Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online before print June 13, 2005, 10.1148/radiol.2361041618
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2361041618v1
236/1/85    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Kinkel, K.
Right arrow Articles by Hricak, H.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Kinkel, K.
Right arrow Articles by Hricak, H.
(Radiology 2005;236:85-94.)
© RSNA, 2005

Evidence-based Practice

Indeterminate Ovarian Mass at US: Incremental Value of Second Imaging Test for Characterization—Meta-Analysis and Bayesian Analysis1

Karen Kinkel, MD, PD, Ying Lu, PhD, Amir Mehdizade, MD, Marie-Françoise Pelte, MD and Hedvig Hricak, MD, PhD

1 From the Departments of Radiology (K.K.), Gynecology and Obstetrics (K.K.), and Clinical Pathology (M.F.P.), University Hospital Geneva, Geneva, Switzerland; Department of Epidemiology and Biostatistics, University of California, San Francisco (Y.L.); and Department of Radiology, Memorial Sloan-Kettering Cancer Center, New York, NY (A.M., H.H.). Received September 19, 2004; revision requested October 25; revision received December 4; accepted December 10. Supported in part by Fondation des Grangettes, Chêne-Bougeries, Switzerland. Address correspondence to K.K., Institut de Radiologie, Clinique des Grangettes, Chemin des Grangettes 7, CH 1224 Chêne-Bougeries, Canton de Genève, Switzerland (e-mail: karen.kinkel{at}grangettes.ch).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
PURPOSE: To compare value of current diagnostic strategies in assessment of changes in posttest probability of ovarian cancer when menopausal status and combination and sequence of diagnostic imaging tests are considered.

MATERIALS AND METHODS: Prevalence of ovarian cancer according to menopausal status in women with an ovarian mass and performance of combined gray-scale and Doppler ultrasonography (US), computed tomography (CT), and nonenhanced magnetic resonance (MR) imaging and contrast material–enhanced MR imaging after indeterminate results at gray-scale US were derived from meta-analysis by using MEDLINE database and institutional data. Study was approved by the institutional review board of University Hospital Geneva, Geneva, Switzerland; informed consent was waived. Posttest probability values were computed through Bayesian analysis and Monte Carlo simulation after initial gray-scale US and secondary combined gray-scale and Doppler US, CT, or MR imaging, while dependence of test results among imaging modalities was considered. Changes in posttest probability were compared among imaging modalities with summary receiver operating characteristic curves.

RESULTS: Prevalence of ovarian cancer was 8.75% in premenopausal women and 32.40% in postmenopausal women with an ovarian mass. After characterization with initial gray-scale US, posttest probability in pre- and postmenopausal women changed, respectively, to 25% and 63% for indeterminate results and to 2% and 7% for benign results. Subsequent use of combined gray-scale and Doppler US, CT, or MR imaging had significant higher positive and lower negative posttest probability than did use of gray-scale US alone. In women with an indeterminate initial US result, posttest probability decreased after secondary testing with benign results for all imaging modalities to 2% in premenopausal women and to 8%–10% in postmenopausal women. After secondary testing for suspicious lesions, posttest probability increased more after nonenhanced (premenopausal women, 70%; postmenopausal women, 92%) or contrast-enhanced MR imaging (premenopausal women, 80%; postmenopausal women, 95%) than it did after combined gray-scale and Doppler US (premenopausal women, 30%; postmenopausal women, 69%) or CT (premenopausal women, 38%; postmenopausal women, 76%) (P < .001).

CONCLUSION: In women with an indeterminate ovarian mass at gray-scale US, MR imaging results contributed to change in probability of ovarian cancer in both pre- and postmenopausal women more than did CT or combined gray-scale and Doppler US results.

© RSNA, 2005


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
The presence of an ovarian mass is one of the leading indications for gynecologic surgery. Ultrasonography (US) is considered the primary imaging modality for confirmation of the ovarian origin of the mass and characterization of the nature of the mass as benign or malignant (1). Often, the initial US examination is performed by a gynecologist trained only in basic gray-scale US. Because the treatment of ovarian cancer differs considerably from the treatment of a benign ovarian mass, the gynecologist tries to increase the diagnostic confidence of the malignant nature of the mass to allow optimal scheduling of the surgical intervention or a consultation with an oncologist prior to surgery. The purpose of secondary imaging after an initial indeterminate US result is therefore to characterize the mass and to assess the extent of possible malignant disease.

Because the performance of a combination of US techniques is better than the performance of basic gray-scale US alone (2), the gynecologist most often sends the patient to a physician who specializes in US and was trained in combined gray-scale and color Doppler imaging. Another option is to send the patient to a physician trained in computed tomography (CT) or magnetic resonance (MR) imaging of the body (3). Best practice guidelines for diagnostic evaluation are of great interest to the medical community. Often, however, they are based on experience or habits rather than on evidence. As far as we know, there have been no studies in which an analysis of the interdependence of test results between a primary and a secondary imaging modality or an analysis of the incremental value of a second imaging modality has been performed. To our knowledge, in no study has the posttest probability of ovarian cancer been calculated when a combination and sequence of diagnostic tests are considered. Since similar criteria are used to diagnose ovarian cancer with all imaging modalities, the incremental value of a second imaging modality might be very small. Moreover, the prior or pretest probability of ovarian cancer changes considerably according to menopausal status, which modifies the final posttest probability of ovarian cancer and affects diagnosis and treatment planning. Thus, the purpose of this study was to compare the value of current diagnostic strategies in the assessment of changes in posttest probability of ovarian cancer when menopausal status and the combination and sequence of diagnostic imaging tests are considered.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 References
 
Study Design
The pretest probability of ovarian cancer in pre- and postmenopausal patients with ovarian masses was assessed with meta-analysis and institutional data. The results of initial gray-scale US in ovarian lesion characterization were updated from a published meta-analysis in which standard meta-analytic methods were used (4). Meta-analysis was used to evaluate the performance of combined gray-scale and Doppler US, CT, and nonenhanced or contrast material–enhanced MR imaging after initial gray-scale US with indeterminate results. Posttest probability of ovarian cancer after initial gray-scale US and after a second imaging test with combined gray-scale and Doppler US, CT, or nonenhanced or contrast-enhanced MR imaging were obtained through Bayesian analysis and Monte Carlo simulation. Summary receiver operating characteristic curves were used to compare posttest probabilities.

