The New England Journal of Medicine Copyright © 2002 by the Massachusetts Medical Society VOLUME 347 D ECEMBER 19, 2002 NUMBER 25 N Engl J Med, Vol. 347, No. 25 · December 19, 2002 · www.nejm.org · 1999 A GENE-EXPRESSION SIGNATURE AS A PREDICTOR OF SURVIVALIN BREAST CANCER M ARC J. VAN DE V IJVER , M.D., P H .D., Y UDONG D. H E , P H .D., L AURA J. VAN ’ T V EER , P H .D., H ONGYUE D AI , P H .D., A UGUSTINUS A.M. H ART , M.S C ., D ORIEN W. V OSKUIL , P H .D., G EORGE J. S CHREIBER , M.S C ., J OHANNES L. P ETERSE , M.D., C HRIS R OBERTS , P H .D., M ATTHEW J. M ARTON , P H .D., M ARK P ARRISH , D OUWE A TSMA , A NKE W ITTEVEEN , A NNUSKA G LAS , P H .D., L EONIE D ELAHAYE , T ONY VAN DER V ELDE , H ARRY B ARTELINK , M.D., P H .D., S JOERD R ODENHUIS , M.D., P H .D., E MIEL T. R UTGERS , M.D., P H .D., S TEPHEN H. F RIEND , M.D., P H .D., AND R ENÉ B ERNARDS , P H .D. A BSTRACT Background A more accurate means of prognos-tication in breast cancer will improve the selection ofpatients for adjuvant systemic therapy. Methods Using microarray analysis to evaluate ourpreviously established 70-gene prognosis profile, weclassified a series of 295 consecutive patients with pri-mary breast carcinomas as having a gene-expressionsignature associated with either a poor prognosis ora good prognosis. All patients had stage I or II breastcancer and were younger than 53 years old; 151 hadlymph-node–negative disease, and 144 had lymph-node–positive disease. We evaluated the predictivepower of the prognosis profile using univariable andmultivariable statistical analyses. Results Among the 295 patients, 180 had a poor-prognosis signature and 115 had a good-prognosis sig-nature, and the mean (±SE) overall 10-year survivalrates were 54.6±4.4 percent and 94.5±2.6 percent, re-spectively. At 10 years, the probability of remainingfree of distant metastases was 50.6±4.5 percent in thegroup with a poor-prognosis signature and 85.2±4.3percent in the group with a good-prognosis signature.The estimated hazard ratio for distant metastases inthe group with a poor-prognosis signature, as com-pared with the group with the good-prognosis signa-ture, was 5.1 (95 percent confidence interval, 2.9 to9.0; P<0.001). This ratio remained significant when thegroups were analyzed according to lymph-node sta-tus. Multivariable Cox regression analysis showed thatthe prognosis profile was a strong independent factorin predicting disease outcome. Conclusions The gene-expression profile we stud-ied is a more powerful predictor of the outcome of dis-ease in young patients with breast cancer than stand-ard systems based on clinical and histologic criteria.(N Engl J Med 2002;347:1999-2009.) Copyright © 2002 Massachusetts Medical Society. From the Divisions of Diagnostic Oncology (M.J.V., L.J.V., D.W.V., J.L.P.,D.A., A.W., A.G., L.D.), Radiotherapy (A.A.M.H., H.B.), Medical Oncology(S.R.), Biometrics (T.V.), Surgical Oncology (E.T.R.), and Molecular Car-cinogenesis (R.B.), Netherlands Cancer Institute, Amsterdam; the Centerfor Biomedical Genetics, Amsterdam (R.B.); and Rosetta Inpharmatics, Kirk-land, Wash. (Y.D.H., H.D., G.J.S., C.R., M.J.M., M.P., S.H.F.). Address re-print requests to Dr. Bernards at the Division of Molecular Carcinogenesis,Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam, theNetherlands, or at [email protected]. van de Vijver, He, and van ’t Veer contributed equally to this article. DJUVANT systemic therapy substantiallyimproves disease-free and overall survival inboth premenopausal and postmenopausalwomen up to the age of 70 years withlymph-node–negative or lymph-node–positive breastcancer. 1,2 It is generally agreed that patients with poorprognostic features benefit the most from adjuvanttherapy. 3,4 The main prognostic factors in breast can-cer are age, tumor size, status of axillary lymph nodes,histologic type of the tumor, pathological grade, andhormone-receptor status. A large number of otherfactors have been investigated for their potential to pre-dict the outcome of disease, but in general, they haveonly limited predictive power. 5 Using complementary DNA (cDNA) microarraysto analyze breast-cancer tissue, Perou et al. identifiedtumors with distinct patterns of gene expression thatthey termed “basal type” and “luminal type.” 6 Thesesubgroups differ with respect to the outcome of dis-ease in patients with locally advanced breast cancer. 7 In addition, microarray analysis has been used to dis-tinguish cancers associated with BRCA1 or BRCA2 mutations 8,9 and to determine estrogen-receptor sta-tus 6,9,10 and lymph-node status. 11,12 Using inkjet-synthesized oligonucleotide microar-rays, we recently identified a gene-expression profileACopyright © 2002 Massachusetts Medical Society. All rights reserved. Downloaded from www.nejm.org at Stanford University on November 14, 2005 .2000 · N Engl J Med, Vol. 347, No. 25 · December 19, 2002 · www.nejm.org The New England Journal of Medicine that is associated with prognosis in patients with breastcancer. 9 We analyzed only tumors that were less than5 cm in diameter from lymph-node–negative patientswho were younger than 55 years of age. We found thata classification system based on 70 genes outperformedall clinical variables in predicting the likelihood of dis-tant metastases within five years. We estimated that theodds ratio for metastases among tumors with a genesignature associated with a poor prognosis, as com-pared with those having a signature associated with agood prognosis, was approximately 15 using a cross-validation procedure. Even though these results wereencouraging, a limitation of the study was that the re-sults were derived from and evaluated in two groupsof patients selected on the basis of outcome: distantmetastases had developed in one group within fiveyears, and the other group remained disease-free for atleast five years. Therefore, to provide a more accurateestimate of the risks of metastases associated with thetwo gene-expression signatures and to substantiate thatthe gene-expression profile of breast cancer is a clin-ically meaningful tool, we studied a cohort of 295young patients with breast cancer, some of whom werelymph-node–negative and some of whom were lymph-node–positive. METHODS Selection of Patients Tumors from a series of 295 consecutive women with breast can-cer were selected from the