Proteomic Biomarkers in Combination with CA125 for Detection of Epithelial Ovarian Cancer using Pre-diagnostic Serum Samples from the Prostate Lung Colon and Ovary (PLCO) Cancer Screening Trial

doi:10.1002/cncr.26241

Cancer. Author manuscript; available in PMC 2012 June 28.

Published in final edited form as:

Cancer. 2012 January 1; 118(1): 91–100.

Published online 2011 June 29. doi: 10.1002/cncr.26241

PMCID: PMC3385508

NIHMSID: NIHMS384520

Proteomic Biomarkers in Combination with CA125 for Detection of Epithelial Ovarian Cancer using Pre-diagnostic Serum Samples from the Prostate Lung Colon and Ovary (PLCO) Cancer Screening Trial

Lee E. Moore,¹ Ruth M. Pfeiffer,¹ Zhen Zhang,² Karen H. Lu,³ Eric T. Fung,⁴ and Robert C. Bast, Jr.⁵

¹Occupational and Environmental Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH, Bethesda, MD 20852-7240

²Department of Pathology, Johns Hopkins Medical Institutions, Baltimore, MD 21231

³Department of Gynecologic Oncology, University of Texas M.D. Anderson Cancer Center, Houston, TX 77030

⁴Vermillion, Inc., Freemont, CA 94555

⁵Department of Experimental Therapeutics, University of Texas, M.D. Anderson Cancer Center, Houston, TX 77030-4009

Corresponding Author: Lee E. Moore, Ph.D., Occupational and Environmental Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH, Bethesda, MD 20852-7240; Email: moorele/at/mail.nih.gov

The publisher's final edited version of this article is available at Cancer

See other articles in PMC that cite the published article.

Abstract

Background

When detected in early stage (I-II), epithelial ovarian cancer 5-year survival is 70–90%, whereas in late stage (III-IV) 5-year survival slips to < 30%. In our previous report, proteomic biomarkers and cancer antigen 125 (CA125) exhibited a sensitivity of 84% at a specificity of 98% for identifying sera from stage I patients at the time of surgery, significantly improving sensitivity of CA125 alone. The challenge, however, is to detect ovarian cancer prior to clinical diagnosis. This study was part of a large effort to compare different multi-marker biomarker panels for the early detection of ovarian cancer. We evaluated several biomarkers alone and combined with CA125 in pre-diagnostically collected sera from women in the PLCO Cancer Screening Trial.

Methods

Proximal pre-diagnostic sera from 118 ovarian cancer patients and 951 age-matched controls, 8 per case (4-randomly selected from the general population, 2 with CA125 ≥35 U/mL, 2 with positive family history of breast/ovarian cancer), were analyzed using the CA125 immunoassay and SELDI-TOF-MS to measure 7 proteins [Apolipoprotein A1(Apo-A1), truncated transthyretin(TT), transferrin(TRFR), hepcidin(HEPC), β-2-microglobulin(β2M), Connective Tissue Activating Protein III(CTAPIII), and Inter-alpha-trypsin inhibitor heavy chain 4(ITIH4)]. Data were analyzed by two statistical strategies that combined seven markers and CA125 into one predictive score for disease classification.

Results

CA125 was elevated (≥35 U/mL) in 61.5% of 65 cases with CA125 data sampled <12 months prior to cancer diagnosis, however the 7 biomarker levels were not different between cases, and the three control groups individually, or combined. Two panels, combining CA125 and 7 biomarkers failed to improve the sensitivity of CA125 alone.

Conclusion

In contrast to our prior findings using post-diagnostically collected sera, addition of these biomarkers to CA125 did not improve sensitivity for preclinical diagnosis beyond CA125 alone.