Prevalence of Ovarian Cancer in Women with Ovarian Mass
A literature search of English-language abstracted studies that involved human subjects was performed by two physicians (K.K., A.M.) by using the MEDLINE database to determine the prevalence of ovarian cancer in women with an ovarian mass according to menopausal status. Two search algorithms were applied by each physician: "ovarian neoplasm or ovarian cyst or adnexal mass or adnexal lesion and epidemiology and menopause" and the combination of the keywords "ovarian neoplasm or ovarian cyst or adnexal mass or adnexal lesion" and "ultrasonography or sonography or ultrasound."

A total of 83 articles were found for the assessment of the prevalence of ovarian cancer in women with ovarian masses. To include unpublished data, the database of the Department of Clinical Pathology of the University Hospital Geneva, Geneva, Switzerland, was searched with the keyword ovarian cyst for January 1999 to December 2002 (M.F.P.). Clinical and surgical chart review were used to assess the indication for surgery and the menopausal status (A.M.). The study was approved by the institutional review board of the University Hospital Geneva, and informed consent was waived.

Only patients with a known ovarian mass prior to surgery were included. Patients with ovarian cysts discovered incidentally during surgery performed for another reason were excluded. Postmenopausal status was defined as the absence of menstruation during the past 12 months or, in patients with prior hysterectomy, persistent climatic symptoms such as hot flashes.

Included studies met the following criteria: (a) Patients had an adnexal mass that was not discovered during screening. (b) The reference standard was histopathologic findings. (c) Histopathologic data presented were detailed separately for pre- and postmenopausal women.

Two readers (K.K., A.M.) with 12 years and 2 years, respectively, of experience in reading images of adnexal masses each abstracted the data from each article. The data included author and year of publication, sample size, number of malignant ovarian neoplasms (defined as primary, secondary, or borderline neoplasms), and the percentage of neoplasms classified as stage III or IV ovarian cancer as defined according to the International Federation of Gynecology and Obstetrics guidelines for staging. Disagreement was resolved with consensus. The readers were not blinded to the origin of publication, the journal, or the year of publication. The weighted mean percentage of women with ovarian cancer, calculated by using the inverse of the variances, was used to determine the pretest probability of ovarian cancer in pre- and postmenopausal women with ovarian masses.

Sensitivity and Specificity of Ovarian Mass Characterization
A literature search of English-language abstracted studies that involved human subjects was performed by two physicians (K.K., A.M.) by using the MEDLINE database and a search algorithm including the keywords "ovarian neoplasm or ovarian cyst or adnexal mass or adnexal lesion" and "ultrasound or sonography or computed tomography or CT or magnetic resonance imaging or MRI." This search was undertaken to determine sensitivity and specificity of ovarian mass characterization with initial gray-scale US and after combined gray-scale and Doppler US, CT, or MR imaging. When we considered the time of clinical introduction of each modality, the searches were designed to include studies that occurred from December 1980 to December 2002 for gray-scale US and CT and from December 1984 to December 2002 for MR imaging and combined gray-scale and Doppler US (US with combination of gray-scale assessment with color Doppler information or Doppler index measurements). Review articles, comments, case reports, and articles without original data were excluded. Additional relevant articles were found through manual checking of the reference lists of retrieved articles.

The studies included in the meta-analysis met the following criteria: (a) Patients had an ovarian mass that was not discovered during screening for ovarian cancer. (b) The reference standard was histopathologic findings. (c) Interpreters of imaging data were blinded to histopathologic findings. (d) The data presented allowed calculation of true-positive, true-negative, false-positive, and false-negative secondary imaging results for either positive or negative results at initial gray-scale US. (e) The data or subsets of data had not been published more than once.

Two readers (K.K., A.M.) each abstracted the data from each article. When disagreement occurred, a third reader (H.H.) reviewed the article, and the disagreement was resolved in consensus. The readers were not blinded to the author, the journal, or the year of publication. The following data were recorded for each article: (a) author and year of publication; (b) sample size; (c) imaging modality and technique (CT or nonenhanced or contrast-enhanced MR imaging); and (d) true-positive, false-negative, true-negative, and false-positive findings for the characterization of a malignant ovarian neoplasm versus a benign ovarian neoplasm. In some studies, only a subgroup of patients fulfilled the inclusion criteria. If data for more than one imaging modality and technique were presented in a study, inclusion and exclusion criteria were applied to each technique.

For gray-scale US, references to studies from a previously published meta-analysis (2) were included. In addition, three references to studies (four data sets) in which gray-scale US was used for ovarian lesion characterization in 758 patients (5–7) and that were published between January 1999 and December 2002 were included. These studies fulfilled the inclusion criterion for allowing calculation of sensitivity and specificity values. Therefore, the total number of patients examined for ovarian lesion characterization increased from 3377 to 4135 for the assessment of gray-scale US alone. For combined gray-scale and Doppler US, references to studies from the previously published meta-analysis were included. In addition, three references to studies (four data sets) in which combined gray-scale and Doppler US were used in 859 patients (5–8) and that were published between January 1999 and December 2002 were included. Among the 11 references to studies in which combined gray-scale and Doppler US were used (5–15), eight studies (5–7,9–13) with a total of 1529 patients in which results of prior gray-scale US were indicated were used to assess sensitivity and specificity in patients with an indeterminate result at gray-scale US.

