Analytical validity of DecisionDx-SCC, a gene expression profile test to identify risk of metastasis in cutaneous squamous cell carcinoma (SCC) patients.

BACKGROUND: To improve identification of patients with cutaneous squamous cell carcinoma (SCC) at high risk for metastatic disease, the DecisionDx-SCC assay, a prognostic 40-gene expression profile (40-GEP) test, was developed and validated. The 40-GEP assay utilizes RT-PCR gene expression analysis on primary tumor biopsy tissue to evaluate the expression of 34 signature gene targets and 6 normalization genes. The test provides classifications of low risk (Class 1), moderate risk (Class 2A), and high risk (Class 2B) of metastasis within 3 years of diagnosis. The primary objective of this study was to validate the analytical performance of the 40-gene expression signature. METHODS: The repeatability and reproducibility of the 40-GEP test was evaluated by performance of inter-assay, intra-assay, and inter-operator precision experiments along with monitoring the reliability of sample and reagent stability for class call concordance. The technical performance of clinical orders from September 2020 through July 2021 for the 40-GEP test was assessed. RESULTS: Patient hematoxylin and eosin (H&E) stained slides were reviewed by a board-certified pathologist to assess minimum acceptable tumor content. Class specific controls (Class 1 and Class 2B) were evaluated with Levey-Jennings analysis and demonstrated consistent and reproducible results. Inter-assay, inter-operator and intra-assay concordance were all ≥90%, with short-term and long-term RNA stability also meeting minimum concordance requirements. Of the 2586 orders received, 93.5% remained eligible for testing, with 97.1% of all tested samples demonstrating actionable class call results. CONCLUSION: DecisionDx-SCC demonstrates a high degree of analytical precision, yielding high concordance rates across multiple performance experiments, along with exhibiting robust technical reliability on clinical samples.

Analytical Validation and Performance of a 7-Gene Next-Generation Sequencing Panel in Uveal Melanoma.

INTRODUCTION: Gene expression profiling (GEP) is widely used for prognostication in patients with uveal melanoma (UM). Because biopsy tissue is limited, it is critical to obtain as much genomic information as possible from each sample. Combined application of both GEP and next-generation sequencing (NGS) allows for analysis of RNA and DNA from a single biopsy sample, offers additional prognostic information, and can potentially inform therapy selection. This study evaluated the analytical performance of a targeted custom NGS panel for mutational profiling of 7 genes commonly mutated in UM. METHODS: One hundred five primary UM tumors were analyzed, including 37 formalin-fixed paraffin-embedded (FFPE) and 68 fine-needle aspiration biopsy specimens. Sequencing was performed on the Ion GeneStudio S5 platform to an average read depth of >500X per region of interest. RESULTS: The 7-gene panel achieved a positive percent agreement of 100% for detection of both single-nucleotide variants and insertions/deletions, with a technical positive predictive value of 98.8% and 100%, respectively. Intra-assay and inter-assay concordance studies confirmed the assay's reproducibility and repeatability. DISCUSSION/CONCLUSION: The 7-gene panel is a robust, highly accurate NGS test that can be successfully performed, along with GEP, from a single small-gauge needle biopsy sample or FFPE specimen.

Gene Expression Profiling in Uveal Melanoma: Five-Year Prospective Outcomes and Meta-Analysis.

INTRODUCTION: The prognostic 15-gene expression profile (15-GEP) test for uveal melanoma (UM) predicts metastatic risk based on primary tumor biology. Here we report outcomes from a prospective registry of 15-GEP-tested patients, and a meta-analysis with published cohorts. OBJECTIVES: Management and 5-year clinical outcomes following 15-GEP testing were evaluated. METHODS: Eighty-nine patients with 15-GEP results were prospectively enrolled at four centers. Physician-recommended management plans were collected, and clinical outcomes tracked every 6 months. RESULTS: Eighty percent of Class 1 (low-risk) patients underwent low-intensity management; all Class 2 (high-risk) patients underwent high-intensity management (p < 0.0001). Median follow-up for event-free patients was 4.9 years. Five Class 1 (10%) and 23 Class 2 (58%) tumors metastasized (p < 0.0001). Five-year Class 1 and 2 metastasis-free survival rates were 90% (81-100%) and 41% (27-62%; p < 0.0001), and melanoma-specific survival rates were 94% (87-100%) and 63% (49-82%; p = 0.0007). Class 2 was the only independent predictor of metastasis and was associated with increased risk for metastasis and mortality by meta-analysis. CONCLUSIONS: UM patient management is guided by 15-GEP testing. Class 2 patients were managed more intensely, in accordance with an observed metastatic rate of >50%; Class 1 patients were safely spared intensive surveillance, resulting in appropriate utilization of healthcare resources.