INTRODUCTION

Despite recent advances in ovarian cancer treatment, new methods of early detection remain of paramount importance as they could clearly improve long-term cancer survival. Currently, more than two-thirds of ovarian cancer cases are detected at an advanced stage, resulting in poor overall five-year survival rates of 10–30%.¹ This is in stark contrast to stage I/IIa patients that have approximately a 90% five-year survival. Longitudinal studies are ongoing in several countries to evaluate screening strategies using CA125 and/or transvaginal sonography (TVS) and their impact on overall cancer detection and mortality. Recently, through screening with CA125 and TVS in the Prostate, Lung, Colorectal, and Ovarian (PLCO) Trial, investigators were able to identify both early and late-stage malignancies, however, the predictive value of each was surprisingly low due to a high proportion of false-positive diagnoses.² Other existing serum markers that have been identified thus far have not proven adequate sensitivity or specificity for screening. New biomarkers that could improve early detection for ovarian cancer are thus critically needed. Such methods could be used alone, or in combination with existing methods, to improve diagnostic performance.

Proteomic methods have been used to detect a number of promising serum biomarkers that can distinguish patients with early stage ovarian cancer from healthy individuals.³ A seven biomarker panel that included transthyretin (TT), apolipoprotein A1 (Apo-A1), beta-2-microglobulin (β2M), transferrin (TRFR), hepcidin (HEPC), connective tissue activating protein III (CTAPIII), and inter-alpha-trypsin inhibitor heavy chain 4 (ITIH4), has been evaluated not only for early detection,⁴ but also for distinguishing malignant from benign pelvic masses.⁵ Immunoassays for Apo-A1, TT and ITIH4 in combination with CA125 improved upon the sensitivity of CA125 alone for detecting early stage disease.⁴ Sensitivity increased from 65% with CA125 alone to 74% with the addition of Apo-A1, TT and ITIH4, when specificity was held constant at 97%. Subsequently, Moore, et al⁶ replicated these findings using post-diagnostic/pre-treatment serum samples from ovarian cancer cases and hospitalized controls diagnosed with benign ovarian cysts or digestive diseases. The specificity of a panel of TT markers and Apo-A1 was 96.5% but sensitivity was only 52.34%. However, when combined with CA125, the specificity remained high (94.3%) and 78.6% sensitivity was achieved. In another recent study⁷, a biomarker panel comprised of Apo-A1, TT and CTAPIII achieved a sensitivity of 84% at a specificity of 98% for detecting early stage ovarian cancer.

All prior studies have been performed with sera obtained at the time of clinical diagnosis. Ideally, a screening strategy should detect ovarian cancer before it becomes symptomatic or readily palpable. Consequently, as part of a large multi-center study to compare different multi-marker panels for the early detection of ovarian cancer, we evaluated the discriminatory power of a seven biomarker panel using prospectively collected pre-diagnostic serum samples from 118 asymptomatic women enrolled in the PLCO Cancer Screening Trial.⁸ For each case, eight matched healthy controls were used as a comparison group. In addition to 4 population controls that had been described in a previous report⁹, our analysis also included two high risk comparison groups; two matched controls who ever had an elevated CA125 level ≥35U/ml, and two women with a family history of breast or ovarian cancer. The sensitivity and specificity of this expanded marker panel alone and in combination with CA125 was evaluated for the ability to discriminate women with ovarian cancer from healthy women using pre-diagnostic serum samples, to determine whether this combined marker panel could perform better than serum CA125 alone, as was previously observed in our investigations using post-diagnostic serum samples and controls with benign ovarian or digestive diseases ⁶.

