Predictive Blood-Based Biomarkers in Patients with Epithelial Ovarian Cancer Treated with Carboplatin and Paclitaxel with or without Bevacizumab: Results from GOG-0218

Epithelial ovarian cancer (EOC) is the leading cause of gynecologic cancer-related death in the United States (1). Several antiangiogenic agents (bevacizumab, nintedanib, cediranib, pazopanib) have demonstrated clinical efficacy and improved progression-free survival (PFS; refs. 2–7). In specific subset analyses, these agents have demonstrated increases in overall survival (OS) within selected patients with ovarian cancer as well (8–10). On June 13, 2018, the FDA approved bevacizumab for use in combination with carboplatin and paclitaxel, followed by bevacizumab maintenance therapy, as a first-line treatment for women with advanced ovarian cancer (11). However, bevacizumab and the other antiangiogenic agents have significant side effects and expense, not all patients respond to the treatment, and ultimately resistance to the treatment develops. Given the projected increase in the global burden of cancer and limited health care resources, it is imperative to conduct research to define patients with EOC that will benefit from antiangiogenic-specific therapy. Rationally directed antiangiogenic therapy in women with EOC can maximize benefit, while minimizing toxicity and cost of unnecessary treatment (12).

GOG-0218 is the pivotal phase III 3-arm placebo-controlled randomized clinical trial that evaluated the efficacy of first-line chemotherapy with and without the antiangiogenic agent, bevacizumab (3). In this study, subjects were randomized to one of three treatment arms. All arms included standard intravenous (IV) chemotherapy with carboplatin at an area under the curve of 6 and paclitaxel 175 mg/m², for cycles 1 through 6, and a study maintenance treatment for cycles 2 through 22. The cycles were delivered every 3 weeks. Arm A represented the control treatment that included chemotherapy combined with placebo in cycles 2 through 22. Arm B comprised of concurrent chemotherapy and bevacizumab (15 mg/kilogram of body weight IV) for cycles 2 through 6 followed by placebo cycles 7 through 22. In arm C, bevacizumab was administered with chemotherapy for cycles 2–6 and continued through cycle 22. The randomized trial design with a placebo control arm allowed for the identification of factors that specifically predict which patients will and will not benefit from bevacizumab treatment. The GOG-0218 trial included the acquisition of pretreatment plasma specimens that were available for analysis. The GOG-0218 dataset has previously been evaluated for blood- and tissue-based markers (13). No prognostic or predictive association was seen for any of the markers evaluated, including VEGF-A, VEGFR-2, NRP-1, or MET. However, when comparing IHC-based tumor CD31 microvascular density (MVD) along with tumor VEGF-A levels [>quartile (Q)3 vs. ≤Q3], these markers demonstrated prognostic and potential predictive value for PFS and OS in the concurrent and maintenance bevacizumab arm (CPBB) compared with the placebo control arm (CPP).

While the IHC biomarker analysis appears promising, interest in blood-based markers exists due to ease of sample collection and longitudinal analyses, reproducible and quantifiable results, and circumvention of tumor tissue heterogeneity concerns (14–16). Multiplex ELISA technology allows for efficient evaluation of multiple soluble markers simultaneously. We have developed and optimized a protein multiplex array for the evaluation of key angiogenic and inflammatory markers, termed the Angiome. The Angiome multiplex array has been approved by the NCI Biomarker Review Committee as an integrated biomarker for use in NCTN and ETCTN studies. While the full Angiome array evaluates 26 unique protein markers, we prioritized seven markers [IL6, Ang-2, osteopontin (OPN), stromal cell-derived factor-1 (SDF-1), VEGF-D, IL6R, and GP130] that had been previously shown to be predictive of benefit from antiangiogenic therapies in other solid tumors and/or associated with prognostic outcomes with EOC or implicated with ovarian carcinogenesis (17–24) for the primary inferential analysis. The main objectives of this study were to evaluate whether the plasma Angiome components (IL6, Ang-2, OPN, SDF-1, VEGF-D, IL6R, and GP130) were associated with a therapeutic bevacizumab advantage for PFS and/or OS in women with advanced EOC treated on GOG-0218.