Analytic validity of DecisionDx-Melanoma, a gene expression profile test for determining metastatic risk in melanoma patients.

BACKGROUND: The DecisionDx-Melanoma test provides prognostic information for patients with cutaneous melanoma (CM). Using formalin-fixed paraffin-embedded primary tumor tissue, the RT-PCR-based test classifies patients into a low- (Class 1) or high-risk (Class 2) category for recurrence based on expression of 31 genes. The current study was designed to assess the analytical validity of this test. METHODS: Inter-assay, inter-instrument, and inter-operator studies were performed to evaluate reliability of the 31-gene expression test results, sample stability and reagent stability. From March 2013 through June 2016, the gene expression test was performed on 8244 CM tumors. De-identified data from Pathology Reports were used to assess technical success. RESULTS: Robust sample and reagent stability was observed. Inter-assay concordance on 168 specimens run on 2 consecutive days was 99% and matched probability scores were significantly correlated (R2 = 0.96). Inter-instrument concordance was 95%, and probability scores had a correlation R2 of 0.99 (p < 0.001). From 8244 CM specimens submitted since 2013, 85% (7023) fulfilled pre-specified tumor content parameters. In these samples with sufficient tumor requirements, the technical success of the test was 98%. CONCLUSION: DecisionDx-Melanoma is a robust gene expression profile test that demonstrates strong reproducibility between experiments and has high technical reliability on clinical samples.

Gene expression profiling in uveal melanoma: technical reliability and correlation of molecular class with pathologic characteristics.

BACKGROUND: A 15-gene expression profile test has been clinically validated and is widely utilized in newly diagnosed uveal melanoma (UM) patients to assess metastatic potential of the tumor. As most patients are treated with eye-sparing radiotherapy, there is limited tumor tissue available for testing, and technical reliability and success of prognostic testing are critical. This study assessed the analytical performance of the 15-gene expression test for UM and the correlation of molecular class with pathologic characteristics. METHODS: Inter-assay, intra-assay, inter-instrument/operator, and inter-site experiments were conducted, and concordance of the 15-gene expression profile test results and associated discriminant scores for matched tumor samples were evaluated. Technical success was determined from de-identified clinical reports from January 2010 - May 2016. Pathologic characteristics of enucleated tumors were correlated with molecular class results. RESULTS: Inter-assay concordance on 16 samples run on 3 consecutive days was 100%, and matched discriminant scores were strongly correlated (R2 = 0.9944). Inter-assay concordance of 46 samples assayed within a one year period was 100%, with an R2 value of 0.9747 for the discriminant scores. Intra-assay concordance of 12 samples run concurrently in duplicates was 100%; discriminant score correlation yielded an R2 of 0.9934. Concordance between two sites assessing the same tumors was 100% with an R2 of 0.9818 between discriminant scores. Inter-operator/instrument concordance was 96% for Class 1/2 calls and 90% for Class 1A/1B calls, and the discriminant scores had a correlation R2 of 0.9636. Technical success was 96.3% on 5516 samples tested since 2010. Increased largest basal diameter and thickness were significantly associated with Class 1B and Class 2 vs. Class 1A signatures. CONCLUSIONS: These results show that the 15-gene expression profile test for UM has robust, reproducible performance characteristics. The technical success rate during clinical testing remains as high as first reported during validation. As molecular testing becomes more prevalent for supporting precision medicine efforts, high technical success and reliability are key characteristics when testing such limited and precious samples. The performance of the 15-gene expression profile test in this study should provide confidence to physicians who use the test's molecular classification to inform patient management decisions.

Identification of high-risk cutaneous melanoma tumors is improved when combining the online American Joint Committee on Cancer Individualized Melanoma Patient Outcome Prediction Tool with a 31-gene expression profile-based classification.