MATERIALS AND METHODS

Patient Population

Phase III studies were coordinated by PLCO investigators.^2,8 Between 1998 and 2006, 118 cases of invasive ovarian, primary peritoneal and fallopian tube cancers were identified and histologically confirmed. Women with tumors of borderline malignancy were excluded as well as those with a history of cancer at baseline. Controls were selected from a pool of healthy individuals who remained cancer-free, and were matched to cases by five year age categories at blood draw, calendar year of blood draw and calendar year (in categories) of cancer diagnosis. For each case, eight healthy controls were selected. Four controls per case were randomly selected from a pool of all eligible controls and are referred to as general population controls. Two controls per case were selected among women who reported having a positive family history of breast or ovarian cancer. Two additional controls per case were selected who were healthy, but at some time point during the PLCO Trial sampling period had an elevated serum CA125 level (≥35 units/ml), allowing examination of the biomarker panel among a high-risk group of women who would be more likely to be targeted for ovarian cancer screening than healthy women in the general population. These women with at least one positive CA125 test result continued to undergo annual screening. Sixty replicate pairs were randomly inserted into the batches for blinded QC, ten pairs had CA125>25 units/ml. For all cases, the serum sample most proximate to the case diagnosis, was selected for this investigation. Biomarker information was missing from one case and CA125 levels were missing from one of 51 cases diagnosed more than one year prior to blood draw, leaving 50 cases in each group for analysis. Two cases were missing CA125 levels in the group diagnosed within one year of blood draw, leaving 65 of 67 cases for analysis. Two controls with high CA125 levels during the PLCO Trial and six population controls had missing biomarker data. Additional details of the study design have been previously described.⁸ The collection of biospecimens was approved by the NCI Special Studies Institutional Review Board (OH-C-N041) at the U.S. National Institutes of Health, and by the local Institutional Review Board for each screening site. Informed consent was obtained from all subjects who provided blood samples to be stored for future research. Approval for this study to use biological specimens to evaluate biomarkers for early detection of ovarian cancer was granted through a peer review process administered by the PLCO Etiologic and Early Marker Studies (EEMS) program.

Sample Storage and Processing

All aliquoting for Phase III studies investigating early detection markers for ovarian cancer was coordinated by the PLCO using a common sampling plan for several studies. The serum sample closest and prior to diagnosis was selected for laboratory analysis. The interval from the date of blood draw to diagnosis ranged from 12 to 2898 days, with 67 samples collected within one year. Blood was drawn prior to case diagnosis and was processed within two hours. Other details of sample storage and processing have been described. During processing, all specimens were organized so that equal proportions of cases, high-risk subgroup controls, general population controls, and QC replicates were included per batch of 96 samples. Aliquots were shipped to MD Anderson Cancer Center (MDACC, Houston, TX) and Vermillion, Inc. (Freemont, CA) for processing via FedEx on dry ice/overnight and stored at −70 °C before processing.

Biomarker Analysis

CA125 was measured and recorded by PLCO investigators using the CA125II radioimmunoassay (Centocor, Inc., Malvern, PA) according to manufacturer’s instructions.^2,8 Apo-A1 (mg/dL), TT (mg/dL) and ITIH4 (pg/uL), were measured as described previously.^4,6,10 Levels of CTAPIII, TRFR, HEPC (pg/uL), and β2M (Total Ion Current (TIC)) were estimated based on a relative quantification method that compared peak heights observed in study samples to references on a standard curve. All samples were analyzed using mass spectrometry assays using surface enhanced laser desorption/ionization mass spectrometry as described previously (Bio-Rad Laboratories, Inc., Hercules, CA).

Data Acquisition

Arrays were read in a PCS 4000 Protein Chip Reader, and a time-lag focus-in, linear laser desorption/ionization-time-of-flight mass spectrometer. Instruments were calibrated daily. Spectra were acquired in the positive-ion mode. Sampling rate was set at 800 MHz. Ions were extracted using 3.4 –kV extraction pulse and accelerated to a final velocity using 25-KV acceleration potential. The system employed a pulsed nitrogen laser at a repetition rate of 20Hz. Laser pulse energy of 1.500 to 2,000 nJ was delivered into a 100-uM diameter area. This illuminated area was rastered across a 2-mm diameter sample spot. An automated analytic protocol was used to control the data acquisition process throughout the analysis. Each spectrum averaged at least 1,000 laser shots and was externally calibrated against a mixture of known peptides/proteins.

