Chapter  130:  Management of Adnexal Mass

Limitations of the Literature

The main limitation in the literature was the failure to adequately describe relevant patient characteristics, including the presence or absence of symptoms, and variable reporting of menopausal status. Inadequate sample size, lack of blinding, and failure to account for observer variability were also common limitations.

Limitations of the Report

The report did not include non-English publications. We did not include non-U.S. studies in our review of the prevalence of different types of adnexal mass. Given the heterogeneity of studies, pooled estimation of sensitivity and specificity may not be appropriate. The NIS does not include outpatient procedures, and our coding algorithm may have missed some complications.

Sources

The primary sources of literature were MEDLINE^® (1966-September 2004) and the Cochrane Database of Systematic Reviews. Searches of these databases were supplemented by reviews of reference lists contained in all included articles and in relevant review articles and meta-analyses.

Search Strategies

The basic search strategy used the National Library of Medicine's Medical Subject Headings (MeSH) key word nomenclature developed for MEDLINE^® and was adapted for use in the other databases. The searches were limited to the English language. The texts of the three major search strategies are given in Appendix A.^* The searches yielded a total of 677 citations, whose records are maintained in a ProCite²⁰ database.

Abstract and Full-text Screening

Paired researchers from the Duke research team independently reviewed a set of abstracts and classified each as “include” or “exclude” according to study-specific criteria, which they developed. An abstract was included if at least one of the paired reviewers recommended that it be included. A total of 445 abstracts were included for the further “full-text review” stage. Interrater reliability for include/exclude decisions was tested by having 10 pairs of readers review 138 abstracts. Agreement was good to excellent (kappa 0.66 to 0.95).

At the full-text review stage, the paired researchers independently reviewed a set of the articles, and indicated a decision to “include” or “exclude” the article for the data abstraction stage. When a pair of reviewers arrived at a different opinion about whether to include an article, they were asked to reconcile the difference. Detailed inclusion and exclusion screening criteria were developed by research question and are listed below.

Full-text Screening Criteria

Initially, the patient population was limited to peri- and postmenopausal women, and only articles that provided data specifically by age or menopausal status were included. After initial discussion with the expert panel, the search was expanded to include premenopausal women.

Question 1. Background clarifications were as follows:

• (1)

  The search should be limited to (a) screening studies and (b) case series of women with an undiagnosed mass (not just women who went to laparoscopy/path diagnosis).

• (2)

  Pathology list:

  • a

    Benign

    • i

      Uterine leiomyoma

    • ii

      Nonneoplastic cysts, such as:

      • 1

        Follicular (functional) cysts

      • 2

        Corpus luteal (functional) cysts

      • 3

        Theca lutein cysts

      • 4

        Simple cysts

      • 5

        Peritoneal inclusion cysts

      • 6

        Paraovarian cysts

      • 7

        Hemorrhagic cysts

      • 8

        Endometrial cyst

    • iii

      Polycystic ovary disease

    • iv

      Cystic teratoma (dermoid cyst)

    • v

      Hydrosalpinx,

    • vi

      Cystadenoma

    • vii

      Fibroma

  • b

    Malignant ovarian neoplasms

    • i

      Adenocarcinoma

    • ii

      Others

  • c

    Tumors of low malignant potential

Screening criteria for Question 1 were:

• (1)

  undiagnosed mass (regardless of whether symptomatic or asymptomatic; detected by palpation or ultrasound imaging);

• (2)

  exclude if n < 50; if n ≥ 50, write n on decision sheets;

• (3)

  histology diagnosis;

• (4)

  screened women without mass (case series or cohort) or women with adnexal mass (case series).

Question 2. Screening criteria were as follows:

• (1)

  comparison of bimanual pelvic examination to a reference standard;

• (2)

  n ≥ 20;

• (3)

  able to construct 2-by-2 table for test characteristics.

Question 3. Screening criteria were as follows:

• (1)

  undiagnosed mass (regardless of whether symptomatic or asymptomatic; detected by palpation or ultrasound imaging) or screening population;

• (2)

  disease status distinguishes malignant from non-malignant;

• (3)

  must have 20 or more subjects;

• (4)

  disease status must be verified by histology or negative surgery (laparoscopy/laparotomy);

• (5)

  test is ultrasound, CT, MRI, PET, serum CA-125, or bimanual pelvic exam;

• (6)

  able to construct 2-by-2 table for test characteristics.

Question 4. Screening criteria were as follows:

• (1)

  patients with cancer;

• (2)

  studies with scoring, risk score, combined modality approach;

• (3)

  assesses predictive value of two or more variables (radiographic, patient characteristics or CA-125) using multivariable model;

• (4)

  screening studies;

• (5)

  n ≥ 50.

Question 5. Screening criteria were as follows:

• (1)

  n ≥ 50;

• (2)

  histology or followup interval = at least 9 months;

• (3)

  outcome = continued negative test with no clinical evidence of developing ovarian cancer.

Question 6. Screening criteria were as follows:

• (1)

  procedure = operative laparoscopy for adnexal mass, with or without biopsy;

• (2)

  addresses complications of procedure (morbidity or mortality);

• (3)

  n ≥ 100 for morbidity.

Question 7. Screening criterion was as follows: article described mathematical or computer model of natural history of ovarian cancer.

Summaries of the results of the abstract screening and full-text review are provided in Tables 2 and 3. A list of excluded articles by reason for exclusion is found in Appendix B.^*

Test Characteristics and Confidence Intervals

For test characteristics, a Microsoft Excel® spreadsheet was developed which calculated appropriate test characteristics (sensitivity, specificity, negative predictive value, positive predictive value) for individual studies if studies provided enough data to input (a) values for individual cells of a 2-by-2 table, (b) the prevalence of disease and values for sensitivity and specificity, or (c) sufficient data to solve for two equations involving sensitivity, specificity, or predictive values. Ninety-five percent confidence intervals were automatically estimated using the approximate formula for proportions:

, where p = point estimate of proportion, N = total sample size.

Prevalence and Event Rates and Confidence Intervals

For Questions 3 and 6, prevalence of different mass types, and morbidity and mortality rates, were also calculated using the above formula. For studies where the numerator of a particular proportion was 0, the upper bound was estimated using the formula:

, where p = 2/(N + 2).

Estimation of Pooled Sensitivity and Specificity

For Questions 2, 3, and 4, we used two complementary methods for assessing diagnostic test performance: (1) summary receiver operating characteristic (ROC) analysis; and (2) independently combined sensitivity and specificity values. We calculated pooled sensitivity and specificity estimates, along with 95 percent confidence intervals and summary ROC curves, using Meta-Stat 0.6, a shareware program for performing meta-analyses of diagnostic tests.²² In this software, logits of sensitivity and specificity values are pooled, using a random-effects model weighted by the inverse of the variance.²³

We combined the sensitivity and specificity values of the tests across studies using a random-effects model to estimate the average values. A random-effects model incorporates both the within-study variation (sampling error) and between-study variation (true treatment-effect differences) into the overall treatment estimate. It gives a wider confidence interval than the fixed-effect model (which considers only within-study variability) when estimates are based on heterogeneous results.

When each is combined separately, sensitivity and specificity tend to underestimate the true test sensitivity and specificity; however, they can provide an indication of the approximate test operating point for most of the studies.

Summary ROC curves are a potentially useful graphical summary of the diagnostic test performance data. In brief, each study provides a pair of sensitivity and specificity values to the analysis. After logistic transformation of data, a linear model is fitted to the observed studies using regression analysis. This best-fit model can then be transformed back to ROC space and plotted as curve. A summary ROC curve can be thought of as an ROC curve that describes joint changes in sensitivity and specificity with changes in cutoff values. The ideal position of an ROC curve is near the upper left corner. The area under the curve (AUC) is another summary measure of the degree of discrimination of a test.

The summary ROC method assumes that the variability in the reported sensitivity and specificity values from different studies is due to different cutoff values (explicit or implicit) being applied.²⁴ However, the summary ROC curve can summarize studies whose variability may be due to other sources of variation, since the summary ROC curve no longer ties specific cutoff values to specific intervals of the curve. One can think of a summary ROC curve as an overall estimate of the discrimination ability of a test.

When there is little variability in the test results - i.e., when studies appear to be operating at similar thresholds and report similar results - summary ROC analysis provides little additional information. In this case, separately averaged sensitivity and specificity values across studies will give similarly useful summary information. However, where there is substantial variability in test results, the separately averaged sensitivity and specificity values tend to have wide confidence intervals and have means that do not characterize any of the studies. In this case, SROC curves provide a more suitable analysis framework.

Estimates of National Rates of Surgery for Adnexal Mass

The Nationwide Inpatient Sample (NIS) is a public access database maintained by AHRQ. The NIS represents a stratified sample of approximately 20 percent of all discharges from U.S. hospitals; data for the year 2000 contain administrative discharge data from hospitals in 28 states, while 2001 contains data from 33 states.²⁵ Weights are provided in order to allow estimation of national data based on this sample. We used data from 2000 and 2001 to provide supplemental data on the frequency of diagnostic laparoscopy and exploratory laparotomy for Question 6. Because previous work has shown that administrative data may lack sufficient clinical detail to compare outcomes,²⁶ we did not attempt to directly compare complication rates between these procedures, or between diagnoses.

The search was limited to women 15 years and older, who had one of the following International Classification of Diseases, Ninth Revision (ICD-9) diagnostic codes: 183.x (malignant neoplasm of the ovary and other uterine adnexa), 220.x (benign neoplasms of the ovary); 620.x (ovarian cysts); 752.11 (para-ovarian cysts); 614.0, 614.1, 614.2, 614.6 (adnexal masses secondary to pelvic inflammatory disease); 789.33, 789.34, 789.39 (abdominal masses arising in the left or right lower quadrant, or other nonspecified site); and V655 (normal findings after diagnostic evaluation).

In order to avoid overestimation of complication rates due to other procedures, we then excluded patients who had an ICD-9 diagnosis code for hysterectomy (68.x). Procedures were then classified as laparoscopy only (54.21), laparoscopy with conservative ovarian surgery (65.3x, 65.4x, 65.5x, 65.6x), laparoscopy with oophorectomy (65.0×, 65.2x), or laparotomy (54.11) alone, with conservative ovarian surgery (same codes), or with oophorectomy (same codes).

A discharge status of “Dead” indicated in-hospital mortality. Complications of surgery or hospitalization were indicated by diagnosis codes of E870 through E876.

Model of Natural History of Ovarian Cancer

We used a Markov state-transition model to explore the impact of alternate assumptions about the natural history of ovarian cancer. The original model was developed as a graduate school project by Karen Hoffman, MD, and further refined in collaboration with two of the authors of this report (Drs. Kulasingam and Myers).