BACKGROUND: A significant proportion of patients with American Joint Committee on Cancer (AJCC)-defined early-stage cutaneous melanoma have disease recurrence and die. A 31-gene expression profile (GEP) that accurately assesses metastatic risk associated with primary cutaneous melanomas has been described. OBJECTIVE: We sought to compare accuracy of the GEP in combination with risk determined using the web-based AJCC Individualized Melanoma Patient Outcome Prediction Tool. METHODS: GEP results from 205 stage I/II cutaneous melanomas with sufficient clinical data for prognostication using the AJCC tool were classified as low (class 1) or high (class 2) risk. Two 5-year overall survival cutoffs (AJCC 79% and 68%), reflecting survival for patients with stage IIA or IIB disease, respectively, were assigned for binary AJCC risk. RESULTS: Cox univariate analysis revealed significant risk classification of distant metastasis-free and overall survival (hazard ratio range 3.2-9.4, P < .001) for both tools. In all, 43 (21%) cases had discordant GEP and AJCC classification (using 79% cutoff). Eleven of 13 (85%) deaths in that group were predicted as high risk by GEP but low risk by AJCC. LIMITATIONS: Specimens reflect tertiary care center referrals; more effective therapies have been approved for clinical use after accrual. CONCLUSIONS: The GEP provides valuable prognostic information and improves identification of high-risk melanomas when used together with the AJCC online prediction tool.

Validation of a Gene Expression Test for Mesothelioma Prognosis in Formalin-Fixed Paraffin-Embedded Tissues.

A molecular test performed using fresh-frozen tissue was proposed for use in the prognosis of patients with pleural mesothelioma. The accuracy of the test and its properties was assessed under Clinical Laboratory Improvement Amendments-approved guidelines using FFPE tissue from an independent multicenter patient cohort. Concordance studies were performed using matched frozen and FFPE mesothelioma samples. The prognostic value of the test was evaluated in an independent validation cohort of 73 mesothelioma patients who underwent surgical resection. FFPE-based classification demonstrated overall high concordance (83%) with the matched frozen specimens, on removal of cases with low confidence scores, showing sensitivity and specificity in predicting type B classification (poor outcome) of 43% and 98%, respectively. Concordance between research and clinical methods increased to 87% on removal of low confidence cases. Median survival times in the validation cohort were 18 and 7 months in type A and type B cases, respectively (P = 0.002). Multivariate classification adding pathologic staging information to the gene expression score resulted in significant stratification of risk groups. The median survival times were 52 and 14 months in the low-risk (class 1) and intermediate-risk (class 2) groups, respectively. The prognostic molecular test for mesothelioma can be performed on FFPE tissues to predict survival, and can provide an orthogonal tool, in combination with established pathologic parameters, for risk evaluation.

Epigenetic reprogramming and aberrant expression of PRAME are associated with increased metastatic risk in Class 1 and Class 2 uveal melanomas.

BACKGROUND: We previously identified PRAME as a biomarker for metastatic risk in Class 1 uveal melanomas. In this study, we sought to define a threshold value for positive PRAME expression (PRAME+) in a large dataset, identify factors associated with PRAME expression, evaluate the prognostic value of PRAME in Class 2 uveal melanomas, and determine whether PRAME expression is associated with aberrant hypomethylation of the PRAME promoter. RESULTS: Among 678 samples analyzed by qPCR, 498 (73.5%) were PRAME- and 180 (26.5%) were PRAME+. Class 1 tumors were more likely to be PRAME-, whereas Class 2 tumors were more likely to be PRAME+ (P < 0.0001). PRAME expression was associated with shorter time to metastasis and melanoma specific mortality in Class 2 tumors (P = 0.01 and P = 0.02, respectively). In Class 1 tumors, PRAME expression was directly associated with SF3B1 mutations (P < 0.0001) and inversely associated with EIF1AX mutations (P = 0.004). PRAME expression was strongly associated with hypomethylation at 12 CpG sites near the PRAME promoter. MATERIALS AND METHODS: Analyses included PRAME mRNA expression, Class 1 versus Class 2 status, chromosomal copy number, mutation status of BAP1, EIF1AX, GNA11, GNAQ and SF3B1, and genomic DNA methylation status. Analyses were performed on 555 de-identified samples from Castle Biosciences, 123 samples from our center, and 80 samples from the TCGA. CONCLUSIONS: PRAME is aberrantly hypomethylated and activated in Class 1 and Class 2 uveal melanomas and is associated with increased metastatic risk in both classes. Since PRAME has been successfully targeted for immunotherapy, it may prove to be a companion prognostic biomarker.

Clinical Performance and Management Outcomes with the DecisionDx-UM Gene Expression Profile Test in a Prospective Multicenter Study.

