Objective
The purpose of this study was to investigate the relationship between imaging and the multivariate index assay (MIA) in the prediction of the likelihood of ovarian malignancy before surgery.
Study Design
Subjects were recruited in 2 related prospective, multiinstitutional trials that involved 44 sites across the United States. Women had ovarian imaging, biomarker analysis, and surgery for an adnexal mass. Ovarian tumors were classified as high risk for solid or papillary morphologic condition on imaging study. Biomarker and imaging results were correlated with surgical findings.
Results
Of the 1110 women who were enrolled with an adnexal mass on imaging, 1024 cases were evaluable. There were 255 malignant and 769 benign tumors. High-risk findings were present in 46% of 1232 imaging tests and 61% of 1024 MIA tests. The risk of malignancy increased with rising MIA scores; similarly, the likelihood of malignancy was higher for high-risk, compared with low-risk, imaging. Sensitivity and specificity for the prediction of malignancy were 98% (95% CI, 92–99) and 31% (95% CI, 27–34) for ultrasound or MIA; 68% (95% CI, 58–77) and 75% (95% CI, 72–78) for ultrasound and MIA, respectively. For computed tomography scan or MIA, sensitivity was 97% (95% CI, 92–99) and specificity was 22% (95% CI, 16–28); the sensitivity and specificity for computed tomography scan and MIA were 71% (95% CI, 62–79) and 70% (95% CI, 63–76). Only 1.6% of ovarian tumors were malignant when both tests indicated low risk. A logistic regression model to predict risk of malignancy is presented.
Conclusion
An understanding of how pelvic imaging influences the MIA score can help clinicians better interpret the malignant risk of an ovarian tumor.
Ovarian cancer is the leading cause of gynecologic cancer death in the United States, and fewer than 40% of women diagnosed with ovarian cancer will be cured. One of the recognized challenges is how to identify at-risk ovarian tumors for referral before the initial surgery. More than 15 years ago, the National Institutes of Health released a consensus statement that declared that a woman with an ovarian mass at high risk for malignancy should be given the option of having her surgery performed by a gynecologic oncologist. Many subsequent ovarian cancer publications have established that outcomes are improved with the involvement of a specialist ; yet, 2 of every 3 women in the United States are not referred to a gynecologic oncologist for their primary ovarian cancer surgery. There are several plausible explanations for the low referral rate; among them is that the low sensitivity of existing algorithms fails to alert the evaluating physician before surgery. This may be particularly important for premenopausal women who rarely are considered to be at risk for ovarian malignancy but who account for up to 20% of all ovarian cancers.
In 2006, Myers et al, who published a pooled statistical analysis for algorithms that were used to evaluate an adnexal mass, concluded that a combined strategy of imaging with biomarker was superior to either one alone. Until recently, cancer antigen 125 (CA125) has been the most used biomarker to evaluate women with an ovarian tumor. Unfortunately, the sensitivity of CA125 is reported to be 50% in early-stage disease and has a 20-25% false-negative rate in advanced-stage cancers. In premenopausal women, CA125 has a sensitivity of 50-74%, with a specificity reportedly as low as 26% for ovarian malignancy. OVA1 (multivariate index assay [MIA]) is a sensitive biomarker test specifically for use in the preoperative evaluation of ovarian tumors. In the leading publication, physician assessment plus MIA identified 86% of malignancies that were missed by CA125, and its clinical performance was consistent in early- and late-stage cancers. These findings were confirmed recently with a subsequent prospective investigation by Bristow et al. There are circumstances in which an ovarian tumor has a high-risk MIA score but a low-risk imaging study. In this situation, there are no published data to assist providers in making informed decisions about surgery.
This study was undertaken to better understand the relationship between ovarian imaging and the MIA in the preoperative evaluation of an adnexal mass.
