Can presurgical ultrasound predict survival in women with ovarian masses?

Abstract

Purpose:

Determine the ability of ultrasound to predict survival and detect more aggressive tumors in women with ovarian masses.

Materials and methods:

IRB approval was obtained. Total 167 patients who presented with adnexal mass/masses were included. These were documented as benign or malignant on ultrasound. Age, date of diagnosis and date of death, type of tumor and tumor marker CA-125 values were recorded. CA125 value < 35 H U/ml was considered normal. All cases underwent surgery. Pathologic findings were considered as reference standard. The 2×2 cross-tabulations were used to correlate dichotomized CA125, US diagnosis (benign vs malignant), and pathologic status. Difference of distributions was tested by Wilcoxon rank sum test, association tested by Fisher’s exact test. All tests were two-sided & p-values of ≤0.05 considered statistically significant. Kaplan-Meir curves were generated to estimate survival.

Results:

There was statistically significant difference in patients with benign versus malignant tumors based on pathology (p <0.0001) and ultrasound (p<0.0003). Sensitivity, specificity, positive predictive value, negative predictive value & accuracy of ultrasound was 55%, 86%, 90% and 46%, and 81%. Patients diagnosed as having malignant tumors based on ultrasound had statistically significant worse overall survival. Probability of survival based on pathological diagnosis of malignancy was statistically significant at p<0.0003, based on ultrasound at p<0.0001 & by CA125 at p< 0.041.

Conclusion:

Patients who had ultrasound-based prediction of ovarian malignancy, had overall worse survival probability (p<0.0001), compared to CA-125- or pathology-based prediction.

Introduction

Ovarian cancer is the 13^th leading cause of cancer-related death in the US, accounts for 1.3% of all new cancer cases in the US ¹ and is the leading cause of mortality from gynecologic cancers ². About 11.7 of 100,000 women are diagnosed with ovarian cancer per year, and 7.4 die from it ¹. Five-year survival may reach 90% if detected at FIGO stage 1 ³, but due to the non-specific symptoms of ovarian cancer, most women present with advanced stage ^4,5, with overall 5-year survival <46.5% according to the National Cancer Institute, Cancer Research UK, and an international survival study ^1,3,6. Efforts for decreasing mortality are guided toward finding an effective screening strategy ^7–11, and some recent trials have shown promising results. However, none of the expert groups or professional societies recommend screening for ovarian cancer ^12–14. The current approach for reducing mortality is early diagnosis and management.

Certain tumor biomarkers and behaviors have been shown to be associated with poor clinical outcomes ^15,16, thus indicating more aggressive tumors. It is possible that developing targeted therapies related to the specific biomarkers or considering aggressive treatments for these tumors could also reduce mortality. Recent studies have already shown that aggressive treatment of advanced stage cancers can reduce mortality ^17–19. The question arises as to how we can identify aggressive tumors that may benefit from aggressive treatment and/or who may also benefit from biomarkers screening? As ultrasound (US) is the initial modality of choice when woman presents with adnexal mass ^5,20. In this study we used simple US parameters derived from the models that are already known to be more successful in diagnosing ovarian cancer. We hypothesized that ultrasound imaging parameters would be a biomarker for detecting biologically aggressive ovarian cancers which in turn would correlate with decreased survival.

Materials and methods

After IRB approval, radiology database was searched from 2000–2016, for women undergoing grey-scale pelvic US with color-doppler for suspected adnexal mass or to rule out adnexal mass. It included all locally performed studies as well the outside studies uploaded in our system for review or interpretation. We excluded women who didn’t undergo surgery for mass removal. Total of 182 cases were randomly collected. Patient demographics, age, imaging and pathology findings, and CA125 values were recorded. Another 15 women were excluded because the preoperative US was not available for review in the database. A total of 167 ultrasounds were read by experienced radiologists who were blinded to the CA125 values and pathology results. Most of the images reviewed were acquired by transvaginal technique. If the mass could not be seen or was not seen in entirety on transvaginal ultrasound images, then transabdominal ultrasound images were also reviewed for complete information of the mass.