Uveal melanoma management is challenging due to its metastatic propensity. DecisionDx-UM is a prospectively validated molecular test that interrogates primary tumor biology to provide objective information about metastatic potential that can be used in determining appropriate patient care. To evaluate the continued clinical validity and utility of DecisionDx-UM, beginning March 2010, 70 patients were enrolled in a prospective, multicenter, IRB-approved study to document patient management differences and clinical outcomes associated with low-risk Class 1 and high-risk Class 2 results indicated by DecisionDx-UM testing. Thirty-seven patients in the prospective study were Class 1 and 33 were Class 2. Class 1 patients had 100% 3-year metastasis-free survival compared to 63% for Class 2 (log rank test p = 0.003) with 27.3 median follow-up months in this interim analysis. Class 2 patients received significantly higher-intensity monitoring and more oncology/clinical trial referrals compared to Class 1 patients (Fisher's exact test p = 2.1 × 10(-13) and p = 0.04, resp.). The results of this study provide additional, prospective evidence in an independent cohort of patients that Class 1 and Class 2 patients are managed according to the differential metastatic risk indicated by DecisionDx-UM. The trial is registered with Clinical Application of DecisionDx-UM Gene Expression Assay Results (NCT02376920).

Clinical impact of a 31-gene expression profile test for cutaneous melanoma in 156 prospectively and consecutively tested patients.

OBJECTIVE: DecisionDx-Melanoma * is a 31-gene expression profile test that predicts the risk of metastasis in patients with primary cutaneous melanoma (CM). This study was designed to ascertain clinical management changes determined by the test outcome, which classifies CM patients being at low (Class 1) or high (Class 2) risk for recurrence. RESEARCH DESIGN AND METHODS: Medical charts were reviewed from 156 CM patients from six institutions (three dermatology and three surgical oncology practices) who were consecutively tested between May 2013 and December 2015. Clinical management data that were compiled and compared before and after receipt of the 31-gene expression test result included frequency of physical exams, frequency and modality of imaging, and referrals to surgical and medical oncologists. RESULTS: Forty-two percent of patients were Stage I, 47% were Stage II and 8% were Stage III. Overall, 95 patients (61%) were Class 1 and 61 (39%) were Class 2. Documented changes in management were observed in 82 (53%) patients, with the majority of Class 2 patients (77%) undergoing management changes compared to 37% of Class 1 patients (p < 0.0001 by Fisher's exact test). The majority (77/82, 94%) of these changes were concordant with the risk indicated by the test result (p < 0.0001 by Fisher's exact test), with increased management intensity for Class 2 patients and reduced management intensity for Class 1 patients. CONCLUSIONS: Molecular risk classification by gene expression profiling has clinical impact and influences physicians to direct clinical management of CM patients. The vast majority of the changes implemented after the receipt of test results were reflective of the low or high recurrence risk associated with the patient's molecular classification. Because follow-up data was not collected for this patient cohort, the study is limited for the assessment of the impact of gene expression profile based management changes on healthcare resource utilization and patient outcome.

Gene expression profiling for molecular staging of cutaneous melanoma in patients undergoing sentinel lymph node biopsy.

BACKGROUND: A gene expression profile (GEP) test able to accurately identify risk of metastasis for patients with cutaneous melanoma has been clinically validated. OBJECTIVE: We aimed for assessment of the prognostic accuracy of GEP and sentinel lymph node biopsy (SLNB) tests, independently and in combination, in a multicenter cohort of 217 patients. METHODS: Reverse transcription polymerase chain reaction (RT-PCR) was performed to assess the expression of 31 genes from primary melanoma tumors, and SLNB outcome was determined from clinical data. Prognostic accuracy of each test was determined using Kaplan-Meier and Cox regression analysis of disease-free, distant metastasis-free, and overall survivals. RESULTS: GEP outcome was a more significant and better predictor of each end point in univariate and multivariate regression analysis, compared with SLNB (P < .0001 for all). In combination with SLNB, GEP improved prognostication. For patients with a GEP high-risk outcome and a negative SLNB result, Kaplan-Meier 5-year disease-free, distant metastasis-free, and overall survivals were 35%, 49%, and 54%, respectively. LIMITATIONS: Within the SLNB-negative cohort of patients, overall risk of metastatic events was higher (∼30%) than commonly found in the general population of patients with melanoma. CONCLUSIONS: In this study cohort, GEP was an objective tool that accurately predicted metastatic risk in SLNB-eligible patients.

Intended use for a neoadjuvant chemoradiation response prediction test for locally advanced esophageal adenocarcinoma: a survey analysis of thoracic surgeons in the US.