Table 1 shows the number of references to studies that fulfilled the inclusion criteria and specifies the reasons for exclusion for CT and MR imaging characterization of ovarian masses. For CT, the main reason for exclusion was the absence of lesion characterization, since a majority of articles were about staging. References to only three studies fulfilled the inclusion criteria (16–18), and in these studies, data in a total of 161 lesions in 161 patients were provided. For MR imaging, in 11 articles (17–27) that were included, four data sets for nonenhanced MR imaging (369 patients) and ten data sets for contrast-enhanced MR imaging (773 patients) were provided. Table 2 indicates the number of lesions and the sensitivity and specificity of each included CT and nonenhanced and contrast-enhanced MR imaging data set.


View this table:
[in this window]
[in a new window]

  		TABLE 1. Results of MEDLINE Search and Study Selection for CT and MR Imaging Characterization of Ovarian Masses

 

View this table:
[in this window]
[in a new window]

  		TABLE 2. Characterization of Ovarian Masses at CT and MR Imaging with Prior US Results

 
Statistical Analysis
The meta-analytic method used in the study was based on the weighted average of prevalence, as well as the mean sensitivity and specificity values that were based on the summary receiver operating characteristic curves for meta-analysis (28). Statistical analyses were performed by using statistical software packages (SAS, version 8.2, SAS Institute, Cary, NC; S-plus, Insightful, Seattle, Wash). A goodness-of-fit test, which was based on the analysis of deviance for generalized linear models (29), was used to evaluate the appropriateness of using the summary receiver operating characteristic curve analysis (ie, the appropriateness of the logit-transformed linear relationship, D = a + bS, with D = log[TPR/FNR] – log[FPR/TNR] and S = log[TPR/FNR] + log[FPR/TNR], where TPR signifies true-positive results; FNR, false-negative results; FPR, false-positive results; and TNR, true-negative results). Our data supported this linear relationship for all the modalities, and summary receiver operating characteristic curve analysis was used in the summarization of the utility of all techniques.

With conventional summary receiver operating characteristic curve analysis, the Q* point (the point where sensitivity equals specificity) is used as the summary of performance. The Q* points for most modalities, however, were not at the center of the observed sensitivity and specificity. As a result, Q* points were not a proper summary of diagnostic performance. To accurately describe diagnostic performance, particularly of CT and MR imaging as the secondary diagnostic tools, we used the mean point M to summarize sensitivity and specificity of techniques. In the conventional summary receiver operating characteristic curve analysis, there is a linear relationship between D and S. The center point for this transformed linear regression line is at mean S and mean D. We transformed this mean point on the regression line to the summary receiver operating characteristic curve to obtain the mean sensitivity and specificity. The mean sensitivity and specificity are always at the center of the observed data and are used as summary statistics for each diagnostic test.

Unlike the Q* point, the estimated mean sensitivity and specificity have no explicit formula for estimation of their variance. We used the Monte Carlo method (30). We assumed that each reported value for sensitivity, specificity, and prevalence was random samples generated from the corresponding binomial distributions. First, we randomly selected studies to be included in the experiments with replacements (ie, all studies were the candidates in each random selection regardless of whether it was selected previously). Once we had the same number of studies as had been reported in the large sample, we generated random samples of positive cases, as well as true-positive, false-negative, false-positive, and true-negative cases, according to the corresponding binomial distribution and observed sample size in the included studies. Random samples from each simulation experiment were used to estimate the variations of mean sensitivity and specificity according to the summary receiver operating characteristic curves. We performed 2000 simulations to calculate the variances, as well as the empirical 95% confidence intervals (CIs) for sensitivity and specificity.

For determination of the posttest probability of second imaging tests, we used the Monte Carlo method to generate samples on the basis of the prevalence of ovarian cancer and the joint distribution of initial gray-scale US and subsequent imaging tests (gray-scale and Doppler US or body imaging [defined as gray-scale US followed by CT or nonenhanced or contrast-enhanced MR imaging]) reported in the reviewed articles (30). These Monte Carlo samples were used to estimate variances of sensitivity and specificity of a second imaging test on the basis of positive gray-scale US results. Our approach is similar to the Efron bootstrap approach for estimation of empirical variances, except that we generated random samples from the empirical distributions instead of resampling the original data with replacements (30). A covariate adjustment analysis was performed, as previously reported, to identify whether imaging results were affected by the year of publication of the study (2).

Bayesian Analysis and Clinical Utility
Positive and negative likelihood ratios (LRs) characterize the clinical utility of a test and are used to estimate the posttest probability of disease by means of Bayesian analysis. LRs for gray-scale US with positive or negative findings were obtained from the average weighted mean of sensitivity and specificity of this meta-analysis. The formula for a positive LR is sensitivity/(1 – specificity); the formula for a negative LR is (1 – sensitivity)/specificity.

The posttest probability of ovarian cancer was calculated for premenopausal and postmenopausal patients with ovarian masses. The weighted mean percentage of the prevalence of ovarian cancer in pre- and postmenopausal patients with ovarian masses was used as the pretest probability of ovarian cancer. The posttest probability was calculated first by converting the pretest probability into pretest odds by using the following equation: odds = probability/(1 – probability). Next, posttest odds were calculated by multiplying together the pretest odds and the LR. The posttest odds were converted into probability by using the following equation: probability = odds/(odds + 1). Since an LR of 1.0 indicates that the posttest probability is not greater than the pretest probability, clinically useful tests should have a high positive LR (>5.0, "good for confirming disease") and a low negative LR (<0.2, "good for ruling out disease") (31).

When more than one imaging test was applied, the second test was only applied in subjects who had positive results from the first test. Conditional sensitivity and specificity of the second test are the sensitivity and specificity for these subjects with positive results and were estimated through meta-analysis of studies with conditional information. The posttest probability was computed by using combined LR derived from combined sensitivity (sensitivity of method 1 multiplied by conditional sensitivity of method 2 for subjects with positive results according to method 1) and specificity (specificity of method 1 plus [1 minus specificity of method 1] times conditional specificity of method 2 for subjects with positive results according to method 1). Thus, we did not assume independence of the two tests. The Monte Carlo method was used again to estimate the empirical 95% CIs for Bayesian LR and positive and negative posttest probabilities. Different from a general Bayesian approach, our analysis only used the Bayes formula in the derivation of the posttest probability. We had an estimate of the posttest probability that was based on summarized data in previous studies. Therefore, a P value is appropriately used to assess the chance of nonzero differences between two posttest probability values.