The masses were classified as benign or malignant based on different variables. If the masses were indeterminate, they were categorized as malignant for the purpose of data analysis to separate and avoid influencing the data from clearly benign ovarian masses. The variables collected and used to evaluate the mass include solid components >7mm, mural nodules, septa, septal thickness, vascularity of the solid components, nodular septa, purely solid tumor, acoustic shadow, ascites, and calcifications. A mass was considered malignant if it had any or combination of the following: solid component>7mm, intramural nodules with flow on Doppler, septal thickness >3mm, vascularity within the septa, or with associated ascites. These parameters were used since they are simple and reproducible. Cystic masses with shadowing echogenicity/calcifications, tip of the iceberg sign, hyperechoic lines and dots²¹ without obvious above features were diagnosed as mature teratomas and categorized benign. CA125 levels <35U/ml were considered normal.

Statistical analysis:

Summary statistics of CA125 values were provided in means, standard deviations, and range by US or pathologic status. CA125 distributions were compared by Wilcoxon rank sum test by malignant status according to pathology reports. Contingency table was used to correlate among dichotomized CA125, US, and pathologic status, and tested by Fisher’s exact test. The McNemar’s test was applied to test diagnostic accuracies based on CA125 alone and US alone. Accuracy, sensitivity, specificity, PPV, and NPV of ultrasound and CA125 were estimated using pathology as gold standard. Independent US parameters were cross tabulated with pathology status and significance calculated using Pearson’s chi-square test or Fisher’s Exact test.

Overall survival was estimated using Kaplan-Meier method and compared by log-rank test between benign and malignant lesions. Univariate Cox proportional hazards regression which contains only one explanatory variable was performed for every imaging explanatory variable, age, and continuous CA125. The proportionality assumption was checked for each explanatory variable in the model. Backward elimination steps starting with all significant explanatory variables (ascites, solid/vascular, septa/thick/3mm, complicated/ascites, and solid/GE/7mm) from the univariate models were implemented to get a final multivariate Cox model. All tests were two-sided and p-values of 0.05 or less were considered statistically significant. Statistical analysis was carried out using SAS version 9.2 (SAS Institute, Cary, NC).

Results

The mean age of women in our study was 50.6 years and age range was 16 to 83 years. Based on pathology reports, 29 (17%) cases had ovarian cancer, 138 (83%) cases had benign mass or no mass at all. At a cut off value of <35U/ml, CA125 diagnosis of malignant versus benign was significantly associated with pathology status (P value < 0.0001) with sensitivity and specificity of 52% and 87%, respectively. Ultrasound diagnosis of malignant versus benign was also significantly associated with pathology status (P value < 0.0001), with sensitivity, specificity, negative predictive value and positive predictive value of 55%, 86%, 90% and 46%, respectively. Overall accuracy of US was 81%. The individual variables that were significantly associated with malignant tumors are solid component >7mm (P <0.001 by Pearson’s chi-square test), vascularity of solid component (P=0.019 by Pearson’s chi-square test), ascites (P=0.032 by Fisher’s Exact test), purely solid and solid cystic structure (P = <0.001 by Fisher’s Exact test). About 52% (n=15/29) of malignant tumors had solid component >7mm compared to only 16% (n=22/137) of benign tumors. Of the solid components in malignant tumors, 81% (n=13/16) were vascular, compared to only 42% (n=8/19) in benign tumors. Ascites was found in 14% (n=4/29) of malignant and only 3% (n=4/138) of benign tumors. Other variables were also associated with malignancy, but the relationship didn’t reach statistical significance (table 1).

Table 1. Cross tabulations of nodularity, septations, and size with pathology.

Values presented as N (col %) in column Total and N (row %) in columns Benign and Malignant. P-values: a=Pearson’s chi-square test, b=Fisher’s Exact test. For example, out of total 166 patients (1 missing), 37 (22.3%) were solid component > 7mm, 15 (40.5%) of them were malignant, which was significantly greater than 10.9%, the percentage of malignant in solid component < 7mm patients.