OBJECTIVE: A multi-analyte immunohistochemistry (IHC) based test (DecisionDx-EC * *DecisionDx-EC, Castle Biosciences, Incorporated, Friendswood, TX. Hereafter referred to as the "IHC-based test". ) was developed and proved able to accurately predict response to neoadjuvant chemoradiotherapy (neoCTRT) in esophageal adenocarcinoma (EC) patients with locoregional disease. A survey was conducted with surgeons attending the 2013 Society of Thoracic Surgeons Annual Conference to assess their intention to change standard of care patient treatment based on predictive results obtained from this test. RESEARCH DESIGN AND METHODS: Fifty-seven thoracic surgeons were provided a questionnaire regarding the current clinical management of locoregional EC patients, and their intention to change management strategy if a patient is predicted by the multi-analyte IHC-based test to experience pathological complete response (pathCR) or extremely resistant to neoCTRT (exCTRT). RESULTS: Forty-four out of 46 enrolled respondents indicated that they administer neoCTRT followed by surgery to treat locally advanced EC. Fifteen (32%) respondents presently prescribing neoCTRT to EC patients acknowledged that a pathCR prediction provided by the multi-analyte IHC-based test would change their current regimens. Conversely, 28 surgeons (61%) would be willing to adjust their current strategy for patients predicted to be resistant to neoCTRT, significantly more than the case where patients are predicted to be pathCR (p = 0.01). Twenty-five percent (25%) of surgeons willing to adjust treatment in response to a patient being a potential exCTRT chose to remove CTRT from the strategy, and instead would choose surgery alone for these patients. CONCLUSIONS: Results from the current survey show that patients' response to neoCTRT predicted by the multi-analyte IHC-based test has a significant influence on the decision-making process in the clinic. Nearly twice as many surgeons stated an intention to change strategy with the knowledge that the patient is likely to have extreme resistance to planned treatments than when one is a potential responder. The current survey study is limited by its 'intended use' nature, therefore does not reflect physicians' real action in the clinic upon implementation of this test. However, survey results strongly suggest that use of the IHC-based predictive test is of great interest to physicians, and would likely contribute to more individualized treatment for patients with distinct sensitivity to neoCTRT.

Development of a prognostic genetic signature to predict the metastatic risk associated with cutaneous melanoma.

PURPOSE: The development of a genetic signature for the identification of high-risk cutaneous melanoma tumors would provide a valuable prognostic tool with value for stage I and II patients who represent a remarkably heterogeneous group with a 3% to 55% chance of disease progression and death 5 years from diagnosis. EXPERIMENTAL DESIGN: A prognostic 28-gene signature was identified by analysis of microarray expression data. Primary cutaneous melanoma tumor tissue was evaluated by RT-PCR for expression of the signature, and radial basis machine (RBM) modeling was performed to predict risk of metastasis. RESULTS: RBM analysis of cutaneous melanoma tumor gene expression reports low risk (class 1) or high risk (class 2) of metastasis. Metastatic risk was predicted with high accuracy in development (ROC = 0.93) and validation (ROC = 0.91) cohorts of primary cutaneous melanoma tumor tissue. Kaplan-Meier analysis indicated that the 5-year disease-free survival (DFS) rates in the development set were 100% and 38% for predicted classes 1 and 2 cases, respectively (P < 0.0001). DFS rates for the validation set were 97% and 31% for predicted classes 1 and 2 cases, respectively (P < 0.0001). Gene expression profile (GEP), American Joint Committee on Cancer stage, Breslow thickness, ulceration, and age were independent predictors of metastatic risk according to Cox regression analysis. CONCLUSIONS: The GEP signature accurately predicts metastasis risk in a multicenter cohort of primary cutaneous melanoma tumors. Preliminary Cox regression analysis indicates that the signature is an independent predictor of metastasis risk in the cohort presented.

Tumor associated seizures in glioblastomas are influenced by survival gene expression in a region-specific manner: a gene expression imaging study.

Tumor associated seizures (TAS) are common and cause significant morbidity. Both imaging and gene expression features play significant roles in determining TAS, with strong interactions between them. We describe gene expression imaging tools which allow mapping of brain regions where gene expression has significant influence on TAS, and apply these methods to study 77 patients who underwent surgical evaluation for supratentorial glioblastomas. Tumor size and location were measured from MRI scans. A 9-set gene expression profile predicting long-term survivors was obtained from RNA derived from formalin-fixed paraffin embedded tissue. A total of 32 patients (42%) experienced preoperative TAS. Tumor volume was smaller (31.1 vs. 58.8 cubic cm, p<0.001) and there was a trend toward median survival being higher (48.4 vs. 32.7 months, p=0.055) in patients with TAS. Although the expression of only OLIG2 was significantly lower in patients with TAS in a groupwise analysis, gene expression imaging analysis revealed regions with significantly lower expression of OLIG2 and RTN1 in patients with TAS. Gene expression imaging is a powerful technique that demonstrates that the influence of gene expression on TAS is highly region specific. Regional variability should be evaluated with any genomic or molecular markers of solid brain lesions.