The study design and outcomes of GOG-0218 have been previously reported (3). The clinical study was approved by the Institutional Review Boards at all participating centers and conducted in accordance with the Declaration of Helsinki guidelines. All patients provided written informed consent. Eligible patients had histologically confirmed stage III or stage IV EOC and had undergone primary debulking surgery. Institutional review board approved, written informed consent was obtained from patients who opted to participate in the translational components of GOG-0218. This exploratory retrospective analysis was approved by the NRG Oncology Translational Science Committee, had a prespecified analysis plan, and conforms to the reporting guidelines established by the REMARK criteria (25).

Available plasma specimens previously collected from women registered to GOG-0218, treated on either the CPP or CPBB arms, who were eligible and provided consent to the use of their specimens and clinical information for future cancer research were evaluated. Baseline peripheral blood samples were collected into EDTA anticoagulant vacutainers and centrifuged within 2 hours of collection at 3,500 × g at 4°C for 10 minutes. Plasma was aliquoted into cryovials, snap-frozen, and shipped on dry ice for storage at −80°C at the centralized GOG Tissue Bank. Samples were subsequently shipped to the Duke Molecular Reference Laboratory (Durham, NC), thawed on ice, realiquoted on the basis of specific assay requirements, and stored at −80°C. All sample and data handling procedures were fully compliant with the Health Insurance Portability and Accountability Act of 1996, and the study was conducted under the Duke Institutional Review Board approval.

Plasma samples from GOG-0218 were assessed using multiplex array technology (CiraScan platform from Aushon BioSystems Inc.). Samples were analyzed for seven biomarkers (IL6, Ang-2, OPN, SDF-1, VEGF-D, IL6R, and GP130) following the manufacturer's protocols. Plasma samples were thawed on ice, centrifuged at 20,000 × g for 5 minutes to remove precipitate, and subsequently loaded onto multiplex plates with standard protein controls as previously reported (26). Samples and standards were incubated at room temperature for 1 hour shaking at 450 rpm (Lab- Line Titer Plate Shaker, Model 4625, Barnstead). Plates were washed, biotinylated secondary antibody was added, and plates were incubated for 30 minutes. After washes, streptavidin-HRP was added, plates were incubated for 30 minutes, washed, and SuperSignal substrate was added. Images were taken within 10 minutes, followed by image analysis using the array analyst software. All marker data represent the average of duplicate measures multiplied by dilution, and all analyses were conducted while blinded to clinical outcome.

The primary objective was to determine whether components of the plasma Angiome panel (IL6, Ang-2, OPN, SDF-1, VEGF-D, IL6R, and GP130) were predictive of a therapeutic advantage in PFS of bevacizumab treatment in women with advanced EOC treated on GOG-0218. The secondary objective was to determine whether these biomarkers were predictive of OS. The predictive analyses for both PFS and OS were conducted using a protein marker by treatment interaction term within the framework of a Cox proportional hazards model adjusting for baseline covariates of age, stage/debulking status, and performance status as additive effects. Results presented included hazard ratios (HR) and associated confidence intervals (CI), along with P values for Wald test for interaction. An exploratory analysis was performed to identify markers prognostic of survival outcomes for women with advanced EOC. The prognostic analyses for PFS and OS were performed using the Cox proportional hazards models adjusted for baseline covariates. PFS analyses were stratified by treatment, due to the differences in outcome, which were previously observed.

Prior to performing the analyses, biomarkers were inspected for outliers, defined as any values either less than Q1-1.5xIQR or greater than Q3+1.5xIQR, where Q1 and Q3 are the first and third quartiles, respectively, and IQR is the interquartile range. The similarity among biomarkers at baseline was analyzed using hierarchical clustering. Biomarkers were natural log-transformed for the predictive and prognostic analyses, and analyzed as continuous measures.