Comparison of Imaging Strategies after Initial Indeterminate Gray-Scale US Results
Comparisons of different imaging strategies after initial indeterminate gray-scale US results were based on the positive and negative combined posttest probability values. Both estimates, as well as the significance level of differences, were evaluated by using the Monte Carlo method. For comparisons of strategies, all simulated samples were generated by the joint distribution of paired modalities to consider the effect of a positive correlation between two imaging modalities. These joint multinomial distributions were derived in two steps. First, we used the observed marginal distributions for available studies. Second, we estimated the correlation on the basis of studies with multiple modalities. When we combined the information, we obtained joint distributions and generated samples accordingly for each study with multiple modalities. We then evaluated differences in probability according to sample estimates and obtained empirical two-sided P values that were based on the null hypothesis of a zero difference. We repeated the experiment 2000 times. We calculated proportions of differences in posttest probability values between two strategies that were either less than or greater than zero. We used the relationship between 95% CIs and P values to derive two-sided P values. A two-sided P value for one parameter will be calculated as twice the smaller proportion of positive and negative differences in values for that parameter among all experiments.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 References
 
Prevalence of Ovarian Cancer according to Menopausal Status
Nine references to studies fulfilled the inclusion criteria for the prevalence of ovarian cancer in patients with a persistent ovarian mass according to menopausal status (6,32–39) (Table 3). In addition, we included the clinical, surgical, and histologic findings in patients with a persistent ovarian mass from our institutional database. Among the 400 patients who underwent surgery at our institution between 1999 and 2002, 285 (71%) were in the premenopausal group and 115 (29%) were in the postmenopausal group. The number of ovarian cancers in this population was 14 (5%) of 285 in the premenopausal group (International Federation of Gynecology and Obstetrics stage III or IV in six [43%] of 14) and 42 (37%) of 115 in the postmenopausal group (International Federation of Gynecology and Obstetrics stage III or IV in 26 [62%] of 42).


View this table:
[in this window]
[in a new window]

  		TABLE 3. Data Abstraction for Prevalence of Ovarian Cancer in Patients with Ovarian Mass according to Menopausal Status

 
Among the total of 2827 patients (1458 premenopausal women and 1369 postmenopausal women) with a diagnosis of an ovarian mass, the number of women with ovarian cancers was 149 in the premenopausal patient group and 460 in the postmenopausal patient group. The weighted mean percentage of ovarian cancer in women with persistent ovarian masses differed according to menopausal status (P < .001) and was 8.75% (95% CI: 6%, 11%) in premenopausal women and 32.40% (95% CI: 29%, 35%) in postmenopausal women. The weighted mean percentages of International Federation of Gynecology and Obstetrics stage III or IV masses was lower in premenopausal women (33.11%; 95% CI: 21%, 45%) than it was in postmenopausal women (54.80%; 95% CI: 35%, 74%) (P < .001).

Posttest Probability of Ovarian Cancer
After gray-scale US in patients with an ovarian mass and no second imaging test.—Simulated means of sensitivity and specificity with 95% CIs of gray-scale US were 87% (95% CI: 85%, 90%) and 75% (95% CI: 72%, 78%), respectively. LRs were 3.52 (95% CI: 3.16, 3.92) for positive test results and 0.17 (95% CI: 0.14, 0.2) for negative test results. In premenopausal patients with an ovarian mass and a pretest probability of ovarian cancer of approximately 9%, the posttest probability increased to 25% (95% CI: 19%, 30%) after an indeterminate result at gray-scale US, whereas the probability decreased to 2% (95% CI: 1%, 2%) after a benign result. In postmenopausal patients with an ovarian mass and a pretest probability of ovarian cancer of about 32%, the posttest probability increased to 63% (95% CI: 58%, 67%) after an indeterminate result and decreased to 7% (95% CI: 6%, 9%) after a benign result. Because of the low posttest probability in both pre- and postmenopausal women after a benign gray-scale US assessment, results of secondary imaging tests were not evaluated in these patient groups.

After indeterminate gray-scale US results followed by combined gray-scale and Doppler US.—The distribution of true-positive, false-negative, true-negative, and false-positive results of combined gray-scale and Doppler US are summarized and detailed according to initial gray-scale US results in the first two rows (correctly classified and misclassified with combined gray-scale and Doppler US) of data in Table 4. The second (correctly classified with positive status) and the last (misclassified with negative status) columns in Table 4, concerning ovarian lesions with indeterminate results at initial gray-scale US, represent true-positive and false-positive findings at initial gray-scale US. Findings at combined gray-scale and Doppler US confirmed 328 (98%) of 336 true-positive findings at initial gray-scale US, but eight true-positive findings were converted to false-negative findings (eight [2%] of 336). Conversely, among the 269 false-positive findings at gray-scale US, 121 (45%) false-positive findings were converted to true-negative results and 148 (55%) false-positive results were confirmed at combined gray-scale and Doppler US.


View this table:
[in this window]
[in a new window]

  		TABLE 4. Imaging Results according to Final Histopathologic Findings and Initial Gray-Scale US Results in Women with Ovarian Masses

 
Simulated means of sensitivity and specificity for initial indeterminate results at gray-scale US followed by combined gray-scale and Doppler US were 84% (95% CI: 81%, 87%) and 82% (95% CI: 79%, 85%), respectively. LRs were 4.69 (95% CI: 3.98, 5.65) for positive and 0.19 (95% CI: 0.16, 0.23) for negative test results (Table 5). In premenopausal patients with an indeterminate ovarian mass at gray-scale US and a probability of ovarian cancer of approximately 25%, the posttest probability increased to 30% (95% CI: 23%, 37%) after an indeterminate result at combined gray-scale and Doppler US and decreased to 2% (95% CI: 1%, 2%) after a benign result. In postmenopausal patients with an indeterminate ovarian mass at gray-scale US and a probability of ovarian cancer of approximately 63%, the posttest probability increased to 69% (95% CI: 65%, 74%) after an indeterminate result at combined gray-scale and Doppler US and decreased to 8% (95% CI: 7%, 10%) after a benign result.


