Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

A population-based study of tumor gene expression and risk of breast cancer death among lymph node-negative patients

A population-based study of tumor gene expression and risk of breast cancer death among lymph... Introduction The Oncotype DX assay was recently reported to Results After adjusting for tumor size and grade, the predict risk for distant recurrence among a clinical trial Recurrence Score was associated with risk of breast cancer population of tamoxifen-treated patients with lymph node- death in ER-positive, tamoxifen-treated and -untreated patients negative, estrogen receptor (ER)-positive breast cancer. To (P = 0.003 and P = 0.03, respectively). At 10 years, the risks for confirm and extend these findings, we evaluated the breast cancer death in ER-positive, tamoxifen-treated patients performance of this 21-gene assay among node-negative were 2.8% (95% confidence interval [CI] 1.7–3.9%), 10.7% patients from a community hospital setting. (95% CI 6.3–14.9%), and 15.5% (95% CI 7.6–22.8%) for those in the low, intermediate and high risk Recurrence Score Methods A case-control study was conducted among 4,964 groups, respectively. They were 6.2% (95% CI 4.5–7.9%), Kaiser Permanente patients diagnosed with node-negative 17.8% (95% CI 11.8–23.3%), and 19.9% (95% CI 14.2– invasive breast cancer from 1985 to 1994 and not treated with 25.2%) for ER-positive patients not treated with tamoxifen. In adjuvant chemotherapy. Cases (n = 220) were patients who both the tamoxifen-treated and -untreated groups, approximately died from breast cancer. Controls (n = 570) were breast cancer 50% of patients had low risk Recurrence Score values. patients who were individually matched to cases with respect to age, race, adjuvant tamoxifen, medical facility and diagnosis year, and were alive at the date of death of their matched case. Using an RT-PCR assay, archived tumor tissues were analyzed for expression levels of 16 cancer-related and five reference Conclusion In this large, population-based study of lymph node- genes, and a summary risk score (the Recurrence Score) was negative patients not treated with chemotherapy, the calculated for each patient. Conditional logistic regression Recurrence Score was strongly associated with risk of breast methods were used to estimate the association between risk of cancer death among ER-positive, tamoxifen-treated and - breast cancer death and Recurrence Score. untreated patients. and free from distant metastasis at 10 years [2]. Chemother- apy in this group of patients, especially among patients with Introduction estrogen receptor (ER)-positive disease treated with adjuvant Approximately 65% of women currently diagnosed with inva- hormonal therapy, offers only a modest improvement in 10- sive breast cancer have lymph node-negative disease at diag- year survival [2,3]. Nonetheless, according to current guide- nosis [1], and 85% of these women are expected to be alive lines the majority of node-negative breast cancer patients CI = confidence interval; ER = estrogen receptor; NSABP = National Surgical Adjuvant Breast and Bowel Project; PgR = progesterone receptor; RT-PCR = reverse transcriptase polymerase chain reaction. Page 1 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. should be considered for chemotherapy [4-6]. This is largely specified primary aim of the study was to assess the degree to because of our limited ability to identify those individual which the Recurrence Score would predict the risk of breast patients who are unlikely to benefit from such treatment. cancer-specific mortality among women with ER-positive, node-negative breast cancer treated with tamoxifen (that is, Multigene assays may provide information on patient progno- patients clinically similar to those included in the study con- sis and response to therapy that is superior and/or comple- ducted by Paik and coworkers [12]), either alone or in multivar- mentary to that from standard histopathological and iate analyses with tumor size and tumor grade. A pre-specified immunohistochemical techniques [7,8]. However, multiple secondary aim was to examine the extent to which the Recur- well conducted and controlled observational studies or clinical rence Score predicts risk of breast cancer-specific mortality trials in independent populations are needed to establish the among ER-positive, node-negative patients not treated with clinical usefulness of these assays [9,10]. tamoxifen. A pre-specified exploratory aim was to examine the association between risk of breast cancer death and Recur- Using a multistep approach, a 21-gene assay (Oncotype DX) rence Score among node-negative, ER-negative patients. was recently developed for use in paraffin-embedded tumor tissue to predict risk for distant recurrence or death in lymph Materials and methods node-negative breast cancer patients [11]. Approximately 250 Study population and design genes, selected from the published literature, genomic data- We conducted a case-control study nested within a cohort of bases, pathway analysis, and from microarray-based gene breast cancer patients. The Northern California Kaiser Perma- expression profiling experiments performed using fresh frozen nente tumor registry, a contributor to the Surveillance, Epide- tissue, were considered as candidates. The final gene list (16 miology, and End Results (SEER) program of cancer cancer-related and five reference genes) and summary score registries, was used to identify all female health plan members (Recurrence Score) algorithm for this assay (Figure 1) were who were diagnosed with lymph node-negative invasive breast developed by analyzing the results of three independent pre- cancer from 1985 to 1994. Northern California Kaiser Perma- liminary breast cancer studies (that is, training sets) con- nente is a nonprofit, integrated health services delivery organ- ducted in a total of 447 patients [12]. ization that provides care for over 3 million members at 14 Northern California Kaiser hospitals and 23 outpatient clinics. Two subsequent studies were conducted to evaluate the per- The Kaiser Permanente membership is racially and ethnically formance of the Recurrence Score in predicting distant recur- diverse and is demographically similar to the general popula- rence-free survival in lymph node-negative breast cancer tion of northern California, although it tends to under-represent patients not treated with chemotherapy. Among 668 the extremes of the socioeconomic spectrum [14-16]. The tamoxifen-treated patients in the National Surgical Adjuvant study was approved by the Kaiser Permanente Institutional Breast and Bowel Project (NSABP) B-14 clinical trial [12], the Review Board. Recurrence Score was strongly predictive of risk for distant recurrence, both alone and in multivariate analyses that also Breast cancer patients were eligible if their nodal status was included patient age, tumor size, tumor grade, and ER and documented to be negative by pathologic assessment, they HER2 status. For patients categorized as low, intermediate, or were diagnosed before age 75 years, and their initial disease high risk based on Recurrence Score, the Kaplan-Meier esti- was not treated with chemotherapy. Patients were excluded mates for the rates of distant recurrence at 10 years (and the for the following reasons: inflammatory or bilateral breast can- associated 95% confidence intervals [CIs]) were 6.8% (95% cer or evidence of metastasis (including lymph nodes) at initial CI 4.0–9.6%), 14.3% (95% CI 8.3–20.3%), and 30.5% (95% diagnosis; prior invasive cancer (breast or other) at diagnosis; CI 23.6–37.4%), respectively. A second study [13] evaluated or unknown/unconfirmed treatment with tamoxifen. the assay among 149 patients not treated with adjuvant ther- apy (hormonal or chemotherapy) at the MD Anderson Cancer Using the Kaiser Permanente tumor registry and electronic Center; it found no clear association between the Recurrence membership databases, patients were followed until death Score and risk for distant recurrence. Rates of distant recur- due to breast cancer, death from another cause, bilateral rence at 10 years were 18% (95% CI 7–30%), 38% (95% CI breast cancer, termination of membership, or December 15–61%), and 28% (95% CI 13–32%) for the low, intermedi- 2002, whichever came first. ate, and high risk categories, respectively. Cases were patients whose first event was death from breast To evaluate the Oncotype DX assay in a third independent cancer. At each case's death, up to three controls were ran- study population of lymph node-negative breast cancer domly selected from patients alive and under follow up (that is, patients not treated with chemotherapy, we conducted a case- incidence density sampling) [17]. Controls were individually control study among a large population of women who were matched to their case with respect to age (within 1 year), race diagnosed and treated at 14 hospitals and clinics within the (non-Hispanic white, Hispanic, Black, Asian), calendar year of Northern California Kaiser Permanente health plan. A pre- diagnosis (exact year), Kaiser Permanente pathology depart- Page 2 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 Table 1 Selected characteristics of the study population Characteristic Cases (n = 220) Controls (n = 570) Matched variables Age at diagnosis (years) <40 17(8%) 23(4%) 40–49 42 (19%) 127 (22%) 50–59 64 (29%) 152 (27%) 60–74 97 (44%) 268 (47%) Race/ethnicity White, non-Hispanic 172 (78%) 457 (80%) White, Hispanic 7 (3%) 13 (2%) Black 20(9%) 47(8%) Asian 21 (10%) 53 (9%) Surgery year 1985–1989 146 (66%) 385 (68%) 1990–1994 74 (34%) 185 (32%) Adjuvant tamoxifen No 156 (71%) 393 (69%) Yes 64 (29%) 177 (31%) Unmatched variables ER status from RT-PCR Positive 168 (76%) 514 (90%) Negative 52 (24%) 56 (10%) Tumor size (cm) ≤1 49 (22%) 193 (34%) 1.1–2 93 (42%) 255 (45%) 2.1–4 72 (33%) 114 (20%) >4 6 (3%) 8 (1%) Tumor grade (differentiation) Well 25 (11%) 175 (31%) Moderate 92 (42%) 261 (46%) Poor 103 (47%) 134 (23%) Recurrence Score Low risk (<18) 57 (26%) 322 (56%) Intermediate risk (18–30) 54 (25%) 108 (19%) High risk (≥31) 109 (50%) 140 (25%) 1 2 Cutoff points based on RT-PCR values: ≤6.5 and >6.5 units. Bloom-Richardson grading criteria, pathologist 1. ER, estrogen receptor; RT-PCR, reverse transcription polymerase chain reaction. ment of origin, and treatment of index breast cancer with grade allowed us to examine whether these measures provide tamoxifen (yes, no). Note that matching on tamoxifen treatment information on risk of breast cancer death that is independent maximized our ability to conduct analyses stratified by of Recurrence Score. Matching requirements were relaxed to tamoxifen treatment. Not matching on tumor size or tumor find up to three eligible controls per case. Matching criteria Page 3 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. were relaxed in the following order: age was relaxed to < 50 Blinding and batching of pathology and laboratory versus ≥ 50 years; calendar year of diagnosis was relaxed to procedures ± 1 year; calendar year of diagnosis was relaxed to ± 2 years; All pathology and laboratory procedures (slide review, section- and pathology facility was dropped as a matching factor. ing of tumor blocks, macro-dissection, RT-PCR assays, and Recurrence Score calculations) were conducted blinded to The medical records of cases and controls were reviewed to the case-control status of patient specimens. In addition, all confirm the initial diagnosis, treatment and cause of death, and batches of pathology materials sent to Kaiser Permanente to obtain information on eligibility criteria. Pathologic tumor Regional Laboratory, NSABP Pathology, and Genomic Health, size was obtained from pathology reports. Information on ER Inc. included a mixture of cases and controls. and progesterone receptor (PgR) status of the index tumor Sample preparation (usually assessed by the ligand-binding assay) was also abstracted from the medical record. For eligible cases and controls, the hematoxylin and eosin stained slides from biopsies and/or surgeries performed at the Cases and controls were selected and eligibility was deter- time of the index diagnosis were reviewed using pre-specified mined before laboratory measurement of gene expression. Of criteria in order to identify the most representative block and to the 402 cases identified as potentially eligible by the tumor evaluate whether sufficient tumor tissue was present. Speci- registry, 269 were determined to be eligible by chart review. mens with no tumor or very little tumor (<5% of the area occu- The eligibility of 27 cases could not be determined because of pied by invasive cancer cells compared to the area occupied incomplete or unavailable medical records and were consid- by other epithelial elements, such as normal epithelium, fibro- ered lost to the study. Similarly, 722 of the 989 controls initially cystic change, or ductal or lobular carcinoma in situ) were matched to eligible cases were determined to be eligible by excluded from the study (n = 18). Specimens with regions on chart review and the eligibility of 29 could not be determined. the slide having prominent nontumor elements (such as Of those eligible by chart review, 31 cases and 91 controls smooth muscle, hemorrhage, fibrosis, hyperplastic epithelium, were lost because of missing tumor blocks. Another four and/or normal breast; but not ductal or lobular carcinoma in cases were lost because we were unable to find at least one situ or necrosis) where the nontumor elements were both suf- matched control. This left 234 cases and 631 controls availa- ble for pathology studies. Table 2 Distributions of tumor size and tumor grade in Recurrence Score categories for all 220 cases and 570 controls Variable (number of patients) Recurrence Score < 18 18–30 ≥30 Correlation (n = 379) (n = 162) (n = 249) (P value) Tumor size (cm) 0.25 (<0.0001) ≤1 (n = 242) 147 (61%) 51 (21%) 44 (18%) 1.1–2 (n = 348) 173 (50%) 65 (19%) 110 (32%) 2.1–4 (n = 186) 55 (30%) 44 (24%) 87 (47%) >4 (n = 14) 4 (29%) 2 (14%) 8 (57%) Tumor grade pathologist 1 0.54 (<0.0001) Well (n = 200) 146 (73%) 44 (22%) 10 (5%) Moderate (n = 353) 199 (56%) 90 (25%) 64 (18%) Poor (n = 237) 34 (14%) 28 (12%) 175 (74%) Tumor size and grade 0.42 (<0.0001) ≤2 cm and well or ≤1 cm and moderate 204 (72%) 60 (21%) 21 (7%) (n = 285) >2 cm and well, or 1.1–2 cm and 143 (41%) 63 (19%) 133 (40%) moderate, or ≤2 cm and poor (n = 330) >2 cm and moderate/poor (n = 175) 41 (23%) 39 (22%) 95 (54%) Spearman rank correlation ficiently localized to be amenable to macro-dissection and suf- were sent to NSABP Pathology for macro-dissection (n = 59). ficiently abundant (>50% of the overall tissue on the slide) Macro-dissection to obtain enriched tumor tissue was per- Page 4 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 formed using a safety blade cleaned with RNaseZAP (Ambion, was less than 2.5 Recurrence Score units. A similarly high Austin, TX, USA) by NSABP Pathology on six (10 µm) sec- concordance (Pearson's r = 0.86) was also observed tions of the region enriched in tumor tissue. For all other spec- between Recurrence Score results from core biopsies and imens, three (10 µm) sections were prepared by Kaiser resections among 49 patients with locally advanced breast Permanente Regional Laboratory, placed into a microcentri- cancer. fuge tube, and sent to Genomic Health, Inc. Recurrence Score cutoff points were pre-specified and classi- Tumor size, grade, and estrogen receptor status fied patients into low risk (Recurrence Score <18), intermedi- When not recorded on the pathology report (7% of reports), ate risk (Recurrence Score 18–30) and high risk (Recurrence tumor size was obtained from a pathology review of all hema- Score ≥31) categories. All methods used for measurement of toxylin and eosin stained slides from all surgeries at diagnosis. gene expression levels and for the calculation of Recurrence Tumor grade, based on review