Data Processing

Raw data obtained from the PCS4000 Protein Chip Reader were first smoothed by a fixed-width moving average filter of 25 data points. Subsequently, a convex hull baseline subtraction algorithm was applied to the smoothed data. Data were then internally normalized using total ion current with the Ciphergen Express 3.0 Software. Peaks corresponding to each biomarker were manually labeled and their intensities recorded from the protein chip array data while blinded to disease status. Samples were analyzed in triplicate. After spectral data were collected, archive spectral data was imported into OvaCalc Software (version 3.2.7) using Protein Chip Data Manager Software to detect peak intensities and perform calculations for data processing including; baseline subtraction (15x expected peak width based on mass), spectral filtering (0.2x expected peak width), starting mass (blanking mass +20%), and normalization to a factor of 1 (1/average raw ion current). After all four data processing steps, data was reported as a “normalized” Total Ion Current (TIC). Six of the seven biomarkers (excluding β2M) were further processed by comparing the TIC normalized to a referent standard curve, to provide absolute quantification.

Statistical Analysis

Levels of the biomarkers across groups were compared using the Kruskall-Wallis median analysis of variance (PROC NPAR1WAY, SAS 9). Logistic regression models were used to estimate odds ratios (ORs) and 95% confidence intervals (95% CIs) of associations between marker levels categorized into quartiles (based on control distributions) and ovarian cancer cases (PROC LOGISTIC). In addition to marker levels, logistic models included CA125 as either a continuous or categorical variable (≥35 U/ml) and age (in categories). Several classification models were built using k-nearest-neighbor classifiers with PROC DISCRIM^11,12. This non-parametric method was used because it does not make assumptions about marker distributions within each group. These models combined the marker levels (as continuous variables) and CA125 (as either continuous or a categorical variable (≥;<35 U/ml). Misclassification rates for the k-nearest neighbor models were assessed using leave one out cross-validation to obtain unbiased estimates of the predictive capability of each model ¹³.

In a second combined marker analysis, a linear prediction model that included only the proteomic biomarkers, was derived using sera from 41 early stage ovarian cancer patients and 99 healthy individuals from the MDACC sample bank. Linear discriminant analysis (LDA)¹⁴ with bootstrap was used to produce a single-valued index¹⁰. For the current study, the unblinded portion of the PLCO samples (n=561 including 59 cases) was used to derive a new model. The new model, a combination of the previously derived linear prediction model, CA125 and the 7 biomarker panel were tested on the remaining blinded portion of the PLCO samples.

RESULTS

Characteristics of study participants are provided in Table 1 for all ovarian cancer cases and for the three control subgroups. Case and control groups were balanced for age, race, smoking history, and use of hormone replacement therapy. Among the 118 ovarian cancer cases, fifty seven (47.3%) had CA125 levels ≥35U/ml. Histologic features of cases are provided for the 67 cases diagnosed within one year of blood draw and for the 51 cases diagnosed greater than one year prior to blood draw. In both groups, high grade serous cancers that presented at late stage (III–IV) were the most prevalent. Elevated CA125 levels (≥35 U/ml) were found in 40 of 65 (61.5%) serum samples from women diagnosed within one year of phlebotomy but only 1 of 50 (2%) of women diagnosed more than one year after the last blood draw.

Table 1

Characteristics of Study Population

In Table 2, means, medians, and standard deviations of the 7 biomarkers markers and CA125 levels are presented. P-values are given for comparison of cases diagnosed with ovarian cancer within one year of diagnosis, with each of the control subgroups and all controls combined. Only CA125 levels were statistically higher among this subset of cases when compared to all controls, and for each control subgroup. This difference was not observed in serum samples from cases diagnosed more than one year prior to diagnosis as both the mean and median levels of CA125 were much lower in this subgroup of cases, compared to those recently diagnosed. Table 3 shows ORs for biomarkers [in quartiles] mutually adjusted and also adjusted for CA125 as a dichotomous variable for the ovarian cancer cases diagnosed within one year prior to blood draw, compared to all controls, and to individual control subgroups. Only levels of CA125 were significantly elevated among all controls combined, and across control subgroups [excluding healthy controls that were selected as having a family history of breast and ovarian cancers, as there were no controls with elevated CA125 levels].