Current clinical practice: differential management of uveal melanoma in the era of molecular tumor analyses.

OBJECTIVE: Assess current clinical practices for uveal melanoma (UM) and the impact of molecular prognostic testing on treatment decisions. DESIGN: Cross-sectional survey and sequential medical records review. PARTICIPANTS: Ophthalmologists who treat UM. METHODS: (A) Medical records review of all Medicare beneficiaries tested by UM gene expression profile in 2012, conducted under an institutional review board-approved protocol. (B) 109 ophthalmologists specializing in the treatment of UM were invited to participate in 24-question survey in 2012; 72 were invited to participate in a 23-question survey in 2014. MAIN OUTCOME MEASURES: Responses analyzed by descriptive statistics, frequency analyses (percentages, Tukey, histograms), and Fisher's exact test. Descriptive presentation of essay answers. RESULTS: The review of Medicare medical records included 191 evaluable patients, 88 (46%) with documented medical treatment actions or institutional policies related to surveillance plans. Of these 88, all gene expression profiling (GEP) Class 1 UM patients were treated with low-intensity surveillance. All GEP Class 2 UM patients were treated with high-intensity surveillance (P<0.0001 versus Class 1). There were 36 (19%) with information concerning referrals after initial diagnosis. Of these 36, all 23 Class 2 patients were referred to medical oncology; however, none of the 13 Class 1 patients were referred (P<0.0001 versus Class 1). Only Class 2 patients were recommended for adjunctive treatment regimens. 2012 survey: 50 respondents with an annual median of 35 new UM patients. The majority of respondents (82%) performed molecular analysis of UM tumors after fine needle biopsy (FNAB); median: 15 FNAB per year; 2014 survey: 35 respondents with an annual median of 30 new UM patients. The majority offered molecular analyses of UM tumor samples to most patients. Patients with low metastatic risk (disomy 3 or GEP Class 1) were generally assigned to less frequent (every 6 or 12 months) and less intensive clinical visits. Patients with high metastatic risk (monosomy 3 or GEP Class 2) were assigned to more frequent surveillance with hepatic imaging and liver function testing every 3-6 months. High-risk patients were considered more suitable for adjuvant treatment protocols. CONCLUSION: The majority of ophthalmologists treating UM have adopted molecular diagnostic tests for the purpose of designing risk-appropriate treatment strategies.

A gene signature to determine metastatic behavior in thymomas.

PURPOSE: Thymoma represents one of the rarest of all malignancies. Stage and completeness of resection have been used to ascertain postoperative therapeutic strategies albeit with limited prognostic accuracy. A molecular classifier would be useful to improve the assessment of metastatic behaviour and optimize patient management. METHODS: qRT-PCR assay for 23 genes (19 test and four reference genes) was performed on multi-institutional archival primary thymomas (n = 36). Gene expression levels were used to compute a signature, classifying tumors into classes 1 and 2, corresponding to low or high likelihood for metastases. The signature was validated in an independent multi-institutional cohort of patients (n = 75). RESULTS: A nine-gene signature that can predict metastatic behavior of thymomas was developed and validated. Using radial basis machine modeling in the training set, 5-year and 10-year metastasis-free survival rates were 77% and 26% for predicted low (class 1) and high (class 2) risk of metastasis (P = 0.0047, log-rank), respectively. For the validation set, 5-year metastasis-free survival rates were 97% and 30% for predicted low- and high-risk patients (P = 0.0004, log-rank), respectively. The 5-year metastasis-free survival rates for the validation set were 49% and 41% for Masaoka stages I/II and III/IV (P = 0.0537, log-rank), respectively. In univariate and multivariate Cox models evaluating common prognostic factors for thymoma metastasis, the nine-gene signature was the only independent indicator of metastases (P = 0.036). CONCLUSION: A nine-gene signature was established and validated which predicts the likelihood of metastasis more accurately than traditional staging. This further underscores the biologic determinants of the clinical course of thymoma and may improve patient management.