of an hematoxylin and eosin Score and gene group scores were identical to those used in slide from the most representative block, was assessed inde- the study conducted by Paik and coworkers [12]. pendently by two board certified (in anatomic and clinical Statistical analysis pathology), assistant professors in departments of pathology, using the modified Bloom-Richardson grading criteria [18]. ER Statistical analyses were conducted entirely by Kaiser Perma- status of the index tumor was unavailable in the 10- to 20-year- nente researchers following a pre-specified plan. As specified, old medical record for a significant proportion of patients results were generated for patient groups, based on ER status (16%), and therefore all tumors were classified as positive or and tamoxifen treatment. negative based on ER expression by RT-PCR (>6.5 and ≤6.5, respectively; cutoff points based on prior studies examining The Recurrence Score was examined as a continuous variable correlation of RT-PCR with immunohistochemistry) in 50-unit increments, for consistency and comparison with [11,13,19]. In our study population, the concordance of ER previously published Recurrence Score findings. In additional status from RT-PCR versus from medical chart information analyses, women were categorized into presumptive low, was moderate (kappa 0.49, 95% CI 0.41–0.56). All but seven intermediate, and high risk groups based on cutoff points (out of a total of 122) of the discordances were patients clas- (Recurrence Score < 18, 18–30 and ≥31, respectively), and sified as ER positive based on RT-PCR and ER negative when categorized according to Recurrence Score quartiles. based on information in the medical chart. When examining the relationships between risk of breast can- RT-PCR assay of gene expression and calculation of the cer death and the expression level of individual genes, the Recurrence Score expression levels were treated as continuous variables. When Gene expression analysis in fixed paraffin-embedded tumor examining gene groups, the proliferation gene group and the tissue was performed by Genomic Health, Inc., as described HER2 gene group scores were treated as continuous varia- previously [11,12]. Briefly, after RNA extraction and DNase I bles both with and without transformation based on a thresh- treatment, total RNA content was measured and the absence old value (see the Recurrence Score algorithm in Figure 1). of DNA contamination was verified. Reverse transcription was performed followed by quantitative RT-PCR reactions in 384 Tumor size was examined both as a continuous variable in 2 well plates using Applied Biosystems PRISM 7900 HT cm units (for consistency with the NSABP B-14 study [12]) instruments (Applied Biosystems, Foster City, CA, USA). and when categorized as ≤1 cm, 1.1–2 cm, 2.1–4 cm, and >4 cm. Tumor grade was examined as a categorical variable (well Expression of each gene was measured in triplicate wells, and differentiated, moderately differentiated, and poorly differenti- then normalized relative to a set of five reference genes (β- ated). The grading used was from the standardized re-review. actin, GAPDH, GUS, RPLPO, and TFRC). Reference-normal- We examined tumor grade separately for pathologist 1 and ized expression measurements range from 0 to 15, where a 1- pathologist 2. unit increase reflects approximately a twofold increase in RNA. Conditional logistic regression was used to calculate odds The potential impact of tumor heterogeneity on Recurrence ratios as estimates of the relative risks for breast cancer death Score results was examined in two small studies (unpublished associated with the Recurrence Score, or a component of the data). In a study of 20 patients and 60 blocks (two to five Recurrence Score (univariate analyses). In addition, we esti- blocks/patient), analysis of variance was used to examine the mated relative risks adjusted for tumor size and tumor grade variability of Recurrence Score between blocks from the same (multivariate analyses). Model parameter estimation was done patient (tissue sections did not undergo macro-dissection). by maximum likelihood, and 95% Wald's confidence limits The standard deviation in Recurrence Scores (as a continuous were calculated. Statistical significance was assessed via the value) between blocks was 3.0 Recurrence Score units. For likelihood ratio test [20]. In multivariate analyses, reported P 16 of the 20 patients, the standard deviation between blocks Page 5 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. Figure 1 values are for the addition of the given factor to a model includ- ing all other factors. For women with ER-positive tumors, analyses were performed separately for those treated and not treated with tamoxifen. Because ER status was not a matching factor but was associ- ated with our outcome, fewer than half of the ER-negative cases (n = 16) had matched controls who were also ER-neg- ative. We therefore conducted conditional logistic regression analyses among those patients not treated with tamoxifen and generated relative risk estimates for ER-negative patients using terms for interaction with ER status. Finally, we con- ducted analyses of the full study population, using conditional logistic regression and interaction terms for ER status and tamoxifen therapy to obtain relative risk estimates for the differ- ent pre-specified patient groups characterized by these fac- tors. However, because results were not materially different, we present only those results from analyses conducted within patient groups. Methods developed by Langholz and Borgan [21] for nested case-control data were used to estimate the absolute risk of breast cancer death at 10 years and the corresponding 95% CIs. Estimates were calculated for subgroups of ER-positive patients, stratified on tamoxifen treatment and on Recurrence Score, tumor size, and tumor grade. In addition, 10-year risks for breast cancer death were calculated for subgroups of patients, when cross-classified by tumor size and grade and by Recurrence Score and tumor size and grade. The original Langholz and Borgan estimator assumed simple random sam- pling of controls from the set of known eligibles. In this study, we could not confirm eligibility without a review of the medical records, and therefore we sampled potentially eligible controls until up to three were confirmed for study inclusion. Thus, a slight modification to the original absolute risk estimator was necessary in order to reflect our sampling scheme (Additional file 1). T Tw wen enty ty-one -one gen gene e pa pane nel a l an nd c d ca alc lculation ulation of rec of recu urren rrence ce sc score ore. (a) The For comparison with the absolute risk estimates in the Kaiser final gene list (16 cancer-related and five reference genes) and sum- population, the NSABP provided us with the Kaplan-Meier mary score (Recurrence Score) algorithm for this assay were devel- estimates for the probability of breast cancer death at 10 years oped by analyzing the results of three independent preliminary breast among the 668 NSABP B-14 [12] tamoxifen-treated patients cancer studies (that is, training sets) with a total of 447 patients [11]. The Recurrence Score, on a scale from 0 to 100, is derived from the with node-negative breast cancer (previously published reference-normalized expression measurements in four steps. In the results included estimates for distant recurrence, relapse-free first step the expression for each gene is normalized relative to the survival, and overall survival). expression of the five reference genes (b-actin, GAPDH, GUS, RPLPO, and TFRC). Reference-normalized expression measurements range from 0 to 15, where a 1-unit increase reflects approximately a twofold Results increase in RNA. (b) In the second step the HER2 Group Score, the Characteristics of cases and controls ER Group Score, the Proliferation Group Score, and the Invasion Among 4,964 potentially eligible lymph node-negative invasive Group Score are calculated from individual gene expression measure- breast cancer patients, we identified 234 eligible cases and ments. (c) In the third step the Recurrence Score unscaled (RS ) is selected 631 controls with available tumor blocks. After loss calculated using coefficients that were pre-defined based on regres- sion analysis of gene expression and recurrence in the three training of 7.9% of specimens because of insufficient tumor and 1% studies (Providence, Rush, and NSABP B-20 [12]). A plus sign indi- because of failed RT-PCR analysis, a total of 220 cases and cates increased expression is associated with increased recurrence 570 controls were available for statistical analyses. A total of risk. A minus sign indicates that increased expression is associated 142 cases (64.6%) had three controls each, 66 cases with decreased recurrence risk. Page 6 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 Table 3 Relative risks associated with recurrence score among ER-positive patients, stratified by treatment with tamoxifen Score Cases Controls RR (95% CI) P value Tamoxifen treated (55 cases and 150 controls) Recurrence Score Continuous 55 (100%) 150 (100%) 7.6 (2.6–21.9) <0.0001 Pre-specified categories Low risk (<18) 16 (29%) 95 (63%) 1.0 reference Intermediate risk (18–30) 22 (40%) 35 (23%) 4.0 (1.8–8.8) High risk (≥31) 17 (31%) 20 (13%) 6.2 (2.4–15.8) Quartiles 0.0004 1st (0–8.25) 5 (9%) 38 (25%) 1.0 reference 2nd (8.26–14.43) 6 (11%) 37 (25%) 1.0 (0.3–3.7) 3rd (14.44–20.95) 15 (27%) 38 (25%) 2.9 (1.0–8.9) 4th (20.96–78.78) 53 (53%) 37 (25%) 5.8 (2.0–16.6) Tamoxifen untreated (110 cases and 251 controls) Recurrence Score Continuous 110 (100%) 251 (100%) 4.1 (2.1–8.1) <0.0001 Pre-specified categories <0.0001 Low risk (<18) 40 (36%) 160 (64%) 1.0 reference Intermediate risk (18–30) 32 (29%) 47 (19%) 2.7 (1.5–5.0) High risk (≥31) 38 (35%) 44 (18%) 3.3 (1.8–5.9) Quartiles <0.0001 1st (0–7.53) 11 (10%) 63 (25%) 1.0 reference 2nd (7.54–14.25) 19 (17%) 64 (25%) 1.5 (0.7–3.4) 3rd (14.26–21.86) 23 (28%) 62 (25%) 2.1 (0.9–4.9) 4th (21.87–85.82) 57 (52%) 62 (25%) 5.3 (2.5-1.3) 1 2 Conditional logistic regression models include Recurrence Score variables only. Relative risks for Recurrence Score calculated with regard to an increment of 50 units (chosen to be consistent with and comparable with previous studies). Quartiles based on distribution among controls treated with tamoxifen. Quartiles based on distribution among controls not treated with tamoxifen. CI, confidence interval; ER, estrogen receptor; RR, relative risk. (30.0%) had two controls each, and 12 cases (5.5%) had one non-Hispanic white. Approximately two-thirds of patients were control each. The distribution of factors available from the diagnosed with their breast cancer during the first 5 years of tumor registry (age, tumor size, race, diagnosis year, ER sta- the accrual period (1985–1989). Overall, one-third of patients tus, and tamoxifen treatment) among the evaluable cases and were treated with adjuvant tamoxifen. Before 1989, 11% of controls was generally similar to the distribution of these fac- patients were treated with tamoxifen; from 1989 to 1994, tors among the 239 potentially eligible cases and controls 58% of patients were treated with tamoxifen. Among those who were lost to the study. However, both lost cases and lost treated with tamoxifen, the median duration was 4 years; controls were slightly more likely than evaluable patients to be approximately 10% had a year of treatment or less. Compared younger, to be not white, or to have smaller tumors. with controls, cases more commonly had tumors that were ER negative, larger, or more poorly differentiated. Cases were also Breast cancer deaths occurred a median of 4.9 years after more likely to have tumors with higher Recurrence Score val- diagnosis. Among the cases and controls, the median tumor ues. Approximately 50% of patients had Recurrence Score size was 1.5 cm (range 0.2–7.0 cm). Cases and controls were values in the low risk category (that is, Recurrence Score similar with respect to matching factors including age, race, <18). diagnosis year, and treatment with tamoxifen (Table 1). Nearly three-quarters of the study population was aged 50 years or For pre-specified analyses stratified by ER status and older at diagnosis (median age 58 years) and about 80% were tamoxifen therapy, there were 55 cases and 150 matched Page 7 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. Table 4 Relative risks of breast cancer death associated with tumor size, grade, and recurrence score among ER-positive patients, stratified by treatment with tamoxifen Score Cases Controls RR 95% CI P value Tamoxifen treated (55 cases and 150 controls) Multivariate analyses without Recurrence Score Tumor size (cm) 0.009 Continuous 55 (100%) 150 (100%) 2.6 (1.2–5.5) Grade (pathologist 1) 0.007 Well 6 (11%) 50 (33%) 1.0 reference Moderate 28 (51%) 69 (46%) 2.8 (1.1–7.4) Poor 21 (38%) 31 (21%) 4.5 (1.6–12.3) Multivariate analyses with Recurrence Score Tumor size (cm) 0.013 Continuous 55 (100%) 150 (100%) 2.5 (1.2–5.4) Grade (pathologist 1) 0.126 Well 6 (11%) 50 (33%) 1.0 reference Moderate 28 (51%) 69 (46%) 2.4 (0.9–6.4) Poor 21 (38%) 31 (21%) 2.8 (0.9–9.3) Recurrence Score 0.003 Continuous 55 (100%) 150 (100%) 5.3 (1.6–17.2) Tamoxifen untreated (110 cases and 251 controls) Multivariate analyses without Recurrence Score Tumor size (cm) 0.035 Continuous 110 (100%) 251 (100%) 1.8 (1.0–3.1) Grade (pathologist 1) <0.001 Well 17 (15%) 83 (33%) 1.0 reference Moderate 54 (49%) 131 (52%) 2.0 (1.1–3.6) Poor 39 (35%) 37 (15%) 4.7 (2.3–9.9) Multivariate analyses with Recurrence Score Tumor size (cm) 0.036 Continuous 110 (100%) 251 (100%) 1.8 (1.0–3.1) Grade (pathologist 1) 0.018 Well 17 (15%) 83 (33%) 1.0 reference Moderate 54 (49%) 131 (52%) 1.8 (1.0–3.4) Poor 39 (35%) 37 (15%) 3.2 (1.4–7.2) Recurrence Score 0.025 Continuous 110 (100%) 251 (100%) 2.4 (1.1–5.2) 1 2 Conditional logistic regression models include tumor size and tumor grade only. Relative risks for tumor size calculated with regard to an 3 4 increment of 2 cm. Conditional logistic regression models include tumor size, tumor grade, and Recurrence Score. Relative risks for Recurrence Score calculated with regard to an increment of 50 units (chosen to be consistent with and comparable to previous studies). CI, confidence interval; ER, estrogen receptor; RR, relative risk. controls who were tamoxifen treated and had ER-positive tamoxifen. There were 16 cases with matched controls (n = tumors according to RT-PCR assay. There were 110 cases 19) who had ER-negative tumors according to RT-PCR assay and 251 matched controls who had ER-positive tumors and were not treated with tamoxifen (out of a total of 52 ER- according to RT-PCR assay and were not treated with negative cases and 56 ER-negative controls). In addition, a Page 8 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 Figure 2 Figure 3 Rel in Rel ca a ative risks ( tive risks ( lculation of RRs) of dea RRs) of dea the Recurren t th a h a ce ssoci ssoci Score ated w ated wiith gene group scores used th gene group scores used in calculation of the Recurrence Score. Findings are stratified by Rela genes Relative ri tive risks (RRs sks (RRs) for ) for dea deat th a h associ ssociated ated with with expressio expression n of of sin sing gle le tamoxifen treatment and ER status. The position of each symbol indi- genes. Findings are stratified by tamoxifen treatment and ER status. cates the RR. The length of the horizontal line through the symbol indi- The position of each symbol indicates the RR. The length of the hori- cates the 95% CI. The blue boxes indicate the RRs and 95% CIs for zontal line through the symbol indicates the 95% CI. The blue boxes ER-positive patients treated with tamoxifen, the green pyramids indi- indicate the RRs and 95% CIs for ER-positive patients treated with cate the RRs and 95% CIs for ER-positive patients not treated with tamoxifen, the green pyramids indicate the RRs and 95% CIs for ER- tamoxifen, and the red downward pointing triangles indicate the RRs positive patients not treated with tamoxifen, and the red downward and 95% CIs for ER-negative patients not treated with tamoxifen. CI, pointing triangles indicate the RRs and 95% CIs for ER-negative confidence interval; ER, estrogen receptor; RR, relative risk. patients not treated with tamoxifen. CI, confidence interval; ER, estro- gen receptor; RR, relative risk. ble, when categorized into quartiles, or when categorized small number of ER-negative patients were treated with based on pre-specified cutoff points (Table 3). tamoxifen (nine cases and three controls). In ER-positive patients not treated with tamoxifen, the risk of Distribution of Recurrence Score risk categories by breast cancer death was also positively associated with tumor size and tumor grade Recurrence Score (Table 3). As expected, the association of The distributions of tumor size and/or tumor grade in patients Recurrence Score with risk of breast cancer death appeared categorized on the basis of the Recurrence Score as low risk to be stronger among ER-positive patients treated with (Recurrence Score <18), intermediate risk (Recurrence Score tamoxifen than among those not treated with tamoxifen. 