Factor	Total	Benign	Malignant	p-value
	(N=167)	(N=138)	(N=29)
solid component >7mm*				<0.001 a
N	129(77.7)	115(89.1)	14(10.9)
Y	37(22.3)	22(59.5)	15(40.5)
solid component vascular*				0.019 a
N	14(40.0)	11(78.6)	3(21.4)
Y	21(60.0)	8(38.1)	13(61.9)
intramural nodules present				0.29a
N	143(85.6)	120(83.9)	23(16.1)
Y	24(14.4)	18(75.0)	6(25.0)
nodule vascular*				0.62b
N	18(75.0)	14(77.8)	4(22.2)
Y	6(25.0)	4(66.7)	2(33.3)
septa present				0.93a
N	122(73.1)	101(82.8)	21(17.2)
Y	45(26.9)	37(82.2)	8(17.8)
septa >2*				0.39b
N	16(36.4)	15(93.8)	1(6.3)
Y	28(63.6)	22(78.6)	6(21.4)
Septa thickness >3mm*				0.69b
N	27(61.4)	23(85.2)	4(14.8)
Y	17(38.6)	13(76.5)	4(23.5)
Septa thickness vascular*				0.26b
N	25(56.8)	22(88.0)	3(12.0)
Y	19(43.2)	14(73.7)	5(26.3)
ascites				0.032 b
N	159(95.2)	134(84.3)	25(15.7)
Y	8(4.8)	4(50.0)	4(50.0)
complicated ascites*				0.99b
N	5(62.5)	3(60.0)	2(40.0)
Y	3(37.5)	1(33.3)	2(66.7)
Pure solid*				<0.001 b
PC	48(37.5)	45(93.8)	3(6.3)
PS	14(10.9)	8(57.1)	6(42.9)
SC	43(33.6)	29(67.4)	14(32.6)
Sep C	23(18.0)	22(95.7)	1(4.3)
acoustic shadow*				0.99b
N	164(99.4)	135(82.3)	29(17.7)
Y	1(0.61)	1(100.0)	0(0.0)

^*Data not available for all subjects. Missing values: solid component >7mm = 1, solid component vascular = 132, nodule vascular = 143, septa >2 = 123, Septa thickness >3mm = 123, Septa vascular = 123, complicated ascites = 159, Pure solid = 39, acoustic shadow = 2.

The overall survival was 92% at 2 years and 86% at 5 years. Based on pathology, 2- and 5-year survival for malignant tumors was 84% and 60%, compared to 94% and 91% for benign lesions (P=0.031). Based on US diagnosis, 2- and 5-year survival for malignant tumors was 76% and 64% compared to 96% and 91% for benign lesions (P < 0.0001). At 10 years, survival was 60% for pathology diagnosed malignant tumors, compared to 25% for US-diagnosed malignancies regardless of pathology status (table 2).

Table 2.

Univariate overall survival analysis results for survival time from the surgery date.

Variable	Level	N	Event	Median (95%CI)	Rate at 2 Years (95%CI)	Rate at 5 Years (95%CI)	P-value
All Patients		160	20	NA (NA, NA)	0.92 (0.87, 0.97)	0.86 (0.79, 0.93)
Ultrasound	Benign	128	9	NA (NA, NA)	0.96 (0.92, 1)	0.91 (0.84, 0.98)	<0.0001
	Malignant	32	11	7.4 (2.7, NA)	0.76 (0.61, 0.95)	0.64 (0.47, 0.89)
Pathology	Benign	133	13	NA (NA, NA)	0.94 (0.9, 0.99)	0.91 (0.85, 0.97)	0.031
	Malignant	27	7	NA (3.9, NA)	0.84 (0.7, 1)	0.60 (0.39, 0.94)
CA125	Benign	130	14	NA (NA, NA)	0.92 (0.87, 0.98)	0.89 (0.83, 0.96)	0.174
	Malignant	30	6	NA (3.9, NA)	0.92 (0.81, 1)	0.67 (0.44, 1)

P-values by log-rank test. N is the total number of patients, event is the number of death. Median (95% CI) is the median survival (year) and its corresponding 95% CI. Rate at × years (95% CI) is the survival rate after × years with its corresponding 95% CI.

Analysis of survival relationship with individual US variables has revealed that solid component >7mm, and ascites are the only variables with statistically significant association with poor survival (P= 0.038 and <0.0001, respectively). When solid component >7mm is present, 5- and 10-year survival was around 75% and 55% compared to 90% and 75% when it’s not present. For ascites, survival was 43% and 0%, respectively, compared to 90% and 78% without ascites. Association with other variables did not reach statistical significance.

Univariate Cox proportional hazard model concluded that 5 variables (ascites, solid component/vascular, septa/thick/3mm, complicated ascites, and solid/GE/7mm) were significantly associated with poor survival, however, the backward elimination process resulted in a multivariate Cox model with only two significant variables, solid component/vascular and ascites. After continuous variables like age and CA125 were incorporated to the model, CA-125 increase of 10 units were also included with small hazard ratio of 1.004 (p=0.0231).

Discussion

Ovarian cancer is the leading cause of death from gynecologic caner, and the 13th leading cause of cancer death in the US ^1,2. Despite advancement in detection and management, the mortality remains high with 5-year survival <46.5% ¹. Due to the low incidence of ovarian cancer in the general population, developing an effective screening method was faced by many challenges; a specificity high enough (>99.6) is needed to achieve PPV of at least 10%, which is translated to 10 surgeries performed for each cancer detected successfully ⁷. The observation of progressively rising CA125 levels in women with ovarian cancer compared to stable or declining levels of falsely elevated CA125 has led to the development of the risk of ovarian cancer algorithm (ROCA), which achieved specificity and PPV of 99.8% and 19% ^4,22,23. This was very promising and was utilized in a multimodal approach in combination with secondary US in the biggest ovarian cancer screening trial to date, the UK Collaborative trial of Ovarian Cancer Screening (UKCTOCS), which could achieve a specificity and PPV of 99.8% and 35.1%, respectively ^24,25. To recommend a screening strategy however, a significant mortality reduction must be demonstrated, the UKCTOCS results showed only 15% reduction in mortality, with more significant reduction in the years 7–14 of the screening ^24,25. As promising as it may sound, until today, none of the professional societies have recommended screening for ovarian cancer due to minimal impact on mortality based on available parameters so far ^12–14. Currently, early diagnosis and management of women suspected to have ovarian cancer remains the priority to reduce mortality. Once the cancer is diagnosed, effective treatment is the only hope. Studies have shown that aggressive treatment of advanced stage cancers can reduce mortality ^17–19. By the same token, aggressive treatment of the more aggressive tumors may also help reduce mortality. Certain tumor biomarkers like SNAI2 and estrogen receptors have been strongly correlated with poor outcome and virulent behavior ^15,16. Our approach to impact mortality is from US features based detection of more aggressive cancers. In the future, such aggressive ovarian tumors may be considered for aggressive treatment based on US features alone and/or US may give a lead towards screening for above biomarkers and receive different treatments or even targeted therapies accordingly if such associations are found.

For preoperative diagnosis of adnexal masses, US is the main diagnostic modality used in daily practice ^5,20, with combined grey-scale US and color doppler being the most effective method ^26–28. We derived the US variables for our study, from the models that are simple enough for daily use and have been shown to be some of the most successful in the literature for preoperative diagnosis of ovarian malignancy. Those include the simple rules (SR) and logistic regression 2 (LR2) models ^29–31 which were developed and validated by The International Ovarian Tumor Analysis (IOTA) group, and also shown in a meta-analysis study to have the pooled sensitivity and specificity as 93%, 81%, 92% and 83% respectively. In addition, we used variables that are consistently associated with malignancy in the literature, derived from many different studies and morphologic scoring systems ^32–38. To ensure internal control in our study, and to ensure that our results are as accurate as possible when compared to the literature, we also obtained CA125 and pathology status and compared with the US parameters which we selected for the study. Using these parameters, US was significantly associated with pathology status (P< 0.0001) with an overall accuracy of 81%. This is consistent with the literature, although direct comparison with other US accuracy studies was not possible due to various techniques and approaches used. Of the US variables we used; solid component >7mm, blood flow in the solid component, ascites, purely solid structure, and solid cystic structure were significantly associated with malignancy status. Similar to others in literature, solid component was the most predictive of malignancy ³⁹, and flow in the solid component is the most important doppler feature (fig. 1 and 2) ⁴⁰. Solid component and vascular distribution had the most diagnostic impact in Alcazar study. All the morphologic scoring systems except Alcazar included septa in their models, however the presence of septa wasn’t associated with malignancy in our study (fig. 3 and 4). Neither thickness nor vascularity of the septa predicted malignancy. Weber et al (1998) also considered septal thickness as insignificant. Many times, more than two thick and vascular septa have led us to give malignant diagnosis for tumors that were in fact benign cystadenoma, cystadenofibroma, endometriotic cyst, and Brenner’s tumor in our study. Those were the source of false positives in our study, as well as many other studies in the literature ³⁸.

Figure 1:

A 59-year-old female with crampy abdominal pain & postmenopausal bleeding. Gray scale and color ultrasound images through the pelvis, shows a complex left ovarian mass with large vascular isoechoic solid component (arrows) & small anechoic cystic focus (arrowhead), diagnosed as malignancy based on US features. CA125 was 187.8. Following surgery, the mass was proven to be high grade ovarian cancer on pathology.

