Blood and Urine Markers for Ovarian Cancer: A Comprehensive Review

Kathryn L. Terrya, Patrick M. Slussb, Steven J. Skatesc,d, Samuel C. Mokd, Bin Yed, Allison F. Vitonisa and Daniel W. Cramera Ob-Gyn Center Epidemiology Center, Department of Obstetrics, Gynecology, and Reproductive Biology, Brigham and Women’s Hospital, Boston, MA, USA Reproductive Endocrinology Unit Laboratory, Department of Medicine, Massachusetts General Hospital, Boston, MA, USA Biostatistics Center, Massachusetts General Hospital, Boston, MA, USA Laboratory of Gynecologic Oncology, Department of Obstetrics, Gynecology, and Reproductive Biology, Brigham and Women’s Hospital, Boston, MA, USA


Introduction
The obvious appeal of a strategy for the early detection of ovarian cancer is based upon the tendency for the disease to present at advanced stages associated with poor survival. If the diagnosis could be largely shifted to stage I associated with survival close to 90%, then the overall mortality for this disease could be dramatically altered without any advances in therapy. Whether a screening strategy can be successfully implemented for ovarian cancer depends upon the natural history of the disease, especially the length of time ovarian cancer remains in a pre-clinical phase prior to stage I, and the ability to develop a convenient and suitably-performing screening test acceptable to the target population. Preclinical disease refers to a stage in a disease process before which the disease has produced symptoms, which lead an individual to seek diagnosis or treatment. In the case of cancers like cervix or colon, pre-clinical disease can be equated with well-characterized precursor lesions known to precede the invasive cancer. But, for ovarian cancer, precursor lesions are largely unknown 1 Supported by NCI Early Detection Research Network Grant #U01CA086381.
as is the length of time that the disease remains in a pre-clinical phase.
Despite this lack of understanding, the public health importance of ovarian cancer demands active investigation of approaches to screening. Although these approaches might include pelvic examination, sonography, or even cytologic approaches, this review will focus on blood or urine markers for ovarian cancers as methods most likely to be adaptable to screening both general and high-risk populations. Table 1 summarizes ovarian cancer markers for which sensitivity and specificity estimates were available or where mean levels of markers for cases and controls were available. Sensitivity, the proportion of all those with disease who are screen positive, and specificity, the proportion of all those without disease who are screen negative, are chosen as the key performance characteristics. Unlike related measures such as the predictive values of a positive or negative test, which depend upon the prevalence of the disease in the population, sensitivity and specificity are more stable test parameters. We have also attempted to organize markers by structural or functional characteristics in order that some generalizations about broader categories of markers may be possible, acknowledging that there may be disagreement about how best to categorize a particular marker.