18–30), or high risk (Recurrence Score ≥31) are shown in Table 2. The Recurrence Score was associated with tumor Larger tumors and higher grade tumors were associated with size and even more so with tumor grade. Nonetheless, a an increased risk of breast cancer-specific mortality in both number of patients had large (>2 cm) and/or moderately or tamoxifen-treated and -untreated ER-positive patients (Table poorly differentiated tumors with low risk Recurrence Score 4). When the Recurrence Score (continuous) was added to values. In addition, a small percentage had small (≤1 cm) and/ these multivariate models, it provided information on risk that or well differentiated tumors with high risk Recurrence Score was independent of tumor size and tumor grade. This was also values. The concordance in the assessment of tumor grade true when the Recurrence Score was categorized into quar- between the two pathologists was moderate (kappa 0.53, tiles or when it was categorized based on pre-specified cutoff 95% CI 0.49–0.59). points (not shown). Note that even though there were differ- ences in the assessment of grade between the two patholo- Relative risks for breast cancer death: ER-positive gists, similar to observations from other studies [22-26], the patients results were not materially different when tumor grade assess- In ER-positive patients treated with tamoxifen, the risk of ments from pathologist 2, rather than pathologist 1, were used breast cancer death was positively and strongly associated (not shown). with Recurrence Score when analyzed as a continuous varia- Page 9 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. Table 5 Ten-year risk of death in relation to Recurrence Score and tumor size and grade among ER-positive patients, stratified by treatment with tamoxifen Risk classifier Cases Controls 10-Year risk % 95% CI Tamoxifen treated Recurrence Score (55 cases and 150 controls) Low (<18) 29% 63% 2.8 (1.7–3.9) Intermediate (18–30) 40% 23% 10.7 (6.3–14.9) High (≥31) 31% 13% 15.5 (7.6–22.8) Tumor size (55 cases and 150 controls) ≤1 cm 16% 31% 3.6 (1.7–5.5) 1.1–2 cm 44% 43% 5.8 (3.7–7.8) >2 cm 40% 26% 9.3 (5.5–13.1) Tumor grade (55 cases and 150 controls) Well 11% 33% 2.1 (0.7–3.5) Moderate 51% 46% 6.9 (4.7–9.0) Poor 38% 21% 9.9 (6.0–13.7) Tumor size and grade (55 cases and 150 controls) ≤2 cm and well or ≤1 cm and moderate 15% 43% 2.5 (1.2–3.9) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 49% 39% 7.2 (4.9–9.5) >2 cm and moderate/poor 36% 18% 11.5 (6.6–16.2) Recurrence Score, tumor size and grade Recurrence Score low <18 (16 cases and 95 controls) ≤2 cm and well or ≤1 cm and moderate 6% 46% 0.4 (0.0–1.0) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 63% 39% 4.1 (2.0–6.2) >2 cm and moderate/poor 31% 15% 6.9 (1.8–11.7) Recurrence Score intermediate 18–30 (22 cases and 35 controls) ≤2 cm and well or ≤1 cm and moderate 32% 51% 6.8 (2.6–10.9) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 45% 29% 21.9 (5.0–35.8) >2 cm and moderate/poor 23% 20% 13.5 (1.2–24.2) Recurrence Score high ≥31 (17 cases and 20 controls) ≤2 cm and well or ≤1 cm and moderate 0% 10% 12.2 (0.0–31.0) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 41% 60% 10.9 (2.9–18.3) >2 cm and moderate/poor 59% 30% 28.9 (7.4–45.3) Tamoxifen untreated Recurrence Score (110 cases and 251 controls) Low (<18) 36% 64% 6.2 (4.5–7.9) Intermediate (18–30) 29% 19% 17.8 (11.8–23.3) High (≥31) 35% 18% 19.9 (14.2–25.2) Tumor size (110 cases and 251 controls) ≤1 cm 29% 39% 8.5 (6.0–10.9) 1.1–2 cm 41% 45% 10.4 (7.8–12.9) >2 cm 30% 16% 16.4 (11.4–21.0) Page 10 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 Table 5 (Continued) Ten-year risk of death in relation to Recurrence Score and tumor size and grade among ER-positive patients, stratified by treatment with tamoxifen Tumor grade (110 cases and 251 controls) Well 15% 33% 5.5 (3.4–7.4) Moderate 49% 52% 11.2 (8.6–13.8) Poor 35% 15% 22.5 (15.9–28.6) Tumor size and grade (110 cases and 251 controls) ≤2 cm and well or ≤1 cm and moderate 27% 50% 6.2 (4.3–8.0) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 45% 37% 14.2 (10.8–17.4) >2 cm and moderate/poor 27% 13% 19.1 (13.2–24.7) Recurrence Score, tumor size and grade Recurrence Score low <18 (40 cases and 160 controls) ≤2 cm and well or ≤1 cm and moderate 35% 63% 3.0 (1.7–4.3) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 48% 31% 11.0 (7.0–14.8) >2 cm and moderate/poor 17% 6% 20.2 (6.1–32.2) Recurrence Score intermediate 18–30 (32 cases and 47 controls) ≤2 cm and well or ≤1 cm and moderate 34% 38% 17.0 (8.2–24.9) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 38% 34% 20.9 (10.2–30.4) >2 cm and moderate/poor 28% 28% 22.0 (9.3–32.9) Recurrence Score high ≥31 (38 cases and 44 controls) ≤2 cm and well or ≤1 cm and moderate 13% 14% 25.0 (7.9–39.0) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 50% 59% 21.0 (13.8–27.5) >2 cm and moderate/poor 37% 27% 25.3 (15.1–34.3) CI, confidence interval; ER, estrogen receptor. The relationships between the expression of the individual Relative risks for breast cancer death: ER-negative genes that comprise the Recurrence Score and risk of breast patients cancer death were generally similar in tamoxifen-treated and - After adjusting for tumor size and tumor grade, the risk of untreated ER-positive patients (Figure 2). The risk for death breast cancer death was positively associated with the Recur- was positively associated with expression of each of the five rence Score (continuous variable in 50-unit increments) proliferation genes (cyclin B , Ki-67, MYLBL2, STK15, and among ER-negative patients (relative risk 6.2, 95% CI 1.2– survivin). Positive associations were also observed for genes 31.8). Risk for death was also strongly associated with the in the invasion group. The risk for death was negatively asso- proliferation gene group (relative risk 5.1, 95% CI 2.0–13.5). ciated with expression of the ER-related genes (ER, PgR, For some individual genes, associations appeared to be more BCL2, and SCUBE2). However, the associations between strongly positive in the ER-negative than in the ER-positive expression of the ER-related genes and risk were generally patients (that is, Cyclin B , Ki-67, STK15, survivin). For a few stronger for the tamoxifen-treated patients, especially with genes, associations were in different directions for ER-nega- respect to the quantitative expression of the ER gene. Risk for tive and ER-positive patients (that is, BAG1, GSTM1, PgR). death due to breast cancer was not statistically significantly Note that estimates for ER-negative patients were often impre- associated with CD68, HER2, or GRB7 gene expression in cise because of small numbers. patients treated or untreated with tamoxifen. Absolute risk of breast cancer death at 10 years: ER- The associations between expression of the gene group positive patients scores (as calculated in the Recurrence Score) and risk of The risks for breast cancer death at 10 years for ER-positive breast cancer death were generally stronger than those for patients treated with tamoxifen were 2.8% (95% CI 1.7– individual genes (Figure 3). 3.9%), 10.7% (95% CI 6.3–14.9%), and 15.5% (95% CI 7.6–22.8%) for patients with Recurrence Score values in the low, intermediate, and high risk categories, respectively (Table 5). Categories based on tumor size and/or grade identified a Page 11 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. set of patients at similarly low risk of breast cancer death, as well as prognosis. This is most likely explained by the fact although the Recurrence Score was able to identify a substan- that the expression of ER-related genes was more strongly tially larger subgroup of patients. For example, 63% of the con- associated with breast cancer-specific mortality in the trols were identified as low risk by the Recurrence Score, tamoxifen-treated patients than in those not treated with whereas only 31% of controls had a tumor less than 1 cm. tamoxifen (Figure 2), and is also consistent with the estab- Cross-classifying patients by Recurrence Score and tumor lished relationship between ER status of the tumor (by ligand size and grade resulted in very imprecise estimates (that is, binding or immunohistochemistry assay) and response to large CIs). Nonetheless, results suggest that among patients tamoxifen [4,28]. Our results are also consistent with findings with Recurrence Score values categorized as low risk, tumor from a study conducted among participants of the NSABP B- size and grade provide additional risk prediction information. 14 clinical trial, which randomized patients to tamoxifen versus placebo [29]. In this population, the tamoxifen benefit varied by For ER-positive patients not treated with tamoxifen, the abso- Recurrence Score and was greatest for those with low Recur- lute risks for death at 10 years were higher than those for ER- rence Score values. As expected, the strong association positive patients treated with tamoxifen. The risks for breast between the quantitative expression of the ER gene and cancer death were 6.2% (95% CI 4.5–7.9%), 17.8% (95% CI tamoxifen benefit was largely responsible for this finding. 11.8–23.3%), and 19.9% (95% CI 14.2–25.2%) for patients in the low, intermediate, and high risk Recurrence Score Currently, adjuvant hormonal and/or cytotoxic chemotherapy groups, respectively. Again, when cross-classifying patients are recommended for most women with early-stage invasive by Recurrence Score and tumor size and grade, an improve- breast cancer. Treatment decisions are based on axillary node ment in risk prediction was suggested primarily for those with status, age, tumor size, histologic tumor type, tumor grade, Recurrence Score values categorized as low risk. hormone receptor status (ER, PgR), and coexisting medical conditions [4]. Hormonal therapy is recommended for nearly all women with ER-positive tumors. Although tamoxifen is gen- Discussion In this large, population-based study of lymph node-negative erally well tolerated, a significant proportion of women experi- patients not treated with chemotherapy, we found that the ence hot flashes and leg cramps, and up to 20% do not Recurrence Score was strongly associated with risk of breast complete a 5-year course of tamoxifen therapy [30-32]. cancer death among ER-positive patients treated with Despite its potential for serious adverse effects [33], cytotoxic tamoxifen. We also found that the Recurrence Score was chemotherapy has been recommended for most women with strongly associated with risk of breast cancer death among lymph node-positive disease and for node-negative patients ER-positive patients not treated with tamoxifen and among ER- with tumors greater than 1 cm or with unfavorable pathology negative patients. In addition, we found that these associa- [4-6]. Very little information is available to support the use of tions remained after accounting for tumor size and grade, and other clinical or biologic factors in selecting patients for adju- that the Recurrence Score was able to identify a larger subset vant chemotherapy [4]. Most patients with node-negative dis- of patients with low risk of breast cancer death than was pos- ease who receive chemotherapy will not derive benefit, sible with either of these standard prognostic indicators. because they would not go on to have a recurrence even with- out such treatment. New prognostic and predictive tests are Several limitations should be considered when interpreting our needed to better individualize therapy and confine systemic results. We lacked ER status from the medical record for a treatment, especially cytotoxic chemotherapy, to those substantial proportion of patients, and we therefore classified patients who are most likely to benefit. ER status based on gene expression. However, the estimates of relative risk were not materially changed when analyses A growing number of studies suggest that multigene expres- were restricted to the 84% of patients with ER status from the sion assays may be able to provide important prognostic and/ charts (data not shown). Because of the diagnosis years of the or predictive information for breast cancer patients [34-42]. A study, only approximately 30% of patients were treated with variety of approaches and technologies are being used to tamoxifen. Although this is consistent with what has been select genes and characterize expression (for example, cDNA reported for other patient populations during this period [27], microarray chips, RT-PCR) in different types of pathology it limited the numbers of tamoxifen-treated patients for analy- specimens (for example, fresh frozen tissue, formalin-fixed par- sis. Given that the cases and controls were matched with affin embedded tissue). No matter what the approach or tech- respect to tamoxifen treatment, we could not directly examine nology, multiple, well conducted confirmatory studies using whether the Recurrence Score is able to identify patients who standardized methodologies will be needed before the clinical are likely to respond to tamoxifen therapy. We did find a usefulness of any of these assays can be established. stronger association between the Recurrence Score and risk of breast cancer death among patients treated with tamoxifen This is the third study to evaluate the performance of the than among those untreated with tamoxifen, suggesting that Recurrence Score among patients not treated with systemic the Recurrence Score captures response to tamoxifen therapy chemotherapy [12,13]. The three studies used identical pre- Page 12 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 specified scores and laboratory methods and were conducted participating in a clinical trial and among similar patients from among patients who were independent of those used for gene the community setting. In both study populations, the Recur- selection and Recurrence Score algorithm development. The rence Score was able to identify a large subset of patients results of the first study, conducted by Esteva and coworkers (approximately 50% or more) who were at very low risk of [13], differ substantially from our results and from those of the breast cancer death at 10 years. In both studies it was also NSABP B-14 study [12]. In that study no association was observed that although the Recurrence Score was correlated found between the Recurrence Score and risk for distant with tumor size and grade, there were a number of patients recurrence among a series of 149 node-negative patients who with large and/or moderately or poorly differentiated tumors were treated without adjuvant hormonal therapy or chemother- with low risk Recurrence Score values. Results from our study, apy at the MD Anderson Cancer Center between 1978 and and to a lesser extent the NSABP B-14 study, suggest