View this table:
[in this window]
[in a new window]

  		TABLE 5. Performance of Imaging Modalities for Ovarian Lesion Characterization after Indeterminate Gray-Scale US Results

 
After indeterminate gray-scale US result followed by body imaging.—Table 4 indicates total numbers of true-positive, false-negative, true-negative, and false-positive findings for CT and nonenhanced and contrast-enhanced MR imaging according to prior gray-scale US results. For true-positive findings at initial gray-scale US, subsequent CT or nonenhanced or contrast-enhanced MR imaging results confirmed the indeterminate test results in 98% (54 of 55), 89% (140 of 158), and 95% (278 of 294) of lesions, respectively. For false-positive findings at initial gray-scale US, the indeterminate test result was converted into a benign test result in 72% (13 of 18), 78% (114 of 147), and 75% (181 of 241) of lesions, respectively, at subsequent CT or nonenhanced or contrast-enhanced MR imaging.

Table 5 indicates simulated means of sensitivity, specificity, positive and negative LRs, and posttest probability values for the combination of initial indeterminate gray-scale US results followed by subsequent CT or nonenhanced or contrast-enhanced MR imaging. The posttest probability values for a benign test result after indeterminate gray-scale US in premenopausal women decreased to 2% with any other second imaging modality and ranged from 9% to 10% for postmenopausal women. Posttest probability values for an indeterminate test result demonstrated greater variability according to the imaging modality and ranged from 38% for CT to 80% for contrast-enhanced MR imaging in premenopausal women and from 76% for CT to 95% for contrast-enhanced MR imaging in postmenopausal women. Year of publication of the reference did not influence the performance of imaging techniques in ovarian mass characterization.

Comparison of Diagnostic Strategies
The Figure illustrates posttest probability of ovarian cancer according to the prevalence of ovarian cancer in the study population (pretest probability), the imaging strategy, and a positive or negative combined test result. Both secondary imaging strategies (ie, combined gray-scale and Doppler US and body imaging) had significantly higher posttest probability values for positive test results and lower posttest probability values for negative test results than did the strategy of no further imaging. This result was valid for both pre- and postmenopausal patients (Table 6). When we compared combined gray-scale and Doppler US with body imaging, differences in posttest probability were not significant for negative test results. For positive test results, however, the strategy of body imaging showed significantly higher posttest probability values than did the option of combined gray-scale and Doppler US when MR imaging was used as the modality for body imaging. There was no difference between CT and the option of combined gray-scale and Doppler US. Within the body imaging strategy, we compared CT with nonenhanced and contrast-enhanced MR imaging. With a suspicious test result, posttest probability values increased more after nonenhanced or contrast-enhanced MR imaging than they did after CT (P < .001). Differences between nonenhanced and contrast-enhanced MR imaging were not significant (P = .133).



View larger version (39K):
[in this window]
[in a new window]
[Download PPT slide]
  		Curves based on means of simulation of posttest probability of ovarian cancer according to pretest probability of ovarian cancer with different imaging strategies for ovarian lesion characterization by using gray-scale US assessment alone (green curve) or combined gray-scale and Doppler US (orange curve), gray-scale US followed by CT (blue curve), and gray-scale US followed by nonenhanced (black curve) or contrast-enhanced (red curve) MR imaging. Indeterminate test results are shown with curves in the left upper outer quadrant, whereas benign results are indicated in the lower right quadrant. Gray-scale US assessment alone with a benign test result has the lowest curve and posttest probability of ovarian cancer, and therefore obviates subsequent testing. Distances between curves are greater for indeterminate results than they are for benign results; the lowest posttest probability is that of gray-scale US alone. All other imaging strategies show higher values of posttest probability; gray-scale US followed by nonenhanced and contrast-enhanced MR imaging are the imaging modalities at the top. The mean weighted pretest probability of ovarian cancer is indicated with the left dashed line for premenopausal women and with the right dashed line for postmenopausal women.

 

View this table:
[in this window]
[in a new window]

  		TABLE 6. Differences in Posttest Probability for Pairwise Comparison of Imaging Strategies in Women with Ovarian Masses

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
References
 
Ovarian lesion identification and characterization are key findings that are used as the basis to initiate specific treatment options. Findings of this study indicate that an indeterminate result of basic gray-scale US is insufficient evidence to lead to laparotomy for staging and planned surgical treatment for ovarian cancer. The positive LR of 3.52 of gray-scale US assessment alone is not sufficient to rule in ovarian cancer (31). A negative gray-scale US result with a negative LR of 0.17, however, is excellent for ruling out ovarian cancer. Therefore, a patient with a persistent ovarian mass and a negative (ie, benign) finding at gray-scale US does not need to undergo any further imaging investigation before laparoscopy.

Most gynecologists and primary care physicians outside a university or tertiary care center do not use or apply color Doppler US findings for diagnosis of ovarian cancer because of medium-quality US equipment or insufficient training in combined gray-scale and Doppler US. This situation prompted us to ask how to proceed with a patient who demonstrates indeterminate findings at gray-scale US. Our study provides evidence that the posttest probability in a patient who has an indeterminate gray-scale US result and undergoes another imaging study changes significantly from the posttest probability in a patient who does not undergo further imaging studies. This result is valid for both negative and positive test results of any second imaging modality. Our study findings also indicate that sending the patient with an indeterminate ovarian mass at gray-scale US for combined gray-scale and Doppler US is less efficient than sending the patient for MR imaging.