The model simulates a cohort beginning at age 40 distributed across cancer stages. Subjects progress from no cancer through the stages of ovarian cancer to death. Each cycle is 12 months long. The original model design is illustrated in Figure 3; subsequent modifications include removal from the at-risk population by undergoing oophorectomy for another cause, and allowing some Stage I cancers the possibility of progressing directly to Stage III. Model variables and the ranges over which they were varied are outlined in Table 4. Probability of progressing from no cancer to Stage I cancer varies by age and is based on age-adjusted ovarian cancer incidence rates. Because the probability of progression (or duration of time within a stage) is unknown, probability of progression from Stage I to II, from Stage II to III, and from Stage III to Stage IV was adjusted to reflect incidence distribution across stages. Within the model, subjects may die from causes other than ovarian cancer. The probability of dying from a cause other than ovarian cancer varies by age and was constructed from CDC National Vital Statistics reports and Surveillance, Epidemiology, and End Results (SEER) data.^{27, 28} Probability of clinical diagnosis is based on the annual report of the International Federation of Gynecology and Obstetrics (FIGO).²⁹ Five-year survival rates gathered by SEER 1992-98 were used to predict probability of dying from ovarian cancer.²⁷ SEER localized disease corresponds to Stage I cancer, regional disease corresponds to Stage II cancer, and distant disease corresponds to Stage III/IV ovarian cancer. The model was constructed in DATA 4.0.³⁰

Abbreviations for probabilities are described in Table 4, below.

Approach

We included studies in the U.S. population with more than 50 women and limited the literature search to screening studies and case series where results were provided for all women with an undiagnosed mass, not just those with subsequent positive additional tests.²¹ Studies of adnexal mass in which the gold standard is applied only to those with positive tests results would underestimate the prevalence of disease and cause a substantial bias.

Results

Twenty articles met the inclusion criteria and are described in the Evidence Table 1 (Appendix D ^*).³¹–⁵⁰

Detailed prevalences for specific tumor types are provided in Evidence Table 1. The included studies can be divided into two groups. The first group includes four reports from a large screening study in Kentucky (Table 5). The prevalence of malignancy ranged from 0.09 to 0.18 percent. In postmenopausal women, the prevalence of malignancy was 0.09 to 0.18 percent, borderline tumors were not reported, and the prevalence of benign tumors was 0.08 to 1.3 percent. In a population that included either postmenopausal women or those with a family history of breast, ovarian, or colorectal cancer, the prevalence of malignancy was 0.10 to 0.11 percent, of borderline tumors 0.02 percent, and of benign tumors 1.1 to 1.2 percent.

The most common malignant tumor types include primary ovarian carcinoma, such as serous and mucinous cystadenocarcinoma, granulosa cell tumors, and undifferentiated adenocarcinoma. Borderline tumors were less common, such as serous low malignant potential (0.02 percent). The most common benign tumors were serous cystadenoma (0.4 to 0.7 percent), paratubal cyst (0.1 to 0.16 percent), endometrioma (0.03 to 0.3 percent), and mature teratoma (0.02 to 0.08 percent).

The majority of U.S. studies with histological diagnosis of all masses (n = 16) were case series of women with undiagnosed adnexal mass undergoing laparotomy (Table 6). The prevalence of malignancy ranged from 5.7 to 57.5 percent, the range of borderline tumors was 1.4 to 11.2 percent, and the prevalence of benign tumors was 40 to 100 percent. All tumor types were over-represented because patients had an undiagnosed adnexal mass, and the clinical presentation was not well described in the majority of studies. Most studies included both premenopausal women and postmenopausal women and did not provide results separately. The one study that included only postmenopausal women⁴¹ found only benign tumors.

Discussion

Estimating the age-specific prevalence of specific adnexal tumor types from the available literature is difficult. The best data come from a series of reports from a large screening study; overall prevalence of masses was 1 to 2 percent, with benign masses outnumbering malignant by 4- to 10-fold. Because patients with negative screening test results did not undergo definitive diagnostic procedures in these studies, the prevalence estimates are dependent on the sensitivity of the screening tests used (and the completeness of followup among test negatives). In addition, there is a potential bias in that premenopausal women enrolling in the screening study were at higher risk than average because of family history; in addition, postmenopausal women may have been more likely to enroll because of concerns based on family history, vague symptoms, or other reasons which would affect relative prevalence compared to the general population.

Estimates of prevalence in studies with 100 percent histologic diagnosis are inevitably biased by the clinical factors that determine which patients ultimately undergo surgery. These can include the presence and nature of symptoms (patients with symptoms referable to a mass would likely undergo surgery sooner than those with asymptomatic masses, all other things being equal); other findings (for example, the presence of ascites); patient anxiety; the diagnostic algorithms used (for example, the duration of followup for persistence); and the nature of the practice (malignancies will be more frequent in a gynecologic oncology practice compared to a general gynecology practice).

As mentioned previously, we did not include studies from outside the United States. Given differences in ethnic backgrounds (affecting genetic risks), observed differences in cancer incidence, and differences in clinical practice between countries, and the almost universal failure of studies to describe the clinical history leading to the diagnosis of adnexal mass, inclusion of these studies would not have allowed a more precise estimate of prevalence of different types of adnexal masses in the U.S. population.

Summary

In four reports from a large U.S. screening study, the prevalence of adnexal masses detected by ultrasound among postmenopausal women was 0.8 to 1.3 percent, and the prevalence of malignancy 0.09 to 0.18 percent (i.e., 9 to 18 per 10,000). Prevalence of different pathologies varies widely among case series. There are no data on the relative prevalence of different pathologies among women with asymptomatic masses compared to women with symptomatic masses.

Approach

Articles were sought which evaluated the ability of the bimanual examination to detect adnexal masses, and/or to discriminate benign from malignant masses. Preference was given to studies where there was histological confirmation of the diagnosis, but an alternative reference standard (such as followup) was allowed for screening studies. Data allowing calculation of sensitivity and specificity had to be provided.

Our rationale for including the pelvic examination was based on its role in the initial evaluation of adnexal masses. Some asymptomatic women will have a mass detected as part of a “routine” physical examination; others will have a mass detected as part of an examination performed because of symptoms. The postexamination probability of malignancy is a function of the prevalence of cancer and the sensitivity and specificity of the bimanual examination; these probabilities, in turn, will affect the positive and negative predictive values of additional tests such as cancer antigen 125 (CA-125) and imaging studies. Because the pelvic examination will be the first test performed, either as a screening test or as a diagnostic test, knowledge of its test characteristics is important for evaluating subsequent diagnostic tests.

Results of Literature Search and Screening

We identified 14 studies that met the inclusion criteria.^{42, 51}–⁶³ Nine studies provided data on discrimination between benign and malignant masses,^{42, 51, 53}–^{57, 62, 63}, four on the ability of the bimanual examination to detect any adnexal mass,^{52, 59}–⁶¹ and one provided data on both discrimination between benign and malignant and ability to detect masses.⁵⁸ All 14 studies are summarized in Evidence Table 2 (Appendix D ^*).

Study Characteristics

Types of data incorporated. Two of the studies^{54, 56} included history or clinical impression as part of the “test;” results were not provided separately for examination alone.

Types of study population. Ten of the 14 studies were performed prior to surgery for an adnexal mass, while four were from screening studies.^{51, 52, 57, 58}

Reporting of study populations. Of the screening studies, Andolf et al.⁵² was performed in women over 40 considered at high risk of ovarian cancer because of symptoms or risk factors; Grover and Quinn⁵⁷ was performed in asymptomatic volunteers 25 and older, but described menopausal status; Adonakis et al.⁵¹ was performed in women over 45; and Jacobs et al.⁵⁸ was done entirely in a postmenopausal population.

Seven of the 11 preoperative studies reported menopausal status, but only two reported on test characteristics specifically by menopausal status.^{55, 56} None reported race/ethnicity, and none reported the clinical route by which patients had come to surgery (detection of an asymptomatic mass, symptoms, etc.).

Methodology. The methodological quality of the included studies was as follows:

Reference standard. Of the preoperative studies, all but one⁴² had operative confirmation of findings. Ultrasound was used as the reference standard in the four screening studies, with 12-month followup examinations or questionnaires.

Verification bias. In the study by Roman et al.,⁴² 26 women with non-palpable masses did not undergo definitive diagnosis.

Test reliability. Only one study⁶⁰ provided direct data on test reliability. Grover and Quinn,⁵⁷ Ong et al.,⁵⁹ Schutter et al.,⁶³ and Buckshee et al.⁵⁴ used a single examiner. The other studies did not address the issue of test reliability.

Sample size. None of the reports had a priori sample size calculations.

Use of appropriate statistical tests. All reports used appropriate techniques for calculating test characteristics.

Blinding. Only two studies^{54, 60} explicitly stated whether examiners were blinded to prior history or other findings.

Definition of positive and negative test. Nine of 14 studies reported their definitions of a positive test, although the precision of the definitions was quite variable (from “a mass 5 cm or more in diameter” to “larger than normal”); others relied on “clinical impression.”

Results

Table 7 and Figure 4 present the results of studies that evaluated the sensitivity of the bimanual examination for detecting an adnexal mass. The studies of Padilla et al.^{60, 61} are particularly striking for the low sensitivity, since the examinations were performed under anesthesia, when, presumably, patient discomfort would not be a limiting factor. Both studies suggested a relationship between experience and accuracy; medical students performed worse than residents, who performed worse than attending physicians. Although these differences were not statistically significant, the studies were underpowered to detect significant differences. Obesity, defined as a body mass index greater than 30, had a significant negative impact on sensitivity, while increasing uterine size increased sensitivity, possibly by elevating the adnexae out of the pelvis.

When sensitivity and specificity were combined separately using a random-effects model, the pooled sensitivity was 0.45 (95% confidence interval [CI], 0.28 to 0.68), and the pooled specificity was 0.90 (0.80 to 0.96).

Table 8 and Figure 5 show the test characteristics for discriminating benign from malignant masses. Using a random-effects model, pooled sensitivity was 0.72 (95% CI, 0.49 to 0.88) and specificity was 0.92 (0.80 to 0.97). When only the three screening studies were included, pooled sensitivity was 0.58 (95% CI, 0.21 to 0.88), pooled specificity 0.98 (0.97 to 0.98).

For both types of studies, there appears to be a trend towards decreased specificity as prevalence increases, although the number of studies is small and the confidence intervals are wide. The extreme differences in sensitivity in the two largest studies (0 and 100 percent) prevent even a qualitative assessment of any relationship between prevalence and sensitivity.

The two studies that stratified results by menopausal status^{55, 56} found lower sensitivity and higher specificity for discriminating benign from malignant masses in premenopausal women compared to postmenopausal women (Table 8).

Discussion

Despite the common recommendation for routine pelvic examination, we found surprisingly little literature on its accuracy. Based on the literature we did identify, its sensitivity for detecting adnexal masses appears fairly low. Sensitivity for detecting normal adnexa is also low, as demonstrated in a recent study of examinations under anesthesia.⁶⁴ Although sensitivity for distinguishing a malignant mass from a benign one is somewhat better, these results need to be interpreted with caution, since most of the studies were done in preoperative patients, who would already have a higher probability of having a malignancy. In the four large screening studies, there was a total of only five malignancies, with the bimanual detecting 0 percent, 66 percent, and 100 percent in the three individual studies where ovarian cancer was detected; the fourth had one case of a low malignant potential tumor and two endometrial cancers. Pooled sensitivity for the three screening studies that addressed discrimination between benign and malignant masses was considerably lower than for all studies combined (and was similar to the pooled sensitivity of the studies that examined the ability to detect any adnexal mass).