Epithelial sialomucins
The epithelial sialomucins are currently the most important group of markers for ovarian cancer. These are heavily glycosylated proteins with relatively high molecular weights and include the only approved marker for monitoring ovarian cancer, CA-125, also known as MUC16. Although these proteins are transmembrane in location, they can be found in the serum. Most of the markers included in this category have been identified through approaches in which human cancer cells are used as an antigenic stimulus in animals to raise antibodies, which can then efficiently detect the antigen in human serum. A review of Table 1 suggests that CA 125 remains the best-documented and bestperforming single marker among the epithelial sialomucins currently described [2,6,47,48,54,63,91,94,98,101,102,108,112,116]. Sensitivity estimates vary from as low as 27% in some studies for early-staged disease to better than 90% for late-staged disease. Sensitivity is also greater for serous and endometrioid ovarian tumors compared to mucinous or clear cell tumors. CA 15-3 (MUC1) is a biomarker described for breast (and other cancers) but is also relevant to ovarian cancer. It appears to be expressed in ovarian cancer types similar to CA 125 with promising sensitivity and specificity estimates in limited studies [7,24,48,80,102]. The nomenclature on CA 15-3 is confusing with a number of other terms describing markers similar or identical to CA 15-3 including human milk fat globulin 1 and 2 (HMFG1/2), CASA, OSA, and several others. The remaining sialomucins listed in Table 1 are not serious candidates to replace CA 125 based upon usually poorer and certainly more limited sensitivity and specificity estimates. Rather these are better viewed as markers, which may prove complementary to CA 125. Thus CA 50, CA 54-61, CA 195, and CA 19-9 all appear to have greater sensitivity for mucinous tumors [17,29,47,88] while STN and TAG-72 have better sensitivity for clear cell tumors [31,54]. Combinations of markers involving the sialomucins are discussed in more detail in Section 12.
In general there are several limitations common to members of this family of markers. First, based on their molecular weight, urine would not be a good medium to use in screening for these markers. Secondly, the high molecular weight and differing glycosylation patterns (epitopes) combine to produce the diagnostic problem that different antibodies can be generated and contribute to a lack of consistency between assays using different antibodies [99]. Thirdly, these markers may be secreted by a variety of epithelial tumors including breast, colon, ovary and pancreas as well as normal adult epithelial tissue including that from the breast, gastrointestinal and genitourinary tracts. Finally, a number of benign conditions can be associated with elevation of these markers. In particular, CA125 can be elevated in early pregnancy and with endometriosis, fibroids, or infections of the genital tract [99]. These would, of course, affect the specificity of the test and the possibility of false positive results. One strategy, besides combining markers, to improve the specificity associated with use of these markers would be to use serial testing. Elevated but declining marker levels would indicate a transient condition associated with marker production. Elevated but stable levels might indicate chronic but benign conditions associated with marker production, while elevated and increasing levels are more likely indicative of cancer [87,112]. Another approach to improve specificity would be to combine the marker with an imaging test such as pelvic ultrasounda strategy found in a randomized trial in England to improve the median survival of screen detected cases [113].

Proteases, inhibitors, and cleavage products
Based on the number described, the proteases and related peptides may be the next most valuable group of markers for ovarian cancer. Within this group, the kallikreins, a family of serine proteases, have captured attention because of the importance of prostate-specific antigen (PSA) for detecting prostate cancer in men. PSA was originally discovered through techniques similar to those employed for the sialomucins in which crude or partially purified extracts of normal and cancerous prostate tissues were used to immunize rabbits [100]. Later, PSA was identified to be a member of the human kallikrein family prompting interest in other members of this group. The importance of this family of proteins to ovarian cancer has been subsequently demonstrated through techniques used to identify genes, which are over expressed in ovarian cancer cells. Indeed, the entire family of kallikreins map to a region on chromosome 19q shown to be amplified in ovarian cancers [19,20,85,110].
Currently the best performing of these markers include kallikrein 6 (protease M) and kallikrein 10 with sensitivities reported between 47 and 75% and specificities between 90 and 100% [21,56]. Sensitivity was improved when kallikrein 10 was combined with CA 125 [58]. It is not known whether combining several kallikreins will improve sensitivity and specificity, but this seems unlikely since they appear to be upregulated as group based on their tight genetic clustering. Prostasin is another serine protease, whose gene was discovered to be upregulated in ovarian cancer. When combined with CA 125, its sensitivity to detect ovarian cancer was improved from 51% to 92% with a combined specificity of 94%, similar to what was observed for kallikrein 10 [62].
The matrix metalloproteinases (MMP's) are another family of proteases that may be useful in ovarian cancer screening or prognosis. Unlike the kallikreins, genes for the MMP's are not clustered in the same chromosomal region. MMP-2 was reported to have 66% sensitivity and 100% specificity in a study that needs confirmation [83]. Other proteases described for ovarian cancer include urokinase-type plaminogen activator [97], matriptase [70], and thrombin [16]. Cathepsin L is a cysteine protease. Its level in the serum was found to be significantly elevated in ovarian cancer but no sensitivity estimate was provided [68]. The same paper described a lower false positive rate for this marker compared to CA 125 or CA 72-4 but null correlations suggesting it might be complementary to these markers.
For several of the proteases, their complementary inhibitors have been identified and may also be useful for ovarian cancer screening. Matriptase and its inhibitor hepatocyte growth factor activator inhibitor 1, [70] and urokinase and its inhibitor, plasminogen activator inhibitor type I [97], are examples of complementary pairs. Another protease inhibitor, HE4 was identified through gene upregulation studies and may prove useful for ovarian cancer screening [37]. Its complementary protease has apparently not been identified. Finally, thrombin, its inhibitor, anti-thrombin III, as well as a degradation product D-Dimer are examples of a triad of protease, inhibitor, and degradation product which may all be applicable for ovarian cancer screening [16] and provide a useful paradigm for research on other sets of proteases, inhibitors, and cleavage products. Indeed, cleavage products of proteases may prove to be a large and important group of markers for ovarian and other types of cancers. It is very likely that many of the low molecular weight "peaks" being identified through mass spectrometry (see Proteomic Patterns below) represent cleavage products of unidentified proteases.
Based on their biochemical role, it would seem likely that the proteases would be active in invasion and metastasis, and these markers do appear to have higher sensitivity for advanced stages of disease. A potential advantage of this category of markers based on their lower molecular weight is that urine might be a useful medium for screening. Indeed the MMPs have been found in urine for a number of cancers [66].

