic criteria for discriminating between benign and malignant ovarian tumors on ultrasound scan2–9. As a result, women are often referred for additional investigations, such as measurement of tumor markers and magnetic resonance imaging, in order to clarify the nature of ovarian lesions. Many women are also offered surgery because of concerns about the possibility of malignancy, which is costly and carries the risk of complications.

The risk of malignancy index (RMI)—a scoring system based on a combination of demographic and ultrasound data with measurement of serum CA125—has been widely adopted in the UK to facilitate triage of women with ovarian tumors for referral to tertiary gynecological oncology units5, 10. Although the RMI is a relatively simple test to use in clinical practice, its false negative and false positive rates are significant9. In addition, the results are not immediately available and the test is relatively costly, as it involves the use of serum biochemistry.

Recent studies have shown that a detailed sonographic examination of an ovarian tumor by an expert in gynecological ultrasonography may be superior to all previously described diagnostic models for the noninvasive diagnosis of ovarian cancer11. However, in routine clinical practice most scans are performed by sonographers, many of whom have no particular expertise in the characterization of ovarian tumors.

We have recently described a new morphological ultrasound feature, the ‘ovarian crescent sign’ (OCS), which depends on the fact that healthy ovarian tissue can be seen adjacent to the cyst within the ipsilateral ovary12. The initial report showed that this morphological ultrasound sign has the potential to become a simple and effective way of excluding an invasive ovarian malignancy without the need for a detailed morphological assessment of the tumor or the use of serum biochemistry.

In this study we prospectively compared the value of the OCS and that of the RMI for the diagnosis of ovarian cancer in a large group of women with an ultrasound diagnosis of ovarian tumor.

Methods

This prospective observational study was carried out at the Gynaecological Assessment Unit at King's College Hospital, which is a tertiary referral gynecological ultrasound center. All women referred to the unit underwent a pelvic ultrasound scan, which involved a detailed examination of the uterus and the adnexa. All examinations were performed by gynecologists with a special interest in gynecological ultrasound, using an Aloka SSD-5000 machine (Aloka Co, Tokyo, Japan). Transvaginal and transabdominal scans were performed on each patient to ensure a complete examination of the entire abdominal cavity. All women with a diagnosis of an ovarian tumor were invited to join the study. Simple cysts, defined as anechoic unilocular lesions with regular walls and no internal septations or solid parts13, were all excluded from the study.

The women's age and menopausal status were recorded in all cases. Menopausal status was defined as more than 1 year of amenorrhea or age > 50 in the case of prior hysterectomy. The following morphological ultrasound information was recorded in each case: site and volume of cyst, presence of septations and solid areas within the cyst, metastases and ascites. Healthy ovarian tissue adjacent to the tumor—the OCS—was defined as visible hypoechogenic tissue with or without ovarian follicles enclosed within the ovarian capsule encircling the tumor and located adjacent to the cyst wall, which could not be separated from the cyst when applying a moderate amount of pressure (Figure 1)12.

Figure 1.

A complex ovarian cyst with healthy ovarian tissue (arrow) clearly visible adjacent to the wall of the cyst (the ovarian crescent sign).

A blood sample was taken from each patient to assess the serum CA125 level with the Immuno-1 analyzer (Bayer Diagnostics, Tarrytown, NY, USA). The RMI was calculated as originally described by Jacobs et al.5 in 1990: RMI = U (ultrasound score) × M (menopausal status) × serum CA125 (kU/L). Each of the following gray-scale morphological features was given one point when present: bilateral lesions, multilocular lesions, solid areas, intra-abdominal metastases and ascites. If the sum of these points was 0 or 1, an ultrasound score U = 1 was given, while a sum of ≥ 2 was given a score U = 3. Premenopausal women were given a score M = 1 and postmenopausal women were given a score M = 3. In accordance with our regional cancer center protocol an RMI > 200 was taken as diagnostic of ovarian cancer.

In women with ovarian tumors, indications for surgery were suspected ovarian malignancy, clinical symptoms attributable to the adnexal cyst, or a patient's request to have the ovarian tumor removed. Women who did not have an indication for surgical treatment were managed expectantly and were invited to attend for a follow-up ultrasound scan 6–8 weeks later. The cyst was classified as benign on the follow-up visit if the woman remained asymptomatic and there was no increase in the size of the cyst, and these women attended for another follow-up scan 3 months later. All women in whom the cyst increased in size were offered surgery. However, cysts that resolved spontaneously were classified as being functional in nature. In women who underwent surgery the ultrasound findings were compared with the histology, which was classified according to the World Health Organization (WHO) guidelines14. Ovarian malignancies were staged according to the classification of the International Federation of Gynecology and Obstetrics (FIGO)15.