that 1995, who had potentially 5 or more years of follow up, and for combining Recurrence Score, tumor grade, and tumor size whom archived tissue was available. In contrast to the NSABP provides better risk classification than any one of these factors B-14 study, patients in the study conducted by Esteva and alone. Other studies, either retrospective or prospective, will coworkers were not treated with tamoxifen. In contrast to our be needed to confirm our findings among lymph node-nega- study, the outcome of interest was distant recurrence instead tive, ER-positive patients not treated with tamoxifen and of breast cancer death. Although these may explain some of among ER-negative patients. Two studies have been per- the differences observed, it is also possible that the study by formed [43,44] and others are ongoing to assess the relation- Esteva and coworkers included a nonrepresentative group of ship between the Recurrence Score and the magnitude of patients. Patients with poorly differentiated tumors had better chemotherapy benefit. Areas for future research also include prognosis than those with well differentiated tumors, and there the examination of whether the Recurrence Score assay pro- was a suggestion that patients with ER-negative tumors did vides prognostic or predictive information for patients treated better than those with ER-positive tumors. with other hormonal agents or for patients with node-positive breast cancer, whether test performance can be further opti- Our relative risk estimates for the Recurrence Score in ER- mized by even more individualized dissection techniques, and positive, tamoxifen-treated patients are generally quite similar whether the inclusion of additional genes or the inclusion of to those observed in the NSABP B-14 study. Relative risks standard measures (tumor size, grade) may enhance risk pre- associated with expression of individual genes were also very diction overall or for selected patient subgroups. similar. In the NSABP B-14 study, the 10-year risk of breast cancer death was 3.1% (95% CI 1.2–5.0%) in the low risk Competing interests group, 12.2% (95% CI 6.7–17.6%) in the intermediate risk The following authors received support for study-related activ- group, and 27.0% (95% CI 20.4–33.6%) in the high risk ities from Genomic Health, Inc., but have no other competing group. Although very similar to our findings for the low and financial or nonfinancial interests: LAH, MJK, AC, NTB, DG, intermediate risk groups, the 10-year risk for the NSABP B-14 CPQ, BL, and LF. The remaining co-authors (SS, CA, MP, JB, high risk group was higher than ours. Patients in the two stud- MW, DW, and JH) are employees or consultants for Genomic ies had a comparable age distribution, and although those in Health, Inc. the NSABP B-14 study were more likely to have larger tumors, their tumors were also more likely to be well differentiated. Authors' contributions Therefore, there is uncertainty regarding the extent to which LAH participated in the design of the study, directed the data differences in the distribution of prognostic factors in the two analysis, and drafted the manuscript. SS participated in the study populations may explain the difference observed in design of the study, in interpreting results, and in writing the absolute risk estimates for the high risk group. In the Kaiser manuscript. MKJ and AC performed statistical analyses and population, tumor size and tumor grade remained statistically participated in interpreting results. NTB and CA participated in significantly associated with risk of breast cancer death in the development of the study methods and coordinated data most multivariate models that also included the Recurrence collection. MP participated in the development of the study Score, whereas only tumor grade remained independently methods and in data collection. JB participated in the develop- associated with risk in the NSABP B-14 study. Tumor size was ment of the study methods, in interpreting results, and in writ- determined by pathology in the Kaiser study and by clinical ing the manuscript. MW, DW, and JH participated in the examination in the NSABP B-14 study. Clinical examination of design of the study and in interpreting results. DG participated tumor size is generally less accurate and could have resulted in data collection and in interpreting results. CPQ participated in attenuated relative risk estimates. in the study design, co-directed the data analysis, and partici- pated in interpreting results. BL provided methods for estimating cumulative risks and participated in interpreting Conclusion The Recurrence Score has now been shown to be strongly results. LF participated in interpreting results. All authors read associated with risk of breast cancer-specific mortality among and approved the final manuscript. lymph node-negative, ER-positive, tamoxifen-treated patients Page 13 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. 7. Bast RC Jr, Hortobagyi GN: Individualized care for patients with Additional files cancer: a work in progress. N Engl J Med 2004, 351:2865-2867. 8. Chang JC, Hilsenbeck SG, Fuqua SA: Genomic approaches in The following Additional files are available online: the management and treatment of breast cancer. Br J Cancer 2005, 92:618-624. 9. Ransohoff DF: Lessons from controversy: ovarian cancer Additional File 1 screening and serum proteomics. J Natl Cancer Inst 2005, A Word document providing additional detail regarding 97:315-319. 10. Simon R: When is a genomic classifier ready for prime time? the statistical analyses for estimating 10-year recurrence Nat Clin Pract Oncol 2004, 1:4-5. probabilities. 11. Cronin M, Pho M, Dutta D, Stephans JC, Shak S, Kiefer MC, Este- See http://www.biomedcentral.com/content/ ban JM, Baker JB: Measurement of gene expression in archival paraffin-embedded tissues: development and performance of supplementary/bcr1412-S1.doc a 92-gene reverse transcriptase-polymerase chain reaction assay. Am J Pathol 2004, 164:35-42. 12. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, Baehner FL, Walker MG, Watson D, Park T, et al.: A multigene assay topre- dict recurrence of tamoxifen-treated, node-negative Acknowledgements breastcancer. N Engl J Med 2004, 351:2817-2826. We should like to thank Kaiser Permanente patients, as well as person- 13. Esteva FJ, Sahin AA, Cristofanilli M, Coombes K, Lee SJ, Baker J, Cronin M, Walker M, Watson D, Shak S, et al.: Prognostic role of nel at the Permanente Medical Group Regional Laboratory and at Kaiser a multigene reverse transcriptase-PCR assay in patients with Permanente Medical Center pathology departments. We also gratefully node-negative breast cancer not receiving adjuvant systemic acknowledge Soonmyung Paik in the Division of Pathology at NSABP therapy. Clin Cancer Res 2005, 11:3315-3319. for performing tissue macrodissection and Fredrick L Baehner and Tracy 14. Gordon NP, Kaplan GA: Some evidence refuting the HMO 'favorable selection' hypothesis: the case of Kaiser George for tumor grading (work funded by Genomic Health, Inc). We Permanente. Adv Health Econ Health Serv Res 1991, 12:19-39. also thank Edith Perez (consult paid by Genomic Health), Clifford Hudis 15. Hiatt RA, Friedman GD: The frequency of kidney and urinary (consult paid by Genomic Health), and Noel Weiss (consult paid by Kai- tract diseases in a defined population. Kidney Int 1982, ser Permanente) for their thoughtful comments on the manuscript. 22:63-8. 16. Krieger N: Overcoming the absence of socioeconomic data in medical records: validation and application of a census-based The study was supported by Genomic Health, Inc. The lead author methodology. Am J Public Health 1992, 82:703-710. (LAH) and several co-authors (MKJ, AC, NTB, DB, LF, and CPQ) are 17. Rothman K, Greenland S: Case-control studies. In Modern Epi- employees of Kaiser Permanente, which received study funds from a demiology Edited by: Rothman KJ, Greenland S. Philadelphia: contract with Genomic Health, Inc. One co-author (BL) was a consultant Lippincott-Raven; 1998:93-114. 18. Elston CW, Ellis IO: Pathological prognostic factors in breast and received fees from Kaiser Permanente. In addition to providing cancer. I. The value of histological grade in breast cancer: financial support, Genomic Health, Inc. co-authors (SS, CA, MP, JB, experience from a large study with long-term follow-up. His- MW, DW, and JH) participated in study design; in the collection, analy- topathology 1991, 19:403-410. sis, and interpretation of data; and/or in the writing of the manuscript. All 19. Cobleigh M: Tumor gene expression predicts distant disease- free survival (DDFS) in breast cancer patients with 10 or more the Genomic Health authors hold stock in the company. Genomic positive nodes: high throughout RT-PCR assay or paraffin- Health, Inc. did not participate in the decision to submit the manuscript embeded tumor tissues [abstract]. Prog Proc Am Soc Clin for publication. As part of the funding contract agreement, this was the Oncol 2003, 22:850. independent decision of Kaiser Permanente study investigators. 20. Breslow NE, Day NE: The analysis of case-control studies. In Statistical Methods in cancer research Volume 1. Edited by: Davis W. Lyon, France: International Agency for Research on Cancer; References 1980:205-210. 1. Ghafoor A, Jemal A, Ward E, Cokkinides V, Smith R, Thun M: 21. Langholz B, Borgan O: Estimation of absolute risk from nested Trends in breast cancer by race and ethnicity. CA Cancer J case-control data. Biometrics 1997, 53:767-774. Clin 2003, 53:342-355. 22. Robbins P, Pinder S, de Klerk N, Dawkins H, Harvey J, Sterrett G, 2. Fisher B, Jeong JH, Bryant J, Anderson S, Dignam J, Fisher ER, Ellis I, Elston C: Histological grading of breast carcinomas: a Wolmark N: National Surgical Adjuvant Breast and Bowel study of interobserver agreement. Hum Pathol 1995, Project randomised clinical trials: Treatment of lymph-node- 26:873-879. negative, oestrogen-receptor-positive breast cancer: long- 23. Davis BW, Gelber RD, Goldhirsch A, Hartmann WH, Locher GW, term findings from National Surgical Adjuvant Breast and Reed R, Golouh R, Save-Soderbergh J, Holloway L, Russell I, et al.: Bowel Project randomised clinical trials. Lancet 2004, Prognostic significance of tumor grade in clinical trials of adju- 364:858-868. vant therapy for breast cancer with axillary lymph node 3. Early Breast Cancer Trialists' Collaborative Group: Polychemo- metastasis. Cancer 1986, 58:2662-2670. therapy for early breast cancer: an overview of the randomised 24. Hopton DS, Thorogood J, Clayden AD, MacKinnon D: Observer trials. Lancet 1998, 352:930-942. variation in histological grading of breast cancer. Eur J Surg 4. Eifel P, Axelson JA, Costa J, Crowley J, Curran WJ Jr, Deshler A, Oncol 1989, 15:21-23. Fulton S, Hendricks CB, Kemeny M, Kornblith AB, et al.: National 25. Theissig F, Kunze KD, Haroske G, Meyer W: Histological grading Institutes of Health Consensus Development Conference of breast cancer. Interobserver, reproducibility and prognostic Statement: adjuvant therapy for breast cancer, November 1–3, significance. Pathol Res Pract 1990, 186:732-736. 2000. J Natl Cancer Inst 2001, 93:979-989. 26. Meyer JS, Alvarez C, Milikowski C, Olson N, Russo I, Russo J, 5. Goldhirsch A, Glick JH, Gelber RD, Coates AS, Senn HJ: Meeting Glass A, Zehnbauer BA, Lister K, Parwaresch R: Cooperative highlights: International Consensus Panel on the Treatment of Breast Cancer Tissue Resource: Breast carcinoma malignancy Primary Breast Cancer. Seventh International Conference on grading by Bloom-Richardson system vs proliferation index: Adjuvant Therapy of Primary Breast Cancer. J Clin Oncol reproducibility of grade and advantages of proliferation index. 2001, 19:3817-3827. Mod Pathol 2005, 18:1067-1078. 6. Carlson RW, Edge SB, Theriault RL: NCCN: Breast cancer. Can- 27. Harlan LC, Abrams J, Warren JL, Clegg L, Stevens J, Ballard-Bar- cer Control 2001, 8:54-61. bash R: Adjuvant therapy for breast cancer: practice patterns of community physicians. J Clin Oncol 2002, 20:1809-1817. Page 14 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 28. Harvey JM, Clark GM, Osborne CK, Allred DC: Estrogen receptor status by immunohistochemistry is superior to the ligand- binding assay for predicting response to adjuvant endocrine therapy in breast cancer. J Clin Oncol 1999, 17:1474-1481. 29. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, Watson D, Bry- ant J, Costantino J, Wolmark N: Expression of the 21 genes in the Recurrence Score assay and tamoxifen clinical benefit in the NSABP study B-14 of node negative, estrogen receptor positive breast cancer [abstract]. J Clin Oncol 2005, 23:6s. 30. Partridge AH, Wang PS, Winer EP, Avorn J: Nonadherence to adjuvant tamoxifen therapy in women with primary breast cancer. J Clin Oncol 2003, 21:602-606. 31. Demissie S, Silliman RA, Lash TL: Adjuvant tamoxifen: predic- tors of use, side effects, and discontinuation in older women. J Clin Oncol 2001, 19:322-328. 32. Fink AK, Gurwitz J, Rakowski W, Guadagnoli E, Silliman RA: Patient beliefs and tamoxifen discontinuance in older women with estrogen receptor-positive breast cancer. J Clin Oncol 2004, 22:3309-3315. 33. Partridge AH, Winer EP: Long-term complications of adjuvant chemotherapy for early stage breast cancer. Breast Dis 2004, 21:55-64. 34. van 't Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M, Peterse HL, van der Kooy K, Marton MJ, Witteveen AT, et al.: Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002, 415:530-536. 35. van de Vijver MJ, He YD, van't Veer LJ, Dai H, Hart AA, Voskuil DW, Schreiber GJ, Peterse JL, Roberts C, Marton MJ, et al.: A gene- expression signature as a predictor of survival in breast cancer. N Engl J Med 2002, 347:1999-2009. 36. Huang E, Cheng SH, Dressman H, Pittman J, Tsou MH, Horng CF, Bild A, Iversen ES, Liao M, Chen CM, et al.: Gene expression pre- dictors of breast cancer outcomes. Lancet 2003, 361:1590-1596. 37. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pollack JR, Ross DT, Johnsen H, Akslen LA, et al.: Molecular por- traits of human breast tumours. Nature 2000, 406:747-752. 38. Sotiriou C, Neo SY, McShane LM, Korn EL, Long PM, Jazaeri A, Martiat P, Fox SB, Harris AL, Liu ET: Breast cancer classification and prognosis based on gene expression profiles from a pop- ulation-based study. Proc Natl Acad Sci USA 2003, 100:10393-10398. 39. Bertucci F, Houlgatte R, Benziane A, Granjeaud S, Adelaide J, Tagett R, Loriod B, Jacquemier J, Viens P, Jordan B, et al.: Gene expression profiling of primary breast carcinomas using arrays of candidate genes. Hum Mol Genet 2000, 9:2981-2991. 40. Pittman J, Huang E, Dressman H, Horng CF, Cheng SH, Tsou MH, Chen CM, Bild A, Iversen ES, Huang AT, et al.: Integrated mode- ling of clinical and gene expression information for personal- ized prediction of disease outcomes. Proc Natl Acad Sci USA 2004, 101:8431-8436. 41. Rouzier R, Wagner P, Morandi P, Pusztai L: Gene expression profiling of primary breast cancer. Curr Oncol Rep 2005, 7:38-44. 42. Chang J, Powles TJ, Allred DC, Ashley SE, Clark GM, Makris A, Assersohn L, Gregory RK, Osborne CK, Dowsett M: Biologic markers as predictors of clinical outcome from systemic ther- apy for primary operable breast cancer. J Clin Oncol 1999, 17:3058-3063. 43. Gianni L, Zambetti M, Clark K, Baker J, Cronin M, Wu J, Mariani G, Rodriguez J, Carcangiu M, Watson D, et al.: Gene expression profiles in paraffin-embedded core biopsy tissue predict response to chemotherapy in women with locally advanced breast cancer. J Clin Oncol 2005, 23:7265-7277. 44. Paik S, Tang G, Shak S, Kim C, Baker J, Kim W, Cronin M, Baehner FL, Watson D, Bryant J, Costantino JP, Geyer CE Jr, Wickerham DL, Wolmark N: Gene expression and benefit of chemotherapy in women with node negative, estrogen receptor positive breast cancer. J Clin Oncol 2006, 24:2019-2027. Page 15 of 15 (page number not for citation purposes) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Breast Cancer Research Springer Journals