Both types of studies show a trend toward decreased specificity as the prevalence of abnormality increases - this may reflect a greater degree of suspicion on the part of the examiner, based on other findings, and a greater likelihood of calling an examination abnormal. This is supported by the finding of the two studies which stratified results by menopausal status, which found higher sensitivity and lower specificity in postmenopausal women compared to premenopausal women.^{55, 56} Because examiners were unblinded, and were likely aware of the higher prevalence of malignancy among postmenopausal women, they may have been more likely to assign a diagnosis of malignancy among those patients. Future studies need to pay stricter attention to blinding examiners to other information. In theory, this bias should also result in higher sensitivity as prevalence increases, although, because of the small number of studies, the small numbers of subjects in most studies, and the diametrically opposed findings of the two largest studies, we were unable to recognize any relationship.

In the two studies that addressed the effect of experience on test characteristics,^{60, 61} there appeared to be a relationship between increasing experience and increased sensitivity (specificity did not change); however, even attending physicians achieved a sensitivity of only 28 percent. Based upon the available literature, the bimanual examination does not appear to be a sensitive test for detecting the presence of adnexal masses and appears to have limited ability to discriminate benign from malignant masses. Although specificity was somewhat better, positive predictive values will still be quite low in low prevalence settings, as discussed under Question 7. This will, in turn, lower the positive predictive value of diagnostic tests performed in patients referred on the basis of a pelvic examination. These tests are discussed in detail in the next section.

Approach

This section considers the various evaluation modalities that are described in the literature and would be available to a clinician to aid in the work-up of an adnexal mass after it has been diagnosed. We focused our search on articles whose primary reference standard was histopathology. Ideally this reference standard would be applied to all test negatives. However, we accepted a repeat negative test (such as imaging) conducted at least 6 months later as an acceptable alternative. We did include some studies that were from population-based screening samples, and these will be considered in a separate section below. The evaluation modalities investigated can be divided into several general categories. Imaging studies will be divided by technological mode (ultrasound, MRI, etc.). Ultrasound studies will be divided into those that evaluate adnexal morphology (either by an explicit scoring system or by descriptive standards), those that measure vascular flow in the mass (Doppler), and those that evaluate these modalities in combination. Serum studies will focus primarily on CA-125, as this is the most common marker in both the literature and in clinical practice. However, other serum markers will be discussed as well. Finally, the studies for which it was possible to stratify by menopausal status will be discussed where appropriate.

Results of Literature Search and Screening

Two hundred and five articles were identified for abstraction. Of these, 153 met the inclusion criteria and were abstracted into Evidence Table 2 (Appendix D ^*).^{31, 33}–^{36, 39, 42}–^{44, 46, 47, 49}–^{56, 58, 62, 63, 65}–¹⁹⁵

Ultrasound Morphology

Conventional grey scale ultrasonography is the most common imaging modality used to differentiate benign from malignant adnexal masses. Especially with the advent of high-frequency transvaginal probes, the quality of the images allows description of the gross anatomic features of the lesion. This is, however, limited by the great variability of macroscopic characteristics of both benign and malignant masses. Furthermore, the technique is operator dependent. To overcome these limitations, morphologic scoring systems have been developed. Such scoring systems are based on specific ultrasound parameters each with several scores according to determined features and with a cutoff value to categorize masses as either malignant or benign.

Table 9 describes the details of the most commonly used scoring systems. Briefly, the following scores are suggestive of malignancy: Sassone¹⁵⁹ greater than 9, DePriest³⁶ greater than or equal to 5, Ferrazzi⁹³ greater than 9, and for Lerner¹³¹ greater than or equal to 3. Although the development of all the scoring systems was motivated to improve the reproducibility of morphological measurements, only the scoring system by Lerner et al. based the categories on a multivariate logistic analysis.

Reproducibility of tests. Timmerman et al.¹⁹⁶ evaluated the subjective assessment of ultrasonographic images for discriminating between malignant and benign masses. Three hundred consecutive patients were evaluated with TVUS by six different operators, and both diagnostic accuracy and interassessor agreement were calculated. The operators had varied experience in TVUS - from approximately 300 to 15,000 scans. The two most experienced operators agreed 92 percent of the time. The accuracy of the least experienced operators ranged from 82 to 87 percent (p = 0.0001). Overall, 65 percent of all the masses were correctly classified by all six operators. Interassessor agreement was greater between the most experienced operators as well (kappa = 0.852). When comparing experienced with less experienced operators, the kappa ranged from 0.581 to 0.737. This is similar to the kappa reported by Yamashita et al.¹⁹² among five operators, 0.62 (± 0.02) with TVUS. Interassessor agreement was not calculated between the less experienced operators. None of the included articles described operator experience, and only a few addressed interobserver variability. Although operator experience appears to correlate with accuracy, the specialty training of the unltrasonographer does not. In a meta-analysis of both morphologic and color Doppler tests in the evaluation of adnexal masses, Kinkel et al.¹⁹⁷ found no difference between radiologists and gynecologists in the performance of ultrasound.

TVUS versus abdominal ultrasound. Of the 122 articles that evaluated adnexal masses via ultrasound (through either ultrasound morphology or Doppler measurements), only five articles exclusively used transabdominal imaging.^{52, 58, 116, 133, 198} Fifty-nine articles used TVUS exclusively and 51 used a combination of TVUS and abdominal ultrasound. There were seven articles for which the ultrasound modality was unknown. In the majority of the articles that used a combination of TVUS and abdominal ultrasound, TVUS was the “method of choice.” The most common reasons cited for also including abdominal ultrasound were patient refusal of transvaginal scans, virginity, poor image quality, and very large masses. Although a few articles reported how many women had which type of ultrasound, none of the articles reported their results such as to permit a stratification by TVUS or abdominal ultrasound. We therefore elected to group all ultrasound studies together regardless of TVUS or abdominal imaging.

Trials identified. We identified 69 articles comprising 73 ultrasound morphology assessments. Despite the availability of published scoring systems, most of the studies based their diagnoses on either descriptive assessments of adnexal masses or used a modified or unique scoring system. Only 13 studies explicitly used Sassone's criteria, six used DePriest's, and three used Ferrazzi's, Finkler's, Lerner's, and Valentin's respectively. When a scoring system other than an established criterion was used, it was not always clear how it had been developed or modified. Details of the tests and their evaluative performance are provided in Table 10. Assessments of adnexal morphology by ultrasound which were either a unique or modified or unclear scoring system are labeled “other” with a brief description when possible. It is also important to note that not all of the established scoring systems were employed using the original cutpoints. For example, Caruso et al.⁸³ and Itakure et al.¹¹⁵ both used a cutpoint of > 7 for the DePriest scoring system, where the original description used ≥ 5.

Results. The results of pooled sensitivity and specificity using a random-effects model, along with the range of sensitivity and specificity reported in individual studies, are shown in Table 10. Included studies are shown in Figures 6–10. There was a great range in test results, especially in the studies not using established scoring systems. This most likely reflects the heterogeneity of the tests themselves. There was little concrete difference among the established scoring systems. Overall the tests achieved relatively higher levels of sensitivity and negative predictive value (NPV) in the diagnosis of malignancy than specificity or positive predictive value (PPV). With the exception of four studies, the NPV was above 0.80, with the majority of tests above 0.90. The PPV in the majority of studies was below 0.50. In general, there was a trade-off between sensitivity and specificity, both in the individual studies of a specific scoring system, and in pooled results of all studies of a scoring system - as sensitivity increases, specificity decreases.

Comparing the figures, studies using the Sassone criteria show greater variability in sensitivity compared to variability in specificity (Figure 6), while those using the DePriest criteria (Figure 7) show greater variability in specificity and a relatively narrow range of sensitivity. Figure 10, which depicts a variety of other studies, suggests trade-offs between sensitivity and specificity; different morphology methods for discriminating benign from malignant have different thresholds, resulting in the sensitivity/specificity trade-off.

Three articles compared different scoring systems within the same study population. Caruso et al.⁸³ examined 112 women with adnexal masses comparing Sassone, DePriest, and Valentin scores. All performed similarly, displaying a sensitivity and NPV of 1.00, a range of specificity of 0.61 to 0.75, and a range of PPV of 0.35 to 0.48. Alcazar et al.⁶⁹ also compared the performance of Sassone, DePriest, and Ferrazzi. There were no significant differences between these scoring systems when receiver operating characteristic (ROC) curves were compared. The area under the curve (AUC) was 0.89 for Sassone, 0.92 for DePriest, and 0.90 for Ferrazzi. Ferrazzi et al.⁹³ evaluated 261 masses collected in three different centers. They compared ROC curves for scores based on Sassone, Granberg, DePriest, and Lerner's criteria and compared it with a scoring system they developed. The AUC ranged from 0.72 to 0.75 for the previously established systems. Their new scoring system (Ferrazzi) performed better, with an AUC of 0.84 (p < 0.0001). However, subsequent comparisons have not reaffirmed its superior functioning. When the Ferrazzi scoring system was compared to both Sassone and DePriest,⁶⁹ its performance was almost identical.

In spite of different designs, all the scoring systems performed similarly when compared within the same study population. It has been suggested that the poor performance of scoring systems with regard to their PPV is due to the misclassification of dermoid tumors.¹⁹⁷ Dermoids share many of the features that are characterized as “malignant” in scoring systems. The Alcazar study proposes a scoring system that was developed in part to correct this. Although this scoring system does perform well in its initial application, it has not been independently verified. The authors conclude, “a completely reliable differentiation of malignant masses cannot be obtained by sonographic imaging alone.”⁶⁹

Stratification by menopausal status. Of the 69 articles identified that addressed the assessment of adnexal morphology by ultrasound, only 13 contained data that either directly reported test characteristics by menopausal status or contained enough information to enable the stratification of results. Six were studies in a 100 percent postmenopausal patient population. Seven were studies that allowed comparison by menopausal status within the study population. They are presented in Table 11. The only significant difference in test performance appears to be in regards to the PPV. With the exception of Roman et al.,⁴² the PPV is slightly higher in postmenopausal women. This likely reflects the higher prevalence of ovarian malignancy after menopause. Aside from the PPV, the performance of ultrasound in the morphological assessment of adnexal masses does not appear to be significantly changed by menopausal status.

Ultrasound Doppler Studies

Color Doppler scanning allows the assessment of tumor vascularity. Malignant neoplasms have active blood vessel creation (angiogenesis) compared to normal or benign neoplasms due, in part, to their increased metabolic activity. Overall, malignancies display an increased vascularity with decreased peripheral blood flow resistance and increased blood flow velocity compared with benign tissue.^{152, 200} Doppler signal analysis can separate high-resistance and low-resistance vessels and has therefore been investigated as a separate test modality, as well as in combination with ultrasound morphological evaluation in the evaluation of adnexal masses.

The most common flow criteria are the resistance index (RI), the pulsatility index (PI), and the maximum systolic velocity. RI is defined as the difference between peak systolic and maximum enddiastolic flow velocity, divided by peak systolic flow velocity. Usually the lowest measured RI from a series of measurements is reported from different arteries. PI is defined as the difference between peak systolic and enddiastolic flow velocity, divided by the time-averaged flow velocity. The maximum systolic velocity is the maximum flow recorded in any visualized artery.

In order to make a measurement of either RI or PI or maximum systolic velocity, an artery must be identified on ultrasound. The inability to identify an artery in the mass means that the test cannot be performed. Therefore, not every individual included in the study population is captured with the assessment of these color Doppler modalities. Another limitation of these measurements is that the range observed in malignant masses overlaps with that observed in benign masses. For example, in Lin et al.,¹³² discussed in more detail below, the RI for malignant masses ranged from 0.23 to 0.82. Although they did not report a range for the benign masses, there were eight benign tumors with a RI < 0.4. This overlap limits the effectiveness of any threshold and, perhaps, contributes to the different thresholds reported in the literature.