Cytokines, targets, and acute phase reactants
Arguably next in importance for ovarian cancer screening are the cytokines, which may prove to have good sensitivity for ovarian cancer, although it seems likely that their specificity may suffer from levels being elevated with other types of cancers as well as inflammatory conditions. Perhaps the best-known marker for ovarian cancer in this category is macrophage colony stimulating factor (MCSF), which was reported to have a sensitivity of 75 to 100% even for early stage disease and specificity of 92% to 98% in healthy controls [98,102,103]. Tumor necrosis factor (TNF)alpha is a cytokine that inhibits growth by increasing apoptosis. Consequently, one might expect its levels to be reduced in individuals with cancer. However, investigators reported no difference in TNF-alpha levels between individuals with ovarian cancer and those with benign tumors [75]. Yet, a chemical relative, TNFrelated apoptosis-inducing ligand was found to have a ten fold increase in ovarian cancer compared to normal ovarian epithelium, suggesting that increased levels may indicate the presence of ovarian cancer [55]. Interleukin (IL) 6 levels were significantly increased in serum and peritoneal fluid in ovarian cancer and correlated with stage of disease, although no sensitivity and specificity estimates provided [65,92]. Additional markers we include in this category are the soluble form of the IL receptors and acute phase reactant proteins. Soluble IL 2 receptor alpha had a sensitivity of 95% but very poor specificity improved by combining the marker with CA 125 [44]. The acute phase reactant protein alpha haptoglobin has also been identified as a potential marker for ovarian cancer. Interestingly, intact haptoglobin was identified years previously as a screening marker with modest sensitivity and specificity for multiple cancers. The fragment of haptoglobin alpha was rediscovered through mass spectrometry and found to have improved sensitivity and specificity when combined with CA-125. Finally, C-reactive protein levels were reported to be 50 mg/l in 20% stage I/II and 30% of stage III/IV ovarian cancers and to be associated with poorer survival [53].