Materials and Methods
Subjects were enrolled prospectively at 44 sites across the United States ( Figure 1 ) and included primary care women’s health clinics, obstetrics and gynecology groups, gynecologic oncology practices, community and university hospitals, and health maintenance organizations. These data were merged from 2 published national trials. Both trials had identical inclusion and exclusion criteria. The inclusion criteria included female age ≥18 years, a documented ovarian tumor with planned surgery within 3 months of imaging, agreeable to phlebotomy, and signed informed consent. The exclusion criteria were age <18 years, no planned surgical intervention, declined phlebotomy, or a malignancy diagnosis in the last 5 years, with the exception of a nonmelanoma skin cancer. Menopause was defined as the absence of menses for at least 12 months or age ≥50 years when not stated. Institutional review board approval was obtained from each site. All data were collected on standardized case report forms.
The methods for blood collection and specimen handling have been reported previously. Biomarker measurements were performed by Quest Diagnostics, Inc (Chantilly, VA); blinded validation testing was done at Johns Hopkins Medical Institutions (Baltimore, MD) and Specialty Laboratories (Valencia, CA).
The MIA test
The OVA1 test, which has been cleared by the Food and Drug Administration and is commercially available (Quest Diagnostics, Madison NJ), incorporates CA125-II, transferrin, transthyretin (prealbumin), apolipoprotein A1, and beta 2 microglobulin. The OvaCalc software program (Vermillion Inc, Austin, TX) combines the values for each assay and uses a proprietary algorithm to generate an ovarian malignancy risk index score for each. The numeric result ranges from 0.0–10.0, with the following clinical report:
Premenopausal: low risk for malignancy, <5.0; high risk for malignancy, ≥5.0; Postmenopausal: low risk for malignancy, <4.4; high risk for malignancy, ≥4.4.
Ovarian imaging
Preoperative imaging results, which included computed tomography (CT), ultrasound scans, or magnetic resonance imaging, were collected prospectively and analyzed retrospectively. Enrolling physicians were allowed to choose the type of imaging to be performed. Magnetic resonance imaging results were omitted from the analysis because of low numbers (n = 43). High-risk imaging criteria were selected based on univariate analysis of the study group. The following variables are statistically predictive of ovarian malignancy ( P < .001 for each): solid tumor components or papillary ovarian morphologic condition (odds ratio [OR], 4.2; 95% confidence interval [CI], 3.0–5.8), ascites (OR, 8.0; 95% CI, 5.3–12.1), and metastatic implants (OR, 28.3; 95% CI, 9.9–80.8). Ascites and metastatic implants are correlated highly with advanced disease; however, because the MIA test is not indicated for use in women with clinical evidence of advanced ovarian cancer, ascites and metastatic implants were omitted intentionally from the analysis. For this study, high-risk imaging was defined as any complex ovarian tumor with evidence of solid or papillary components. The low-risk category included unilocular or septate cystic ovarian tumors without high-risk findings. Tumor volume was not recorded in the original datasets. Study participants were permitted to have >1 imaging test, and each was considered an independent event. All imaging reports were reviewed individually by the 2 primary authors (S.T.G., F.T.U.).
Model for predicting ovarian malignancy
With the use of a logistic regression model, the relationship between the risk of malignancy and the MIA score was examined, and the first order (linear) term was highly significant ( P < .001) after the inclusion of menopausal status in the model. The addition of a quadratic term or higher order for MIA to the logistic model did not contribute to the fit (χ 2 test, 2.8; df = 1; P = .10). The model was extended to incorporate the radiologic findings (low risk, high risk), menopausal status, and the MIA score. A cutoff was selected to obtain a similar sensitivity to MIA alone (92.9%). The model uses the same prevalence of malignancy as measured in this study population (25%).
Statistical methods
Data were forwarded to Applied Clinical Intelligence (Bala Cynwyd, PA) for statistical evaluation. Biomarker and imaging results were stratified based on menopausal status, type of imaging, pathologic diagnosis, and stage of malignancy. Clinically relevant criteria such as sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), percentage malignant, and ORs were calculated. The adjusted PPVs and NPVs were determined with the prevalence (25%) of the entire study population. For comparative purposes, the PPV and NPV have been calculated at a lower prevalence of disease (10%). These calculations assume that the population would be unchanged in all aspects, except for prevalence of malignancy. Ninety-five percent CIs were constructed where appropriate. Logistic regression was used to model the risk of malignancy with the use of the actual MIA score in combination with menopausal status and imaging assessments. Statistical analysis was performed with SAS software (version 9.2; SAS Institute Inc, Cary, NC).