Reproducibility of tests. Timmerman et al.¹⁹⁶ (discussed above under ultrasound morphology) included Doppler measurements in its analysis of interobserver variability and experience. In short, operators with more experience (300 versus 15,000 scans) had greater accuracy (92 percent versus 82 to 87 percent, p = 0.0001). Interassessor agreement was also greater between the most experienced operators (kappa = 0.852) compared with the less experienced operators (range 0.581 to 0.737). None of the articles evaluating color Doppler described operator experience, nor did any address interobserver variability specifically in regards to Doppler measurement.

Trials identified. Fifty-six articles were identified that described color Doppler analysis, comprising a description of 65 tests. Thirty-two articles evaluated RI, 20 PI, and six the maximum systolic velocity. These are the most common flow criteria measured in the literature and presumably in clinical practice as well. Other Doppler parameters were described in the literature sometimes in conjunction with either RI or PI or maximum systolic velocity but were not included in this table. The other articles included 10 that involved the visualization of flow within the mass,^{70, 71, 104, 105, 119, 137, 160, 161, 168, 182} two that involved counting the total number of arteries (either > 4¹⁵² or > 3¹⁹⁹), and one that measured the absence of a diastolic notch.¹³⁷

Results. Table 12 details the test characteristics of RI, PI, and the maximum systolic velocity in the evaluation of an adnexal mass, again using pooled values from a random-effects model. For RI the range reported was from ≤ 0.8 to < 0.4, with < 0.4 being the most common. For PI the range was relatively narrower from < 1.5 to < 1.0 with the majority of studies using either ≤ 1.0 or < 1.0. The reported range was greatest in the assessment of maximum systolic velocity, where there were also the fewest studies from > 30 cm/second to > 10 cm/second. As the threshold for RI decreases from ≤ 0.8 to < 0.4, the sensitivity and NPV decrease, and the specificity and PPV increase. This is seen most clearly in studies that evaluated a series of RI cutpoints with the same study population.^{132, 176}

Lin et al.¹³² evaluated 370 women with adnexal masses who were scheduled for surgery at a single institution. They reported outcomes based on RI cutpoints of 0.4, 0.5, and 0.6. For RI < 0.4, the sensitivity, specificity, PPV, and NPV were 0.69, 0.97, 0.89, and 0.91, respectively. For RI < 0.5, they were 0.79, 0.92, 0.77, and 0.93. And for < 0.6, they were 0.91, 0.86, 0.68, and 0.98. The authors conclude that the 0.4 cutpoint yields the highest concordance rate between Doppler prediction and histopathologic diagnosis. This conclusion, however, is based more on clinical impression, as ROC curve analysis was not performed.

The range of Doppler study performance is listed in Table 12 and shown in Figures 11–13. Overall there was great heterogeneity of performance results. The range of sensitivity was largest for RI. This range did not appear to be secondary to differences in RI thresholds; however, the < 0.4 threshold did appear to narrow specificity results. In spite of the large variation in thresholds described for maximum systolic velocity, the range of test characteristics was somewhat narrower than that for RI, probably because there were fewer studies identified that used this measurement. Again, there is a trade-off between sensitivity and specificity, although this appears greatest for maximum velocity.

Table 13 compares the characteristics of Doppler studies that did not use measurement or calculation of Doppler waveforms. They relied instead on either the presence of vascularity within the mass (yes/no) or on a direct count of vessels seen. These tests seem to perform as well as the RI or PI in terms of sensitivity, although specificity varies quite widely (Figure 13). Valentin¹⁸² measured both the PI (< 1.0) and the presence of color lakes visible on Doppler in the same study population. Of 151 patients, PI was measured in 135, indicating that for 16 individuals, no artery was visualized within the mass. The sensitivity reported for the PI was 0.83, specificity 0.34, PPV 0.20, and NPV 0.91. Simply documenting the presence or absence of visible color lakes on Doppler yielded a sensitivity of 0.88, a specificity of 0.67, a PPV of 0.33, and a NPV of 0.97. Not only did the direct visualization test perform better, but because its outcome was a simple binary outcome (present or absent), the results included the entire study population (n = 151). Prompeler et al.¹⁵² measured RI, maximum systolic velocity, as well as the number of arteries visualized in the mass. Their data for the simple counting of arteries also performs as well if not better than the calculated tests such as RI or PI. In a random-effects model, pooled sensitivity for the presence or absence of blood flow within a mass was 0.88 (95% CI, 0.80 to 0.92) and pooled specificity 0.78 (95% CI, 0.65 to 0.87)

Stratification by menopausal status. Out of a total of 56 studies identified that evaluated color Doppler, only 11 contained data that either directly reported test characteristics by menopausal status or contained enough information to enable the stratification of results. Two of these studies were in a 100 percent postmenopausal population, and nine enabled comparison by menopausal status within the same study population (Table 14). When comparing test performance within the same study population stratified by menopausal status, the PPV of the test is significantly increased in the postmenopausal group. In Salem et al.,¹⁵⁸ the PPV increased only from 0.20 in the premenopausal group to only 0.47 in the peri- and postmenopausal group. This may be a reflection of how they defined peri- and postmenopause (which was not clearly stated by the authors). After stratifying the reported results by age (> 45), the PPV is 0.73. This increase in PPV among postmenopausal women appears to be greater in the context of Doppler studies than that observed with ultrasound morphology. This finding differs from the one meta-analysis on the subject. Kinkel et al.¹⁹⁷ did a systematic review of both ultrasound morphology and Doppler in the detection of malignant masses. Although they noted a difference in outcomes dependent on menopausal status, this difference did not reach statistical significance. Interestingly, there was a difference in terms of Doppler test performance by year of publication with better results demonstrated by earlier studies (p = 0.005), a result that was independent of sample size.

Combined Ultrasound Morphology and Doppler

A limiting feature of ultrasound morphologic assessments has been felt to be the high rate of false positive test results.¹⁹⁶ Color Doppler, in contrast, has displayed a slightly higher PPV, especially in the earlier studies.¹⁹⁷ There have, therefore, been attempts to combine ultrasound morphology and Doppler studies in a single test.

Trials identified. Of all the articles that investigated the use of either ultrasound morphology or color Doppler in the evaluation of an adnexal mass, nine articles containing a total of 13 tests described a combination ultrasound morphology and Doppler modality.^{65, 79, 91, 100, 123}–^{125, 130, 201}

Results. There is a large range in the reported study performance (sensitivity ranges from 0.71 to 0.98, specificity from 0.6 to 1.0. The relevant studies are shown in Figure 15; all but two had both sensitivity and specificity above 0.80. Pooled sensitivity in a random-effects model was 0.89 (95% CI, 0.81 to 0.93) and pooled specificity 0.91 (0.80 to 0.96). Both of these values were higher than the pooled values for any morphology or Doppler method alone.

Stratification by menopausal status. There were two studies that analyzed combined ultrasound morphology and Doppler in 100 percent post menopausal patient populations. Kurjak et al.¹²⁶ reported a combined sensitivity, specificity, PPV, and NPV of 0.90, 0.94, 0.90, and 0.94, respectively. Their combined test consisted of RI < 0.41 and an ultrasound morphology scoring system unique to them. Veunto et al.¹⁸⁶ in a population-based screening study reported a sensitivity, specificity, PPV, and NPV of 1.00, 0.83, 0.006, and 1.00, respectively. Given that these two studies are of greatly different design, it is hard to compare them directly. Comparing Kurjak et al. to the range of combined ultrasound and Dopper studies, it appears that in the postmenopausal group, the test has a better performance. However, this test performance may reflect patient selection criteria for the study that was not clearly explained. Combination modalities as a screening tool for ovarian cancer had a high false positive rate (as seen in the PPV of 0.006¹⁸⁶).

3D Versus 2D Ultrasound

We identified five studies that analyzed 3D ultrasound. Four are listed in Table 15. The fifth, by Cohen et al.,³³ was not included because it compared 2D ultrasound with 2D plus some component of a 3D exam (possibly 3D Doppler) that was not clearly stated in the article. Overall, 3D ultrasound appears superior to 2D especially in regards to sensitivity and PPV performance. We were unable to stratify these results by menopausal status. Test reliability and variability were not addressed specifically in terms of 3D ultrasound.

Other Imaging Modalities

Although ultrasound remains the most common imaging modality in the evaluation and diagnosis of adnexal masses, newer technologies such as MRI, CT, and positron emission tomography (PET) have been studied as well. These modalities may not be as readily available to the clinician as ultrasound, and there is less literature devoted to them than to ultrasound; however, they are included in this review because of growing interest both clinical and research in their use. Further, despite refinements in ultrasound morphology scoring systems or Doppler measurements, the overall performance of ultrasound in the evaluation of the adnexal mass may be relatively fixed by the technology itself. Therefore it is necessary to investigate other imaging modalities and see how they compare with ultrasound.

Reproducibility of tests. Unlike ultrasound, MRI, CT, and PET images are not operator dependent in terms of obtaining the images. There is, however, the potential for interobserver variability in their analysis. There are no standardized morphological scoring systems for any imaging modality other than ultrasound. We identified two articles that directly addressed the issue of test reproducibility for either MRI and/or CT in the evaluation of adnexal masses. Buist et al.,⁷⁸ however, reported a series of 64 women who were evaluated by both MRI and CT and reviewed by two different radiologists. They reported a kappa value for the interobserver reliability for distinguishing between benign and malignant disease of 0.28 for CT and 0.41 for MRI. Yamashita et al.¹⁹² also calculated kappa values for interobserver variability among five radiologists. They showed far greater agreement: for precontrast MRI, kappa = 0.71 (± 0.02); for contrast-enhanced MRI, kappa = 0.73 (± 0.02).

Trials identified. We identified 17 articles comprising 22 tests. There were 15 articles for MRI, three for CT, and three for PET and one that used a combined CT/MRI test. There were two articles that investigated nuclear medicine technologies in the evaluation of adnexal masses. These, however, were not included in the review given the experimental nature of such tests at this time. The PET studies were all performed also using tracer 18-Fluorodeoxyglucose (FDG) with the test measuring uptake of FDG in the lesion.

Results. The results of MRI, CT, and PET modalities are summarized in Table 16. All of the articles describing CT and PET and most of the articles describing MRI either used descriptive criteria for differentiating malignant from benign appearing lesions or did not report the criteria used. Only two articles for MRI used a scoring system, slightly different from each other, which increases the difficulty in comparing studies. To date, there are no standardized scoring systems for any imaging modality other than ultrasound.

The range of test performance of MRI, CT, and PET are shown in Table 16. Table 17 includes, for comparison, the test performance for ultrasound morphology, color Doppler (all the modalities), and ultrasound morphology and Doppler combined. Tian et al.¹⁷⁷ was excluded from this table because there was no description how CT and MRI were combined for a single test result (in series versus in parallel). Overall the sensitivity for MRI, CT, and PET are similar to that of combined ultrasound morphology and Doppler and less heterogenous than either modality separate. The specificity, however, is equivalent to either test separate and wider than the tests combined, with the exception of FDG-PET. However, the comparatively narrow range of both CT and PET results could be secondary to the relatively few studies that use these modalities. There is a large range of results for PET PPVs and a small range for CT, again possibly reflecting the paucity of studies. The range of NPVs for MRI is comparable to that for combined ultrasound morphology and Doppler and better than either CT or PET. Overall MRI appears similar in performance to combined ultrasound. More research is needed to accurately assess the performance range of CT and PET.

Another way to compare imaging modalities is by looking at studies that compare imaging modalities within the same study population. These are listed in Table 17. There may be a small benefit in performance of MRI over ultrasound, especially in terms of PPV. There is no evidence to support the superiority of any single modality, although FDG-PET appears inferior to the rest.

Only two studies compared pre- and postcontrast enhancement with MRI.^{111, 192} Contrast enhancement improved evaluative performance in both studies, particularly sensitivity. In Hricak et al. the sensitivity increased from 0.87 to 0.95, specificity from 0.75 to 0.79, PPV from 0.78 to 0.83, and NPV 0.84 to 0.94.¹¹¹ These results are similar to those of Yamashita et al.¹⁹² in Table 17.

Stratification by menopausal status. None of the studies describing MRI, CT, or PET reported results either by menopausal status or in data that would allow menopausal status to be stratified.

Serum Markers: CA-125

The concept of using tumor markers as either screening or diagnostic tests for ovarian cancer is dependent upon identifying an abnormal level of a particular marker in serum, reflecting a systemic effect of disease in the ovary. The most extensively investigated ovarian cancer associated antigen is CA-125. This antigen is recognized by a murine monoclonal antibody produced using an ovarian cancer cell line as an immunogen. Elevated levels are detected in approximately 80 percent of ovarian carcinomas at the time of diagnosis;^{136, 167} however, elevated serum levels have also been reported in a variety of benign conditions, potentially affecting specificity. In addition, CA-125 is not as commonly elevated in non-epithelial ovarian cancers. Because these stromal and germ cell tumors are proportionately more common in pre-menopausal women, the sensitivity of CA-125 may it is not as sensitive in premenopausal women.³

Reproducibility of tests. Only one study included specific information regarding the inter- and intra-assay coefficients of variation.⁶⁶ They were < 7.5 percent and < 5.3 percent, respectively. The sensitivity of the assay in this study was < 5 U/ml.

Trials identified. We identified 66 studies that investigated the use of CA-125 as a serum marker in the evaluation of an adnexal mass. One study was a population-based screening study that employed CA-125 as part of the screening triage.⁵¹ Forty-six studies in total used 35 U/ml as a threshold - in 37 it was the only threshold used, whereas in five, both 35 U/ml and another threshold were reported for the same patient population. There were 24 studies that reported a threshold other than 35 U/ml ranging from >20 U/ml to >100 U/ml. In addition to the five studies that reported 35 U/ml and an additional level, there were four other studies that reported two threshold levels within the same study population. All but one of the studies were case series. Although there were a few studies that compared CA-125 results from operative cases with normal controls, only the data from the operative series were included in the 2-by-2 tables. The clinical presentation of the cases was rarely described. Some of the series were drawn from oncology clinics

Results. At the most commonly used threshold of 35 U/mL, the pooled sensitivity of CA-125 for discriminating benign from malignant lesions was 0.78 (95% CI, 0.75 to 0.81), and the pooled specificity 0.78 (95% CI, 0.71 to 0.82). Individual study sensitivities ranged from 0.45 to 1.0, and specificities from 0.46 to 0.99; see Figure 16, where the trade-off between sensitivity and specificity resulting from different thresholds is clearly seen. Not including the one screening study in this series,⁵¹ the studies ranged in size from 52 to 429 individuals. Unlike ultrasound morphology assessments, the range of CA-125 performance is not influenced by the heterogeneity of evaluative modalities. However, the results of performance have, overall, a similarly broad range. This most likely reflects heterogeneity of study populations. As very few studies actually reported how patients were diagnosed with masses, it is impossible to accurately stratify these results by patient characteristics. As with ultrasound measurements (both morphology and Doppler), the narrowest range of CA-125 test performance was with NPV, making this, perhaps, the most reliable part of the test itself.

The only screening study identified for CA-125 in our literature search⁵¹ included 2000 women. The sensitivity in this study was 1.00, specificity 0.99, PPV 0.17, and NPV 1.00. Few of the other studies achieved this degree of sensitivity, specificity, or NPV, although overall the PPV was higher. In the presence of an adnexal mass, the false negative rate increases compared with a screened population reflecting the fact that benign gynecologic disease can cause elevation of CA-125.

The most common threshold other than 35 U/ml was 65 U/ml. Most of the studies using 65 U/ml as a threshold were from Asia. The probable heterogeneity of study populations makes comparisons between these levels limited. Looking at the studies that reported results for different levels of CA-125 for within the same study population,^{87, 98, 134, 136, 147, 148, 162, 167, 180} in the higher threshold measurement, the specificity and PPV are higher, the sensitivity is lower, and the NPV is only slightly lower.

Stratification by menopausal status. Of the 59 studies we identified that examined CA-125, only nine contained data that either directly reported test characteristic by menopausal status or contained enough information to enable the stratification of results. One study was conducted exclusively in a postmenopausal population.⁶³ The studies are listed in Table 18.

The incidence of ovarian cancer is higher in postmenopausal women relative to benign gynecologic conditions, which also increase CA-125 levels. This should translate into a greater accuracy of CA-125 test performance in this population. Indeed, all test parameters except NPV are both higher and the range narrower in postmenopausal women. The lowest PPV was 0.73, with the remaining above 0.85, which is significantly higher than the range of PPV observed in studies that did not stratify their results by menopausal status. The NPV is lower in the postmenopausal population, despite the higher sensitivity, because of a greater prevalence of cancer in this population. CA-125 is consistently more helpful in discriminating benign from malignant lesions in postmenopausal women compared with premenopausal women.

Other Serum Markers

The fact that CA-125 is < 35 U/ml in 20 percent of women with early stage ovarian cancer, has motivated research into other serum based tests. We identified 13 articles that described a total of 17 different sera studies in women with an adnexal mass. Some studies investigated the performance of other tumor-associated antigens such as tumor-associated glycoprotein 72 (TAG-72) or CA-19-9. Although most of the tumor-associated antigens achieved specificities of approximately 0.82 to 0.92, the sensitivity, PPV, and NPV were overall lower than those reported for CA-125. Two studies investigated carcinoembryonic antigen (CEA),^{114, 157} and although they employed slightly different thresholds, the sensitivity reported in both (0.16 and 0.22) are so poor as to lead both authors to conclude that assessment of CEA in the evaluation of an adnexal mass is not helpful. Roman et al.⁴² investigated whether the addition of human chorionic gonadotropin (hCG), alpha-fetoprotein (AFP), and lactate dehydrogenase (LDH) to CA-125 improved the test performance. In their series the sensitivity of CA-125 alone was 0.67, the specificity was 0.71, PPV 0.35, and NPV 0.90. The addition of the other three tests did not change the test results very much. The combined test (defined as any of the markers positive) sensitivity was 0.72, its specificity was 0.70, PPV 0.36, and NPV 0.94. AFP, hCG, and LDH do not appear to improve the diagnostic performance of CA-125.

Gadducci et al. investigated the role of D-Dimer in a series of 121 women with adnexal masses.⁹⁶ The sensitivity for D-Dimer alone was 0.91, the specificity was 0.83, the PPV 0.82, and the NPV 0.92 - making D-Dimer one of the best performing tests identified in our review. Stratifying by menopausal status showed a greater performance in premenopausal women where the sensitivity, specificity, PPV and NPV were 1.00, 0.91, 0.75, and 1.00 respectively (n = 57). For postmenopausal women they were 0.89, 0.65, 0.85, and 0.72, respectively. Chalas et al. investigated the role of elevated platelets in 241 women.³¹ The specificity and PPV were similar to that reported for D-Dimer (0.84 and 0.83, respectively), but the sensitivity and NPV were significantly lower (0.56 and 0.59). These two studies are intriguing, but the results need to be established in future studies to better assess their possible contribution to the evaluation of adnexal masses.

Aside from D-Dimer, none of the studies contained information making stratification by menopausal status possible. In conclusion, none of the sera markers investigated in this review appears to perform better than CA-125, with the possible exception of D-Dimer in the premenopausal population.

Population-based Studies

Almost all of the studies identified were case series. There were, however, 13 population-based screening studies included in this review. They are listed in Table 19. Although all of the women included in these studies did not have a diagnosis of an adnexal mass at the time of enrollment, these studies are included here because they highlight some important issues about test performance. The strongest studies from a methodological perspective were those by Marchetti et al.,¹⁴⁰ Vuento et al.,¹⁸⁶ DePriest et al.,³⁴ Adonakis et al.,⁵¹ and Tailor et al.¹⁷¹ Marchetti, Vuento, Tailor and DePriest all used ultrasound as a screening modality. In all of these studies, the PPV was low, ranging between 0.006 to 0.07. Screening with CA-125 yielded a slightly higher PPV of 0.17.⁵¹ Tailor et al.¹⁷¹ offered followup screening within the same populations. In the first screening episode, which captured the total study population of 2,500 women, the test characteristics were similar to those reported in the other screening studies. The test characteristics improved, however, with subsequent screening. Women who had a negative screen were offered either a 12- or 6-month repeat ultrasound (depending on individual risk factors). Nine hundred and ninety-eight women received a second ultrasound screening. For this subset, the PPV improved to 0.21. For women screened greater than two times, the PPV was 0.25. However, not all women offered additional screening returned for the ultrasound. This potential bias was not discussed by the authors, and it is unclear how it may have influenced the performance of repeat screening. The three studies by Kurjak et al. each had various biases that could have accounted for their markedly different reported test performances. One did not report followup on test negatives and therefore included no false negative in the series,¹²⁶ another study population was an undescribed subset of a larger still incomplete screening series,¹²⁷ and the last study did not describe inclusion criteria.¹²⁸ Van Nagell et al.⁴⁹ screened 14,469 women with ultrasound. They reported their results 12 months from the time of screening. However they note that four women were diagnosed with cancer greater than 12 months after screening. These women had all screened negative and were included in their analysis as true negatives. Reclassifying these individuals as false negatives changes the sensitivity from 0.81 to 0.68.

Methodological Issues

In reviewing the literature on evaluation modalities, numerous methodological problems consistently reduced our ability to draw conclusions about the performance of various tests both individually and in comparison with each other. Some of these problems concerned study design, others related to statistical issues.

Patient population. With the exception of the 13 population-based screening studies, all of the articles were case series. Some were consecutive and others non-consecutive. Some were based on operative cases within a specific time frame at one or several institutions, whereas others were referral series, often located in oncology clinics. The path to diagnosis was almost never described, making it difficult to asses the generalizability of the results. Further, age was the only patient characteristic that was reliably documented. Other characteristics, such as family history, were almost never included. This has several implications. The overrepresentation of operative cases especially from academic facilities, likely overrepresents the prevalence of malignancy in the study populations when compared with the population of women with adnexal masses in general. It also exaggerates the performance of the evaluative modalities, especially in regards to sensitivity and PPV. Finally, it limits the generalizability of the evidence.

Definition of malignant. There was inconsistency between studies regarding whether the malignant classification included any malignancy or whether it included only ovarian malignancies. The inclusion of all malignancies would exaggerate the test's specificity and PPV at the expense of its sensitivity and NPV. From a practical standpoint, this difference may not be that problematic, as all malignancies are important. However, this classification bias increased the heterogeneity of test performance and limits generalizability. Finally, almost all of the articles that reported series containing tumors of low malignant potential (LMP) (also called borderline) classified these tumors as malignant. This changes the reported performance of the various evaluative modalities in these studies. There were three studies identified where stratification by LMP was possible. These are listed in Table 20. Classifying LMP tumors as malignant increases the specificity and PPV relative to classifying them as benign, while decreasing the specificity and NPV. Overall, PPV tended be somewhat low (even in populations with high prevalences of disease). The inclusion of LMP tumors into the malignant category inflated this measurement somewhat. Obviously, because of uncertainty about the natural history of LMP tumors, the most appropriate way of classifying them as part of diagnostic test evaluation is also uncertain. Given this uncertainty, ideally investigators would report results using alternative methods of classifying LMP tumors.

Variability in test criteria. Of the 69 articles that evaluated ultrasound morphology, only 31 used established scoring criteria; 38 used a novel method. This resulted in a great heterogeneity of tests for ultrasound morphology and contributed to the range in performance noted. Many of the studies employed purely descriptive analysis to arrive at a benign versus malignant diagnosis. This limits the reproducibility of those results. Many of the scoring systems and descriptive categories had never been independently verified, and the paucity of details regarding what constituted a positive test makes such verification impossible. In terms of ultrasound evaluation by color Doppler, there was also a range of reported thresholds. Some of the variability in test criteria reflects the limitations of ultrasound technology. However, such differences limited the comparability between studies.

Masses as numerator. While most studies examined persons as the unit of 2-by-2 analysis, there were many studies that analyzed their data by masses. Even though the number of persons in the study was usually reported, it was often impossible to reconfigure the 2-by-2 table to refer to persons not masses. This was especially true in the radiology literature. This influenced the comparability between studies.

Menopausal status. Most of the studies did describe the patient population in terms of age. We were able to calculate the proportion of menopausal patients in most studies. However, the results were rarely reported in a way that allowed stratification by menopausal status. Where stratification was possible, a difference in test performance was seen. The heterogeneity in test performances was magnified by the different proportions of pre- and postmenopause in the different study populations.

Sample size. Few studies discussed sample size issues, potentially leading to inappropriate conclusions, especially regarding comparability of test characteristics.

Failure to account for observer variability. No studies attempted to account for the effects of observer variation on the precision of estimates, although a few did calculate interobserver coefficients. For tests where the thresholds for normal and abnormal were based on either qualitative assessments (such as descriptions of ultrasound morphology) or quantitative measures (such as ultrasound morphology scores), this variability will have implications for the precision of sensitivity and specificity.

Prevalence and predictive value. We did not limit our analysis of test characteristics to studies from the United States. As the incidence of ovarian cancer is different in different countries, this influences the range of predictive values reported in the literature. Locations with low disease prevalence will have low PPVs compared with higher prevalence areas. The heterogeneity of study locations influenced the range of reported test characteristics and somewhat limits the comparability of the results.

Summary

Table 21 summarizes the pooled sensitivity and specificity estimates for CA-125 and the various imaging modalities.

The use of established scoring systems in the evaluation of an adnexal mass by ultrasound morphology appears to perform slightly better than simple descriptive assessment. However, there does not appear to be a benefit of one scoring system over another. Based on small numbers of studies, 3D ultrasound shows some improvement over 2D. Although the pooled sensitivity and specificity of MRI was the highest of any imaging modality, its performance was less consistent in studies where it was directly compared to other modalities such as CT and ultrasound.

Color Doppler assessment by RI, PI, and maximum systolic velocity are not superior to the more simple assessment of the presence or absence of arterial vessels within the mass. The efficacy of RI, PI, and maximum systolic velocity are hampered by the overlap in values of these measurements between benign and malignant masses.

Combined ultrasound morphology and color Doppler assessments have higher sensitivity and specificity compared to either alone. Although ultrasound morphologic evaluation by a gynecologist appears to be as reliable as that performed by a radiologist, there was no evidence of Doppler measurements done outside of the context of a radiology referral.

In postmenopausal women, an elevated CA-125 is useful for helping rule in ovarian cancer.

Qualitatively, there was a consistent trade-off across all tests between sensitivity and specificity.

The relatively low PPVs in all of the tests are particularly striking given that many of the included studies were done in preoperative patients; the likely “screening” done prior to a decision for surgery suggests that the PPV of a particular test in the initial evaluation of an adnexal mass is likely to be even lower.

Approach

Explicit scoring systems were sought in the medical literature from among all studies of diagnostic assessment of adnexal or pelvic masses. We considered only scoring systems that combined data from more than one category of the following types of information: (1) imaging findings; (2) patient risk factors; and (3) laboratory data. Clinical prediction rules that utilized data entirely from only one category (for example, ultrasound based morphological indices⁵⁶) are described as part of Question 3.

Imaging findings could include: (1) ultrasound based tests, such as transabdominal or transvaginal 2D ultrasound or Doppler ultrasound; (2) radiographic tests, such as CT; or (3) other imaging studies, such as MRI or PET scans.

Patient risk factors include menopausal status, age, or other risk factors.

Laboratory data was primarily CA-125, but we recorded data on other serum tumor markers as well.

Results of Literature Search and Screening

We identified 36 studies that met the inclusion criteria.^{42, 48, 51}–^{53, 55, 62, 63, 66, 72, 86, 87, 97, 103, 105, 116, 134, 135, 138, 139, 147, 169, 178, 180, 181, 185, 202}–²¹¹ These are described in Evidence Table 4 (Appendix D ^*).

Study Characteristics

Scoring systems identified. The scoring systems were of several types. The most common were models developed using statistical modeling techniques such as logistic regression (or artificial neural networks) to develop estimates for predicted probability of malignancy. Such estimates were then used to construct clinical prediction rules (e.g., the Risk of Malignancy Index [RMI], which calculates a numeric score based on CA-125 level multiplied by a menopausal score and an ultrasound morphology score) and decision thresholds (e.g., for RMI, the most common threshold is 200). Other scoring systems used simple combinations of criteria based on individual modalities, which were then combined using Boolean and or or (e.g. CA-125 > 65 U/ml and ultrasound morphology score > 10 points). Some models were validated in separate populations from the data set used to develop the scoring systems either described as part of its initial development, or in subsequent publications by the original developers or others.

Types of data incorporated. The most common scoring systems used ultrasound, CA-125 and menopausal status. Some type of ultrasound data was used in all 36 publications; studies varied with regard to the type of ultrasound technology that was used. All used 2D ultrasound to evaluate morphology, some using transabdominal and many using transvaginal probes. Studies that used Doppler ultrasound used a variety of parameters, including measures as simple as detection of flow, or as complex as specific indices derived from Doppler-measured flow rates, such as the RI or PI. Many described scoring rules based on combinations of features of morphology (Finkler score) or combined morphology and blood flow.

CA-125 was a component of the scoring system in 30 reports; other serum tumor markers included CA-72-4, incorporated into two reports,^{53, 63} and the markers AFP, LDH, and hCG, were used in one report.⁴² All studies that used these other serum markers also used CA-125.

Menopausal status was incorporated into scoring systems of 19 reports. The definition of menopausal status varied across studies, and in a few cases age was used as a proxy for clinically determined menopausal status. Three studies included only postmenopausal women,^{62, 63, 135} and thus could not use this variable in the scoring system.

Physical examination was a component of scoring systems in six reports.^{42, 51}–^{53, 62, 63}

Type of study populations. Most study populations were case series assembled at the time of referral for surgery and collected either at the point of preoperative ultrasound imaging or preoperative surgical evaluation. No studies were based in primary care clinical populations. One study described evaluation of adnexal masses detected during an ovarian cancer screening program.⁵¹

Reporting of study populations. Menopausal status of the study populations was described in 28 of the 36 reports; three reports included only postmenopausal women.^{62, 63, 135}

Age was reported for the study population as a mean or median in 18 of 36 studies; it was reported in categories in one additional study. Symptom status was seldom described in the candidate reports.

Race/ethnicity was not reported in any of the studies.

Risk factors for ovarian cancer (besides menopausal status and age, describe above) were not reported, except in one study that reported the proportion of the study population that was nulliparous versus multiparous.¹³⁸

Methodology. The methodological quality of the included studies may be described as follows:

Reference standard (handling of borderline). Some studies, particularly those assembled at the time of ultrasound investigation rather than surgery, encountered women with masses due to simple cysts with low risk of malignancy. Two studies allowed use of an operative report in lieu of histopathology as a reference standard,^{87, 116} and one used clinical followup without surgery as an alternate reference standard.⁴⁸

Verification bias. Fourteen studies failed to verify disease status for all or a significant sample of test-negative women.

Test reliability. Only nine studies provided data on the reliability of test assessments.

Sample size. Only 11 of the reports described a priori recruitment targets or sample size calculations. We excluded studies with fewer than 50 women; however, some studies report subgroup analyses with fewer than 50 women, for example, the subset of postmenopausal women in Strigini et al.¹⁶⁹

Use of appropriate statistical tests. The majority of reports (n = 28) used appropriate statistical analysis of the diagnostic data; however seven reports reported inadequate analyses.

Blinding. None of the reports described the use of techniques to blind investigators to the disease status of study patients.

Definition of positive and negative test. Most studies (n = 24) provided a priori definitions of a positive and negative test result; studies failed to meet this criterion most often when no explicit threshold was set a priori, but it was set based on study data.

Explicit validation method. Half of the reports (18/36) used some explicit validation method; many of the reports replicated previously described scoring systems in a new population. In many cases, these studies described new scoring systems which were not always validated.

The most common validation method was replication in a separate population. Two studies used validation techniques within a single study population: one split-sample,²⁰⁹ and one bootstrap.²⁰⁵

Diagnostic Accuracy of Scoring Systems

This section considers the diagnostic accuracy of the RMI (Jacobs 1990) and subsequent replications and refinements (RMI2, RMI3, Jacobs 1993, and Timmerman models).

RMI. The first scoring system based on a statistical model was published in 1990;¹¹⁶ it has been replicated in 11 subsequent clinical populations.^{55, 72, 87, 139, 147, 180, 204, 206}–^{208, 210} The diagnostic performance in these 12 studies is shown in Figure 17.

The RMI is a clinical prediction rule based on ultrasound, CA-125, and menopausal status data defined as follows:

RMI = U × M × CA-125

where ultrasound (transabdominal) is scored 1 point for each of the following characteristics: multilocular cyst, evidence of solid areas, evidence of metastases, presence of ascites, and bilateral lesions.

U = 0 for ultrasound score of 0

    = 1 for ultrasound score of 1

    = 3 for ultrasound score ≥ 2

CA-125 = Serum CA-125 in U/ml

Menopausal status

M = 1 if premenopausal

     = 3 if postmenopausal

In the initial report, Jacobs et al.¹¹⁶ used the cutoff value of 200. At this cutpoint, sensitivity was 85 percent and specificity was 97 percent among a population of 143 women undergoing surgical investigation for an adnexal mass. The performance of the initial model (study 6 in Figure 17) has, in most studies, failed to be equaled in subsequent attempts at validation. Three of the subsequent 11 studies have similar performance (studies 7, 9, 10 in Figure 17).^{72, 87, 208} It is notable that these three studies have fewer quality features (n ≤ 4) than the other eight studies (n ≥ 5 of 7 quality features).

When sensitivity and specificity are combined separately using a random-effects model, the pooled sensitivity is 0.78 (95% CI, 0.72 to 0.84) and the pooled specificity is 0.90 (0.81 to 0.95).

RMI2. In 1996, Tingulstad et al.¹⁸⁰ reported a refinement to the original RMI scoring system, commonly referred to as RMI2. RMI2 is defined identically to RMI except that new weights were used for the ultrasound and menopause components as follows:

U = 1 for ultrasound score of 0–1

    = 4 for ultrasound score ≥ 2

M = 1 if premenopausal

     = 4 if postmenopausal

A cutoff value of 200 was also recommended for RMI2. Like the RMI, the RMI2 scoring system has been replicated.^{134, 139, 147, 207} The original report of RMI2 found sensitivity of 0.8 and specificity of 0.92. Subsequent validation studies have performed no better. These validation studies all exhibited five or more quality features. The pooled sensitivity of all five studies is 0.77 (0.71 to 0.82), and pooled specificity 0.89 (0.85 to 0.91). The summary ROC curve is shown in Figure 18.

RMI3. Subsequently, a further refinement to the RMI and RMI2 was reported by Tingulstad et al.²¹¹ This third scoring system is defined identically to RMI and RMI2 except that new weights were used for the ultrasound and menopause components as follows:

U = 1 for ultrasound score of 0–1

    = 3 for ultrasound score ≥ 2

M = 1 if premenopausal

     = 3 if postmenopausal

A cutoff value of 200 was also recommended for RMI3. The RMI3 scoring system has been replicated in one additional study.¹³⁹ The original report of RMI3 found sensitivity of 0.71 and specificity of 0.92, while the validation study reported very similar performance, with sensitivity of 0.74 (0.65 to 0.83) and specificity of 0.91 (0.83 to 0.99).

Tailor and subsequent replications. Tailor et al.²⁰⁹ reported a scoring system based on an artificial neural network method that was based on a small population of 67 women total, 15 of whom had malignancies. Unlike the RMI family of systems described above, this system did not include CA-125, but considered age, menopausal status, and a variety of ultrasound morphological features and Doppler indices. While this system reported using 52 cases as a training set and 15 cases as a test set, the performance of the system was reported only for the study population as a whole: sensitivity 0.93 (95% CI, 0.81 to 1.0) and specificity 1.0 (0.94 to 1.0). Subsequently four studies have replicated this system showing markedly poorer diagnostic performance (Figure 19) when applied to separate populations, consistent with over-fitting in the initial model development.^{185, 203, 204, 207}

Twenty other scoring systems have been described, none of which has been as extensively replicated as the systems described above. Five of these other scoring systems have been validated in one other population as shown in Table 22; each of the systems was based on ultrasound morphology, CA-125, Doppler, and menopausal status. The models were: Timmerman LR1,^{178, 210} Timmerman AAN1,^{178, 207} Timmerman AAN2,^{178, 207} Timmerman LR2,^{178, 207} and Jacobs 1993.^{207, 212}

In each case, the initial diagnostic performance described by the system significantly degrades on replication in another population (Figure 20).

Ten additional systems were described in seven reports.^{42, 48, 53, 63, 181, 203, 205} Most of these studies used logistic regression or artificial neural network modeling methods to derive a new model. One used bootstrap validation techniques,²⁰⁵ but none was validated in another study population. One of these studies²⁰³ reported on newly fitted logistic regression models created by forcing variables that were include in previously described scoring systems.^{178, 209, 213} Aslam et al.²⁰⁴ constructed three separate models based on each possible pairwise combination of the three previously described models. The diagnostic performance of these miscellaneous unvalidated models is shown in Figure 21.

Thirteen further reports describe the diagnostic performance of simple rules for combining single test or single modalities into a decision rule.^{42, 51, 52, 62, 63, 66, 86, 97, 103, 105, 135, 138, 169} None of these criteria has been validated in another population. Each of these studies used dichotomous rules for two or more tests (or modalities) and then combined them using a simple rule like “malignant if any test positive” (Boolean or) or “malignant if all tests positive” (Boolean and). Some of the studies reported diagnostic performance of several different simple rules.

Twelve of these studies used ultrasound and CA-125, five incorporated physical exam, two included other serum tumor markers^{42, 63} and one used age over 50 years.¹³⁸

Six of these studies reported results for postmenopausal women separately: in three studies, the entire study population was postmenopausal^{62, 63, 135} while three studies reported diagnostic performance for the postmenopausal subgroup separately.^{103, 105, 169} The diagnostic performance of 18 simple combination rules in these six studies is shown in Figure 22.

In contrast, the diagnostic performances of 17 simple combination rules in studies that include both premenopausal and postmenopausal women in the study population are shown in Figure 23.

The results show a wide range of sensitivity and specificity. This variation reflects differences in decision thresholds (e.g., CA-125 > 35 U/ml versus CA-125 > 65 U/ml) and in the rules for combining tests (e.g., use of Boolean or versus and when combining results of two or more tests).

Discussion

No scoring systems were both developed and validated expressly for evaluating adnexal masses in postmenopausal women. Existing scoring systems that have been validated have all been developed in mixed pre- and postmenopausal populations. Those scoring systems that have been described in populations of postmenopausal women were neither rigorously developed (they consist of simple combination rules) nor validated in other populations.

The highest demonstrated specificity obtained with these scoring systems appears to be in the range of 90 to 95 percent and, at this range of specificity, the sensitivity appears to be in the range of 65 to 80 percent. However, as suggested by the performance in the few populations of postmenopausal women studied, the same degree of sensitivity and specificity is unlikely to be possible. Reliable estimates of the diagnostic performance of scoring systems cannot be determined from these studies.

This review of scoring systems demonstrates several important limitations of predictive models and has important implications for the clinical usefulness of these models and the future research in this area of inquiry. First, validation in an external population is critical to obtain accurate estimates of diagnostic performance, because all modeling techniques lead to overestimation of diagnostic performance in the data from which it was derived. This overestimation of diagnostic performance is clearly demonstrated by comparing the development and validation studies described for RMI, Tailor, and other scoring systems (Figures 17–20). The studies described here suffer from being relatively small for modeling; reliable variable selection and parameter estimation requires at least 10 to 15 cases (in this case, ovarian malignancies) for every term selected in a predictive model. Few, if any, met this statistical rule of thumb. This limitation is particularly apparent in the case of the Tailor model, where subsequent studies demonstrated a high degree of overestimation of the original model. Third, these studies used populations that were identified following referral for surgery in most cases, after some filtering had already occurred. Furthermore, these studies failed to describe the initial presentation (symptomatic or asymptomatic, palpable or non-palpable mass) of women eventually enrolled. Thus, the applicability of these studies to women in primary care, where an adnexal mass is often first noted, is uncertain.

Approach

For each study we sought to identify a population of patients with a screening abnormality which was “probably benign” and which the authors felt did not meet criteria for immediate surgical intervention. We then attempted to define the outcomes of further testing in the defined population, including the results of subsequent testing and final clinical outcome as defined by a pathology report or extended clinical followup. The interpretation of results is limited by the narrow scope of Question 5. Specifically, it is often difficult to identify a subgroup of patients with a screening abnormality which could be defined as a “suspected benign lesion” within larger screening studies. Often, results are not stratified with respect to these sub-populations, making it difficult to calculate sensitivity and specificity of the followup testing. In addition, by definition, it is also difficult to estimate the “sensitivity” of a followup regimen. We assumed that this refers to detection of cancer as part of the followup regimen, and that women with cancer diagnosed outside of the followup were “false negatives.”

Results

We identified nine articles meeting the criteria for this question;^{40, 127, 135, 145, 214}–²¹⁸ these are summarized in Table 23, with details in Evidence Table 5 (Appendix D ^*). Five were population-based screening studies of asymptomatic, postmenopausal patients without known ovarian masses;^{40, 127, 145, 214, 217} one was a voluntary screening program.²¹⁶ All addressed to some degree the use of interval ultrasound for detecting malignant masses. Although several used CA-125 as part of their followup, none reported any results based on the use of interval CA-125 in a population with adnexal lesions. None addressed the effects of changing the interval of testing on sensitivity and predictive value; the disparate nature of the studies prohibited any inferences on the effect of test interval on sensitivity.

Menon et al.¹⁴⁵ performed a large prospective screening study of 22,000 postmenopausal women older than 45 years. Initial screening consisted of CA-125; patients with CA-125 ≥ 30 underwent endovaginal ultrasound evaluation. Results were interpreted as normal (ovarian volume < 8.8 ml/normal morphology), equivocal (volume < 8.8 ml, abnormal morphology), or abnormal (volume ≥ 8.8 ml). Normal morphology was defined as uniform hypoechogenicity and smooth outline. Abnormal morphology was defined as simple cysts or complex lesions. Patients with normal scans were triaged to repeat CA-125 every 3 months for a year and subsequently returned to yearly screening; median followup was 6.8 years, with loss to followup not reported. Patients with abnormal scans were referred to a gynecologist for consideration of surgical intervention. Patients with equivocal scans were triaged to repeat ultrasound at 6-week intervals until a scan could be classified as normal or abnormal. Of 741 patients who were triaged to ultrasound, 20 (2.7 percent) index cancers were identified. We focused on the group of patients with “equivocal” scans who were triaged to interval testing in an attempt to answer the study question. There were 17 equivocal scans. Of these, nine had simple cysts which were followed and did not result in a cancer diagnosis (true negatives). One patient died of pneumonia prior to her first repeat ultrasound, and one died of advanced ovarian cancer prior to her first repeat ultrasound; this cancer death could possibly be considered a false negative for the followup strategy, although it could also be considered a false negative from the original study since the death occurred within 6 weeks of the initial scan. Six patients were scheduled for surgery following an equivocal scan, presumably due to abnormal followup ultrasound. One of these had ovarian cancer (true positive), and the other five had benign disease (false positive). Because the number of equivocal scans was so small, and because the classification “equivocal” does not necessarily imply that the lesions were felt to be “suspected benign” as designated in Question 5, it is not possible to calculate the sensitivity and specificity of prolonged monitoring strategies using this study. The authors do not draw any conclusions regarding the appropriateness of interval testing.