A population-based study of tumor gene expression and risk of breast cancer death among lymph node-negative patients

Download PDF
15 pages

Loading next page...
 
/lp/springer-journals/a-population-based-study-of-tumor-gene-expression-and-risk-of-breast-anJN0FSLA1

References (57)

Publisher
Springer Journals
Copyright
2006 Habel et al.; licensee BioMed Central Ltd.
eISSN
1465-542X
DOI
10.1186/bcr1412
Publisher site
See Article on Publisher Site

Abstract

Introduction The Oncotype DX assay was recently reported to Results After adjusting for tumor size and grade, the predict risk for distant recurrence among a clinical trial Recurrence Score was associated with risk of breast cancer population of tamoxifen-treated patients with lymph node- death in ER-positive, tamoxifen-treated and -untreated patients negative, estrogen receptor (ER)-positive breast cancer. To (P = 0.003 and P = 0.03, respectively). At 10 years, the risks for confirm and extend these findings, we evaluated the breast cancer death in ER-positive, tamoxifen-treated patients performance of this 21-gene assay among node-negative were 2.8% (95% confidence interval [CI] 1.7–3.9%), 10.7% patients from a community hospital setting. (95% CI 6.3–14.9%), and 15.5% (95% CI 7.6–22.8%) for those in the low, intermediate and high risk Recurrence Score Methods A case-control study was conducted among 4,964 groups, respectively. They were 6.2% (95% CI 4.5–7.9%), Kaiser Permanente patients diagnosed with node-negative 17.8% (95% CI 11.8–23.3%), and 19.9% (95% CI 14.2– invasive breast cancer from 1985 to 1994 and not treated with 25.2%) for ER-positive patients not treated with tamoxifen. In adjuvant chemotherapy. Cases (n = 220) were patients who both the tamoxifen-treated and -untreated groups, approximately died from breast cancer. Controls (n = 570) were breast cancer 50% of patients had low risk Recurrence Score values. patients who were individually matched to cases with respect to age, race, adjuvant tamoxifen, medical facility and diagnosis year, and were alive at the date of death of their matched case. Using an RT-PCR assay, archived tumor tissues were analyzed for expression levels of 16 cancer-related and five reference Conclusion In this large, population-based study of lymph node- genes, and a summary risk score (the Recurrence Score) was negative patients not treated with chemotherapy, the calculated for each patient. Conditional logistic regression Recurrence Score was strongly associated with risk of breast methods were used to estimate the association between risk of cancer death among ER-positive, tamoxifen-treated and - breast cancer death and Recurrence Score. untreated patients. and free from distant metastasis at 10 years [2]. Chemother- apy in this group of patients, especially among patients with Introduction estrogen receptor (ER)-positive disease treated with adjuvant Approximately 65% of women currently diagnosed with inva- hormonal therapy, offers only a modest improvement in 10- sive breast cancer have lymph node-negative disease at diag- year survival [2,3]. Nonetheless, according to current guide- nosis [1], and 85% of these women are expected to be alive lines the majority of node-negative breast cancer patients CI = confidence interval; ER = estrogen receptor; NSABP = National Surgical Adjuvant Breast and Bowel Project; PgR = progesterone receptor; RT-PCR = reverse transcriptase polymerase chain reaction. Page 1 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. should be considered for chemotherapy [4-6]. This is largely specified primary aim of the study was to assess the degree to because of our limited ability to identify those individual which the Recurrence Score would predict the risk of breast patients who are unlikely to benefit from such treatment. cancer-specific mortality among women with ER-positive, node-negative breast cancer treated with tamoxifen (that is, Multigene assays may provide information on patient progno- patients clinically similar to those included in the study con- sis and response to therapy that is superior and/or comple- ducted by Paik and coworkers [12]), either alone or in multivar- mentary to that from standard histopathological and iate analyses with tumor size and tumor grade. A pre-specified immunohistochemical techniques [7,8]. However, multiple secondary aim was to examine the extent to which the Recur- well conducted and controlled observational studies or clinical rence Score predicts risk of breast cancer-specific mortality trials in independent populations are needed to establish the among ER-positive, node-negative patients not treated with clinical usefulness of these assays [9,10]. tamoxifen. A pre-specified exploratory aim was to examine the association between risk of breast cancer death and Recur- Using a multistep approach, a 21-gene assay (Oncotype DX) rence Score among node-negative, ER-negative patients. was recently developed for use in paraffin-embedded tumor tissue to predict risk for distant recurrence or death in lymph Materials and methods node-negative breast cancer patients [11]. Approximately 250 Study population and design genes, selected from the published literature, genomic data- We conducted a case-control study nested within a cohort of bases, pathway analysis, and from microarray-based gene breast cancer patients. The Northern California Kaiser Perma- expression profiling experiments performed using fresh frozen nente tumor registry, a contributor to the Surveillance, Epide- tissue, were considered as candidates. The final gene list (16 miology, and End Results (SEER) program of cancer cancer-related and five reference genes) and summary score registries, was used to identify all female health plan members (Recurrence Score) algorithm for this assay (Figure 1) were who were diagnosed with lymph node-negative invasive breast developed by analyzing the results of three independent pre- cancer from 1985 to 1994. Northern California Kaiser Perma- liminary breast cancer studies (that is, training sets) con- nente is a nonprofit, integrated health services delivery organ- ducted in a total of 447 patients [12]. ization that provides care for over 3 million members at 14 Northern California Kaiser hospitals and 23 outpatient clinics. Two subsequent studies were conducted to evaluate the per- The Kaiser Permanente membership is racially and ethnically formance of the Recurrence Score in predicting distant recur- diverse and is demographically similar to the general popula- rence-free survival in lymph node-negative breast cancer tion of northern California, although it tends to under-represent patients not treated with chemotherapy. Among 668 the extremes of the socioeconomic spectrum [14-16]. The tamoxifen-treated patients in the National Surgical Adjuvant study was approved by the Kaiser Permanente Institutional Breast and Bowel Project (NSABP) B-14 clinical trial [12], the Review Board. Recurrence Score was strongly predictive of risk for distant recurrence, both alone and in multivariate analyses that also Breast cancer patients were eligible if their nodal status was included patient age, tumor size, tumor grade, and ER and documented to be negative by pathologic assessment, they HER2 status. For patients categorized as low, intermediate, or were diagnosed before age 75 years, and their initial disease high risk based on Recurrence Score, the Kaplan-Meier esti- was not treated with chemotherapy. Patients were excluded mates for the rates of distant recurrence at 10 years (and the for the following reasons: inflammatory or bilateral breast can- associated 95% confidence intervals [CIs]) were 6.8% (95% cer or evidence of metastasis (including lymph nodes) at initial CI 4.0–9.6%), 14.3% (95% CI 8.3–20.3%), and 30.5% (95% diagnosis; prior invasive cancer (breast or other) at diagnosis; CI 23.6–37.4%), respectively. A second study [13] evaluated or unknown/unconfirmed treatment with tamoxifen. the assay among 149 patients not treated with adjuvant ther- apy (hormonal or chemotherapy) at the MD Anderson Cancer Using the Kaiser Permanente tumor registry and electronic Center; it found no clear association between the Recurrence membership databases, patients were followed until death Score and risk for distant recurrence. Rates of distant recur- due to breast cancer, death from another cause, bilateral rence at 10 years were 18% (95% CI 7–30%), 38% (95% CI breast cancer, termination of membership, or December 15–61%), and 28% (95% CI 13–32%) for the low, intermedi- 2002, whichever came first. ate, and high risk categories, respectively. Cases were patients whose first event was death from breast To evaluate the Oncotype DX assay in a third independent cancer. At each case's death, up to three controls were ran- study population of lymph node-negative breast cancer domly selected from patients alive and under follow up (that is, patients not treated with chemotherapy, we conducted a case- incidence density sampling) [17]. Controls were individually control study among a large population of women who were matched to their case with respect to age (within 1 year), race diagnosed and treated at 14 hospitals and clinics within the (non-Hispanic white, Hispanic, Black, Asian), calendar year of Northern California Kaiser Permanente health plan. A pre- diagnosis (exact year), Kaiser Permanente pathology depart- Page 2 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 Table 1 Selected characteristics of the study population Characteristic Cases (n = 220) Controls (n = 570) Matched variables Age at diagnosis (years) <40 17(8%) 23(4%) 40–49 42 (19%) 127 (22%) 50–59 64 (29%) 152 (27%) 60–74 97 (44%) 268 (47%) Race/ethnicity White, non-Hispanic 172 (78%) 457 (80%) White, Hispanic 7 (3%) 13 (2%) Black 20(9%) 47(8%) Asian 21 (10%) 53 (9%) Surgery year 1985–1989 146 (66%) 385 (68%) 1990–1994 74 (34%) 185 (32%) Adjuvant tamoxifen No 156 (71%) 393 (69%) Yes 64 (29%) 177 (31%) Unmatched variables ER status from RT-PCR Positive 168 (76%) 514 (90%) Negative 52 (24%) 56 (10%) Tumor size (cm) ≤1 49 (22%) 193 (34%) 1.1–2 93 (42%) 255 (45%) 2.1–4 72 (33%) 114 (20%) >4 6 (3%) 8 (1%) Tumor grade (differentiation) Well 25 (11%) 175 (31%) Moderate 92 (42%) 261 (46%) Poor 103 (47%) 134 (23%) Recurrence Score Low risk (<18) 57 (26%) 322 (56%) Intermediate risk (18–30) 54 (25%) 108 (19%) High risk (≥31) 109 (50%) 140 (25%) 1 2 Cutoff points based on RT-PCR values: ≤6.5 and >6.5 units. Bloom-Richardson grading criteria, pathologist 1. ER, estrogen receptor; RT-PCR, reverse transcription polymerase chain reaction. ment of origin, and treatment of index breast cancer with grade allowed us to examine whether these measures provide tamoxifen (yes, no). Note that matching on tamoxifen treatment information on risk of breast cancer death that is independent maximized our ability to conduct analyses stratified by of Recurrence Score. Matching requirements were relaxed to tamoxifen treatment. Not matching on tumor size or tumor find up to three eligible controls per case. Matching criteria Page 3 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. were relaxed in the following order: age was relaxed to < 50 Blinding and batching of pathology and laboratory versus ≥ 50 years; calendar year of diagnosis was relaxed to procedures ± 1 year; calendar year of diagnosis was relaxed to ± 2 years; All pathology and laboratory procedures (slide review, section- and pathology facility was dropped as a matching factor. ing of tumor blocks, macro-dissection, RT-PCR assays, and Recurrence Score calculations) were conducted blinded to The medical records of cases and controls were reviewed to the case-control status of patient specimens. In addition, all confirm the initial diagnosis, treatment and cause of death, and batches of pathology materials sent to Kaiser Permanente to obtain information on eligibility criteria. Pathologic tumor Regional Laboratory, NSABP Pathology, and Genomic Health, size was obtained from pathology reports. Information on ER Inc. included a mixture of cases and controls. and progesterone receptor (PgR) status of the index tumor Sample preparation (usually assessed by the ligand-binding assay) was also abstracted from the medical record. For eligible cases and controls, the hematoxylin and eosin stained slides from biopsies and/or surgeries performed at the Cases and controls were selected and eligibility was deter- time of the index diagnosis were reviewed using pre-specified mined before laboratory measurement of gene expression. Of criteria in order to identify the most representative block and to the 402 cases identified as potentially eligible by the tumor evaluate whether sufficient tumor tissue was present. Speci- registry, 269 were determined to be eligible by chart review. mens with no tumor or very little tumor (<5% of the area occu- The eligibility of 27 cases could not be determined because of pied by invasive cancer cells compared to the area occupied incomplete or unavailable medical records and were consid- by other epithelial elements, such as normal epithelium, fibro- ered lost to the study. Similarly, 722 of the 989 controls initially cystic change, or ductal or lobular carcinoma in situ) were matched to eligible cases were determined to be eligible by excluded from the study (n = 18). Specimens with regions on chart review and the eligibility of 29 could not be determined. the slide having prominent nontumor elements (such as Of those eligible by chart review, 31 cases and 91 controls smooth muscle, hemorrhage, fibrosis, hyperplastic epithelium, were lost because of missing tumor blocks. Another four and/or normal breast; but not ductal or lobular carcinoma in cases were lost because we were unable to find at least one situ or necrosis) where the nontumor elements were both suf- matched control. This left 234 cases and 631 controls availa- ble for pathology studies. Table 2 Distributions of tumor size and tumor grade in Recurrence Score categories for all 220 cases and 570 controls Variable (number of patients) Recurrence Score < 18 18–30 ≥30 Correlation (n = 379) (n = 162) (n = 249) (P value) Tumor size (cm) 0.25 (<0.0001) ≤1 (n = 242) 147 (61%) 51 (21%) 44 (18%) 1.1–2 (n = 348) 173 (50%) 65 (19%) 110 (32%) 2.1–4 (n = 186) 55 (30%) 44 (24%) 87 (47%) >4 (n = 14) 4 (29%) 2 (14%) 8 (57%) Tumor grade pathologist 1 0.54 (<0.0001) Well (n = 200) 146 (73%) 44 (22%) 10 (5%) Moderate (n = 353) 199 (56%) 90 (25%) 64 (18%) Poor (n = 237) 34 (14%) 28 (12%) 175 (74%) Tumor size and grade 0.42 (<0.0001) ≤2 cm and well or ≤1 cm and moderate 204 (72%) 60 (21%) 21 (7%) (n = 285) >2 cm and well, or 1.1–2 cm and 143 (41%) 63 (19%) 133 (40%) moderate, or ≤2 cm and poor (n = 330) >2 cm and moderate/poor (n = 175) 41 (23%) 39 (22%) 95 (54%) Spearman rank correlation ficiently localized to be amenable to macro-dissection and suf- were sent to NSABP Pathology for macro-dissection (n = 59). ficiently abundant (>50% of the overall tissue on the slide) Macro-dissection to obtain enriched tumor tissue was per- Page 4 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 formed using a safety blade cleaned with RNaseZAP (Ambion, was less than 2.5 Recurrence Score units. A similarly high Austin, TX, USA) by NSABP Pathology on six (10 µm) sec- concordance (Pearson's r = 0.86) was also observed tions of the region enriched in tumor tissue. For all other spec- between Recurrence Score results from core biopsies and imens, three (10 µm) sections were prepared by Kaiser resections among 49 patients with locally advanced breast Permanente Regional Laboratory, placed into a microcentri- cancer. fuge tube, and sent to Genomic Health, Inc. Recurrence Score cutoff points were pre-specified and classi- Tumor size, grade, and estrogen receptor status fied patients into low risk (Recurrence Score <18), intermedi- When not recorded on the pathology report (7% of reports), ate risk (Recurrence Score 18–30) and high risk (Recurrence tumor size was obtained from a pathology review of all hema- Score ≥31) categories. All methods used for measurement of toxylin and eosin stained slides from all surgeries at diagnosis. gene expression levels and for the calculation of Recurrence Tumor grade, based on review of an hematoxylin and eosin Score and gene group scores were identical to those used in slide from the most representative block, was assessed inde- the study conducted by Paik and coworkers [12]. pendently by two board certified (in anatomic and clinical Statistical analysis pathology), assistant professors in departments of pathology, using the modified Bloom-Richardson grading criteria [18]. ER Statistical analyses were conducted entirely by Kaiser Perma- status of the index tumor was unavailable in the 10- to 20-year- nente researchers following a pre-specified plan. As specified, old medical record for a significant proportion of patients results were generated for patient groups, based on ER status (16%), and therefore all tumors were classified as positive or and tamoxifen treatment. negative based on ER expression by RT-PCR (>6.5 and ≤6.5, respectively; cutoff points based on prior studies examining The Recurrence Score was examined as a continuous variable correlation of RT-PCR with immunohistochemistry) in 50-unit increments, for consistency and comparison with [11,13,19]. In our study population, the concordance of ER previously published Recurrence Score findings. In additional status from RT-PCR versus from medical chart information analyses, women were categorized into presumptive low, was moderate (kappa 0.49, 95% CI 0.41–0.56). All but seven intermediate, and high risk groups based on cutoff points (out of a total of 122) of the discordances were patients clas- (Recurrence Score < 18, 18–30 and ≥31, respectively), and sified as ER positive based on RT-PCR and ER negative when categorized according to Recurrence Score quartiles. based on information in the medical chart. When examining the relationships between risk of breast can- RT-PCR assay of gene expression and calculation of the cer death and the expression level of individual genes, the Recurrence Score expression levels were treated as continuous variables. When Gene expression analysis in fixed paraffin-embedded tumor examining gene groups, the proliferation gene group and the tissue was performed by Genomic Health, Inc., as described HER2 gene group scores were treated as continuous varia- previously [11,12]. Briefly, after RNA extraction and DNase I bles both with and without transformation based on a thresh- treatment, total RNA content was measured and the absence old value (see the Recurrence Score algorithm in Figure 1). of DNA contamination was verified. Reverse transcription was performed followed by quantitative RT-PCR reactions in 384 Tumor size was examined both as a continuous variable in 2 well plates using Applied Biosystems PRISM 7900 HT cm units (for consistency with the NSABP B-14 study [12]) instruments (Applied Biosystems, Foster City, CA, USA). and when categorized as ≤1 cm, 1.1–2 cm, 2.1–4 cm, and >4 cm. Tumor grade was examined as a categorical variable (well Expression of each gene was measured in triplicate wells, and differentiated, moderately differentiated, and poorly differenti- then normalized relative to a set of five reference genes (β- ated). The grading used was from the standardized re-review. actin, GAPDH, GUS, RPLPO, and TFRC). Reference-normal- We examined tumor grade separately for pathologist 1 and ized expression measurements range from 0 to 15, where a 1- pathologist 2. unit increase reflects approximately a twofold increase in RNA. Conditional logistic regression was used to calculate odds The potential impact of tumor heterogeneity on Recurrence ratios as estimates of the relative risks for breast cancer death Score results was examined in two small studies (unpublished associated with the Recurrence Score, or a component of the data). In a study of 20 patients and 60 blocks (two to five Recurrence Score (univariate analyses). In addition, we esti- blocks/patient), analysis of variance was used to examine the mated relative risks adjusted for tumor size and tumor grade variability of Recurrence Score between blocks from the same (multivariate analyses). Model parameter estimation was done patient (tissue sections did not undergo macro-dissection). by maximum likelihood, and 95% Wald's confidence limits The standard deviation in Recurrence Scores (as a continuous were calculated. Statistical significance was assessed via the value) between blocks was 3.0 Recurrence Score units. For likelihood ratio test [20]. In multivariate analyses, reported P 16 of the 20 patients, the standard deviation between blocks Page 5 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. Figure 1 values are for the addition of the given factor to a model includ- ing all other factors. For women with ER-positive tumors, analyses were performed separately for those treated and not treated with tamoxifen. Because ER status was not a matching factor but was associ- ated with our outcome, fewer than half of the ER-negative cases (n = 16) had matched controls who were also ER-neg- ative. We therefore conducted conditional logistic regression analyses among those patients not treated with tamoxifen and generated relative risk estimates for ER-negative patients using terms for interaction with ER status. Finally, we con- ducted analyses of the full study population, using conditional logistic regression and interaction terms for ER status and tamoxifen therapy to obtain relative risk estimates for the differ- ent pre-specified patient groups characterized by these fac- tors. However, because results were not materially different, we present only those results from analyses conducted within patient groups. Methods developed by Langholz and Borgan [21] for nested case-control data were used to estimate the absolute risk of breast cancer death at 10 years and the corresponding 95% CIs. Estimates were calculated for subgroups of ER-positive patients, stratified on tamoxifen treatment and on Recurrence Score, tumor size, and tumor grade. In addition, 10-year risks for breast cancer death were calculated for subgroups of patients, when cross-classified by tumor size and grade and by Recurrence Score and tumor size and grade. The original Langholz and Borgan estimator assumed simple random sam- pling of controls from the set of known eligibles. In this study, we could not confirm eligibility without a review of the medical records, and therefore we sampled potentially eligible controls until up to three were confirmed for study inclusion. Thus, a slight modification to the original absolute risk estimator was necessary in order to reflect our sampling scheme (Additional file 1). T Tw wen enty ty-one -one gen gene e pa pane nel a l an nd c d ca alc lculation ulation of rec of recu urren rrence ce sc score ore. (a) The For comparison with the absolute risk estimates in the Kaiser final gene list (16 cancer-related and five reference genes) and sum- population, the NSABP provided us with the Kaplan-Meier mary score (Recurrence Score) algorithm for this assay were devel- estimates for the probability of breast cancer death at 10 years oped by analyzing the results of three independent preliminary breast among the 668 NSABP B-14 [12] tamoxifen-treated patients cancer studies (that is, training sets) with a total of 447 patients [11]. The Recurrence Score, on a scale from 0 to 100, is derived from the with node-negative breast cancer (previously published reference-normalized expression measurements in four steps. In the results included estimates for distant recurrence, relapse-free first step the expression for each gene is normalized relative to the survival, and overall survival). expression of the five reference genes (b-actin, GAPDH, GUS, RPLPO, and TFRC). Reference-normalized expression measurements range from 0 to 15, where a 1-unit increase reflects approximately a twofold Results increase in RNA. (b) In the second step the HER2 Group Score, the Characteristics of cases and controls ER Group Score, the Proliferation Group Score, and the Invasion Among 4,964 potentially eligible lymph node-negative invasive Group Score are calculated from individual gene expression measure- breast cancer patients, we identified 234 eligible cases and ments. (c) In the third step the Recurrence Score unscaled (RS ) is selected 631 controls with available tumor blocks. After loss calculated using coefficients that were pre-defined based on regres- sion analysis of gene expression and recurrence in the three training of 7.9% of specimens because of insufficient tumor and 1% studies (Providence, Rush, and NSABP B-20 [12]). A plus sign indi- because of failed RT-PCR analysis, a total of 220 cases and cates increased expression is associated with increased recurrence 570 controls were available for statistical analyses. A total of risk. A minus sign indicates that increased expression is associated 142 cases (64.6%) had three controls each, 66 cases with decreased recurrence risk. Page 6 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 Table 3 Relative risks associated with recurrence score among ER-positive patients, stratified by treatment with tamoxifen Score Cases Controls RR (95% CI) P value Tamoxifen treated (55 cases and 150 controls) Recurrence Score Continuous 55 (100%) 150 (100%) 7.6 (2.6–21.9) <0.0001 Pre-specified categories Low risk (<18) 16 (29%) 95 (63%) 1.0 reference Intermediate risk (18–30) 22 (40%) 35 (23%) 4.0 (1.8–8.8) High risk (≥31) 17 (31%) 20 (13%) 6.2 (2.4–15.8) Quartiles 0.0004 1st (0–8.25) 5 (9%) 38 (25%) 1.0 reference 2nd (8.26–14.43) 6 (11%) 37 (25%) 1.0 (0.3–3.7) 3rd (14.44–20.95) 15 (27%) 38 (25%) 2.9 (1.0–8.9) 4th (20.96–78.78) 53 (53%) 37 (25%) 5.8 (2.0–16.6) Tamoxifen untreated (110 cases and 251 controls) Recurrence Score Continuous 110 (100%) 251 (100%) 4.1 (2.1–8.1) <0.0001 Pre-specified categories <0.0001 Low risk (<18) 40 (36%) 160 (64%) 1.0 reference Intermediate risk (18–30) 32 (29%) 47 (19%) 2.7 (1.5–5.0) High risk (≥31) 38 (35%) 44 (18%) 3.3 (1.8–5.9) Quartiles <0.0001 1st (0–7.53) 11 (10%) 63 (25%) 1.0 reference 2nd (7.54–14.25) 19 (17%) 64 (25%) 1.5 (0.7–3.4) 3rd (14.26–21.86) 23 (28%) 62 (25%) 2.1 (0.9–4.9) 4th (21.87–85.82) 57 (52%) 62 (25%) 5.3 (2.5-1.3) 1 2 Conditional logistic regression models include Recurrence Score variables only. Relative risks for Recurrence Score calculated with regard to an increment of 50 units (chosen to be consistent with and comparable with previous studies). Quartiles based on distribution among controls treated with tamoxifen. Quartiles based on distribution among controls not treated with tamoxifen. CI, confidence interval; ER, estrogen receptor; RR, relative risk. (30.0%) had two controls each, and 12 cases (5.5%) had one non-Hispanic white. Approximately two-thirds of patients were control each. The distribution of factors available from the diagnosed with their breast cancer during the first 5 years of tumor registry (age, tumor size, race, diagnosis year, ER sta- the accrual period (1985–1989). Overall, one-third of patients tus, and tamoxifen treatment) among the evaluable cases and were treated with adjuvant tamoxifen. Before 1989, 11% of controls was generally similar to the distribution of these fac- patients were treated with tamoxifen; from 1989 to 1994, tors among the 239 potentially eligible cases and controls 58% of patients were treated with tamoxifen. Among those who were lost to the study. However, both lost cases and lost treated with tamoxifen, the median duration was 4 years; controls were slightly more likely than evaluable patients to be approximately 10% had a year of treatment or less. Compared younger, to be not white, or to have smaller tumors. with controls, cases more commonly had tumors that were ER negative, larger, or more poorly differentiated. Cases were also Breast cancer deaths occurred a median of 4.9 years after more likely to have tumors with higher Recurrence Score val- diagnosis. Among the cases and controls, the median tumor ues. Approximately 50% of patients had Recurrence Score size was 1.5 cm (range 0.2–7.0 cm). Cases and controls were values in the low risk category (that is, Recurrence Score similar with respect to matching factors including age, race, <18). diagnosis year, and treatment with tamoxifen (Table 1). Nearly three-quarters of the study population was aged 50 years or For pre-specified analyses stratified by ER status and older at diagnosis (median age 58 years) and about 80% were tamoxifen therapy, there were 55 cases and 150 matched Page 7 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. Table 4 Relative risks of breast cancer death associated with tumor size, grade, and recurrence score among ER-positive patients, stratified by treatment with tamoxifen Score Cases Controls RR 95% CI P value Tamoxifen treated (55 cases and 150 controls) Multivariate analyses without Recurrence Score Tumor size (cm) 0.009 Continuous 55 (100%) 150 (100%) 2.6 (1.2–5.5) Grade (pathologist 1) 0.007 Well 6 (11%) 50 (33%) 1.0 reference Moderate 28 (51%) 69 (46%) 2.8 (1.1–7.4) Poor 21 (38%) 31 (21%) 4.5 (1.6–12.3) Multivariate analyses with Recurrence Score Tumor size (cm) 0.013 Continuous 55 (100%) 150 (100%) 2.5 (1.2–5.4) Grade (pathologist 1) 0.126 Well 6 (11%) 50 (33%) 1.0 reference Moderate 28 (51%) 69 (46%) 2.4 (0.9–6.4) Poor 21 (38%) 31 (21%) 2.8 (0.9–9.3) Recurrence Score 0.003 Continuous 55 (100%) 150 (100%) 5.3 (1.6–17.2) Tamoxifen untreated (110 cases and 251 controls) Multivariate analyses without Recurrence Score Tumor size (cm) 0.035 Continuous 110 (100%) 251 (100%) 1.8 (1.0–3.1) Grade (pathologist 1) <0.001 Well 17 (15%) 83 (33%) 1.0 reference Moderate 54 (49%) 131 (52%) 2.0 (1.1–3.6) Poor 39 (35%) 37 (15%) 4.7 (2.3–9.9) Multivariate analyses with Recurrence Score Tumor size (cm) 0.036 Continuous 110 (100%) 251 (100%) 1.8 (1.0–3.1) Grade (pathologist 1) 0.018 Well 17 (15%) 83 (33%) 1.0 reference Moderate 54 (49%) 131 (52%) 1.8 (1.0–3.4) Poor 39 (35%) 37 (15%) 3.2 (1.4–7.2) Recurrence Score 0.025 Continuous 110 (100%) 251 (100%) 2.4 (1.1–5.2) 1 2 Conditional logistic regression models include tumor size and tumor grade only. Relative risks for tumor size calculated with regard to an 3 4 increment of 2 cm. Conditional logistic regression models include tumor size, tumor grade, and Recurrence Score. Relative risks for Recurrence Score calculated with regard to an increment of 50 units (chosen to be consistent with and comparable to previous studies). CI, confidence interval; ER, estrogen receptor; RR, relative risk. controls who were tamoxifen treated and had ER-positive tamoxifen. There were 16 cases with matched controls (n = tumors according to RT-PCR assay. There were 110 cases 19) who had ER-negative tumors according to RT-PCR assay and 251 matched controls who had ER-positive tumors and were not treated with tamoxifen (out of a total of 52 ER- according to RT-PCR assay and were not treated with negative cases and 56 ER-negative controls). In addition, a Page 8 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 Figure 2 Figure 3 Rel in Rel ca a ative risks ( tive risks ( lculation of RRs) of dea RRs) of dea the Recurren t th a h a ce ssoci ssoci Score ated w ated wiith gene group scores used th gene group scores used in calculation of the Recurrence Score. Findings are stratified by Rela genes Relative ri tive risks (RRs sks (RRs) for ) for dea deat th a h associ ssociated ated with with expressio expression n of of sin sing gle le tamoxifen treatment and ER status. The position of each symbol indi- genes. Findings are stratified by tamoxifen treatment and ER status. cates the RR. The length of the horizontal line through the symbol indi- The position of each symbol indicates the RR. The length of the hori- cates the 95% CI. The blue boxes indicate the RRs and 95% CIs for zontal line through the symbol indicates the 95% CI. The blue boxes ER-positive patients treated with tamoxifen, the green pyramids indi- indicate the RRs and 95% CIs for ER-positive patients treated with cate the RRs and 95% CIs for ER-positive patients not treated with tamoxifen, the green pyramids indicate the RRs and 95% CIs for ER- tamoxifen, and the red downward pointing triangles indicate the RRs positive patients not treated with tamoxifen, and the red downward and 95% CIs for ER-negative patients not treated with tamoxifen. CI, pointing triangles indicate the RRs and 95% CIs for ER-negative confidence interval; ER, estrogen receptor; RR, relative risk. patients not treated with tamoxifen. CI, confidence interval; ER, estro- gen receptor; RR, relative risk. ble, when categorized into quartiles, or when categorized small number of ER-negative patients were treated with based on pre-specified cutoff points (Table 3). tamoxifen (nine cases and three controls). In ER-positive patients not treated with tamoxifen, the risk of Distribution of Recurrence Score risk categories by breast cancer death was also positively associated with tumor size and tumor grade Recurrence Score (Table 3). As expected, the association of The distributions of tumor size and/or tumor grade in patients Recurrence Score with risk of breast cancer death appeared categorized on the basis of the Recurrence Score as low risk to be stronger among ER-positive patients treated with (Recurrence Score <18), intermediate risk (Recurrence Score tamoxifen than among those not treated with tamoxifen. 18–30), or high risk (Recurrence Score ≥31) are shown in Table 2. The Recurrence Score was associated with tumor Larger tumors and higher grade tumors were associated with size and even more so with tumor grade. Nonetheless, a an increased risk of breast cancer-specific mortality in both number of patients had large (>2 cm) and/or moderately or tamoxifen-treated and -untreated ER-positive patients (Table poorly differentiated tumors with low risk Recurrence Score 4). When the Recurrence Score (continuous) was added to values. In addition, a small percentage had small (≤1 cm) and/ these multivariate models, it provided information on risk that or well differentiated tumors with high risk Recurrence Score was independent of tumor size and tumor grade. This was also values. The concordance in the assessment of tumor grade true when the Recurrence Score was categorized into quar- between the two pathologists was moderate (kappa 0.53, tiles or when it was categorized based on pre-specified cutoff 95% CI 0.49–0.59). points (not shown). Note that even though there were differ- ences in the assessment of grade between the two patholo- Relative risks for breast cancer death: ER-positive gists, similar to observations from other studies [22-26], the patients results were not materially different when tumor grade assess- In ER-positive patients treated with tamoxifen, the risk of ments from pathologist 2, rather than pathologist 1, were used breast cancer death was positively and strongly associated (not shown). with Recurrence Score when analyzed as a continuous varia- Page 9 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. Table 5 Ten-year risk of death in relation to Recurrence Score and tumor size and grade among ER-positive patients, stratified by treatment with tamoxifen Risk classifier Cases Controls 10-Year risk % 95% CI Tamoxifen treated Recurrence Score (55 cases and 150 controls) Low (<18) 29% 63% 2.8 (1.7–3.9) Intermediate (18–30) 40% 23% 10.7 (6.3–14.9) High (≥31) 31% 13% 15.5 (7.6–22.8) Tumor size (55 cases and 150 controls) ≤1 cm 16% 31% 3.6 (1.7–5.5) 1.1–2 cm 44% 43% 5.8 (3.7–7.8) >2 cm 40% 26% 9.3 (5.5–13.1) Tumor grade (55 cases and 150 controls) Well 11% 33% 2.1 (0.7–3.5) Moderate 51% 46% 6.9 (4.7–9.0) Poor 38% 21% 9.9 (6.0–13.7) Tumor size and grade (55 cases and 150 controls) ≤2 cm and well or ≤1 cm and moderate 15% 43% 2.5 (1.2–3.9) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 49% 39% 7.2 (4.9–9.5) >2 cm and moderate/poor 36% 18% 11.5 (6.6–16.2) Recurrence Score, tumor size and grade Recurrence Score low <18 (16 cases and 95 controls) ≤2 cm and well or ≤1 cm and moderate 6% 46% 0.4 (0.0–1.0) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 63% 39% 4.1 (2.0–6.2) >2 cm and moderate/poor 31% 15% 6.9 (1.8–11.7) Recurrence Score intermediate 18–30 (22 cases and 35 controls) ≤2 cm and well or ≤1 cm and moderate 32% 51% 6.8 (2.6–10.9) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 45% 29% 21.9 (5.0–35.8) >2 cm and moderate/poor 23% 20% 13.5 (1.2–24.2) Recurrence Score high ≥31 (17 cases and 20 controls) ≤2 cm and well or ≤1 cm and moderate 0% 10% 12.2 (0.0–31.0) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 41% 60% 10.9 (2.9–18.3) >2 cm and moderate/poor 59% 30% 28.9 (7.4–45.3) Tamoxifen untreated Recurrence Score (110 cases and 251 controls) Low (<18) 36% 64% 6.2 (4.5–7.9) Intermediate (18–30) 29% 19% 17.8 (11.8–23.3) High (≥31) 35% 18% 19.9 (14.2–25.2) Tumor size (110 cases and 251 controls) ≤1 cm 29% 39% 8.5 (6.0–10.9) 1.1–2 cm 41% 45% 10.4 (7.8–12.9) >2 cm 30% 16% 16.4 (11.4–21.0) Page 10 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 Table 5 (Continued) Ten-year risk of death in relation to Recurrence Score and tumor size and grade among ER-positive patients, stratified by treatment with tamoxifen Tumor grade (110 cases and 251 controls) Well 15% 33% 5.5 (3.4–7.4) Moderate 49% 52% 11.2 (8.6–13.8) Poor 35% 15% 22.5 (15.9–28.6) Tumor size and grade (110 cases and 251 controls) ≤2 cm and well or ≤1 cm and moderate 27% 50% 6.2 (4.3–8.0) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 45% 37% 14.2 (10.8–17.4) >2 cm and moderate/poor 27% 13% 19.1 (13.2–24.7) Recurrence Score, tumor size and grade Recurrence Score low <18 (40 cases and 160 controls) ≤2 cm and well or ≤1 cm and moderate 35% 63% 3.0 (1.7–4.3) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 48% 31% 11.0 (7.0–14.8) >2 cm and moderate/poor 17% 6% 20.2 (6.1–32.2) Recurrence Score intermediate 18–30 (32 cases and 47 controls) ≤2 cm and well or ≤1 cm and moderate 34% 38% 17.0 (8.2–24.9) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 38% 34% 20.9 (10.2–30.4) >2 cm and moderate/poor 28% 28% 22.0 (9.3–32.9) Recurrence Score high ≥31 (38 cases and 44 controls) ≤2 cm and well or ≤1 cm and moderate 13% 14% 25.0 (7.9–39.0) >2 cm and well, or 1.1–2 cm and moderate, or ≤2 cm and poor 50% 59% 21.0 (13.8–27.5) >2 cm and moderate/poor 37% 27% 25.3 (15.1–34.3) CI, confidence interval; ER, estrogen receptor. The relationships between the expression of the individual Relative risks for breast cancer death: ER-negative genes that comprise the Recurrence Score and risk of breast patients cancer death were generally similar in tamoxifen-treated and - After adjusting for tumor size and tumor grade, the risk of untreated ER-positive patients (Figure 2). The risk for death breast cancer death was positively associated with the Recur- was positively associated with expression of each of the five rence Score (continuous variable in 50-unit increments) proliferation genes (cyclin B , Ki-67, MYLBL2, STK15, and among ER-negative patients (relative risk 6.2, 95% CI 1.2– survivin). Positive associations were also observed for genes 31.8). Risk for death was also strongly associated with the in the invasion group. The risk for death was negatively asso- proliferation gene group (relative risk 5.1, 95% CI 2.0–13.5). ciated with expression of the ER-related genes (ER, PgR, For some individual genes, associations appeared to be more BCL2, and SCUBE2). However, the associations between strongly positive in the ER-negative than in the ER-positive expression of the ER-related genes and risk were generally patients (that is, Cyclin B , Ki-67, STK15, survivin). For a few stronger for the tamoxifen-treated patients, especially with genes, associations were in different directions for ER-nega- respect to the quantitative expression of the ER gene. Risk for tive and ER-positive patients (that is, BAG1, GSTM1, PgR). death due to breast cancer was not statistically significantly Note that estimates for ER-negative patients were often impre- associated with CD68, HER2, or GRB7 gene expression in cise because of small numbers. patients treated or untreated with tamoxifen. Absolute risk of breast cancer death at 10 years: ER- The associations between expression of the gene group positive patients scores (as calculated in the Recurrence Score) and risk of The risks for breast cancer death at 10 years for ER-positive breast cancer death were generally stronger than those for patients treated with tamoxifen were 2.8% (95% CI 1.7– individual genes (Figure 3). 3.9%), 10.7% (95% CI 6.3–14.9%), and 15.5% (95% CI 7.6–22.8%) for patients with Recurrence Score values in the low, intermediate, and high risk categories, respectively (Table 5). Categories based on tumor size and/or grade identified a Page 11 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. set of patients at similarly low risk of breast cancer death, as well as prognosis. This is most likely explained by the fact although the Recurrence Score was able to identify a substan- that the expression of ER-related genes was more strongly tially larger subgroup of patients. For example, 63% of the con- associated with breast cancer-specific mortality in the trols were identified as low risk by the Recurrence Score, tamoxifen-treated patients than in those not treated with whereas only 31% of controls had a tumor less than 1 cm. tamoxifen (Figure 2), and is also consistent with the estab- Cross-classifying patients by Recurrence Score and tumor lished relationship between ER status of the tumor (by ligand size and grade resulted in very imprecise estimates (that is, binding or immunohistochemistry assay) and response to large CIs). Nonetheless, results suggest that among patients tamoxifen [4,28]. Our results are also consistent with findings with Recurrence Score values categorized as low risk, tumor from a study conducted among participants of the NSABP B- size and grade provide additional risk prediction information. 14 clinical trial, which randomized patients to tamoxifen versus placebo [29]. In this population, the tamoxifen benefit varied by For ER-positive patients not treated with tamoxifen, the abso- Recurrence Score and was greatest for those with low Recur- lute risks for death at 10 years were higher than those for ER- rence Score values. As expected, the strong association positive patients treated with tamoxifen. The risks for breast between the quantitative expression of the ER gene and cancer death were 6.2% (95% CI 4.5–7.9%), 17.8% (95% CI tamoxifen benefit was largely responsible for this finding. 11.8–23.3%), and 19.9% (95% CI 14.2–25.2%) for patients in the low, intermediate, and high risk Recurrence Score Currently, adjuvant hormonal and/or cytotoxic chemotherapy groups, respectively. Again, when cross-classifying patients are recommended for most women with early-stage invasive by Recurrence Score and tumor size and grade, an improve- breast cancer. Treatment decisions are based on axillary node ment in risk prediction was suggested primarily for those with status, age, tumor size, histologic tumor type, tumor grade, Recurrence Score values categorized as low risk. hormone receptor status (ER, PgR), and coexisting medical conditions [4]. Hormonal therapy is recommended for nearly all women with ER-positive tumors. Although tamoxifen is gen- Discussion In this large, population-based study of lymph node-negative erally well tolerated, a significant proportion of women experi- patients not treated with chemotherapy, we found that the ence hot flashes and leg cramps, and up to 20% do not Recurrence Score was strongly associated with risk of breast complete a 5-year course of tamoxifen therapy [30-32]. cancer death among ER-positive patients treated with Despite its potential for serious adverse effects [33], cytotoxic tamoxifen. We also found that the Recurrence Score was chemotherapy has been recommended for most women with strongly associated with risk of breast cancer death among lymph node-positive disease and for node-negative patients ER-positive patients not treated with tamoxifen and among ER- with tumors greater than 1 cm or with unfavorable pathology negative patients. In addition, we found that these associa- [4-6]. Very little information is available to support the use of tions remained after accounting for tumor size and grade, and other clinical or biologic factors in selecting patients for adju- that the Recurrence Score was able to identify a larger subset vant chemotherapy [4]. Most patients with node-negative dis- of patients with low risk of breast cancer death than was pos- ease who receive chemotherapy will not derive benefit, sible with either of these standard prognostic indicators. because they would not go on to have a recurrence even with- out such treatment. New prognostic and predictive tests are Several limitations should be considered when interpreting our needed to better individualize therapy and confine systemic results. We lacked ER status from the medical record for a treatment, especially cytotoxic chemotherapy, to those substantial proportion of patients, and we therefore classified patients who are most likely to benefit. ER status based on gene expression. However, the estimates of relative risk were not materially changed when analyses A growing number of studies suggest that multigene expres- were restricted to the 84% of patients with ER status from the sion assays may be able to provide important prognostic and/ charts (data not shown). Because of the diagnosis years of the or predictive information for breast cancer patients [34-42]. A study, only approximately 30% of patients were treated with variety of approaches and technologies are being used to tamoxifen. Although this is consistent with what has been select genes and characterize expression (for example, cDNA reported for other patient populations during this period [27], microarray chips, RT-PCR) in different types of pathology it limited the numbers of tamoxifen-treated patients for analy- specimens (for example, fresh frozen tissue, formalin-fixed par- sis. Given that the cases and controls were matched with affin embedded tissue). No matter what the approach or tech- respect to tamoxifen treatment, we could not directly examine nology, multiple, well conducted confirmatory studies using whether the Recurrence Score is able to identify patients who standardized methodologies will be needed before the clinical are likely to respond to tamoxifen therapy. We did find a usefulness of any of these assays can be established. stronger association between the Recurrence Score and risk of breast cancer death among patients treated with tamoxifen This is the third study to evaluate the performance of the than among those untreated with tamoxifen, suggesting that Recurrence Score among patients not treated with systemic the Recurrence Score captures response to tamoxifen therapy chemotherapy [12,13]. The three studies used identical pre- Page 12 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 specified scores and laboratory methods and were conducted participating in a clinical trial and among similar patients from among patients who were independent of those used for gene the community setting. In both study populations, the Recur- selection and Recurrence Score algorithm development. The rence Score was able to identify a large subset of patients results of the first study, conducted by Esteva and coworkers (approximately 50% or more) who were at very low risk of [13], differ substantially from our results and from those of the breast cancer death at 10 years. In both studies it was also NSABP B-14 study [12]. In that study no association was observed that although the Recurrence Score was correlated found between the Recurrence Score and risk for distant with tumor size and grade, there were a number of patients recurrence among a series of 149 node-negative patients who with large and/or moderately or poorly differentiated tumors were treated without adjuvant hormonal therapy or chemother- with low risk Recurrence Score values. Results from our study, apy at the MD Anderson Cancer Center between 1978 and and to a lesser extent the NSABP B-14 study, suggest that 1995, who had potentially 5 or more years of follow up, and for combining Recurrence Score, tumor grade, and tumor size whom archived tissue was available. In contrast to the NSABP provides better risk classification than any one of these factors B-14 study, patients in the study conducted by Esteva and alone. Other studies, either retrospective or prospective, will coworkers were not treated with tamoxifen. In contrast to our be needed to confirm our findings among lymph node-nega- study, the outcome of interest was distant recurrence instead tive, ER-positive patients not treated with tamoxifen and of breast cancer death. Although these may explain some of among ER-negative patients. Two studies have been per- the differences observed, it is also possible that the study by formed [43,44] and others are ongoing to assess the relation- Esteva and coworkers included a nonrepresentative group of ship between the Recurrence Score and the magnitude of patients. Patients with poorly differentiated tumors had better chemotherapy benefit. Areas for future research also include prognosis than those with well differentiated tumors, and there the examination of whether the Recurrence Score assay pro- was a suggestion that patients with ER-negative tumors did vides prognostic or predictive information for patients treated better than those with ER-positive tumors. with other hormonal agents or for patients with node-positive breast cancer, whether test performance can be further opti- Our relative risk estimates for the Recurrence Score in ER- mized by even more individualized dissection techniques, and positive, tamoxifen-treated patients are generally quite similar whether the inclusion of additional genes or the inclusion of to those observed in the NSABP B-14 study. Relative risks standard measures (tumor size, grade) may enhance risk pre- associated with expression of individual genes were also very diction overall or for selected patient subgroups. similar. In the NSABP B-14 study, the 10-year risk of breast cancer death was 3.1% (95% CI 1.2–5.0%) in the low risk Competing interests group, 12.2% (95% CI 6.7–17.6%) in the intermediate risk The following authors received support for study-related activ- group, and 27.0% (95% CI 20.4–33.6%) in the high risk ities from Genomic Health, Inc., but have no other competing group. Although very similar to our findings for the low and financial or nonfinancial interests: LAH, MJK, AC, NTB, DG, intermediate risk groups, the 10-year risk for the NSABP B-14 CPQ, BL, and LF. The remaining co-authors (SS, CA, MP, JB, high risk group was higher than ours. Patients in the two stud- MW, DW, and JH) are employees or consultants for Genomic ies had a comparable age distribution, and although those in Health, Inc. the NSABP B-14 study were more likely to have larger tumors, their tumors were also more likely to be well differentiated. Authors' contributions Therefore, there is uncertainty regarding the extent to which LAH participated in the design of the study, directed the data differences in the distribution of prognostic factors in the two analysis, and drafted the manuscript. SS participated in the study populations may explain the difference observed in design of the study, in interpreting results, and in writing the absolute risk estimates for the high risk group. In the Kaiser manuscript. MKJ and AC performed statistical analyses and population, tumor size and tumor grade remained statistically participated in interpreting results. NTB and CA participated in significantly associated with risk of breast cancer death in the development of the study methods and coordinated data most multivariate models that also included the Recurrence collection. MP participated in the development of the study Score, whereas only tumor grade remained independently methods and in data collection. JB participated in the develop- associated with risk in the NSABP B-14 study. Tumor size was ment of the study methods, in interpreting results, and in writ- determined by pathology in the Kaiser study and by clinical ing the manuscript. MW, DW, and JH participated in the examination in the NSABP B-14 study. Clinical examination of design of the study and in interpreting results. DG participated tumor size is generally less accurate and could have resulted in data collection and in interpreting results. CPQ participated in attenuated relative risk estimates. in the study design, co-directed the data analysis, and partici- pated in interpreting results. BL provided methods for estimating cumulative risks and participated in interpreting Conclusion The Recurrence Score has now been shown to be strongly results. LF participated in interpreting results. All authors read associated with risk of breast cancer-specific mortality among and approved the final manuscript. lymph node-negative, ER-positive, tamoxifen-treated patients Page 13 of 15 (page number not for citation purposes) Breast Cancer Research Vol 8 No 3 Habel et al. 7. Bast RC Jr, Hortobagyi GN: Individualized care for patients with Additional files cancer: a work in progress. N Engl J Med 2004, 351:2865-2867. 8. Chang JC, Hilsenbeck SG, Fuqua SA: Genomic approaches in The following Additional files are available online: the management and treatment of breast cancer. Br J Cancer 2005, 92:618-624. 9. Ransohoff DF: Lessons from controversy: ovarian cancer Additional File 1 screening and serum proteomics. J Natl Cancer Inst 2005, A Word document providing additional detail regarding 97:315-319. 10. Simon R: When is a genomic classifier ready for prime time? the statistical analyses for estimating 10-year recurrence Nat Clin Pract Oncol 2004, 1:4-5. probabilities. 11. Cronin M, Pho M, Dutta D, Stephans JC, Shak S, Kiefer MC, Este- See http://www.biomedcentral.com/content/ ban JM, Baker JB: Measurement of gene expression in archival paraffin-embedded tissues: development and performance of supplementary/bcr1412-S1.doc a 92-gene reverse transcriptase-polymerase chain reaction assay. Am J Pathol 2004, 164:35-42. 12. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, Baehner FL, Walker MG, Watson D, Park T, et al.: A multigene assay topre- dict recurrence of tamoxifen-treated, node-negative Acknowledgements breastcancer. N Engl J Med 2004, 351:2817-2826. We should like to thank Kaiser Permanente patients, as well as person- 13. Esteva FJ, Sahin AA, Cristofanilli M, Coombes K, Lee SJ, Baker J, Cronin M, Walker M, Watson D, Shak S, et al.: Prognostic role of nel at the Permanente Medical Group Regional Laboratory and at Kaiser a multigene reverse transcriptase-PCR assay in patients with Permanente Medical Center pathology departments. We also gratefully node-negative breast cancer not receiving adjuvant systemic acknowledge Soonmyung Paik in the Division of Pathology at NSABP therapy. Clin Cancer Res 2005, 11:3315-3319. for performing tissue macrodissection and Fredrick L Baehner and Tracy 14. Gordon NP, Kaplan GA: Some evidence refuting the HMO 'favorable selection' hypothesis: the case of Kaiser George for tumor grading (work funded by Genomic Health, Inc). We Permanente. Adv Health Econ Health Serv Res 1991, 12:19-39. also thank Edith Perez (consult paid by Genomic Health), Clifford Hudis 15. Hiatt RA, Friedman GD: The frequency of kidney and urinary (consult paid by Genomic Health), and Noel Weiss (consult paid by Kai- tract diseases in a defined population. Kidney Int 1982, ser Permanente) for their thoughtful comments on the manuscript. 22:63-8. 16. Krieger N: Overcoming the absence of socioeconomic data in medical records: validation and application of a census-based The study was supported by Genomic Health, Inc. The lead author methodology. Am J Public Health 1992, 82:703-710. (LAH) and several co-authors (MKJ, AC, NTB, DB, LF, and CPQ) are 17. Rothman K, Greenland S: Case-control studies. In Modern Epi- employees of Kaiser Permanente, which received study funds from a demiology Edited by: Rothman KJ, Greenland S. Philadelphia: contract with Genomic Health, Inc. One co-author (BL) was a consultant Lippincott-Raven; 1998:93-114. 18. Elston CW, Ellis IO: Pathological prognostic factors in breast and received fees from Kaiser Permanente. In addition to providing cancer. I. The value of histological grade in breast cancer: financial support, Genomic Health, Inc. co-authors (SS, CA, MP, JB, experience from a large study with long-term follow-up. His- MW, DW, and JH) participated in study design; in the collection, analy- topathology 1991, 19:403-410. sis, and interpretation of data; and/or in the writing of the manuscript. All 19. Cobleigh M: Tumor gene expression predicts distant disease- free survival (DDFS) in breast cancer patients with 10 or more the Genomic Health authors hold stock in the company. Genomic positive nodes: high throughout RT-PCR assay or paraffin- Health, Inc. did not participate in the decision to submit the manuscript embeded tumor tissues [abstract]. Prog Proc Am Soc Clin for publication. As part of the funding contract agreement, this was the Oncol 2003, 22:850. independent decision of Kaiser Permanente study investigators. 20. Breslow NE, Day NE: The analysis of case-control studies. In Statistical Methods in cancer research Volume 1. Edited by: Davis W. Lyon, France: International Agency for Research on Cancer; References 1980:205-210. 1. Ghafoor A, Jemal A, Ward E, Cokkinides V, Smith R, Thun M: 21. Langholz B, Borgan O: Estimation of absolute risk from nested Trends in breast cancer by race and ethnicity. CA Cancer J case-control data. Biometrics 1997, 53:767-774. Clin 2003, 53:342-355. 22. Robbins P, Pinder S, de Klerk N, Dawkins H, Harvey J, Sterrett G, 2. Fisher B, Jeong JH, Bryant J, Anderson S, Dignam J, Fisher ER, Ellis I, Elston C: Histological grading of breast carcinomas: a Wolmark N: National Surgical Adjuvant Breast and Bowel study of interobserver agreement. Hum Pathol 1995, Project randomised clinical trials: Treatment of lymph-node- 26:873-879. negative, oestrogen-receptor-positive breast cancer: long- 23. Davis BW, Gelber RD, Goldhirsch A, Hartmann WH, Locher GW, term findings from National Surgical Adjuvant Breast and Reed R, Golouh R, Save-Soderbergh J, Holloway L, Russell I, et al.: Bowel Project randomised clinical trials. Lancet 2004, Prognostic significance of tumor grade in clinical trials of adju- 364:858-868. vant therapy for breast cancer with axillary lymph node 3. Early Breast Cancer Trialists' Collaborative Group: Polychemo- metastasis. Cancer 1986, 58:2662-2670. therapy for early breast cancer: an overview of the randomised 24. Hopton DS, Thorogood J, Clayden AD, MacKinnon D: Observer trials. Lancet 1998, 352:930-942. variation in histological grading of breast cancer. Eur J Surg 4. Eifel P, Axelson JA, Costa J, Crowley J, Curran WJ Jr, Deshler A, Oncol 1989, 15:21-23. Fulton S, Hendricks CB, Kemeny M, Kornblith AB, et al.: National 25. Theissig F, Kunze KD, Haroske G, Meyer W: Histological grading Institutes of Health Consensus Development Conference of breast cancer. Interobserver, reproducibility and prognostic Statement: adjuvant therapy for breast cancer, November 1–3, significance. Pathol Res Pract 1990, 186:732-736. 2000. J Natl Cancer Inst 2001, 93:979-989. 26. Meyer JS, Alvarez C, Milikowski C, Olson N, Russo I, Russo J, 5. Goldhirsch A, Glick JH, Gelber RD, Coates AS, Senn HJ: Meeting Glass A, Zehnbauer BA, Lister K, Parwaresch R: Cooperative highlights: International Consensus Panel on the Treatment of Breast Cancer Tissue Resource: Breast carcinoma malignancy Primary Breast Cancer. Seventh International Conference on grading by Bloom-Richardson system vs proliferation index: Adjuvant Therapy of Primary Breast Cancer. J Clin Oncol reproducibility of grade and advantages of proliferation index. 2001, 19:3817-3827. Mod Pathol 2005, 18:1067-1078. 6. Carlson RW, Edge SB, Theriault RL: NCCN: Breast cancer. Can- 27. Harlan LC, Abrams J, Warren JL, Clegg L, Stevens J, Ballard-Bar- cer Control 2001, 8:54-61. bash R: Adjuvant therapy for breast cancer: practice patterns of community physicians. J Clin Oncol 2002, 20:1809-1817. Page 14 of 15 (page number not for citation purposes) Available online http://breast-cancer-research.com/content/8/3/R25 28. Harvey JM, Clark GM, Osborne CK, Allred DC: Estrogen receptor status by immunohistochemistry is superior to the ligand- binding assay for predicting response to adjuvant endocrine therapy in breast cancer. J Clin Oncol 1999, 17:1474-1481. 29. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, Watson D, Bry- ant J, Costantino J, Wolmark N: Expression of the 21 genes in the Recurrence Score assay and tamoxifen clinical benefit in the NSABP study B-14 of node negative, estrogen receptor positive breast cancer [abstract]. J Clin Oncol 2005, 23:6s. 30. Partridge AH, Wang PS, Winer EP, Avorn J: Nonadherence to adjuvant tamoxifen therapy in women with primary breast cancer. J Clin Oncol 2003, 21:602-606. 31. Demissie S, Silliman RA, Lash TL: Adjuvant tamoxifen: predic- tors of use, side effects, and discontinuation in older women. J Clin Oncol 2001, 19:322-328. 32. Fink AK, Gurwitz J, Rakowski W, Guadagnoli E, Silliman RA: Patient beliefs and tamoxifen discontinuance in older women with estrogen receptor-positive breast cancer. J Clin Oncol 2004, 22:3309-3315. 33. Partridge AH, Winer EP: Long-term complications of adjuvant chemotherapy for early stage breast cancer. Breast Dis 2004, 21:55-64. 34. van 't Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M, Peterse HL, van der Kooy K, Marton MJ, Witteveen AT, et al.: Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002, 415:530-536. 35. van de Vijver MJ, He YD, van't Veer LJ, Dai H, Hart AA, Voskuil DW, Schreiber GJ, Peterse JL, Roberts C, Marton MJ, et al.: A gene- expression signature as a predictor of survival in breast cancer. N Engl J Med 2002, 347:1999-2009. 36. Huang E, Cheng SH, Dressman H, Pittman J, Tsou MH, Horng CF, Bild A, Iversen ES, Liao M, Chen CM, et al.: Gene expression pre- dictors of breast cancer outcomes. Lancet 2003, 361:1590-1596. 37. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pollack JR, Ross DT, Johnsen H, Akslen LA, et al.: Molecular por- traits of human breast tumours. Nature 2000, 406:747-752. 38. Sotiriou C, Neo SY, McShane LM, Korn EL, Long PM, Jazaeri A, Martiat P, Fox SB, Harris AL, Liu ET: Breast cancer classification and prognosis based on gene expression profiles from a pop- ulation-based study. Proc Natl Acad Sci USA 2003, 100:10393-10398. 39. Bertucci F, Houlgatte R, Benziane A, Granjeaud S, Adelaide J, Tagett R, Loriod B, Jacquemier J, Viens P, Jordan B, et al.: Gene expression profiling of primary breast carcinomas using arrays of candidate genes. Hum Mol Genet 2000, 9:2981-2991. 40. Pittman J, Huang E, Dressman H, Horng CF, Cheng SH, Tsou MH, Chen CM, Bild A, Iversen ES, Huang AT, et al.: Integrated mode- ling of clinical and gene expression information for personal- ized prediction of disease outcomes. Proc Natl Acad Sci USA 2004, 101:8431-8436. 41. Rouzier R, Wagner P, Morandi P, Pusztai L: Gene expression profiling of primary breast cancer. Curr Oncol Rep 2005, 7:38-44. 42. Chang J, Powles TJ, Allred DC, Ashley SE, Clark GM, Makris A, Assersohn L, Gregory RK, Osborne CK, Dowsett M: Biologic markers as predictors of clinical outcome from systemic ther- apy for primary operable breast cancer. J Clin Oncol 1999, 17:3058-3063. 43. Gianni L, Zambetti M, Clark K, Baker J, Cronin M, Wu J, Mariani G, Rodriguez J, Carcangiu M, Watson D, et al.: Gene expression profiles in paraffin-embedded core biopsy tissue predict response to chemotherapy in women with locally advanced breast cancer. J Clin Oncol 2005, 23:7265-7277. 44. Paik S, Tang G, Shak S, Kim C, Baker J, Kim W, Cronin M, Baehner FL, Watson D, Bryant J, Costantino JP, Geyer CE Jr, Wickerham DL, Wolmark N: Gene expression and benefit of chemotherapy in women with node negative, estrogen receptor positive breast cancer. J Clin Oncol 2006, 24:2019-2027. Page 15 of 15 (page number not for citation purposes)

Journal

Breast Cancer Research – Springer Journals

Published: Jun 1, 2006

Keywords: Cancer Research; Oncology; Surgical Oncology

There are no references for this article.