Indeed, there is a 10-fold difference among the positive LRs between the two imaging modalities (4.69 for combined gray-scale and Doppler US vs 44.18 for contrast-enhanced MR imaging). One explanation for the lower incremental value of combined gray-scale and Doppler US compared with MR imaging is a greater similarity to the initial test. Combined gray-scale and Doppler US only adds color Doppler US findings to the initial basic gray-scale US findings, whereas MR imaging has the ability to help to identify tissue components, such as blood or fat, according to signal intensity that is independent from the US characteristics of the ovarian mass. In a prospective study of 101 patients undergoing US followed by CT or MR imaging, Satoh et al (40) showed that the added value of the second imaging test mostly consisted of the resultant definitive diagnosis of a cystic teratoma. CT and MR imaging also allowed the diagnosis of endometriosis and gave greater confidence in the characterization of ovarian masses, according to cystic wall irregularities (40).

Our study findings extended previous findings from the Radiology Diagnostic Oncology Group that demonstrated the superiority of MR imaging compared with Doppler US for ovarian lesion characterization in a prospective multimodality study with 280 patients (3). We demonstrated that MR imaging (the body imaging strategy) remains superior even when Doppler US is combined with gray-scale US, an issue not previously studied. Kurtz et al (3) demonstrated little variation between conventional US, CT, and MR imaging with regard to ovarian cancer staging. Therefore, the choice of the secondary imaging modality after an indeterminate US result should be influenced more strongly by the performance of the modality in the differentiation between benign and malignant ovarian lesions than by its performance in the staging of possible ovarian cancer.

Our study findings also demonstrated that within the body imaging strategy, MR imaging should be the imaging modality of choice because of higher posttest probability values compared with those at CT in patients with a positive test result. Differences between CT and MR imaging did not reach statistical significance in previous studies (3,17,18). In this study, a difference of 42% in positive posttest probability in premenopausal women justifies the use of MR imaging instead of CT.

Our study findings confirm prior findings of a 3.5-fold greater prevalence of ovarian cancer in postmenopausal women than there is in premenopausal woman (36,41). Because of the low prevalence of ovarian cancer in postmenopausal patients with noncomplex (nonindeterminate, benign) ovarian cysts (none of 45 patients with persistent unilocular cystic ovarian tumors [42] and none of 28 patients with noncomplex adnexal cysts [43]), we did not evaluate the role of a secondary imaging test when findings at conventional US clearly allowed classification of the ovarian mass as benign. The management of postmenopausal patients with ovarian masses has drastically changed during the past 10 years, from systematic surgery to follow-up with US. Surgical management was recommended only when gray-scale US showed a persistent complex ovarian mass (44).

According to our study findings, the probability that a lesion is ovarian cancer in a premenopausal woman with an indeterminate ovarian mass at gray-scale US decreases to less than 2% with negative contrast-enhanced MR imaging or combined gray-scale and Doppler US results but increases to 80% with positive contrast-enhanced MR imaging results and to only 30% with positive combined gray-scale and Doppler US results. Differences in positive test results according to the chosen secondary imaging strategy remain significant in postmenopausal women. The clinical relevance of differences in the probability of ovarian cancer, however, might be less interesting in postmenopausal women, as both a 69% and a 95% chance of ovarian cancer are likely to generate a decision that leads to laparotomy for staging and to complete surgical excision for ovarian cancer. The Figure aids a clinician in the decision about where to place the threshold for surgical management decisions. If the threshold for immediate laparotomy is at the 33% risk for ovarian cancer, the clinician will order MR imaging for a premenopausal woman and no further imaging for a postmenopausal woman.

Some may question whether we overestimated the pretest probability of ovarian cancer by examining the prevalence of ovarian cancer in patients who had undergone surgery. Indeed, patients in later studies underwent US prior to surgery, and that could have considerably increased the probability of ovarian cancer. Findings in our study show that findings in the earliest study (36), performed before the era of US lesion characterization, indicated a 45% pretest probability of ovarian cancer in postmenopausal women. This value is greater than the values from all other studies in which US was used prior to surgery. Other factors, such as repeated US prior to surgery to decrease the percentage of functional cysts, might have influenced the proportion of ovarian cancer in pre- and postmenopausal women but could not be assessed in this meta-analysis because of missing information about the referral pattern for patients with ovarian masses.

Although imaging techniques might have improved over the years, the year of publication did not influence the reported performance of ovarian mass characterization in studies with CT, MR imaging, and gray-scale or combined US. A limitation of our study is that we did not perform a threshold analysis for differences in posttest probability according to prevalence of ovarian cancer and test result. The large number of comparative pairs of imaging modalities would have considerably decreased the comprehensibility of the study. The Figure graphically illustrates large distances between curves within the range of possible values for pretest probability of ovarian cancer.

Another limitation of our study was the absence of integration of elevated serum CA-125 values into the final estimation of posttest probability of ovarian cancer. Such a study would have required published data, including presentation of imaging results according to CA-125 measurements. Although researchers in several studies compared the diagnostic accuracy of CA-125 measurements with either basic or combined gray-scale and Doppler US findings, none of them detailed US findings according to elevated or nonelevated CA-125 measurements (37,45–48).

In conclusion, our study findings show that findings at subsequent imaging in a patient with an indeterminate ovarian mass at gray-scale US contribute to significant changes in posttest probability of ovarian cancer compared with posttest probability in a patient who undergoes no further imaging. Because of greater changes in posttest probability of ovarian cancer, as determined with Bayesian analysis, MR imaging is preferable to combined gray-scale and Doppler US or to CT for the diagnosis of ovarian cancer before treatment planning.


   FOOTNOTES
 

Abbreviations: CI = confidence interval • LR = likelihood ratio

Authors stated no financial relationship to disclose.