Modesitt et al.⁴⁰ performed a large screening study of 15,106 asymptomatic women at least 50 years old without a history of ovarian cancer. Patients were screened with TVUS. Criteria for abnormality were ovarian volume > 10 ml and any morphologic abnormality, including simple or complex cysts. Patients with abnormal TVUS were triaged to repeat TVUS in 4 to 6 weeks, with Doppler flow ultrasound, CA-125 level, and tumor morphology indexing performed at the second visit. Patients with simple unilocular cysts which were considered likely benign were triaged to repeat TVUS every 3 to 6 months. Mean followup was 6.3 years. Two thousand and seven hundred and sixty-three (2,763) women were diagnosed with 3,259 unilocular cysts. Spontaneous resolution of unilocular cysts occurred in 2,261 (69.4 percent) of lesions. Ten patients subsequently developed ovarian cancer. Seven of these had additional abnormal areas which subsequently developed on TVUS (considered true positives because they were subsequently identified by interval testing). Two developed ovarian cancer after the cyst in question had resolved on sonographic followup (these might be considered false negatives). One patient developed cancer in the ovary opposite the cyst being followed (this might also be considered a false negative). Calculated on a per-patient basis, the sensitivity and specificity of followup testing in the population with a simple unilocular ovarian cyst are 70 percent (95% CI, 41.6 to 98.4 percent) and 100 percent (99.9 to 100 percent), respectively. Because none of the unilocular cysts subsequently developed into a cancer, the sensitivity and specificity improve to 100 percent (57.1 to 100 percent) and 100 percent (99.9 to100 percent), respectively, when calculated on a per-lesion basis. Followup time is a major strength of this study. The authors conclude that unilocular ovarian cysts are associated with a very low risk of malignancy and can be safely followed with serial ultrasound.

Schincaglia et al.²¹⁷ performed a screening study of 3,541 asymptomatic postmenopausal patients. All patients underwent transabdominal ultrasonography with assessment of ovarian volume and morphology. Patients were divided into four groups based on the results of the initial ultrasound. All patients with ovarian volume > 15 ml (Group 4) were referred for repeat “level II” ultrasonography for morphologic assessment and fine needle aspiration (FNA) when feasible. Patients with ovarian volume between 9 and 15 ml (Group 3) were triaged to followup ultrasound at 3 and 6 months. Patients with ovarian volume < 9 cm but a cystic appearance (Group 2) were triaged to followup ultrasound in 6 months. Patients with ovarian volume < 9 ml and homogeneous appearance (Group 1) were considered negative and had no further intervention. Clinical followup at 1 year and pathology results if surgery was performed were considered the reference standard. Two hundred and eighty-three (283) patients (Groups 2 and 3) were deemed appropriate for followup using repeat ultrasound at 3- to 6-month intervals without the need for immediate referral for FNA/surgery. Of these 283 patients, 34 subsequently developed concerning ultrasound findings and were referred for a level II scan and/or possible FNA. The clinical results of this group of 34 are not given separately. Of the 249 who had non-concerning followup scans, none developed cancer with followup of at least 1 year (“true negatives”). Therefore, the specificity of ultrasound followup is 100 percent (95% CI, 98.8 to 100 percent) for patients with an initial abnormal but “probably benign” ultrasound. Sensitivity within this group cannot be calculated with the information given in the publication. The ability to answer Question 5 would be enhanced if specific outcomes of each of the four groups defined by the authors had been given. The study was also limited by the fairly short followup interval and the lack of prior or concurrent validation of the ultrasonographic groups defined in the study.

Kurjak et al.¹²⁷ screened 5,013 women 40 years old or older (30.6% postmenopausal), of whom 404 had simple cysts with a diameter between 2.5 and 5 cm and a resistive index greater or equal to 0.41. These women received a followup scan in 6 months. Investigators reported the results of 88 women for whom the 6-month scan results were available. The definition of change prompting further diagnosis was not explicitly described. Of the 88 women, 18 ultimately underwent surgery based on the findings at 6 months, with one cancer detected and 17 benign lesions. Results stratified by menopausal status were not provided. This study was limited by lack of details on clinical decision rules, and short followup.

Castillo et al.²¹⁴ screened 8,794 postmenopausal women; 215 had simple unilocular cysts less than 10 cm in diameter. Twelve percent of these masses were asymptomatic. These women underwent repeat ultrasound and CA-125 in 3 months, with subsequent followup studies every 6 months. Progression was defined as an increase in diameter of 1 cm or more, regression as a decrease of 1 cm, and resolution as absence of the cyst at 2 consecutive visits 6 months apart. Median followup was 27 months. There was one interval ovarian cancer between studies, and 44 women had benign masses removed. Although this study was among the highest quality studies in terms of reporting of relevant data, it is limited by the relatively small size and the high loss to followup (30.6%).

The remaining four studies^{135, 215, 216, 218} were all small (less than 200 patients), and of variable quality (Table 23). None reported any interval cancers in patients receiving followup, or cancers detected during followup. The study of Valentin et al.²¹⁸ was notable for length of followup (median 3 years) and complete ascertainment of followup status using Swedish cancer and death registry data.

Discussion

There are limited data available to support a global definition of “probably benign” ovarian lesions or to support a specific method of interval testing to identify ovarian malignancy among patients in whom such lesions have been identified. For the most part, studies are limited by small size, variable length of followup, variable definitions of significant change and thresholds for intervention, and methods for followup.

The question of how best to define and evaluate “sensitivity” of followup regimens is a difficult one. Several factors need to be considered. First, interval cancers presenting between the initial study and the first followup visit may well be considered false negatives of the initial study; alternatively, they may reflect a too-long followup interval. Second, given the lack of data on the natural history of ovarian cancer, it is unclear whether cancers developing in benign-appearing lesions represent subclinical cancers present at the time of the initial diagnosis, or new cancers representing malignant transformation of a benign cyst. If the latter, then the ultimate success of any followup regimen may depend as much on the natural history of a given malignancy as on the sensitivity and specificity of the tests used for followup. Finally, cancers identified during followup should ideally have high survival rates (although whether such high survival rates would reflect the efficacy of the followup or the biology of cancers which are associated with benign-appearing cysts is unclear). The number of cancers identified in the reviewed studies was too small to draw any inferences about relative survival.

Overall, only two interval cancers occurred during followup in the studies identified (one prior to the first followup scan), and 10 cancers were identified during followup. As noted, an additional three cancers developed after resolution of a cyst or in the contralateral ovary. The highest quality study⁴⁰ provides good evidence for the safety of prolonged followup with interval TVUS at 3- to 6-month intervals for patients with unilocular ovarian cysts of up to 10 cm in diameter, and the findings of the other studies are consistent with this conclusion.

Approach

We searched the literature for studies that reported the morbidity and mortality of surgical management of adnexal masses. We also used the Nationwide Inpatient Sample (NIS) discharge database, maintained by the Agency for Healthcare Research and Quality (AHRQ), to obtain estimates of morbidity and mortality associated with diagnostic laparoscopy or exploratory laparotomy for a range of diagnoses associated with adnexal masses. The NIS is limited to inpatient procedures and does not cover ambulatory surgical centers, where some adnexal masses are likely to be managed, especially those masses thought to have a low likelihood of cancer. In addition to surgical complications, we also examined articles that provided data on the test characteristics of frozen section pathologic diagnosis; especially in the setting of minimally invasive procedures, false negative results on frozen section might lead to suboptimal surgical management and delayed therapy, while false positive results might lead to more extensive surgery than necessary, with possible implications for increased surgical morbidity and affects on ovarian function.

Results of Literature Search and Screening

We identified 24 articles that met our inclusion criteria;^{32, 37, 41, 219}–²³⁹ these are summarized in Evidence Table 6 (Appendix D ^*). Twenty-two articles reported on the morbidity and mortality of surgical management of adnexal masses.^{32, 37, 41, 219}–^{234, 237}–²³⁹ In addition, two of the included articles reported on the sensitivity and specificity of frozen section;^{220, 236} false negative frozen section results could lead to inadequate surgical management and delayed treatment, while false positive results could lead to more extensive surgery than necessary. Finally, one of the included articles addressed the potential effect of conservative surgery for removal of an ovarian cyst resulting from endometriosis (endometrioma) on subsequent fertility.²³⁵

Methodological Quality of Included Studies

Size of population. None of the papers provided a description of the referral base; two^{32, 37} were limited to gynecologic oncology practices. Lack of information on the referral base prevents assessment of generalizability. Since all of these studies were performed in centers experienced in laparoscopic surgery, the generalizability may well be limited.

Number of cases. Five studies had fewer than 200 cases, with correspondingly wide confidence intervals for reported event rates. Two studies had larger numbers of cases, 683²³⁰ and 757.²¹⁹ However, the study by Marana et al.²³⁰ was limited to women under 40.

Patient selection. None of the studies reported how patients were referred to the surgical practices. All provided criteria for laparoscopic management of masses, based on various criteria to suggest high or low risk of malignancy. We found two trials where patients were randomized to laparoscopy or laparotomy,^{224, 225} but randomization methods were not well described.

Application of reference standard. In this sense, “reference standard” refers to the method by which a complication was diagnosed. Only two studies described followup beyond 8 weeks, but they did not detail whether all patients underwent similar followup protocols.

Results

There were three deaths in one study of 146 patients (all undergoing laparoscopy), and none in any of the other studies (a total of 5,599 patients). Pooling all patients, the mortality was 0.05 percent, with a 95% CI of 0.01 to 0.17 percent.

Table 24 shows the results from individual studies. The two randomized studies^{224, 225} both showed lower morbidity with laparoscopy compared to laparotomy, although only one of them²²⁴ had sufficient power to show a statistically significant difference. Although the study of Deckardt and colleagues²²⁴ was randomized, there were substantial differences in the procedures performed in each arm. Laparoscopy patients tended to undergo more conservative procedures: they were significantly more likely to have cystectomy (60.0 vs. 20.2 percent), less likely to have oophorectomy (0.8 vs. 20.2 percent), and less likely to have bilateral salphingo-oophorectomy (4.0 vs. 21.4 percent). Both studies where laparoscopy was directly compared to laparotomy showed increased complication rates (primarily postoperative complications) among the laparotomy patients. The four non-randomized studies all showed higher morbidity rates with laparoscopy, but there were substantial differences in patient selection criteria.

In series of laparoscopy cases, morbidity rates ranged from 0.9 percent to 22.1 percent (Table 24); series differed widely in their selection criteria for laparoscopic management of the mass. Few stratified results based on menopausal status; in some cases, postmenopausal patients were explicitly excluded. In one study where multivariate analysis was performed to assess for risks of morbidity, performance of additional procedures (hysterectomy) significantly increased the risk of morbidity, while a history of hysterectomy increased the likelihood of conversion to laparotomy (presumably because of increased technical difficulty secondary to postoperative adhesions).²²⁶

Nationwide Inpatient Sample

Table 25 shows the estimated numbers of discharges in the United States in 2000-2001 under each diagnostic class and procedure (standard errors not shown for simplicity). The results illustrate the difficulty in using discharge data to attempt to estimate morbidity and mortality rates for surgical procedures. Both morbidity and mortality are highest for cancer diagnoses, but there is no way to determine the extent to which the underlying disease process contributed to either complications or death; for example, “exploratory laparotomy” or “diagnostic laparoscopy” in many ovarian cancer patients likely represents a “second-look” procedure done to determine response to chemotherapy. Outcomes of these procedures are not relevant to estimating the risks of a primary diagnostic procedure. The laparoscopies that are included in the NIS are likely not representative of all laparoscopies for adnexal masses; since the NIS does not capture surgeries performed at ambulatory surgery centers, the cases within the NIS may represent those for which surgeons had a higher index of suspicion of malignancy, or anticipated higher technical difficulty. Another major limitation is the inability to distinguish between the initial indication for surgery and the final diagnosis. Finally, in order to try to eliminate confounding by additional procedures, we excluded cases in which hysterectomy was performed - however, because hysterectomy is part of the standard initial surgical treatment of ovarian cancer, many cases of initial management are excluded.