Results
Data from 2 large, consecutive prospective clinical trials were combined (OVA1: February 2007-April 2008; OVA500: August 2010-December 2011) to form the study group. Of the 1110 women who were enrolled in the 2 trials, 1024 women were evaluable with imaging results and MIA score. Seventy-five percent of subjects (770/1024 women) were enrolled by general gynecologists; the remaining 25% of subjects were enrolled by gynecologic oncologists. Women were excluded from analysis if surgery was not performed (27 women) or delayed >3 months (6 women), tissue not obtained or pathologic report not forwarded (34 women), blood specimen was unusable (9 women), subject had previous cancer (5 women), imaging was performed out of window for inclusion (4 women), or imaging study did not confirm an adnexal tumor (1 woman). The demographic and clinical characteristics of the study group are summarized in Table 1 . The surgical findings included 769 benign ovarian tumors and 255 malignancies. There were 241 ovarian malignancies (24%), which included 158 epithelial ovarian cancers, 14 nonepithelial ovarian malignancies, 45 ovarian borderline tumors, and 24 malignancies metastatic to the ovary. Of the 14 nonovarian malignancies, 1 patient had a synchronous borderline ovarian tumor and an endometrial cancer; despite a documented adnexal tumor on imaging study, the remainder had pelvic malignancies with normal ovarian histologic condition (uterine, 5 women; retroperitoneal, 2 women; fallopian tube, 1 woman; small bowel, 1 woman; pelvic lymph node, 1 woman; gallbladder, 1 woman; endometrial, 1 woman; leiomyosarcoma, 1 woman). Data in Table 1 for women with ultrasound scan or CT scan exclude subjects with metastatic implants or ascites on imaging study.
| Variable | All women | Women with ultrasound scan a | Women with computed tomography scan b | ||||||
|---|---|---|---|---|---|---|---|---|---|
| All evaluable (n = 1024) | Premenopausal (n = 519) | Postmenopausal (n = 505) | All evaluable (n = 721) | Premenopausal (n = 397) | Postmenopausal (n = 324) | All evaluable (n = 313) | Premenopausal (n = 127) | Postmenopausal (n = 186) | |
| Age, y | |||||||||
| Mean ± SD | 50.4 ± 14.12 | 40.2 ± 8.67 | 60.9 ± 10.50 | 48.6 ± 13.77 | 39.6 ± 8.59 | 59.7 ± 10.46 | 54.1 ± 14.64 | 41.0 ± 8.24 | 63.0 ± 10.86 |
| Median | 49 | 42 | 60 | 47 | 42 | 59 | 53 | 43 | 63 |
| Range (minimum–maximum) | 18–92 | 18–60 | 33–92 | 18–90 | 18–58 | 33–90 | 18–92 | 18–60 | 35–92 |
| Pathologic diagnosis, n (%) | |||||||||
| Benign ovarian conditions | 769 (75.1) | 443 (85.4) | 326 (64.6) | 630 (87.4) | 371 (93.5) | 259 (79.9) | 209 (66.8) | 101 (79.5) | 108 (58.1) |
| Epithelial ovarian cancer | 156 (15.2) | 43 (8.3) | 113 (22.4) | 54 (7.5) | 13 (3.3) | 41 (12.7) | 62 (19.8) | 15 (11.8) | 47 (25.3) |
| Other primary ovarian malignancies c | 16 (1.6) | 8 (1.5) | 8 (1.6) | 8 (1.1) | 3 (0.8) | 5 (1.5) | 2 (0.6) | 1 (0.8) | 1 (0.5) |
| Low malignant potential (borderline) | 45 (4.4) | 13 (2.5) | 32 (6.3) | 18 (2.5) | 6 (1.5) | 12 (3.7) | 27 (8.6) | 5 (3.9) | 22 (11.8) |