Figure 2:

A 51 years old woman with lower abdominal pain & fatigue. Gray scale and color ultrasound images through the pelvis demonstrated right adnexal complex multiseptate cystic mass with thick septae (arrow) and large solid components (arrowhead) with abnormal vascularity, diagnosed as ovarian malignancy based on US. Pathology diagnosis was malignant mixed mullerian tumor of the right ovary following surgery.

Figure 3:

A 77-year-old woman with lower abdominal pain. Gray scale and color ultrasound through the pelvis shows showed right ovarian multiseptate cystic mass with thin septae (arrow) and mural nodule (arrow head) without abnormal vascularity. This was diagnosed as benign, likely a cystadenoma based on US features. CA125 was 8.9. Following surgery, this was proven to be serous cystadenoma on pathology.

Figure 4:

A 47-year-old woman with family history of breast cancer and abnormal pelvic exam. Gray scale and color ultrasound images through the pelvis showed a right ovarian multiloculated cystic mass (arrow heads) with homogenous varying echogenicity & thin septae (arrow) without abnormal vascularity, diagnosed as benign, likely endometrioma based on US features. Pathology revealed mucinous cystadenoma after surgery.

To prove the ability of US to detect more aggressive cancers, we have to demonstrate a direct relationship between US diagnosis of malignancy and poor survival. In our study, when US diagnosis was malignant, survival was significantly decreased compared to pathologic diagnosis. The 2- and 10-year survivals with US diagnosis of malignancy were 76% and 25%, compared to 84% and 60% with pathologic diagnosis of malignancy. In other words, US predicted poorer survival than pathology. Our suggested reason could be that US detects frankly malignant lesions, while pathology can detect subtle foci or early cancer within a mass. As we discussed above, the significance of such finding is that if tumors detected on US were more aggressive and impactful, then maybe more aggressive management should be directed towards or is needed for these patients to reduce mortality. Also, association of these US findings with the poor prognostic biomarkers in ovarian cancers may be studied, or screening for these biomarkers may be done based on US in such cases, to help in developing and utilization of targeted therapies. In respect to the univariate analysis, only solid component >7mm, and ascites were significantly associated with poor survival. CA125 was also associated with poor survival, but the association didn’t reach statistical significance (P=0.174). CA125 has been extensively studied as a predictor of poor survival, and many studies demonstrated strong poor prognostic role of higher CA125 levels in the pre- and post-operative period, and the pre- and post-chemotherapy period ^2,7. The CA125 half-life was also an independent prognostic factor in many studies ². Without relative risk associated with certain values, it’s still unknown how to predict survival using CA125.

Based on the findings from our study, we strongly believe that US combined with other tools like CA125 should be further studied to determine a formal method of predicting survival and detecting aggressive tumors to categorize patients with ovarian cancer so that patients can receive a treatment plan based on their risk stratification. It may also be useful in the future to analyze the changes in the primary tumor with ultrasound during treatment, by assessing changes in the solid components and disappearance of ascites.

Our study has some limitations; first, it’s a retrospective study with small only 182 cases, and low incidence of cancer within our group. In addition, our institution is a quaternary cancer center, thus there could’ve been a selection bias for high risk patients. Some ultrasounds were performed at outside facilities. Though one may argue that ultrasound performs best in the hands of an expert radiologist. The aim of our study was to find a simple approach that is not highly dependent on expertise, as some of the ultrasounds may be performed by technologists and read by a radiologist remotely. These simple ultrasound parameters can be used also in remote places to make diagnosis where expertise is not available.

Conclusion

Ultrasound seems to be a more useful predictor of poor survival than pathology or CA125. This means USS can detect more aggressive tumors, these tumors may benefit from more aggressive treatment approach, which can potentially reduce mortality. Also, as poor outcome was correlated with certain tumor biomarkers that could be screened for or utilized for developing targeted therapy, US could be potentially helpful pointing towards those cancers, screening for such biomarkers or detect the aggressive cancers on its own. Solid component >7mm and ascites were the only US parameters more significantly associated with poor survival in our study. These parameters may be used to assess the aggressiveness of ovarian tumors, and thus may help customize the treatment or management plan according to the risk stratification.