Additional exploratory and sensitivity analyses were conducted. The predictive and prognostic analyses for both PFS and OS were repeated using the Cox rank-score test, which is robust against outliers, as well as with the outliers removed for sensitivity analysis. Kaplan–Meier plots were used to illustrate differences in PFS and OS for IL6 dichotomized at the median as “Low” versus “High.” Cut-off point optimization using conditional inference trees was performed for selected markers for PFS and OS for exploratory purposes. Kaplan–Meier plots were presented for the marker levels dichotomized at the resulting PFS and OS cut-off point values. It is noted that the primary inferential analyses are based on continuous biomarkers. All analyses based on using cut-off points were considered to be exploratory and were exclusively used for the purpose of illustration and generation of hypotheses. The cut-off points were optimized on the basis of data in the control arm, which may exaggerate the predictive effect (27).

The primary analyses, the interaction of seven markers with bevacizumab with respect to PFS, were adjusted for multiple testing so as to control the family-wise error rate at the two-sided 0.05 level. More specifically, the seven analyses were conducted at the Bonferroni adjusted two-sided alpha level of 0.007 = 0.05/7. The secondary and exploratory analyses were not adjusted for multiple testing. The two-sided unadjusted P values and two-sided 95% confidence intervals (95% CI) are presented.

The R statistical environment [R] version 3.4.4 (28), along with extension packages survival (v 2.41-3; ref. 29), partykit (v 1.2-0; ref. 30, 31), and tidyverse (v 1.2.1; ref. 32), were used to conduct the statistical analyses. The Knitr extension package was used for generation of dynamic reports (33).

Of the 1,248 patients who enrolled to either the control or bevacizumab-throughout arm on the parent protocol, baseline EDTA plasma samples were available for analysis in 751 patients: 384 patients in the control arm and 367 in the bevacizumab-throughout arm (Supplementary Fig. S1). The demographics and clinical characteristics of these patients in the biomarker-evaluable population appeared to be similar to those in the intent-to-treat population of the parent study (Table 1).

The relationships between outcome, PFS, and OS, and each of the baseline covariates (age, stage/debulking status, and performance status) are summarized in Supplementary Table S2. Stage/debulking status and performance status were associated with PFS and OS, while age was only associated with OS.

Multiplex analyses demonstrated good sensitivity and coefficients of variation (CV) ranged from 1% to 6% for all markers, with the exception of SDF-1, which was 11.9%. The medians for all biomarkers at baseline were as follows: 264.2 pg/mL for Ang-2, 369.1 ng/mL for GP130, 22.1 pg/mL for IL6, 34.8 ng/mL for IL6R, 986.2 ng/mL for OPN, 3.0 ng/mL for SDF-1, and 1.0 ng/mL for VEGF-D (Supplementary Table S3).

The primary objective was to determine whether any marker (IL6, Ang-2, OPN, SDF-1, VEGF-D, IL6R, and GP130) was predictive of PFS for women treated with bevacizumab on GOG-0218. The secondary objective was to determine whether any of these biomarkers were predictive of OS for women treated with bevacizumab on GOG-0218. The analysis of the interaction with bevacizumab treatment (predictive efficacy) and each biomarker was assessed on the basis of a continuous quantification of the latter and accounted for the following baseline covariates: age, stage/debulking status, and performance status. IL6 was found to be predictive of a therapeutic advantage with bevacizumab for PFS (P = 0.007; Table 2). For illustrative purposes, we present Kaplan–Meier plots with IL6 dichotomized at the median. Patients with high IL6 levels (>median value of 22.1 pg/mL) treated with bevacizumab throughout had longer PFS (14.2 vs. 8.7 months; HR, 0.63; CI, 0.51–0.79) compared with those treated with placebo (Table 3; Fig. 1A). In contrast, there was no improvement in PFS for those with low IL6 levels treated with bevacizumab compared with placebo [16.9 vs. 12.6 months; HR, 0.93; CI, (0.74–1.15); Table 3; Fig. 1A).

The secondary analysis was to determine whether any of the seven markers tested as continuous variables were predictive of OS for women treated with bevacizumab on GOG-0218. As observed for PFS, IL6 was again found to be predictive of a therapeutic advantage with bevacizumab for OS (P = 0.003; Table 2). Patients with high IL6 levels (>median value of 22.1 pg/mL) treated with bevacizumab throughout had longer OS [39.6 vs. 33.1 months; HR, 0.78; CI, (0.62–0.98)] compared with those treated with placebo (Table 3; Fig. 1B). Patients with low IL6 levels treated with bevacizumab had shorter OS [48.5 vs. 50.8 months; HR, 1.09; CI, (0.85–1.39)] compared with those treated with placebo (Table 3; Fig. 1B). In addition to presenting the continuous, log-adjusted marker level results, we also conducted a quartile analysis of IL6 for both PFS and OS, shown in Supplementary Table S4. None of the other markers (Ang-2, OPN, SDF-1, VEGF-D, IL6R, or GP130) were predictive of a therapeutic advantage or disadvantage with bevacizumab.

To further explore and illustrate the interaction between IL6 and bevacizumab with respect to PFS and OS, we used conditional inference trees to determine the optimal cut-off point. In optimizing the cut-off point, we used the control arm data to avoid confounding due to differences in outcome between arms. The optimal cut-off point of IL6 based on PFS was 21.4 pg/mL while the optimal cut-off point of IL6 based on OS was 90.2 pg/mL. The PFS-derived IL6 cut-off point was found to be very similar to the median IL6 value (21.4 pg/mL vs. 22.1 pg/mL, respectively; Table 3); however, the OS-derived IL6 cut-off point was much higher, at the 87th percentile. For the small subset of patients (98/751) with high IL6 based on the OS-derived cut-off point, there was a seemingly large benefit for those treated with bevacizumab compared with the control group for both the PFS endpoint [13.0 vs. 6.8 months; HR, 0.4; CI, (0.26–0.62)] (Table 3; Fig. 2C) and the OS endpoint [36.0 vs. 17.5 months; HR, 0.44; CI, (0.28–0.70); Table 3; Fig. 2D]. In contrast, dichotomizing IL6 by the median or PFS-derived cut-off point resulted in only a 6–7 month difference in median OS when treated with bevacizumab compared with placebo (Table 3; Fig. 2B).

Because it has been well established that both soluble IL6R and GP130 can bind IL6 in vivo (33, 34), we performed an ad hoc analysis to assess the synergistic effects among the treatment, IL6, and either IL6R or GP130 with respect to outcome (PFS and OS). To this end, we employed three-way multiplicative Cox proportional hazards models adjusting for age, stage/debulking status, and performance status as additive effects (Supplementary Table S5). There was no statistical evidence to support any three-way interaction between treatment, IL6, and either IL6 binder, IL6R, or GP130 (Supplementary Tables S6–S9).

Exploratory analyses were also performed to identify whether IL6, Ang-2, OPN, SDF-1, VEGF-D, IL6R, and/or GP130 were prognostic of outcome for women enrolled on GOG-0218. Both IL6 and OPN were found to be negative prognostic markers for PFS [HR, 1.14; CI, (1.07–1.21); P < 0.001)] and [HR, 1.48; CI, (1.28–1.7); P < 0.001], respectively. IL6 and OPN were also found to be negative prognostic markers for OS [HR, 1.17; CI, (1.1–1.26); P < 0.001] and [HR, 1.59; CI, (1.37–1.84); P < 0.001], respectively. No prognostic associations were observed for any of the other biomarkers tested for either PFS or OS (Table 4).

IL6 signaling plays an important role in carcinogenesis across a variety of solid tumors, including ovarian cancer, regulating proliferation, adhesion, invasion, as well as angiogenesis, and immunologic functions (22). Furthermore, IL6 has been shown to be elevated in patients with ovarian cancer exhibiting paraneoplastic thrombocytosis and has been suggested to be a key mediator driving this biology (35). Our findings indicate that IL6 may identify patients most likely to benefit from the addition of bevacizumab to standard-of-care chemotherapy in women with newly diagnosed advanced EOC. Furthermore, IL6 plasma levels were negative prognostic markers for both PFS and OS in women treated on GOG-0218. These paradoxical findings suggest that the patients destined to do worse (high IL6 levels) may be the most likely to benefit from treatment with bevacizumab.

These data confirm previous findings where IL6 levels were both prognostic for survival in EOC (17) as well as predictive of survival benefit in patients with metastatic renal cancer in two independent antiangiogenic therapeutic trials; one evaluating bevacizumab combined with IFNα and the other evaluating pazopanib (36, 37). Moreover, risk score analyses based on IL6 and hepatocyte growth factor suggest that IL6 is a predictive marker of bevacizumab efficacy in patients with pancreatic cancer (26). The range of median IL6 levels in the renal, pancreatic, and our current ovarian study is 13.1–22.1 pg/mL and represents minimal variation. In addition, these median IL6 levels are strikingly similar to IL6 level measured in our phase II clinical trial of nintedanib in women with recurrent ovarian cancer (23.8 pg/mL; ref. 38). In contrast, no association between IL6 levels and treatment response was observed in women with recurrent ovarian, tubal, or peritoneal cancer treated with or without pazopanib in a randomized phase II study (39). The median IL6 levels identified may be specific to disease type and patient population and could also vary due to differences between processing, assays, prior therapy, and patient characteristics.

To identify the optimal cut-off point that would have clinical relevance to direct first-line concurrent and maintenance bevacizumab therapy in women with advanced epithelial ovarian, tubal, and peritoneal cancers, we tested various IL6 cut-off points. These exploratory cut-off points included the median value as well as performing cut-off point optimization analyses based on both PFS and OS. The PFS-derived cut-off point was very similar to the median cut-off point (21.4 pg/mL vs. 22.1 pg/mL) and indicated that high IL6 using either the median or PFS-derived cut-off point was associated with bevacizumab efficacy. On the basis of either the median or PFS cut-off points, women with high IL6 levels treated with bevacizumab appear to have a reduction in the hazard of disease progression and death, respectively. Interestingly, when applying the OS optimized cut-off point (90.2 pg/mL, equivalent to the 87% quantile), we observed that women with high IL6 treated with bevacizumab throughout had the greatest reduction in the hazard of disease progression and death. Using the OS-derived cut-off point, we were able to identify a highly sensitive subset of the population with extremely high levels of IL6 that appear to benefit the most with the addition of first-line bevacizumab therapy. These findings suggest that IL6 may assist in directing first-line bevacizumab therapy to maximize benefit and minimize toxicity. However, these exploratory observations must be viewed with caution as cut-off point optimization may lead to exaggerated estimates of the predictive effects (27). Our findings in the IL6-low group also need to be considered carefully. The CI for treatment effect in the IL6-low subgroup is too wide to confidently rule out clinically important benefit. We also point out that even if these cut-off points are valid for our study, they may not be generalizable to other studies.

Because the biomarkers were chosen a priori and tested for interaction with treatment with respect to clinical outcome, we opted not to internally validate these findings using training–testing splits of the available data or cross-validation. Further testing is needed to validate the role of IL6 as a predictive biomarker before it can be used to direct clinical care. We note that the results presented here, especially the estimated effect sizes, may be sensitive to departures from the proportional hazards assumption. As with any biomarker analysis, one has to be concerned with potential confounding due to baseline covariates. Although we have attempted to address this issue by adjusting for baseline covariates as additive effects in our multivariable models, we cannot rule out confounding. The relationships between outcome, PFS, and OS, and each of the baseline covariates are summarized in Supplementary Table S2. The relationships between IL6 and each of the baseline covariates are illustrated in Supplementary Table S5. The upcoming NRG Oncology GY004 and GY005 studies comparing olaparib and cediranib to standard-of-care therapies include evaluation of IL6 as a predictive biomarker, further extending the potential role for IL6 to guide the use of multiple antiangiogenic therapeutic approaches. Continued evaluation of IL6 in these studies will further refine and optimize a cut-off point for IL6 to be used in prospective testing. Identifying biomarkers to direct therapy is critically important to reduce cost as well as prioritize treatment sequencing in an era where alternative treatment options exist for EOC [such as PARP inhibitors (PARPi) and immunotherapies; ref. 12].

Previously, two separate investigators independently explored molecularly defined tumor subgroups in women who participated in the ICON7 trial (40, 41), an open-label, two-arm study in patients who had stage I to III debulked or any stage IV ovarian cancer. Gourley and colleagues identified two molecularly defined groups, defined as proangiogenic and immune, and reported that the immune signature was prognostic for improved survival outcomes. However, the immune signature subgroup had worse PFS (HR, 1.73; CI, 1.12–2.68) and OS (HR, 2.00; CI, 1.11–3.61) when treated with bevacizumab compared with chemotherapy alone (40). Winterhoff and colleagues stratified patients who participated in ICON7 into four TCGA serous subgroups (proliferative, mesenchymal, immunoreactive, and differentiated subgroups) and reported that median PFS and OS improvements with bevacizumab were not greater in the differentiated and immunoreactive subtypes. In this report, patients with high-grade serous carcinomas of mesenchymal and proliferative subtypes obtained the greatest overall survival benefit from bevacizumab while those with high-grade serous proliferative subtype demonstrated a modest improvement in PFS only (41). Recently, the group updated their data and reported that only patients with the proliferative subtype had a statistically significant benefit from the addition of concurrent and maintenance bevacizumab to standard chemotherapy [median PFS 22.2 vs. 12 months, HR 0.48 (95%CI, 0.3–0.76), P = 0.002; median OS 52.4 vs. 35.3 months, HR 0.54 (95%CI, 0.3–0.9), P = 0.021]. In this updated analysis, there was no significant improvement of PFS or OS with the addition of bevacizumab in the mesenchymal subtype (42). Collinson and colleagues developed a signature using VEGFR-3, α1-acid glycoprotein, mesothelin, and CA-125 that was predictive of bevacizumab response. The signature-positive group demonstrated improved median PFS in the bevacizumab arm (17.9 vs. 12.4 months; P = 0.04), while the signature-negative group had improved PFS in the chemotherapy alone arm (36.3 vs. 20 months, P = 0.006; ref. 43). While none of these studies has been validated, the finding that several of the biomarker groups did not benefit from bevacizumab is worthy of further investigation.

Birrer and colleagues evaluated blood-based and tumor biomarkers from women participating in GOG-0218 and identified IHC-based tumor CD31 microvascular density (MVD) along with tumor VEGF-A levels [>quartile Q3 vs. ≤Q3] as both prognostic and predictive for concurrent and maintenance bevacizumab efficacy (13). Other markers VEGFR-2, neuropilin-1, and MET had no prognostic or predictive association with survival outcomes or bevacizumab efficacy. Buechel and colleagues conducted a GOG-0218 ancillary study evaluating the association between markers of adiposity, subcutaneous fat, and visceral fat density (SFD/VFD) measurements, derived from CT imaging and survival outcomes. Increased SFD and VFD correlated with a significantly increased risk for death (HR per 1-SD increase 1.12; 95% CI, 1.05–1.19; P = 0.0009 and 1.13; 95% CI, 1.05–1.20; P = 0.0006, respectively). High VFD was associated with an increased risk for death in the placebo group (HR per 1-SD increase 1.22, 95% CI, 1.09–1.37), but not in the bevacizumab group. There was no correlation between high VFD and IL6 levels (r = 0.02; P = 0.57; ref. 44). Evaluation of the IL6 blood-based marker in context with the molecular profile, as well as the degree of tumor angiogenesis based on CD31 tumor staining, VEGF-A levels, and SFD/VFD may elucidate the etiology behind the survival outcome difference and improve identification of candidates most likely to benefit from incorporation of bevacizumab to first-line chemotherapy.

The development of validated predictive markers may be greatly affected by the analytic processes employed; quality and variability of sample processing; long-term storage stability, number of freeze–thaw cycles; statistical analysis; and underlying differences in tumor biology. This study was limited by the lack of monitored site-specific sample processing, which may lead to variability, even with a standardized SOP provided to all sites. While the age of plasma samples varied, once samples were received by our laboratory, the samples were all treated similarly; and the assays were performed according to highly standardized methods encompassing sample type tested (all EDTA plasma), the number of freeze–thaw cycles, consistent reagents from single-batch plate printing, and reference standards. Importantly, the assays were performed by research personnel blinded to the clinical data.

The search for biomarkers continues in an effort to provide therapeutic rationale for antiangiogenic therapy in this era of “Precision Medicine” and as cost minimization strategies. The recent FDA approval of bevacizumab for use in combination with carboplatin and paclitaxel, followed by bevacizumab maintenance therapy, for women with newly diagnosed advanced ovarian cancer highlights the importance of identifying a biomarker to direct bevacizumab therapy (11). We have previously reported that rational biomarker-directed bevacizumab therapy reduced the cost of unnecessary treatment in our cost effectiveness analysis (12). Our current study demonstrates that the inflammatory cytokine IL6 may be predictive of therapeutic benefit from bevacizumab when combined with standard paclitaxel and carboplatin chemotherapy, and our preliminary results appear promising. Our findings are consistent with previous findings in which IL6 predicted benefit from bevacizumab and pazopanib in patients with renal cell cancer, highlighting the potential intersection between inflammation and angiogenesis. Further research regarding the mechanistic effects of IL6 and its signaling partners is needed to understand the role of IL6 in ovarian carcinogenesis. Moreover, additional validation studies and integral biomarker directed clinical trials are required to determine whether the plasma biomarker IL6 can accurately identify patients with EOC who may benefit from bevacizumab.

A. Alvarez Secord is an employee/paid consultant for Genentech, AstraZeneca, and Clovis, and reports receiving commercial research grants from Genentech, Amgen, AstraZeneca, Boehringer Ingelheim, Clovis, Eisai, and Exelixis. H.I. Hurwitz is an employee/paid consultant for and reports receiving commercial research grants from Genentech. K.S. Tewari reports receiving speakers bureau honoraria from Merck, Genentech, Tesaro, Clovis, and AstraZeneca, and is an advisory board member/unpaid consultant for GlaxoSmithKline and Genentech. D.M. O'Malley is an employee/paid consultant for AstraZeneca, Clovis, Tesaro, AbbVie, GOG Foundation, Novocure, Immugen, Translational Genomics Marker Therapeutics, Agenus, Roche, Ambry Genetics, OncoQuest, Myriad, Janssen, Eisai, and LEAP Therapeutics, and reports receiving other remuneration from Roche/Genentech. K. Fujiwara is an employee/paid consultant for, reports receiving commercial research grants from, and reports receiving speakers bureau honoraria from Chugai-Roche. M. Boente is an employee/paid consultant for Genentech. R.A. Burger is an advisory board member/unpaid consultant for Amgen, AstraZeneca, Tesaro, Clovis Oncology, Genentech/Roche, Immunogen, Agenus, Gradalis, Janssen R&D, Merck, Morphotek, and VBL Therapeutics. A.B. Nixon is an employee/paid consultant for Tracon Pharma and Eli Lilly, and reports receiving commercial research grants from Seattle Genetics, MedPatco, Genentech, Tracon Pharma, Acceleron Pharma, Leadiant Biosciences, and Sanofi-Aventis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Alvarez Secord, H.I. Hurwitz, R.S. Mannel, K.S. Tewari, A.B. Nixon

Development of methodology: A. Alvarez Secord, M.D. Starr, H.I. Hurwitz, A.B. Nixon

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Alvarez Secord, Y. Liu, M.D. Starr, J.C. Brady, H.A. Lankes, R.S. Mannel, K.S. Tewari, D.M. O'Malley, H. Gray, J.N. Bakkum-Gamez, K. Fujiwara, M. Boente, M.J. Birrer, A.B. Nixon

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Alvarez Secord, K.B. Burdett, K. Owzar, D. Tritchler, A.B. Sibley, M.D. Starr, H.I. Hurwitz, K.S. Tewari, J.N. Bakkum-Gamez, M. Boente, W. Deng, R.A. Burger, M.J. Birrer, A.B. Nixon

Writing, review, and/or revision of the manuscript: A. Alvarez Secord, K.B. Burdett, K. Owzar, A.B. Sibley, Y. Liu, M.D. Starr, H.A. Lankes, H.I. Hurwitz, R.S. Mannel, K.S. Tewari, D.M. O'Malley, H. Gray, J.N. Bakkum-Gamez, K. Fujiwara, M. Boente, W. Deng, R.A. Burger, M.J. Birrer, A.B. Nixon

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Alvarez Secord, A.B. Sibley, M.D. Starr, H.A. Lankes, M. Boente, M.J. Birrer

Study supervision: A. Alvarez Secord, A.B. Nixon

The authors would like to thank the GOG Tissue Bank and administrative staff at NRG. The authors gratefully acknowledge the invaluable contributions of the patients, their families, and the research staff who participated in this study. The following NRG Oncology/Gynecologic Oncology Group member institutions participated in this study: CTSU, University of Oklahoma Health Sciences Center, Gynecologic Oncology Network/Brody School of Medicine, Ohio State University Comprehensive Cancer Center, University of California Medical Center at Irvine-Orange Campus, Fred Hutchinson Cancer Research Center, Mayo Clinic, Abramson Cancer Center of the University of Pennsylvania, Saitama Medical University International Medical Center, Metro-Minnesota CCOP, University of North Carolina at Chapel Hill, Abington Memorial Hospital, Rush University Medical Center, University of Kentucky, Washington University School of Medicine, University of Alabama at Birmingham, Roswell Park Comprehensive Cancer Center, Walter Reed National Military Medical Center, Women's Cancer Center of Nevada, Indiana University Hospital/Melvin and Bren Simon Cancer Center, University of Iowa Hospitals and Clinics, Memorial Sloan Kettering Cancer Center, Cleveland Clinic Foundation, Seoul National University Hospital, Fox Chase Cancer Center, Duke University Medical Center, University of Mississippi Medical Center, University of Chicago, University of Colorado Cancer Center – Anschutz Cancer Pavilion, University of California at Los Angeles Health System, Yale University, The Hospital of Central Connecticut, Northwestern University, Cooper Hospital University Medical Center, Women and Infants Hospital, Mount Sinai School of Medicine, University of New Mexico, University of Hawaii, Case Western Reserve University, Cancer Research for the Ozarks NCORP, Moffitt Cancer Center and Research Institute, University of Texas – Galveston, University of Pittsburgh Cancer Institute, University of Virginia, University of Minnesota Medical Center-Fairview, Wake Forest University Health Sciences, Stony Brook University Medical Center, Saint Vincent Hospital, Wayne State University/Karmanos Cancer Institute, University of Massachusetts Memorial Health Care, Georgia Center for Oncology Research and Education (CORE), State University of New York Downstate Medical Center, MD Anderson Cancer Center, University of Wisconsin Hospital and Clinics, Northern Indiana Cancer Research Consortium, Penn State Milton S. Hershey Medical Center, Fletcher Allen Health Care, Gynecologic Oncology of West Michigan PLLC, Virginia Commonwealth University, University of Cincinnati, Carle Cancer Center, Michigan Cancer Research Consortium Community Clinical Oncology Program, Tufts-New England Medical Center, Scott and White Memorial Hospital, Cancer Research Consortium of West Michigan NCORP, Central Illinois CCOP, Delaware/Christiana Care CCOP, Northern New Jersey CCOP, Virginia Mason CCOP, Tacoma General Hospital, Wisconsin NCI Community Oncology Research Program, New York University Medical Center, Colorado Cancer Research Program NCORP, Saint Louis-Cape Girardeau CCOP, Aurora Women's Pavilion of Aurora West Allis Medical Center, University of Illinois, Evanston CCOP-NorthShore University Health System, Kalamazoo CCOP, Missouri Valley Cancer Consortium CCOP, William Beaumont Hospital, University of Texas Southwestern Medical Center, Kansas City CCOP, Upstate Carolina CCOP, Dayton Clinical Oncology Program, Mainline Health CCOP, Meharry Medical College Minority Based CCOP, Heartland Cancer Research CCOP, and Wichita CCOP. This research was supported by NIH/NCI R21 5R21CA185730; U10 CA027469 (GOG Administrative Office), CA37517 (GOG Statistical Office), CA114793 (GOG Tissue Bank), U10CA180822 (NRG Oncology SDMC), UG1CA189867 (NCORP), U10CA180868 (NRG Oncology Operations), CA196067 (NRG Biospecimen Bank-Columbus), P01CA142538, and U24 CA114793; Foundation for Women's Cancer, Florence and Marshall Schwid Ovarian Cancer Research Grant; The American Association of Obstetricians and Gynecologists Foundation; Duke Gynecologic Oncology Philanthropic Funding; and the support of the NRG Oncology, including the legacy Gynecologic Oncology Group.

The costs of publication of this article were defrayed, in part, by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.