Table 2

Unadjusted group means and median test for marker differences by disease status and across control group subtypes

Table 3

Odds Ratios (OR) and 95% Confidence Intervals (CI) for Individual Markers in Multivariate Models Among Cases Diagnosed < 1 year of Blood Draw, Compared to all Controls and Control Subgroups

In Table 4, the cross-validated sensitivity and specificity estimates for five prediction models are presented for the cases diagnosed within one year of blood draw, and controls. Model 1 classified women only based on having elevated CA125 [<, ≥35 U/ml]. Models 2–4 were estimated using the K-nearest neighbor algorithm. Model 2 only included continuous CA125 levels; Model 3a included all continuous biomarkers levels and CA125 level as a dichotomous variable, comparing cases to all controls combined. Model 3b included all of the above with the exception cases were compared to controls and each control group was treated separately in the classification. Model 4a included all biomarkers and continuous CA125 levels with cases compared to all controls. Model 4b was the same as Model 4a but treated each control group separately in the classification. When CA125 was considered as a dichotomous variable, the sensitivity of the models was higher than when continuous levels were used (61.5; 95% CI: 48.6–73.4% vs. 41.5; 95% CI: 29.4–54.4), however, the specificity was 100% compared to 89.8% (95% CI: 87.7–91.7%) when dichotomized. Models that included the 7 biomarkers and CA125 exhibited a much lower sensitivity than those which included only CA125, whether used as a categorical or continuous marker, although the specificity remained high (≥98.5%) in each case.

Table 4

Cross-validated sensitivity and specificity estimates for various prediction models using the K-nearest neighbor algorithm.

The second analytical approach that combined CA125 and the biomarker panel also failed to improve upon the performance of CA125 alone. For the unblinded training samples, at a fixed specificity of 95%, the sensitivity for CA125 was 32.2% (95% CI: 21.3–44.6) while the sensitivity of the multivariate model was 35.6% (95% CI: 24.3–48.1). Using the same cutoffs, the sensitivity of CA125 in the blinded test set was 35.1% (95% CI: 23.7–47.8), yet the specificity dropped to 92.5% (95% CI 89.9–94.7). The model’s sensitivity on the test set was only 26.3% (95% CI: 16.2–38.5) at a specificity of 93.4% (95% CI: 90.9–95.4).

DISCUSSION

This study was performed in the context of a cooperative trial with other SPORE, EDRN and PLCO investigators that evaluated some 35 different biomarkers in proximal pre-diagnostically collected serum samples from 118 women who subsequently developed ovarian cancer. In evaluating samples obtained prior to conventional diagnosis, the addition seven biomarkers to CA125 in a combined multi-marker panel did not improve the sensitivity over that obtained with CA125 alone at 98% specificity. CA125 was originally used to monitor women known to have ovarian cancer ^15,16 as changes in CA125 levels can track the progression or regression of disease during treatment with up to 90% accuracy. Persistent elevation of CA125 following surgery and chemotherapy indicates the presence of residual disease. When multiple serum samples are evaluated, rising levels of CA125 can detect recurrent ovarian cancer with an average lead time of at least 3 months. CA125 can also be elevated many months prior to cancer diagnosis.¹⁶ In the current study, using one pre-diagnostic serum sample from PLCO Trial participants, CA125 was elevated (≥35 U/mL) in 61% of cases sampled within the 12 months prior to diagnosis and in 31% of the patients sampled more than 1 year prior to diagnosis. Observation of elevated CA125 in a significant fraction of patients >12 months prior to diagnosis can be considered an encouraging outcome.

Despite the observed lead time in the pre-diagnostic samples, the positive predictive value of CA125 was only 3.7% in the PLCO Trial. While the specificity of CA125² can approach 99% in postmenopausal women, given the very low prevalence of ovarian cancer (1 in 2,500 among postmenopausal women), 99.6% specificity is required to achieve a positive predictive value of 10%, meaning that there are less than 10 surgeries for each diagnosed ovarian cancer case. Specificity of CA125 was improved by combining biomarker results with TVS. In the PLCO Trial, if CA125 was elevated and a TVS was abnormal, the positive predictive value increased to 23.5%, although 60% of the invasive cancers would not have been detected.²

Specificity of CA125 can also be increased by studying changes in levels over time. An algorithm using age and change point analysis to determine whether levels have increased beyond the subject’s own baseline when compared to annual determinations¹⁷ is being evaluated in a study of 202,638 British woman (UKCTOCS).¹⁸ TVS result combined with rising CA125 levels with time produced a sensitivity of 89.5%, a specificity of 99.8% and a positive predictive value of 35%. Overall, 48% of the prevalent cancers were detected in stage I or II, twice the fraction expected using conventional diagnostic methods. In the U.S., a similar screening trial of 3,252 postmenopausal women followed annually with CA125 using the same algorithm coordinated by the Ovarian SPORE at MDACC.¹⁹ Less than 1% of patients were referred for TVS each year and less than 3% over multiple years. Overall, the positive predictive value for the entire screen was 37%, consistent with the British study.

Only 80% of invasive ovarian cancers express significant quantities of CA125¹⁵. Consequently, numerous biomarkers have been evaluated to complement or replace CA125. Using multiplex assays, some 96 biomarkers have been tested for the ability to distinguish healthy individuals from women with stage I ovarian cancer.²⁰ A four biomarker panel that included CA125, human epididymous protein 4 (HE4), carcinoembryonic antigen (CEA) and soluble vascular adhesion molecule (sVCAM) produced 86% sensitivity for early stage (I/II) disease and 95% sensitivity for late stage (III/IV) disease at 98% specificity. Similar sensitivity and specificity were observed at the time of conventional diagnosis with the proteomic biomarkers evaluated in this study. A panel of Apo-A1, TT, and CTAPIII produced 87% sensitivity at 98% specificity for distinguishing stage I ovarian cancer from healthy individuals.⁷

The discrepancy in the results in pre-diagnostic samples for the three biomarker panel compared to our previous studies using post-diagnostic samples^4,6 may relate to the fact that Apo-A1, TT and CTAPIII are each acute phase reactants that are down-regulated in response to cancer, and that a significant volume of tumor may be required to trigger this reaction. While stage I disease is by definition limited to the ovaries, the volume of cancer can be substantial, but must be sufficient to produce symptoms or a readily palpable pelvic mass to result in conventional diagnosis. In pre-clinical disease, the volume of tumor may not be sufficient to evoke an acute phase response, and alter levels of proteins associated with this process. Another issue with the markers in our panel (with the exception of CA125) is that they were lower in cancer cases, but also be reduced as a result of storage time, sample processing, and other host factors (described in our previous report) ⁶ which could hinder identification of small volume disease, a goal of the current study. Lastly, the current study clearly attempted to reduce confounding and bias in ways that were not possible in our previous study in which we matched cases to hospital controls rather than healthy women, and we were unable to control sample processing or match on storage time which may have biased results.

Although a marginally higher performance was observed for identification of cases at least six months prior diagnosis using an all-site multi-marker panel (that included CA125, HE4, tumor associated glycoprotein 72 (CA72-4), substance P-like immunoreactivity (SLPI) and β2M) were observed compared to CA125 alone, the increase was not statistically significant.²¹ In addition to the current study, five additional panels were evaluated, none of which improved upon results with CA125 alone⁸. Considering the failure of multiple biomarkers to improve upon CA125 in pre-diagnostic samples, new approaches are badly needed for biomarker discovery. One weakness of the current study is that we were unable to evaluate markers in non-white populations due to a very small number of non-white cases in the PLCO Trial. The results of this combined effort will likely reshape our approach to biomarker discovery and validation. In addition to searching for protein analytes, auto-antibodies might also be sought. Lastly, previous studies have had limited success in identifying and evaluated auto-antibodies of human proteins expressed in bacteria or insect cells. Recent advances in expressing human proteins in human cells could allow identification of new epitopes that are selective for altered tertiary structure and glycosylation status of selected protein targets.

Acknowledgments

This work was supported by funds from the M.D. Anderson SPORE in Ovarian Cancer NCI P50 CA83639 and philanthropic support from Golfers Against Cancer, the Tracey Jo Wilson Foundation and the Mossy Foundation, and by the Intramural Research Program of the NIH, National Cancer Institute, Division of Cancer Epidemiology and Genetics. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.

Footnotes

Financial Disclosures: R.C.B. receives royalties for the discovery of CA125 and has served on the Scientific Advisory Boards of Fujirebio Diagnostics, Inc., and Vermillion, Inc.

References

1. Cannistra SA. Cancer of the Ovary. New Engl J Med. 2004;351:2519–29. [PubMed]

2. Buys SS, Partridge E, Greene MH, et al. Ovarian cancer screening in the Prostate, Lung, Colorectal and Ovarian (PLCO) cancer screening trial: Findings from the initial screen of a randomized trial. American Journal of Obstetrics and Gynecology. 2005;193(5):1630–1639. [PubMed]

3. Kim G, Minig L, Kohn EC. Proteomic profiling in ovarian cancer. Int J Gynecol Cancer. 2009;19:S2–S6. [PubMed]

4. Zhang Z, Bast RC, Jr, Yu Y, Li J, Sokoll LJ, Rai AJ, Rosenzweig JM, Cameron B, Wang YY, Meng XY, Berchuck A, Van Haaften-Day C, Hacker NF, de Bruijn HW, van der Zee AG, Jacobs IJ, Fung ET, Chan DW. Three biomarkers identified detection of early stage ovarian cancer. Cancer Res. 2004;64:5882–90. [PubMed]

5. Hodgall E, Fung ET, Zhang Z, Engelholm SA, Petri AL, Risum S, Lundvall L, Nedergaard L, Pedersen AT, Hartwell D, Yip C, Wang Z, Meng XY, Tjirasoraam V, Lomas L, Chan DW, Hogdall C. Prospective independent validation of a marker panel for distinguishing malignant from benign pelvic masses. Abstract. 39th Annual Meeting of the Society of Gynecologic Oncologists; Gynecol Oncol; 2008. Supplement 1.

6. Moore LE, Fung ET, McGuire M, Rabkin CC, Molinaro A, Wang Z, Zhang F, Wang J, Yip C, Meng X-Y, Pfeiffer RM. Evaluation of apolipoprotein A1 and postranslationally modified forms of transthyretin as biomarkers for ovarian cancer detection in an independent study population. Cancer Epidemiol Biomarkers Prev. 2006;15:1641–46. [PubMed]

7. Clarke CH, Fung ET, Yip C, Joy C, Badgwell D, Coombes KR, Zhang Z, Lu KH, Bast RC. Use of a panel of proteomic markers to improve the sensitivity of CA125 for detecting stage I epithelial ovarian cancer. J Clin Oncol. 2008 May 20;26(suppl):abstr 5542.

8. Zhu CS, Pinsky PF, Cramer DW, Ransohoff DF, Hartge P, Pfeiffer RM, Urban N, Mor G, Bast RC, Jr, Moore LE, Lokshin AE, McIntosh MW, Skates SJ, Vitonis A, Zhang Z, Ward DC, Symanowski JT, Lomakin A, Fung ET, Sluss PM, Scholler N, Lu KH, Marrangoni AM, Patriotis C, Srivastava S, Buys SS, Berg CD. for the PLCO Project Team. A framework for evaluating biomarkers for early detection: validation of biomarker panels for ovarian cancer. Cancer Prev Res. 2011;4(3):375–383. [PMC free article] [PubMed]

9. Cramer DW, Bast RC, Jr, Berg CD, Diamandis EP, Godwin AK, Hartge P, Lokshin AE, Lu KH, McIntosh MW, Mor G, Patriotis C, Pinsky PF, Thornquist MD, Scholler N, Skates SJ, Sluss PM, Srivastava S, Ward DC, Zhang Z, Zhu CS, Urban N. Ovarian cancer biomarker performance in Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial specimens. Cancer Prev Res. 2011;4(3):365–374. [PMC free article] [PubMed]

10. Fung ET, Yip TT, Lomas L, Wang Z, Yip C, Meng XY, Lin S, Zhang F, Chan DW, Weinberger SR. Classification of cancer types by measuring variants of host response proteins using SELDI serum assays. Int J Cancer. 2005;115:783–9. [PubMed]

11. Hand DJ. Discrimination and Classification. NY: John Wiley & Sons, Inc; 1981.

12. SAS Users Guide Version 9. SAS Institute; Cary, NC, USA: 1999.

13. Breiman L, Spector P. Submodel selection and evaluation in regression. The X-random case. International Statistical Review. 1992;60:291–319.

14. Fisher RA. The use of multiple measurements in taxonomic problems. Ann of Eugenics. 1950;1936(7):179–188.

15. Bast RC, Jr, Klug TL, St John E, Jenison E, Niloff JM, Lazarus H, Berkowitz RS, Leavitt T, Griffiths CT, Parker L, Zurawski VR, Jr, Knapp RC. A radioimmunoassay using a monoclonal antibody to monitor the course of epithelial ovarian cancer. New Engl J Med. 1983;309:883–887. [PubMed]

16. Bast RC, Jr, Siegal FP, Runowicz C, Klug TL, Zurawski VR, Jr, Schonholz D, Cohen CJ, Knapp RC. Elevation of serum CA125 prior to diagnosis of an epithelial ovarian carcinoma. Gynecol Oncol. 1985;22:115–120. [PubMed]

17. Skates SJ, Xu FJ, Yu YH, Sjovall K, Einhorn N, Chang Y, et al. Toward an optimal algorithm for ovarian cancer screening with longitudinal tumor markers. Cancer. 1995;76(10 Suppl):2004–2010. [PubMed]

18. Menon U, Gentry-Maharaj A, Hallett R, Ryan A, Burnell M, Sharma A, et al. Sensitivity and specificity of multimodal and ultrasound screening for ovarian cancer, and stage distribution of detected cancers: results of the prevalence screen of the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS) Lancet Oncol. 2009;10(4):327–340. [PubMed]

19. Lu KH, Skates S, Bevers T, Adeyinka O, Newland W, Moore R, Leeds L, Harris L, Harris S, Fritsche H, Bast RC., Jr A prospective U.S. ovarian cancer screening study using the risk of ovarian cancer algorithm (ROCA) Proc Amer Soc Clin Oncol. 2010;28:390s. #5003.

20. Yurkovetsky Z, Lomakin A, Skates S, Menon U, Marrangoni A, Velikokhatnaya L, Nolen B, Winans M, Modugno F, Marks J, Godwin A, Gorelik E, Badgwell D, Bast RC, Jr, Lokshin A. Development of a multimarker assay for early detection of ovarian cancer. J Clin Oncol. 2010;28:2159–66. [PMC free article] [PubMed]