| Nonprimary ovarian malignancies with involvement of the ovaries | 24 (2.3) | 8 (1.5) | 16 (3.2) | 5 (0.7) | 2 (0.5) | 3 (0.9) | 7 (2.2) | 2 (1.6) | 5 (2.7) |
| Nonprimary ovarian malignancies with no involvement of ovaries | 14 (1.4) | 4 (0.8) | 10 (2.0) | 6 (0.8) | 2 (0.5) | 4 (1.2) | 6 (1.9) | 3 (2.4) | 3 (1.6) |
| Stage, n (%) d | |||||||||
| I | 61 (35.5) | 18 (35.3) | 43 (35.5) | 31 (50.0) | 9 (56.3) | 22 (47.8) | 26 (40.6) | 7 (43.8) | 19 (39.6) |
| II | 25 (14.5) | 10 (19.6) | 15 (12.4) | 8 (12.9) | 2 (12.5) | 6 (13.0) | 15 (23.4) | 4 (25.0) | 11 (22.9) |
| III | 76 (44.2) | 20 (39.2) | 56 (46.3) | 22 (35.5) | 5 (31.3) | 17 (37.0) | 21 (32.8) | 4 (25.0) | 17 (35.4) |
| IV | 8 (4.7) | 2 (3.9) | 6 (5.0) | 0 | 0 | 0 | 1 (1.6) | 0 | 1 (2.1) |
| Not given | 2 (1.2) | 1 (2.0) | 1 (0.8) | 1 (1.6) | 0 | 1 (2.2) | 1 (1.6) | 1 (6.3) | 0 |
| Imaging, n (%) | |||||||||
| High risk | 470 (45.9) | 228 (43.9) | 242 (47.9) | 351 (48.7) | 187 (47.1) | 164 (50.6) | 190 (60.7) | 70 (55.1) | 120 (64.5) |
| Low risk | 421 (41.1) | 231 (44.5) | 190 (37.6) | 370 (51.3) | 210 (52.9) | 160 (49.4) | 123 (39.3) | 57 (44.9) | 66 (35.5) |
| Multivariate index assay, n | |||||||||
| High risk | 624 (60.9) | 255 (49.1) | 369 (73.1) | 392 (54.4) | 181 (45.6) | 211 (65.1) | 212 (67.7) | 72 (56.7) | 140 (75.3) |
| Low risk | 400 (39.1) | 264 (50.9) | 136 (26.9) | 329 (45.6) | 216 (54.4) | 113 (34.9) | 101 (32.3) | 55 (43.3) | 46 (24.7) |
a Excludes subjects with ascites or metastatic implants that were identified in ultrasound imaging mode
b Excludes subjects with ascites or metastatic implants that were identified in computed tomography scan imaging mode
c Not epithelial ovarian cancer
d Stage presented for epithelial ovarian cancers and other primary ovarian malignancies.
The MIA was high risk in 61% of all evaluable subjects and had the following statistical performance in predicting malignancy: sensitivity, 92%; specificity, 49%; PPV and NPV, 38% and 95%, respectively. When evaluated in the population of women who were enrolled by only nongynecologic oncologists (n = 770), the test performance yielded the following results: sensitivity, 90%; specificity, 54%; PPV and NPV, 35% and 95%, respectively.
There were 1232 imaging procedures (204 women had multiple imaging modalities) performed with 1024 women, with 99% of participants receiving either ultrasound or CT scans ( Table 2 ). Thirteen percent of subjects (133/1024 women) had radiographic evidence of advanced disease at study enrollment (ascites, 11.5%; metastases, 3.6%; both, 2.1%). Ascites that was found on CT scan compared with ultrasound scan was more likely to represent malignancy (77% vs 53%). When radiologic findings of advanced disease are excluded, 49% of ultrasound scans and 61% of CT scans were categorized as high risk. The high-risk ultrasound imaging criteria had the following statistical performance to predict the likelihood of malignancy: sensitivity, 77%; specificity, 55%; adjusted PPV and NPV, 36% and 88%, respectively. CT scan performance in the prediction of malignancy was sensitivity, 80%; specificity, 49%; adjusted PPV and NPV, 34% and 88%, respectively ( Table 3 ).