The study was approved by the King's Research Ethics Committee, and all patients gave their informed consent.

All statistical analyses were carried out using SPSS version 11 (SPSS Inc., Chicago, Illinois, USA). The statistical significance of differences in continuous variables was determined using Mann–Whitney, Kruskall–Wallis or Student's t-tests depending on the distribution of the data. Proportions were compared using Yates corrected χ2 test. Two-tailed P < 0.05 was considered statistically significant.

The diagnostic accuracy of the tests was assessed using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and positive and negative likelihood ratio (LR + /LR−) measures. Diagnostic performance of the tests was also assessed by examining the areas under receiver–operating characteristics (ROC) curves.

Data were also analyzed using decision-tree analysis. The decision tree was developed using the classification and regression trees method16, which generates binary decision trees. The decision and regression tree was constructed by splitting subsets of the dataset using all predictor variables to create two child nodes repeatedly. The goal was to maximize the positive and negative predictive values in diagnosing invasive ovarian cancer. This allowed a sequential analysis of variables to predict the nature of the ovarian tumor.

Results

One hundred and six consecutive women with ovarian masses were included in the study. The indications for the ultrasound scan were: pain in 38 (35.8%) women, suspected ovarian tumor on a previous routine scan in 37 (34.9%) women, menstrual disorders in 17 (16.0%) women, clinical suspicion of an abdominal mass in seven (6.6%) women, ovarian cancer screening in four (3.8%) women and postmenopausal bleeding in three (2.8%) women. Fifty-five (51.9%) women were managed expectantly and 51 women (48.1%) had surgery. Thirty-one (60.8%) of the women operated on had minimally invasive surgery, while 20 (39.2%) had a laparotomy. The indications for surgery in the 51 women operated on were: symptoms attributable to the adnexal lesion in 25 (49.0%) women, suspected malignancy in 14 (27.5%) women and patient's choice in 12 (23.5%) women. The final diagnoses are listed in Table 1. There were significant differences in types of tumors in women who were managed expectantly in comparison with those who underwent surgery (χ2 = 89.86; P < 0.001).

Table 1. Types of ovarian lesions and management strategies applied in each group
Final diagnosisLaparoscopy n (%)*Laparotomy n (%)*Expectant n (%)Total n
  • *

    Final diagnosis based on histological report.

  • Final diagnosis based on ultrasound appearance.

Dermoid16 (45.7)3 (8.6)16 (45.7)35
Cystadenoma5 (62.5)3 (37.5)8
Endometrioma3 (13.6)1 (4.6)18 (81.8)22
Cystadenofibroma2 (100)2
Fibroma1 (50)1 (50)2
Functional cyst2 (8.7)21 (91.3)23
Borderline2 (40)3 (60)5
Epithelial cancer8 (100)8
Non-epithelial cancer1 (100)1
Total31 (29.2)20 (18.9)55 (51.9)106

None of the women had an invasive tumor on one side and a non-invasive one on the other. FIGO staging and histological types of invasive cancers are listed in Table 2.

Table 2. Histology and stage of invasive ovarian cancer according to International Federation of Gynecology and Obstetrics guidelines
 Stage I n (%)Stage II n (%)Stage III n (%)Stage IV n (%)
Epithelial1 (12.5)3 (37.5)3 (37.5)1 (12.5)
Non-epithelial1 (100)

An RMI > 200 was found in 8/9 (88.9%) women with invasive ovarian cancer and in 8/92 (8.7%) of those with benign lesions. An RMI was < 200 in one case of invasive cancer, which was a Stage II clear cell carcinoma. None of the five women with borderline tumors had an RMI above that threshold. The OCS was absent in all nine cases of invasive ovarian cancer, 3/5 (60%) borderline tumors and 4/92 (4.3%) benign cysts.

False positive results with the RMI typically occurred in women with endometriotic cysts, while the OCS sign was more likely to provide a false positive result in women with benign cystadenomas (Table 3).

Table 3. False-positive cases in the diagnosis of invasive ovarian cancer by the risk of malignancy index (RMI) and ovarian crescent sign (OCS)
DiagnosisRMI > 200 (n)Negative OCS (n)
Functional ovarian cyst10
Dermoid01
Cystadenoma03
Endometrioma70
Borderline tumor03

An RMI > 200 diagnosed invasive ovarian malignancy with a sensitivity of 89% (95% CI, 57–98%), specificity of 92% (95% CI, 85–96%), PPV of 50%, NPV of 99%, LR + of 10.78 (95% CI, 5.34–21.77) and LR– of 0.12 (95% CI, 0.02–0.77). Absence of the OCS gave a sensitivity of 100% (95% CI, 70–100%), specificity of 93% (95% CI, 86–96%), PPV of 56%, NPV of 100%, LR + of 13.86 (95% CI, 6.79–28.29) and LR– of 0, which was significantly better than the RMI (P < 0.01). The area under the ROC curve for the OCS was 0.928 (SE 0.052, asymptotic 95% CI 0.826–1.030) vs. 0.822 (SE 0.081, asymptotic 95% CI 0.662–0.981) for RMI > 200 (Figure 2).

Figure 2.

Receiver–operating characteristics curve comparing the diagnostic performance of the risk of malignancy index (RMI) and the ovarian crescent sign (OCS) for the diagnosis of invasive ovarian cancer. Dashed line, RMI > 200; solid line, OCS.

Decision-tree analysis showed that the diagnosis of invasive ovarian malignancy could be further improved by applying the RMI and OCS in a sequential way. The best result was achieved by using RMI > 200 as a first node and the OCS as a second node. This eliminated all the false positive findings with only a single false negative diagnosis of invasive ovarian cancer in the whole study group (Figure 3).

Figure 3.

Decision-tree analysis for the diagnosis of invasive ovarian cancer based on the sequential use of the risk of malignancy index (RMI) and the ovarian crescent sign (OCS).

Discussion

Our study confirms previous reports5, 12 that showed that the RMI and OCS are fairly good methods for preoperative differentiation between invasive and non-invasive adnexal tumors. Both tests have proven to be very sensitive for the detection of invasive ovarian cancer. However, their false positive rates were well above 5%. False positive rates are particularly important when the tests are applied to low-risk populations diagnosed with ovarian tumors during opportunistic screening for ovarian abnormalities.

The RMI was particularly sensitive to elevations of serum CA125, which was the main reason for high readings in benign pathology, 87.5% of which had ovarian endometriomas. The OCS was typically absent in larger tumors such as cystadenomas and borderline tumors, but it was always easily detectable in endometriotic cysts. Neither the RMI nor the OCS was accurate enough in the diagnosis of ovarian borderline tumors. The RMI failed to diagnose all five borderline tumors in our study population, while the sensitivity of the OCS was only 60%. This is an expected result as borderline tumors exhibit different morphological and biochemical characteristics from invasive cancers and therefore they are unlikely to be detectable using tests that are primarily designed to detect invasive disease. There is some evidence that pattern recognition methods may help to correctly diagnose some borderline tumors17, 18, but this approach has not been prospectively tested as yet.

The RMI has been used in clinical practice for many years, while the OCS is a novel diagnostic test that has been used by a limited number of units. One of the important advantages of the RMI is its relative simplicity in the approach to the ultrasound assessment ovarian tumors. Rather than relying on complex ultrasound tests such as Doppler assessment of intraovarian blood flow or on the meticulous analysis of minute details of tumor architecture, the RMI simply divides ovarian cysts into unilocular and complex. This approach requires minimal ultrasound skills and it may be successfully applied in units without a high level of expertise in gynecological ultrasonography.

Many of the described ultrasound-based diagnostic models are too complex for use in daily practice and they may not be easily transportable between different ultrasound units6–9. The OCS overcomes this problem, and similarly to the RMI, it only requires a modest level of expertise in gynecological ultrasonography. It is therefore likely that the OCS may be successfully used in units with average ultrasound skills that are currently relying on the RMI.

Our study may be criticized for failing to obtain histological confirmation of the nature of the ovarian tumors in all the cases. However, in modern clinical practice it is not acceptable to operate on all asymptomatic women with an incidental ultrasound diagnosis of a benign looking cyst. Nevertheless, all women in our study had follow-up examinations, and non-surgical treatment was offered to women with cysts that were not increasing in size.

Our study shows that false positive results with the RMI and OCS occur in different types of ovarian tumors. Therefore a combination of the two tests, as shown in the decision-tree analysis, may be an effective way of reducing false positive findings.

This study could also be criticized for including a small number of invasive ovarian cancers, especially Stage I cancers, as well as a small number of borderline ovarian tumors. However, women who were included in our study attended for an ultrasound scan not exclusively because of a known ovarian tumor, and only a study performed at a regional cancer center over a long period of time would pick up a higher rate of invasive ovarian malignancies or borderline tumors.

Future work will show whether the proposed sequential use of the RMI followed by the OCS, either during the initial scan or within tertiary referral centers, may be a more effective way of managing women with an ultrasonographic diagnosis of ovarian tumors.

Ancillary