Hormones, growth/inhibition factors
Based upon their role in normal ovarian physiology, epithelial tissue physiology, or tumor physiology, it would seem that a variety of hormones and growth or inhibition factors might be of value for ovarian cancer. However, few are of proven value for epithelial ovarian cancer screening, although several have been useful in immunohistochemical studies that correlate tumor staining properties with prognosis. As a key factor in ovarian physiology, the gonadotropins have been an important focus in animal studies but their measurement in serum does not appear useful in ovarian cancer screening. However, there has been some interest in urinary "gonadotropin fragments" with sensitivity ranging from 73 to 83% and specificity about 92% [15]. The inhibins are also key players in ovarian physiology and are representatives of a large group of proteins known as the transforming growth factor (TGF) beta superfamily, many of which have tumor suppressor or promoter functions. Although the inhibins have not shown particular value in the detection of epithelial ovarian cancer, they may prove useful for the diagnosis of granulosa cell tumors [11,76]. Serum TGF-alpha has also been examined for ovarian cancer with sensitivity estimates of 60-70% for early stage disease and specificity estimates of 89% for healthy controls [13]. Perhaps the current best performing marker in the growth factor category is Mesothelin, which was identified in gene upregulation studies, with a sensitivity of 77% (mostly late stage disease) and specificity of 100% [82]. In addition to these growth or inhibition factors, soluble forms of their receptors have also been studied as potential markers. The epidermal growth factor receptor, called p110, has good specificity at 94% in healthy controls, but its sensitivity declines with age. At ages 20-40, the p110 sensitivity is 73%, but it declines to 33% in older women [4].
Several markers in this family have been used in histochemical staining of tumors and might correlate with prognosis. Epidermal growth factor assessed by immunohistochemical staining of tumors was found to be only marginally predictive of ovarian cancer [72]; and no association between the EGF receptor and stage of disease was found [98]. Other tumor markers of potential interest for "typing" tumors based on their immunohistochemistry include p53 and HER 2 neu. The reported proportion of epithelial ovarian cancers that express p53 varies widely, ranging from 3 to almost 50% [59,78,98]. P53 expression is much lower at early stage than late stage disease [59,78]. P53 expression also varies by histology with the highest proportion of p53 positive cases in the invasive serous group and the lowest in the mucinous group [78]. P53 expression is virtually absent borderline tumors, and no benign tumors stained positively for p53 expression [98]. HER-2 neu is found in less than 25% of newly diagnosed ovarian cancers [36].

Cytokeratins
Further related to epithelial cell biology, the cytokeratins are a group of intermediate filament proteins that compose the cytoskeleton of epithelial cells. Although individual cytokeratins may be relevant to ovarian cancer, more frequently mixtures of them have been studied. The best known of these is tissue (specific) polypeptide antigens (TPA or TPS). A wide range of sensitivities and specificities has been reported but the best performance has been achieved when this marker is combined with CA 125. In several studies, the reported sensitivity was between 79 and 93% and the reported specificity between 89 and 98% [47,84,94,95]. A mixture of two cytokeratins, known as M3/M21 has also been studied as a marker for ovarian cancer with poorer sensitivity but better specificity than TPA [35,92].

Lipids and lipo-proteins
There was considerable excitement and interest in a lipid marker for ovarian cancer, lysophosphatidic acid (LPA) based on a 1998 report [114], which described a sensitivity of 97.9% and specificity of 89.6%. The requirement of this assay for quickly processed and frozen plasma appears to have affected enthusiasm for this marker although a search for more stable forms is underway. Apolipoprotein A has a reasonable sensitivity, ranging from 81-89%, but poor specificity (23-34%). Both sensitivity and specificity varied by age. At ages less than 49, sensitivity is high but specificity is poor. For ages greater or equal to 49, sensitivity decreases but specificity increases [115]. Not shown in Table 1 since serum levels have not been reported are two other apolioproteins, apolipoprotiens E and J (also known as clusterin). Both of these were found to be highly upregulated in large scale serial analysis of gene expression (SAGE) and may prove of value in ovarian cancer screening or prognosis [43].