Author contributions: Guarantors of integrity of entire study, K.K., Y.L.; study concepts and design, all authors; literature research, K.K., A.M.; clinical studies, Y.L., A.M., M.F.P.; data acquisition and analysis/interpretation, all authors; statistical analysis, K.K., Y.L.; manuscript preparation, definition of intellectual content, revision/review, and final version approval, all authors; manuscript editing, K.K., Y.L., H.H.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 

  1. Gynecologic sonography. Report of the ultrasonography task force. Council on Scientific Affairs, American Medical Association. JAMA 1991; 265:2851–2855.[Abstract/Free Full Text]
  2. Kinkel K, Hricak H, Lu Y, Tsuda K, Filly RA. US characterization of ovarian masses: a meta-analysis. Radiology 2000; 217:803–811.[Abstract/Free Full Text]
  3. Kurtz AB, Tsimikas JV, Tempany CM, et al. Diagnosis and staging of ovarian cancer: comparative values of Doppler and conventional US, CT, and MR imaging correlated with surgery and histopathologic analysis—report of the Radiology Diagnostic Oncology Group. Radiology 1999; 212:19–27.[Abstract/Free Full Text]
  4. Moses LE, Shapiro D, Littenberg B. Combining independent studies of a diagnostic test into a summary ROC curve: data-analytic approaches and some additional considerations. Stat Med 1993; 12:1293–1316.[Medline]
  5. Valentin L. Prospective cross-validation of Doppler ultrasound examination and gray-scale ultrasound imaging for discrimination of benign and malignant pelvic masses. Ultrasound Obstet Gynecol 1999; 14:273–283.[CrossRef][Medline]
  6. Schelling M, Braun M, Kuhn W, et al. Combined transvaginal B-mode and color Doppler sonography for differential diagnosis of ovarian tumors: results of a multivariate logistic regression analysis. Gynecol Oncol 2000; 77:78–86.[CrossRef][Medline]
  7. Guerriero S, Alcazar JL, Ajossa S, et al. Comparison of conventional color Doppler imaging and power Doppler imaging for the diagnosis of ovarian cancer: results of a European study. Gynecol Oncol 2001; 83:299–304.[CrossRef][Medline]
  8. Grab D, Flock F, Stohr I, et al. Classification of asymptomatic adnexal masses by ultrasound, magnetic resonance imaging, and positron emission tomography. Gynecol Oncol 2000; 77:454–459.[CrossRef][Medline]
  9. Kurjak A, Predanic M. New scoring system for prediction of ovarian malignancy based on transvaginal color Doppler sonography. J Ultrasound Med 1992; 11:631–638.[Abstract]
  10. Bromley B, Goodman H, Benacerraf BR. Comparison between sonographic morphology and Doppler waveform for the diagnosis of ovarian malignancy. Obstet Gynecol 1994; 83:434–437.[Medline]
  11. Buy JN, Ghossain MA, Hugol D, et al. Characterization of adnexal masses: combination of color Doppler and conventional sonography compared with spectral Doppler analysis alone and conventional sonography alone. AJR Am J Roentgenol 1996; 166:385–393.[Abstract/Free Full Text]
  12. Caruso A, Caforio L, Testa AC, Ciampelli M, Panici PB, Mancuso S. Transvaginal color Doppler ultrasonography in the presurgical characterization of adnexal masses. Gynecol Oncol 1996; 63:184–191.[CrossRef][Medline]
  13. Rehn M, Lohmann K, Rempen A. Transvaginal ultrasonography of pelvic masses: evaluation of B-mode technique and Doppler ultrasonography. Am J Obstet Gynecol 1996; 175:97–104.[CrossRef][Medline]
  14. Brown DL, Doubilet PM, Miller FH, et al. Benign and malignant ovarian masses: selection of the most discriminating gray-scale and Doppler sonographic features. Radiology 1998; 208:103–110.[Abstract/Free Full Text]
  15. Alcazar JL, Jurado M. Using a logistic model to predict malignancy of adnexal masses based on menopausal status, ultrasound morphology, and color Doppler findings. Gynecol Oncol 1998; 69:146–150.[CrossRef][Medline]
  16. Buy JN, Ghossain MA, Sciot C, et al. Epithelial tumors of the ovary: CT findings and correlation with US. Radiology 1991; 178:811–818.[Abstract/Free Full Text]
  17. Buist MR, Golding RP, Burger CW, et al. Comparative evaluation of diagnostic methods in ovarian carcinoma with emphasis on CT and MRI. Gynecol Oncol 1994; 52:191–198.[CrossRef][Medline]
  18. Kubik-Huch RA, Dorffler W, von Schulthess GK, et al. Value of (18F)-FDG positron emission tomography, computed tomography, and magnetic resonance imaging in diagnosing primary and recurrent ovarian carcinoma. Eur Radiol 2000; 10:761–767.[CrossRef][Medline]
  19. Thurnher SA. MR imaging of pelvic masses in women: contrast-enhanced vs unenhanced images. AJR Am J Roentgenol 1992; 159:1243–1250.[Abstract/Free Full Text]
  20. Yamashita Y, Torashima M, Hatanaka Y, et al. Adnexal masses: accuracy of characterization with transvaginal US and precontrast and postcontrast MR imaging. Radiology 1995; 194:557–565.[Abstract/Free Full Text]
  21. Hricak H, Chen M, Coakley FV, et al. Complex adnexal masses: detection and characterization with MR imaging—multivariate analysis. Radiology 2000; 214:39–46.[Abstract/Free Full Text]
  22. Hata K, Hata T, Manabe A, Sugimura K, Kitao M. A critical evaluation of transvaginal Doppler studies, transvaginal sonography, magnetic resonance imaging, and CA 125 in detecting ovarian cancer. Obstet Gynecol 1992; 80:922–926.[Medline]
  23. Medl M, Kulenkampff KJ, Stiskal M, Peters-Engl C, Leodolter S, Czembirek H. Magnetic resonance imaging in the preoperative evaluation of suspected ovarian masses. Anticancer Res 1995; 15:1123–1125.[Medline]