| Imaging method | All evaluable subjects | Subjects with benign disease | Subjects with malignant disease |
|---|---|---|---|
| Ultrasound scan, n (%) | |||
| Imaging mode | 782 | 658 | 124 |
| Positive findings | 412 (52.7) | 309 (47.0) | 103 (83.1) |
| Solid components | 390 (49.9) | 295 (44.8) | 95 (76.6) |
| Ascites | 60 (7.7) | 28 (4.3) | 32 (25.8) |
| Metastatic implants | 7 (0.9) | 0 | 7 (5.6) |
| Included | 721 | 630 | 91 |
| High risk a | 351 (48.7) | 281 (44.6) | 70 (76.9) |
| Low risk | 370 (51.3) | 349 (55.4) | 21 (23.1) |
| Computed tomography scan, n (%) | |||
| n | 407 | 230 | 177 |
| Positive findings | 284 (69.8) | 128 (55.7) | 156 (88.1) |
| Solid components | 260 (63.9) | 118 (51.3) | 142 (80.2) |
| Ascites | 80 (19.7) | 18 (7.8) | 62 (35.0) |
| Metastatic implants | 35 (8.6) | 4 (1.7) | 31 (17.5) |
| Included | 313 | 209 | 104 |
| High risk a | 190 (60.7) | 107 (51.2) | 83 (79.8) |
| Low risk | 123 (39.3) | 102 (44.3) | 21 (20.2) |
a Indicated by the presence of solid tumor or papillary morphologic condition by respective imaging method.
| Variable | All women | Women with ultrasound scan | Women with computed tomography scan | ||
|---|---|---|---|---|---|
| Multivariate index assay (n = 1024) | Multivariate index assay (n = 721) | Ultrasound scan (n = 721) | Multivariate index assay (n = 313) | Computed tomography scan (n = 313) | |
| Sensitivity | |||||
| % | 92.2 | 89.0 | 76.9 | 88.5 | 79.8 |
| n/N | 235/255 | 81/91 | 70/91 | 92/104 | 83/104 |
| 95% CI | 88.2–94.9 | 80.9–93.9 | 67.3–84.4 | 80.9–93.3 | 71.1–86.4 |
| Specificity | |||||
| % | 49.4 | 50.6 | 55.4 | 42.6 | 48.8 |
| n/N | 380/769 | 319/630 | 349/630 | 89/209 | 102/209 |
| 95% CI | 45.9–52.9 | 46.7–54.5 | 51.5–59.2 | 36.1–49.4 | 42.1–55.5 |
| Positive predictive value (observed) | |||||
| % | 37.7 | 20.7 | 19.9 | 43.4 | 43.7 |
| n/N | 235/624 | 81/392 | 70/351 | 92/212 | 83/190 |
| 95% CI | 33.9–41.5 | 16.9–24.9 | 16.1–24.4 | 36.9–50.1 | 36.8–50.8 |
| Positive predictive value (adjusted) | |||||
| % | 37.7 | 37.4 | 36.4 | 33.8 | 34.1 |
| 95% CI | 33.9–41.5 | 34.9–39.9 | 33.2–39.7 | 30.8–36.9 | 30.5–37.8 |
| Negative predictive value (observed) | |||||
| % | 95.0 | 97.0 | 94.3 | 88.1 | 82.9 |
| n/N | 380/400 | 319/329 | 349/370 | 89/101 | 102/123 |
| 95% CI | 92.4–96.7 | 94.5–98.3 | 91.5–96.3 | 80.4–93.1 | 75.3–88.6 |
| Negative predictive value (adjusted) | |||||
| % | 95.0 | 93.3 | 87.9 | 91.8 | 87.9 |
| 95% CI | 92.4–96.7 | 88.5–96.2 | 83.2–91.4 | 86.5–95.1 | 82.9–91.6 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