Oncofetal proteins
Of the traditional oncofetal proteins, carcinogenic embroyonic antigen (CEA) and alpha-fetal protein (AFP), CEA has been the topic of several studies while AFP appears to have diagnostic value only for the germ cell tumors. CEA had very little sensitivity by itself for detecting epithelial ovarian cancer. Although not traditionally considered an oncofetal protein, placenta like alkaline phosphatsase (PLAP), was identified decades ago by similar techniques used to identify CEA and AFP; i.e. studies to identify antigens common to cancer and fetal tissue. PLAP has been the topic of several studies, but like CEA appears to have reasonable performance in screening only when combine with CA 125 with a sensitivity of 93% and a specificity of 83% [94].

Auto-antibodies
The neoplastic process is associated with the entry of a variety of cancer proteins into the circulation that may lead to auto-antibody production. Although a very small number of screening markers currently fit into this category, we think it is important to include this as a distinct and potentially important group of biomarkers for ovarian cancer. Antibodies to p53 have been evaluated but showed poor sensitivity and appear to be of little screening value [30,41]. Rare cases of cerebellar degeneration in association with ovarian cancer have been described and appear to represent a paraneoplastic response to auto-antibodies that react with Purkinje cells [38]. Based on the fact that the cerebellar symptoms often preceded a clinical diagnosis of ovarian cancer, it seems reasonable to speculate that autoantibody formation could precede other clinical symptoms and thus be a potentially good marker for early stage disease [64]. Currently, the only marker studied for its potential as a screening tool is based upon autoantibodies directed against the epithelial cell adhesion molecule (Ep-CAM). This study compared Ep-CAM auto-antibody levels in the sera of 26 healthy controls, 26 women with benign disease, and 26 ovarian cancer patients. Investigators report that the Ep-CAM autoantibody screening showed a sensitivity of 73.1% and a specificity of 80.8% (95% CI = 65.7 to 95.9), while CA125 alone in the same set of sample provided a sensitivity of 86.5% and a specificity of 88.5%. The combination of Ep-CAM auto-antibody and CA125 gave a sensitivity of 90.4% and specificity of 92.3% [51].

Other markers
A variety of other markers not easily classified have been studied in ovarian cancer screening or prognosis. These include several markers identified through biochemical assays, multifunction proteins, and general measures of the degree of gycosylation. Enzymatic tests of potential value for ovarian cancer include alpha-fucosidase, galacotsyltransferase, and LDH [1,8,54,81,86,96]. RNAase A is a biomarker initially assessed for other cancers but found to be elevated in the urine of ovarian cancer patients identified through mass spectrometric approaches [107]. Osteopontin is a multifunction protein identified through gene upregulation studies and found to have higher mean levels in sera from ovarian cancer subjects compared to that from normal women or those with benign disease [50]. Lipid -associated sialic acid provides a more global measure of sialic acid, an amino carbohydrate, that is an important component of tumor glycoproteins and does correlate with advanced ovarian and other types of cancer. However, this test lacks specificity since it may be associated with non-neoplastic conditions as well.

Proteomic patterns
We conclude our review of markers for ovarian cancer with the recent innovative approach of using mass spectrograph patterns of proteins or small molecules as the screening test [73]. This potential advance has been permitted by high throughput mass spectrometry combined with sophisticated biostatistical approaches to identify patterns in the output. When presented with a test set of sera (or other biologic specimens), the algorithm was able to find a pattern that distinguished cancer cases from controls with near perfect accuracy and that was reproducible when presented with specimens from a comparable set of cases and controls. However, as was illustrated in the article describing this approach, when presented with a novel set of alternate controls, the algorithm assigned a "new pattern" for most of these controls. Thus, the transferability of the algorithm to novel specimens as well as the ability to standardize the output from mass spectrometry appear to be current obstacles for adapting this approach to clinical care. Other studies using mass spectrometry have attempted to identify components of the predictive patterns -most of which appear to be in the low molecular weight range, which may make this an ideal tool for the discovery of urinary markers for ovarian cancer.