  24. Komatsu T, Konishi I, Mandai M, et al. Adnexal masses: transvaginal US and gadolinium-enhanced MR imaging assessment of intratumoral structure. Radiology 1996; 198:109–115.[Abstract/Free Full Text]
  25. Reuter M, Steffens J, Schuppler U, et al. Critical evaluation of the specificity of MRI and TVUS for differentiation of malignant from benign adnexal lesions. Eur Radiol 1998; 8:39–44.[CrossRef][Medline]
  26. Rieber A, Nussle K, Stohr I, et al. Preoperative diagnosis of ovarian tumors with MR imaging: comparison with transvaginal sonography, positron emission tomography, and histologic findings. AJR Am J Roentgenol 2001; 177:123–129.[Abstract/Free Full Text]
  27. Huber S, Medl M, Baumann L, Czembirek H. Value of ultrasound and magnetic resonance imaging in the preoperative evaluation of suspected ovarian masses. Anticancer Res 2002; 22:2501–2507.[Medline]
  28. Irwig L, Tosteson AN, Gatsonis C, et al. Guidelines for meta-analyses evaluating diagnostic tests. Ann Intern Med 1994; 120:667–676.[Abstract/Free Full Text]
  29. McCullagh P, Nelder JA. Generalized linear models. In: Cox DR, Hinkley DV, eds. Monographs on statistics and applied probability. Monograph no. 37. London, England: Chapman & Hall, 1983.
  30. Rippley BD. Stochastic simulation. New York, NY: Wiley, 1987.
  31. Jaeschke R, Guyatt GH, Sackett DL. Users' guides to the medical literature. III. How to use an article about a diagnostic test. B. What are the results and will they help me in caring for my patients? The Evidence-Based Medicine Working Group. JAMA 1994; 271:703–707.[Abstract/Free Full Text]
  32. Nowak M, Szpakowski M, Malinowski A, et al. Ovarian tumors in the reproductive age group. Ginekol Pol 2002; 73:354–358.[Medline]
  33. Guerriero S, Mallarini G, Ajossa S, et al. Transvaginal ultrasound and computed tomography combined with clinical parameters and CA-125 determinations in the differential diagnosis of persistent ovarian cysts in premenopausal women. Ultrasound Obstet Gynecol 1997; 9:339–343.[CrossRef][Medline]
  34. Botta G, Zarcone R. Trans-vaginal ultrasound examination of ovarian masses in premenopausal women. Eur J Obstet Gynecol Reprod Biol 1995; 62:37–41.[CrossRef][Medline]
  35. Prompeler HJ, Madjar H, Sauerbrei W. Classification of adnexal tumors by transvaginal color Doppler. Gynecol Oncol 1996; 61:354–363.[CrossRef][Medline]
  36. Koonings PP, Campbell K, Mishell DR Jr, Grimes DA. Relative frequency of primary ovarian neoplasms: a 10-year review. Obstet Gynecol 1989; 74:921–926.[Medline]
  37. Botsis D, Kassanos D, Kalogirou D, Karakitsos P, Liapis A, Antoniou G. Transvaginal color Doppler and CA125 as tools in the differential diagnosis of postmenopausal ovarian masses. Maturitas 1997; 26:203–209.[CrossRef][Medline]
  38. Osmers RG, Osmers M, von Maydell B, Wagner B, Kuhn W. Evaluation of ovarian tumors in postmenopausal women by transvaginal sonography. Eur J Obstet Gynecol Reprod Biol 1998; 77:81–88.[CrossRef][Medline]
  39. Ovadia J, Goldman GA. Ovarian masses in postmenopausal women. Int J Gynaecol Obstet 1992; 39:35–39.[CrossRef][Medline]
  40. Satoh S, Takashima T, Nakano H. Role of imaging modalities in ovarian tumors. Int J Gynaecol Obstet 2002; 76:195–197.[CrossRef][Medline]
  41. Jacobs I, Oram D, Fairbanks J, Turner J, Frost C, Grudzinskas JG. A risk of malignancy index incorporating CA 125, ultrasound and menopausal status for the accurate preoperative diagnosis of ovarian cancer. Br J Obstet Gynaecol 1990; 97:922–929.[Medline]
  42. Bailey CL, Ueland FR, Land GL, et al. The malignant potential of small cystic ovarian tumors in women over 50 years of age. Gynecol Oncol 1998; 69:3–7.[CrossRef][Medline]
  43. Goldstein SR, Subramanyam B, Snyder JR, Beller U, Raghavendra BN, Beckman EM. The postmenopausal cystic adnexal mass: the potential role of ultrasound in conservative management. Obstet Gynecol 1989; 73:8–10.[Medline]
  44. Goldstein SR. Postmenopausal adnexal cysts: how clinical management has evolved. Am J Obstet Gynecol 1996; 175:1498–1501.[CrossRef][Medline]
  45. Kawai M, Kikkawa F, Ishikawa H, et al. Differential diagnosis of ovarian tumors by transvaginal color-pulse Doppler sonography. Gynecol Oncol 1994; 54:209–214.[CrossRef][Medline]
  46. Antonic J, Rakar S. Colour and pulsed Doppler US and tumour marker CA 125 in differentiation between benign and malignant ovarian masses. Anticancer Res 1995; 15:1527–1532.[Medline]
  47. Strigini FA, Gadducci A, Del Bravo B, Ferdeghini M, Genazzani AR. Differential diagnosis of adnexal masses with transvaginal sonography, color flow imaging, and serum CA 125 assay in pre- and postmenopausal women. Gynecol Oncol 1996; 61:68–72.[CrossRef][Medline]
  48. Tingulstad S, Hagen B, Skjeldestad FE, et al. Evaluation of a risk of malignancy index based on serum CA125, ultrasound findings and menopausal status in the pre-operative diagnosis of pelvic masses. Br J Obstet Gynaecol 1996; 103:826–831.[Medline]



This article has been cited by other articles:


Home page
 Am. J. Roentgenol.Home page
S. I. Lee
Radiological reasoning: imaging characterization of bilateral adnexal masses.
Am. J. Roentgenol., September 1, 2006; 187(3 Suppl): S460 - S466.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2361041618v1
236/1/85    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Kinkel, K.
Right arrow Articles by Hricak, H.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Kinkel, K.
Right arrow Articles by Hricak, H.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE