Analysis of Biomarkers and Association With Clinical Outcomes in Patients With Differentiated Thyroid Cancer: Subanalysis of the Sorafenib Phase III DECISION Trial

Histologic subtypes of thyroid cancer are defined by microscopic evaluations of tumor growth patterns, cellular origin, and prognosis. Differentiated thyroid cancer (DTC) includes follicular (FTC) and papillary (PTC) subtypes, which together account for >90% of all thyroid cancer cases (1). The treatment of DTC depends on the clinical and pathologic features of the tumor but generally includes surgery followed by treatment with radioactive iodine (2–4). Although treatment is generally effective, 6–7 patients/million population develop distant metastases (5–7), and two thirds of patients will become refractory to radioactive iodine therapy (8). For patients with locally recurrent or metastatic progressive radioactive iodine–refractory DTC, treatment options were limited before the approval of systemic therapy with sorafenib, an oral multikinase inhibitor (9). Sorafenib has both direct and indirect effects on tumor growth, targeting tumor cell proliferation and angiogenesis through inhibition of a number of kinases, such as VEGFR 1, 2, and 3 (also known as FLT1, KDR, and FLT4, respectively), platelet-derived growth factor receptors (PDGFRA/B), ret proto oncogene (RET, including RET/PTC), cKIT (KIT), and RAF family members (including BRAF V600E; ref. 10). The efficacy of sorafenib in the treatment of metastatic thyroid cancer has been demonstrated in phase II and III clinical trials. In a meta-analysis of seven clinical trials (five phase II and two retrospective), 81% of patients had either partial response or stable disease after sorafenib therapy, with an overall median progression-free survival (PFS) of 18 months (11). In the phase III randomized, double-blind, placebo-controlled DECISION trial, sorafenib 400 mg twice daily significantly improved PFS compared with placebo in patients with locally advanced or metastatic radioactive iodine–refractory DTC [median PFS 10.8 vs. 5.8 months, respectively, HR, 0.59; 95% confidence interval (CI), 0.45–0.76; P < 0.0001; ref. 12].

A number of genetic abnormalities have been implicated in the etiology of DTC, which vary among DTC subtypes. In PTC, genetic abnormalities have been identified in approximately 70% of patients, the most common being BRAF and RAS mutations and RET/PTC rearrangements (The Cancer Genome Atlas). In FTC, abnormalities have been identified in 70%–75% of patients, including RAS mutations and PPARG rearrangements. However, there are limited data regarding the prognostic or predictive value of these and other biomarkers associated with DTC; identification of prognostic biomarkers for disease outcome and predictive biomarkers for treatment benefit with sorafenib in DTC would assist in clinical decision-making and the management of radioactive iodine–refractory DTC.

This investigation is a retrospective, exploratory biomarker analysis conducted to identify prognostic biomarkers for disease outcome and predictive biomarkers for sorafenib treatment benefit in patients with radioactive iodine–refractory DTC from the DECISION trial. We evaluated the relationships between clinical outcomes and pretreatment levels of candidate plasma protein markers, serum thyroglobulin levels at baseline and during treatment, and tumor mutations.

The DECISION study design (NCT00984282, EudraCT: 2009-012007-25) has been previously reported in detail (12, 13). In brief, patients with locally advanced or metastatic radioactive iodine–refractory DTC (papillary, follicular, or poorly differentiated) progressing within the preceding 14 months (by RECIST) and ≥1 measurable lesion were randomly assigned (1:1) to either sorafenib 400 mg orally twice daily or matching placebo. Patients with radioactive iodine–refractory DTC were defined as: (i) having at least one target lesion that did not have iodine uptake; or (ii) having tumors that had iodine uptake and (a) progressed after one radioactive iodine treatment within the past 16 months, (b) progressed after two radioactive iodine treatments within 16 months of each other, the last radioactive iodine treatment administered >16 months ago, or (c) received cumulative radioactive iodine activity ≥ 22.3 GBq (≥600 mCi).

Patients who had received prior therapy for thyroid cancer (targeted therapy, thalidomide, or chemotherapy) were excluded. Study medication was administered continuously until disease progression, unacceptable toxicity, noncompliance, or withdrawal of consent. Patients randomly assigned to placebo were allowed to receive sorafenib upon disease progression. The primary endpoint was PFS assessed every 8 weeks by central independent blinded review using RECIST v1.0, from the date of randomization to the date of radiological progression or death. Radiological progression in bone was modified to include: (i) radiological appearance of new lesions; (ii) ≥20% increase in the sum of the longest diameter of all target lesions, which may include bone lesions if they have measurable soft tissue components; and (iii) bone lesions that require external radiation.

The study was conducted in accordance with International Conference on Harmonization of good clinical practice guidelines, local laws, regulations, and organizations, as well as with the Declaration of Helsinki. Ethics committees/institutional review boards at all participating centers approved the study. All patients provided written informed consent before participating in the study.

Plasma proteins were tested in samples collected at baseline (during screening or before treatment on day 1 of cycle 1). Blood was collected with EDTA as the anticoagulant, stored at −70°C (storage at −20°C was acceptable if no −70°C freezer was available), and shipped on dry ice to a central laboratory after a maximum of 6 weeks. The analysis included 15 candidate mechanistic plasma biomarkers with known or hypothesized relevance to the mechanism of action of sorafenib and/or to thyroid carcinoma pathogenesis or outcomes including: chemokines and soluble domains of receptors implicated in cancer angiogenesis, tumor migration, and metastasis, several of which are direct targets of sorafenib [soluble VEGFRs and ligands (sVEGFR-2, sVEGFR-3, VEGFA, VEGFC, VEGFA₁₂₁, and VEGFA₁₆₅), CXCL12, and transforming growth factor-β1 (TGF-β1)]; cytokines that modulate the development and growth of normal and neoplastic thyroid cells (IL6, IL8, CXCL8, and PDGF-AA); the MET signaling pathway ligand [hepatocyte growth factor (HGF)], elevated levels of which have been detected in thyroid cancer and appear to be associated with increased angiogenesis, tumor growth, and invasive potential (14–16), and which showed a trend toward influencing sorafenib benefit in patients with hepatocellular carcinoma in a phase III trial (17); a hypoxia-induced cell surface protein frequently overexpressed in human tumors [carbonic anhydrase IX (CAIX)], whose hypoxia-induced HIF-1–mediated expression is increased by RET overexpression (18), and which was associated with pN stage and vascular invasion in papillary thyroid carcinoma (19); and proteins involved in tissue invasion, metastasis, and epithelial–mesenchymal transition [the E-cadherin tumor suppressor and vimentin, the latter of which has increased expression levels in papillary thyroid carcinoma with metastasis to the lymph nodes (ref. 20), and expression correlating with presence of BRAFV600E (ref. 21)]. Plasma proteins were tested using commercially available Luminex bead-based multiplex assays and ELISA kits (Supplementary Table S1). Samples were assayed within 30 months of collection. Serum samples for thyroglobulin testing were collected at screening, day 1 of every cycle for the first nine cycles, and day 1 of every other cycle for subsequent cycles, and measured at a central laboratory using IMMULITE 2000 Thyroglobulin (Siemens Diagnostics).

Tumor sample collection and genetic analyses were optional. An archival, formalin-fixed, paraffin-embedded tumor biopsy sample was collected at screening from each patient who consented and used for tumor mutation analysis. Because archival tumor samples were limited, mutational hotspot analysis of candidate genes was conducted. Mutations were examined using the Sequenom OncoCarta 1.0 panel, which tests 238 mutational hotspots in 19 genes (Supplementary Table S1), including genes of specific interest to DTC (BRAF, NRAS, HRAS, and KRAS).

The biomarker analysis population included all patients in the intent-to-treat population with ≥1 biomarker result available at baseline. The study sample size was not determined on the basis of biomarker analyses, but for evaluation of the clinical endpoints. Biomarker analyses were performed using all available data, regardless of the sample size of the biomarker population (12). Biomarkers were analyzed for both prognostic as well as predictive (ability to predict benefit from sorafenib) value. Clinical efficacy variables analyzed included PFS, overall survival (OS), and disease control rate (DCR, defined as complete or partial response or stable disease). Circulating biomarkers (plasma proteins and serum thyroglobulin) were analyzed as both continuous variables and dichotomized or binned variables; optimized dichotomization cut-off points were determined for each protein using a maximum χ² method testing all possible cut-off points between the 25th and 75th percentiles.

Correlative analyses of biomarkers versus outcome were performed using both univariate and multivariate analyses; multivariate models included other clinical variables [age, Eastern Cooperative Oncology Group (ECOG) performance status, sex, race, and histology; for some analyses of thyroglobulin, thyroid-stimulating hormone (TSH) was also included, as specified in tables] and, in models where patients receiving sorafenib were added, treatment group as a covariate. Cox proportional hazards models, as well as Kaplan–Meier analyses (for the continuous outcome variables of PFS and OS) and logistic regression analyses (for the binary outcome variable of DCR), were performed to assess correlations between biomarkers and outcome or treatment effect. For analyses of predictive value, Cox models included a treatment–biomarker interaction term. Analyses of biomarkers for prognostic value were performed among placebo patients alone and, in some cases, in “treatment-independent” analyses, in which the sorafenib and placebo arms were combined.

Because of the skewed values for untransformed biomarker values, all statistical analyses for plasma proteins were conducted on a log₂ scale. P values for the plasma protein analyses were adjusted for multiplicity using a Bonferroni-type correction based on the number of biomarkers tested, and both unadjusted (for purposes of hypothesis generation) and adjusted P values were reported. For plasma protein analyses, statistical significance was defined as P_adj < 0.05 for analyses of VEGFA and HGF (biomarkers with known relevance to sorafenib based on previous studies; refs. 17, 22) and as P_adj < 0.20 for the hypothesis generating analyses of the remaining 13 plasma protein biomarkers. Because serum thyroglobulin analyses were performed as post hoc analyses (not using predefined thresholds for significance, as the plasma protein analyses were), P values from the thyroglobulin analyses were not adjusted for multiplicity.

Multibiomarker approaches were also used to attempt to identify signatures of circulating biomarkers correlating with prognosis or sorafenib benefit using the bootstrap elastic net selection probability method (23). Serum thyroglobulin was additionally examined as a pharmacodynamic biomarker of response to sorafenib, including summarizing change in serum thyroglobulin over time and by response status (i.e., complete or partial response, stable disease, or progressive disease) using linear mixed-effects models.

Baseline plasma biomarker data were available for 395 patients (94.7% of the full randomized study population of 417) and serum thyroglobulin data for 403 patients (96.6%). Patient demographics and baseline clinical characteristics in the plasma protein subpopulation were similar to those in the overall study population, including histologic distribution, and were well balanced between the sorafenib and placebo arms (Table 1). Sorafenib antitumor activity in the plasma protein subpopulation was similar to that of the full study population, with similar HRs for PFS and OS, and similar response rates (Table 1).

Fifteen candidate mechanistic plasma biomarkers with known or hypothesized relevance to sorafenib's mechanism of action or to thyroid carcinoma outcome (Supplementary Table S1) were analyzed for both prognostic as well as predictive (ability to predict benefit from sorafenib) value. In the analyses of plasma proteins as potentially prognostic biomarkers, elevated baseline VEGFA was significantly and robustly associated with poor prognosis for PFS, OS, and DCR. For example, in the dichotomized analyses of VEGFA using the maximum χ² optimized cut-off point and adjusting for covariables, elevated VEGFA at baseline was significantly associated with poor PFS prognosis [combined sorafenib plus placebo analysis: HR = 1.82 (95% CI, 1.38–2.44), P_adj = 0.0007; placebo only: HR = 2.16 (95% CI, 1.47–3.22), P_adj = 0.003; Fig. 1A]. Elevated baseline VEGFA was also associated with poor OS prognosis [combined sorafenib plus placebo analysis: HR = 2.13 (95% CI, 1.37–3.36), P_adj = 0.013; placebo only: HR = 3.12 (95% CI, 1.55–6.85), P_adj = 0.034; Fig. 1B] in dichotomized analyses using the maximum χ² optimized cut-off point and adjusting for covariables [i.e., age, sex, race, ECOG performance status (PS), and histology]. This association between high baseline VEGFA and poor prognosis was also evident in other analyses of PFS and OS (i.e., models using VEGFA as a continuous variable) and with or without inclusion of covariates (age, sex, race, ECOG PS, and histology), as well as in analyses using DCR as the clinical endpoint (all summarized in Supplementary Tables S2 and S3).

Elevated baseline levels of vimentin were associated with poor OS prognosis in analyses of vimentin as a continuous variable [in the combined sorafenib and placebo population, with covariables (i.e., age, sex, race, ECOG PS, and histology) included: HR = 1.35 (95% CI, 1.05–1.79), P = 0.018, P_adj = 0.161]. Elevated baseline levels of VEGFC and TGF-β1 and low baseline levels of E-cadherin were associated with poor prognosis for DCR, with each biomarker meeting the predefined exploratory, multiplicity-adjusted α in multiple analyses of DCR [e.g., VEGFC as a dichotomized variable, combined sorafenib and placebo population, with covariates: OR = 0.17 (95% CI, 0.06–0.42), P_adj = 0.008; TGF-β1 as a continuous variable, combined sorafenib and placebo population, with covariates: OR = 0.63 (95% CI, 0.44–0.88), P_adj = 0.060; E-cadherin as a continuous variable, combined sorafenib and placebo population, without covariates: OR = 1.50 (95% CI, 1.15–1.98), P_adj = 0.028]. Summaries of all prognostic analyses of the plasma biomarkers (for PFS, OS, and DCR) are shown in Supplementary Tables S2 and S3.

In analyses of plasma proteins as potentially predictive biomarkers for sorafenib benefit, using either the maximum χ² methodology for biomarker dichotomization or analyzing the biomarkers as continuous variables, none of the tested plasma proteins significantly predicted sorafenib benefit after multiplicity adjustment. This finding was consistent in models either with or without inclusion of clinical covariates (i.e., age, ECOG PS, sex, race, and histology), and in analyses using PFS, OS, or DCR as the clinical outcome measure. A summary of all predictive analyses of the plasma biomarkers (for PFS, OS, and DCR) is shown in Supplementary Table S4 and examples of associated Kaplan–Meier curves in Supplementary Fig. S1.

Baseline mean thyroglobulin levels were highest in patients with Hürthle cell carcinoma (9,867 ng/mL) and with non-Hürthle cell follicular (8,350 ng/mL) and lowest in those with papillary thyroid cancer (2,938 ng/mL; Supplementary Table S5). Mean thyroglobulin levels at baseline were higher in patients with bone metastases (8,370 ng/mL; n = 108) than in those without (3,602 ng/mL; n = 300; P < 0.001), and were lower in patients with lung metastases (1,792 ng/mL; n = 69) than in those without (5,489 ng/mL; n = 339; P = 0.002). Baseline thyroglobulin levels did not differ significantly between the sorafenib and placebo groups (P = 0.529).

Elevated baseline levels of serum thyroglobulin were associated with poor prognosis for both PFS and DCR. For example, using thyroglobulin as a dichotomized variable with the optimized cut-off point of 1,021 ng/mL (equivalent to the 58th percentile), in the combined sorafenib and placebo population, and including covariables (age, sex, race, ECOG PS, and histology), a higher baseline thyroglobulin was associated with shorter PFS (HR = 2.03; 95% CI, 1.52–2.71; P < 0.0001; Fig. 2A). The optimal thyroglobulin cut-off identified in the PFS analyses ranged from 324 ng/mL (equivalent to the 46th percentile) to 2,882 ng/mL (equivalent to the 71st percentile), depending on the analysis. Similarly, in the analyses of DCR by baseline thyroglobulin levels, for example, the group with higher baseline thyroglobulin using the optimized thyroglobulin cut-off of 2,011 ng/mL (equivalent to the 68th percentile) had lower DCR than those with low thyroglobulin levels at baseline (OR = 0.32; 95% CI, 0.17–0.60; P = 0.01) in the combined sorafenib and placebo population. These associations were shown by models that included combined sorafenib plus placebo or placebo only, with and without clinical covariates (age, sex, race, ECOG PS, and histology), with and without TSH as a covariable, and for models with thyroglobulin as either a continuous or dichotomized variable (summarized in Supplementary Table S6).

Previous analysis of baseline thyroglobulin as a potentially predictive biomarker for sorafenib benefit in the DECISION trial showed that when thyroglobulin was dichotomized using the median, sorafenib significantly improved median PFS irrespective of high or low thyroglobulin levels at baseline (12). The extended analyses described herein support this conclusion. For example, dichotomization of thyroglobulin using the maximum χ² methodology in a model that included covariables showed PFS benefit of sorafenib compared with placebo in both the low and high thyroglobulin groups [low thyroglobulin: HR = 0.58 (95% CI, 0.43–0.79), P = 0.0006; high thyroglobulin: HR = 0.38 (95% CI, 0.23–0.64), P = 0.0002; Supplementary Fig. S2). In the search for an optimal baseline thyroglobulin cut-off point, no cut-off point could be identified that would improve predictive value for sorafenib benefit, regardless of whether thyroglobulin was analyzed as a categorical or continuous variable or with or without the inclusion of clinical covariates (age, sex, race, ECOG PS, and histology; summarized in Supplementary Table S7).

In analyses of serum thyroglobulin as a pharmacodynamic biomarker (tracking changes over time), previous analyses of DECISION showed that thyroglobulin levels decreased from baseline over time in patients receiving sorafenib but increased steadily from baseline in patients receiving placebo, and that among patients receiving sorafenib, thyroglobulin rose in patients with progressive disease, remained below baseline in patients with stable disease, and decreased further and remained low in patients with partial responses (12). Additional analyses of this data confirm that patients who were initially receiving placebo who crossed over to sorafenib treatment because of disease progression also experienced a decrease in serum thyroglobulin over time (Fig. 3). Significantly more patients in the sorafenib group achieved either a 30% or 50% decrease in serum thyroglobulin over time than patients in the placebo group (Fig. 3; Supplementary Table S8): 76.0% of patients in the sorafenib group and 13.6% of patients in the placebo group had a thyroglobulin decrease of ≥30% (P < 0.001), while 58.9% and 7.3%, respectively, achieved a decrease of ≥50% (P < 0.001). A decrease in serum thyroglobulin of ≥30% during treatment correlated with improved best response (Supplementary Table S8) and longer PFS (Fig. 4).

Baseline levels of plasma proteins and serum thyroglobulin were analyzed together to identify multimarker signatures that were either prognostic of disease outcome or predictive of sorafenib benefit. A combined VEGFA and thyroglobulin signature correlated with poor prognosis for PFS (in the combined sorafenib and placebo group, HR = 2.12, P < 0.00001; Fig. 2B). No multimarker signature was identified that correlated with prognosis for OS or had a predictive value for either PFS or OS.

The 256-patient tumor genetic population from DECISION was described in brief previously (12), and more details and analyses are presented herein. The most prevalent mutations identified were BRAF (in 30.1% of the genetic subpopulation, most of whom had papillary histology) and RAS (in 19.5%, most being NRAS mutations, although HRAS or KRAS mutations were also detected). Mutations in PDGFRA, RET (point mutation), EGFR, KIT, FGFR1, and AKT1 were also detected in 1–5 patients each (Table 2). However, all of these six mutations were paired with 1–2 other mutations, and in all but one case, the pairing was with either a BRAF or RAS mutation, suggesting that these less common mutations may not be the primary drivers of DTC. Fifteen patients had multiple (2–3) mutations.

Analysis of the prognostic value of tumor mutations in the placebo cohort (in models without clinical covariates) showed that BRAF mutation was associated with longer PFS than wild-type in the genetic subpopulation (HR = 0.51; 95% CI, 0.32–0.83; P = 0.006), and similar results were observed in the subset of patients with papillary carcinoma (HR = 0.60; 95% CI, 0.35–1.04; P = 0.066; Fig. 5). Conversely, RAS mutation was associated with shorter PFS compared with wild-type (HR = 1.8; 95% CI, 1.08–2.99; P = 0.022; Fig. 5) in models without clinical covariates. As previously reported, these mutations were not independently prognostic in multivariate models that included clinical factors (i.e., age, ECOG PS, sex, race, and histology; ref. 12).

The DECISION study was a double-blind, randomized phase III study that demonstrated that sorafenib significantly improved PFS compared with placebo in patients with progressive DTC refractory to radioactive iodine. The retrospective exploratory biomarker analyses of data from the DECISION study reported herein were conducted to identify prognostic biomarkers for disease outcome and predictive biomarkers for sorafenib treatment benefit in patients with radioactive iodine–refractory DTC. Biomarker evaluations in DECISION included baseline levels of 15 candidate plasma protein biomarkers, both baseline and on-treatment serum thyroglobulin, and tumor mutations.

In the analyses of plasma proteins as potentially prognostic biomarkers, elevated baseline VEGFA was significantly and robustly associated with poor prognosis for PFS, OS, and DCR. This finding was consistent in a wide range of models, including in the presence of clinical covariables, demonstrating that VEGFA is an independently prognostic factor for DTC. While many analyses were performed showing this relationship between VEGFA and prognosis (summarized in Supplementary Table S3), most identified an optimal VEGFA cut-off point of approximately 9.0 pg/mL, corresponding to approximately the 40th percentile of VEGFA levels, to separate the poor prognosis subgroup from the better prognosis group (Fig. 1A and B, and data not shown). VEGFA was not, however, predictive for sorafenib benefit in DTC (i.e., both the low-VEGFA and high-VEGFA groups showed benefit from treatment) and therefore is not useful in identifying which patients should receive sorafenib. High baseline VEGFA was shown to correlate with poor prognosis in phase III trials of sorafenib in different tumor types, including hepatocellular carcinoma and renal cell carcinoma (17, 22). Elevated plasma levels of VEGFA have also been identified as prognostic but not predictive in trials of other antiangiogenic agents, specifically bevacizumab (24). A prospective study of 76 patients with DTC showed that VEGFA levels were significantly higher in patients with extrathyroidal extensions (319 ± 49 vs. 149 ± 32 pg/mL, P = 0.005), lymph node metastases (318 ± 52 vs. 180 ± 35 pg/mL, P = 0.027), and distant metastases (512 ± 104 vs. 194 ± 29 pg/mL, P = 0.0008). VEGFA was also an independent risk factor for distant metastases (OR, 1.04; 95% CI, 1.01–1.07; P = 0.008; ref. 25).

Elevated baseline serum thyroglobulin was found in this study to be independently associated with poor prognosis for PFS and DCR in DTC, although not with OS. The optimal thyroglobulin cut-off point identified in the various PFS analyses performed varied widely, ranging from 324 ng/mL (equivalent to the 46th percentile) to 2,882 ng/mL (equivalent to the 71st percentile; Fig. 2; Supplementary Table S3; and data not shown). All thyroglobulin subgroups benefited from sorafenib treatment for PFS, suggesting that baseline serum thyroglobulin levels do not predict benefit from sorafenib in DTC.

In multimarker models of circulating plasma proteins, the VEGFA/thyroglobulin signature correlated with poor prognosis for PFS, and the combined signature provided a marginally improved PFS prognosis separation over either VEGFA or thyroglobulin alone. Initial changes in serum thyroglobulin over time may be a pharmacodynamic biomarker of tumor response or progression, as thyroglobulin decreased from baseline over time in patients receiving sorafenib (including those patients who were randomized to placebo but then crossed over to open-label sorafenib treatment after progressing), but increased in patients receiving placebo. A decrease in serum thyroglobulin of ≥30% during treatment correlated with improved best response and longer PFS, which supports the previous finding from DECISION showing that changes in thyroglobulin over time reflected best response status (12). Of note, because many patients with stable disease (and even a few with progressive disease) experienced a rapid decrease in thyroglobulin during treatment similar in magnitude to the decrease observed in patients with partial response, individual serum thyroglobulin measurements may not accurately reflect tumor progression, and thus radiologic criteria should remain the standard for determination of treatment duration in DTC.

Findings from our exploratory biomarker analyses are consistent with those reported for the oral multikinase inhibitor lenvatinib, which was also approved for radioactive iodine–refractory DTC (26). In a biomarker assessment of data from a phase II trial in patients with radioactive iodine–DTC, there was a significant association between lower thyroglobulin levels relative to baseline and increased PFS; this association showed significance starting on the fifth day of the 28-day treatment cycle and was maintained through several timepoints into the 15th day of the 28-day cycle of continued therapy (27). Furthermore, analysis of pharmacodynamic biomarkers using data from a phase III trial of lenvatinib in patients with radioactive iodine–DTC found that decreased thyroglobulin levels were associated with tumor shrinkage and overall objective response rate (28). While the thyroglobulin analyses in both DECISION and the phase III lenvatinib study suggest that initial changes in thyroglobulin predict PFS and DCR, these analyses do not address the use of thyroglobulin as a marker of duration of clinical benefit. Therefore, based on available data, changes in thyroglobulin levels should not be used as a determinant of response or progression, but may serve a role in suggesting the timing of tumor surveillance via imaging.

Several other associations between plasma protein biomarkers and prognosis were noted in this analysis. Elevated baseline levels of VEGFC and TGF-β1 and low baseline levels of E-cadherin were associated with poor prognosis for DCR, and elevated baseline levels of vimentin were associated with poor prognosis for OS.

As expected, the most common tumor mutations identified were BRAF and RAS, although less common mutations were also identified in PDGFRA, RET, EGFR, KIT, FGFR1, and AKT1. The presence of a BRAF mutation in this DTC study was associated with better prognosis for PFS, while RAS mutations were associated with worse PFS prognosis, although these mutations were not independently prognostic in multivariate models that included clinical factors. Predictive analyses of the relationship between BRAF and RAS mutations and PFS (previously published data; ref. 12) and OS (data not shown) showed no significant difference in sorafenib benefit based on mutational status. Our finding that BRAF and RAS mutations have prognostic value in this patient population is consistent with other reports (29, 30). A recent biomarker analysis of the SELECT trial in patients with DTC showed that the PFS benefit of lenvatinib versus placebo was maintained regardless of serum levels of BRAF and RAS mutations; neither BRAF nor RAS mutations were predictive or prognostic biomarkers in patients with DTC. However, wild-type BRAF was prognostic for poorer PFS in metastatic progressive DTC in the placebo group (31). One possible explanation for these results is that the BRAF-mutated group may be enriched in PTC, while the RAS-mutated group may be enriched for FTC and poorly DTC (PDTC). Patients with PTC may have a better prognosis than patients with FTC or PDTC.

Limitations of these analyses include their exploratory and hypothesis-generating nature, and that this DECISION trial was powered for evaluation of clinical endpoints only and not biomarker analyses. However, the demographics and baseline clinical characteristics of the subpopulations analyzed were representative of the full DECISION study population, and, at least for the circulating biomarkers, 95%–97% of the full study population was included in the analyses. In addition, analyses of OS by biomarker status are confounded by the cross-over of patients from placebo treatment to sorafenib treatment upon disease progression.

In conclusion, in these exploratory biomarker analyses of data from the phase III DECISION study of sorafenib in patients with locally advanced or metastatic radioactive iodine–DTC, elevated baseline levels of plasma VEGFA, serum thyroglobulin, and a combined thyroglobulin/VEGFA signature were independently associated with poor prognosis. Thyroglobulin changes over time may be a pharmacodynamic biomarker of tumor response or progression. BRAF and RAS mutations were prognostic of PFS in univariate analyses but not independently prognostic in multivariate models. Confirmatory investigations would be needed to verify the clinical utility of the identified prognostic biomarkers.

M.S. Brose reports receiving commercial research grants from Bayer, Eisai, Exelixis, and Sanofi-Genzyme. M. Schlumbeger is an employee/paid consultant for Bayer, Eisai, IPSEN-Exelixis, and Sanofi-Genzyme. M. Jeffers, C. Kappeler, and G. Meinhardt are employees/paid consultants for Bayer. C.E.A. Peña is an employee/paid consultant for and holds ownership interest (including patents) in Bayer.

Conception and design: M.S. Brose, M. Schlumbeger, M. Jeffers, C. Kappeler, C.E.A. Peña

Development of methodology: M.S. Brose, M. Jeffers, C. Kappeler

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.S. Brose, M. Schlumbeger, M. Jeffers, C.E.A. Peña

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Schlumbeger, C. Kappeler, G. Meinhardt, C.E.A. Peña

Writing, review, and/or revision of the manuscript: M.S. Brose, M. Schlumbeger, M. Jeffers, C. Kappeler, G. Meinhardt, C.E.A. Peña

Study supervision: M.S. Brose, M. Schlumbeger, C. Kappeler, G. Meinhardt

The authors thank Scott Wilhelm for scientific advisory support. Scientific writing support was provided by C4 MedSolutions, LLC, a CHC Group company, and funded by Bayer HealthCare Pharmaceuticals. This study was funded by Bayer HealthCare Pharmaceuticals and Onyx, a wholly owned subsidiary of Amgen.

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.