Combinations of markers
In reviewing Table 1 it is apparent that sensitivities and specificities were often provided for combinations of 2 or more markers (usually with CA 125 as one of these). Table 2 highlights these combinations for easier comparison. Some of the better performing combinations of markers include CA 125 and HMFG1/G2 with a sensitivity of 93% and a specificity of 93% [18] as well as CA 125 and CEA with a sensitivity of 93% and a specificity of 93% [94]. CA 125 with TAG72 and CA15-3 increase the specificity to 95% but sacrifice sensitivity [7]. We should note here that the traditional approach to combining multiple markers in most of these studies has been to use fixed cutoffs for each marker. Figure 1 illustrates the limitations of this approach in a simple two marker example using actual data on haptoglobin-alpha and CA 125 from the paper by Ye [106]. Using fixed cutoffs for both markers and declaring a positive screen to occur when either cutoff is exceeded and a negative screen to occur only when both values are below the cutoff, subjects within the box are declared negative and subjects outside the box are declared positive. This rule leads to 6 cases falsely declared negative and 18 controls falsely declared positive.
However a linear or logistic function of the two parameters, represented by the diagonal line, declares a negative screen to occur below the diagonal and a positive screen to occur above the diagonal. This rule leads to 8 cases falsely declared negative and 10 controls falsely declared positive; i.e. a substantial improvement in specificity with only a modest decrease in sensitivity. The problem becomes more complex when the distributions for the markers are not normal or when more than two markers are being examined. Logistic regression models for more than two variables or mixtures of multivariate normal distributions have been suggested. While a full discussion goes beyond the scope of this paper, the topic is considered in more detail elsewhere [33,87].

Discussion
To some extent, the markers identified in Table 1 reflect the dual evolution of cancer model systems and available biologic tools. Some of the earliest models relied upon experimentally-induced tumors in animals and used relatively simple gel and immune based tools to identify cancer antigens not found in adult tissue but often found in fetal tissue from the same organ in the animal model. These led to identification of oncofetal proteins like CEA as some of the earliest cancer biomarkers. Since many early diagnostic tests in the general medical clinic relied upon enzymatic tests, it is not surprising that many of these found a role as early cancer diagnostic tests, including lactate dehydrogenase and acid phosphatase. With the evolution of mammalian cell culturing techniques including capacity for monoclonal anti-body production, the stage was set for a productive era of discovery of human cancer associated antigens, including CA 125 for ovarian cancer. Recognition of the role of tumor promoters and suppressors in cancer formation and the development of powerful new genetic techniques led to discovery of a number of genes up or downregulated in cancer. This in turn has allowed focus on their gene products as potential cancer markers. Most recently the devel-opment of precise and high throughput mass spectrography has shifted the focus to proteomic patterns rather than specific markers. These methods are well suited for the discovery of entirely new biomarkers since the technique begins with no assumptions about the disease model and allows for post-translational modifications to be detected. Most of the biologic models leading to marker development have focused on cancer cell cultures and invasive disease. Thus, it is not surprising that many of the markers identified so far are indicative of cell death, tissue remodeling or invasion, inflammatory response elements, or systemic response markers. These types of markers would seem to be most appropriate in late stage disease as appears to be the case for most of the markers in Table 1. Perhaps, new paradigms will be necessary for detection of markers suited for detection of early stage disease. Methods better suited to detect trace concentrations associated with low tumor load, early genetic changes, or subtle changes related to tumor surveillance cells or the immune system are needed. With regard to the latter, we postulate that the largely unexplored area of auto-antibody formation to tumor products might prove useful for early stage disease. Another technical limitation is our ability to describe and quantitate glycosylation. Aberrant glycosylation has been known as a feature of cancer for many years, especially apparent for the epithelial sialomucins. Thus, advances may await newer techniques to quantitatively and qualitatively describe glycosylation and take this into consideration in marker assays.
Other limitations are also clearly apparent in reviewing the studies cited in Table 1, including those that affect the reliability of the sensitivity and specificity estimates. Precision of the estimates of sensitivity and specificity will be greater for larger versus smaller study populations; but the effect produced by the use of different populations in studies poses a problem less able to be predicted. Case groups may have different mixtures of disease stages and histologic types and grades that affect overall estimates of sensitivity. In addition, although some groups were explicit in stating that case specimens were limited to those collected pre-operatively, not all groups stated this allowing for the possibility that some specimens were collected in the immediate post-operative period. Few studies have used pre-diagnostic sera months or years before clinical diagnosis. Obviously, the types of controls selected will greatly influence the estimated specificity of a particular marker or combination of markers. Inclusion of surgical controls that have other gynecological conditions such as fibroids, endometriosis, or benign ovarian tumors will lead to lower specificity compared to non-surgical controls. In terms of relevance to general population screening, specificity estimates based upon healthy controls are best. Description of the case or controls populations sometimes did not include the proportions that were pre-and postmenopausal. CA 125 is known to vary by menopausal status and it is likely that a number of other markers will vary as well. A final limitation is that assay performance was seldom addressed in detail by including estimates of inter-and intra-assay coefficients of variation. Poor assay performance could seriously compromise the reproducibility of sensitivity and specificity estimates.
Conceding these limitations, we still had the goal in preparing this review to identify individual markers or combination of markers with the best performance. Individual serum markers measured by immunoassay that appeared to perform best (if we require a specificity of at least 90% and sensitivity of 75% or greater) included the following: CA 125, CA 15-3 and its variants, MAM-6, kallikrein 10, thrombin-antithrombin III, mesothelin, IAP, MCSF, SIL-2RA, and TPA. Combinations of markers achieving these objectives include CA 125 in combination with HMFG1/2, in combination with TAG 72 and CA 15-3, or in combination with IAP, or haptoglobin alpha. Clearly, high specificity is demanded if expensive or invasive diagnostics tests are required to follow-up on a positive screening test; and the 90% specificity required above may not be good enough. Singling out some markers with specificities above 98% (while maintaining sensitivities of at least 60%) revealed some of the following markers TAG-72, kallikrein 6 or 10, MMP-2, and Mesothelin. As a suggestion for future combinations to examine, it would be reasonable to combine one or more marker from each category in Table 1 in hopes that complementary markers could be identified to improve sensitivity while selecting additional markers which have excellent specificity. One of the most important characteristics needed for future ovarian cancer markers is high sensitivity at early stages.
Our belief that there may already be a very satisfactory set of markers that exist for ovarian cancer must be tempered by the likelihood that this panel may be positive for benign ovarian diseases and other cancers. Thus, follow-up for a positive test may require a battery of imaging studies. Imaging studies that reveal a complex ovarian mass will likely lead to surgery and in some cases will reveal a benign tumor. We would argue this should not be considered a failure of screening. A woman with a 10 cm complex ovarian tumor requires surgery not only because of the mass's potential to be cancerous but also because of its potential to undergo torsion or rupture producing a surgical emergency. A "positive panel" for ovarian cancer not associated with an obvious ovarian mass will require additional imaging or screening studies that are already recommended on a routine basis including mammogram, colonoscopy, or chest x-ray.
In conclusion, as the search for new markers for ovarian cancer continues, it is important to re-assess what has been already learned in the past three or four decades of previous research on ovarian cancer. Good and likely reliable markers already exist that simply need to be combined together in more convenient assay platforms. New research would be worthwhile and should focus on underdeveloped areas such as autoantibodies to cancer proteins or further descriptors of glycosylation. Mass spectrometry approaches may reveal complex patterns of cleavage products that provide unique cancer signatures; but there will still be a need for identifying these products to determine their role in cancer pathogenesis or progression. Since the focus of research has largely been on serum markers, attention should also be given to the urinary markers, which may be ideally suited to investigation using the mass spectrometric approaches.