Table 13Evidence Table. Clinical utility and validity, H/I ratio

Study, yearMeasureConclusions
Fan, 200679Context: The study is a comparison of classification agreement among 5 different gene expression based predictors (70-genes, Oncotype DX, Wound-response, Intrinsic subtypes and two-gene ratio) using the 295 samples from the consecutive cohort originally used by van de Vijver 2002[reference].Good but not perfect correlation between predictions, albeit surprising with different gene sets. Degree of prediction over and above “standard” clinical stratifiers is not clear - reclassification not done.
  • Cramer V statistics
  • Kaplan-Meier survival analysis
  • Multivariate Cox proportional hazards analysis, adjusting for age, tumor size, tumor grade, ER status, number of lymph node involved
  • All analyses were performed using all the patients, as well as the ER positive patients subset (N=225)
  • NB: coefficients of clinical predictors were allowed to vary between models
  • In multivariate Cox proportional hazards analysis all tests except the two-gene ratio were highly significant predictors of OS and DFS:
 70 genes DFS HR = 3.4 (95%CI = 2.0–6.0), P value < 0.001
 Two-gene ratio DFS HR = 0.91 (95%CI = 0.6–1.3), P value = 0.62
 Oncotype OS HR = 4.3 (95%CI = 2.1–8.9), P value < 0.001
 70 genes OS HR = 4.71 (95%CI = 2.02–11.0), P value < 0.001
 Two-gene ratio OS HR = 1.00 (95%CI = 0.61–1.63), P value = 0.99
 Oncotype DFS HR = 6.14 (95%CI = 1.84–20.4), P value = 0.003
  • ER status, tumor grade, tumor size, and lymph nodes were also significant predictors.
  • Patients classification into dichotomized risks groups proved to be roughly similar, ranging from ~80% to ~40%:
 70-genes and RS yielded a V= 0.60
 70-genes and RS yielded an agreement of ~81% (239/295)
Goetz, 200662Context: Validation of the H/I ratio signature using patients from a randomized trial on tamoxifen treatment.A high 2-gene expression ratio is associated with increased relapse and death in patients with node-negative ER positive breast cancer treated with tamoxifen.
End points:
  • RFS (time from randomization to any event of recurrence, or contralateral breast cancer or death), DFS (time from randomization to any event of recurrence, or contralateral breast cancer, or other cancer, or death), and OS (time from randomization to death)
  • RT-PCR expression values for each gene were normalized using a standard curve obtained analyzing the human universal total RNA (Stratagen, La Jolla, CA); No reference genes were used
  • Specific cut-off points were obtained in the selected population analyzed: none negative (n=130) and node-positive (n=86) patients
  • Cut-off points estimation for the HOXB13:IL-17BR ratio were obtained by minimizing the P value in log-rank tests
  • Multiple testing corrected P values from the log-rank test were obtained by the Lausen and Schumacher method, modify by Altman
  • Log-rank test and univariate Cox proportional hazard models were used to assess whether distributions of RFS, DFS, or OS differed with respect to the following clinical factors: age, extent of surgery, ER status, number of positive nodes, tumor size, Nottingham grade, HER2 status, prior exposure to estrogens
  • Cross-validated HR in Cox proportional hazard models were obtained by the Faraggi-Simon method
  • Multivariate Cox proportional hazard models and the likelihood-ration test were applied to asses the contribution of the dichotomized ratio to the model
  • In multivariate analysis tumor size, nodal status, tumor grade were included in model
  • Nodal status, tumor size and Nottingham grade significantly associated with endpoints
  • All patient in the study:
 RFS, Univariate Cox 1.90 (CI 95% 1.25, 2.90)
 RFS, Univariate F-S 1.62 (CI 95% 1.06, 2.48)
 RFS, Multivariate Cox 1.65 (CI 95% 1.06, 2.57)
 RFS, Multivariate F-S 1.45 (CI 95% 0.93, 2.27)
 DFS, Univariate Cox 2.01 (1.36, 2.96)
 DFS, Univariate F-S 1.69 (1.14, 2.51)
 DFS, Multivariate Cox 1.77 (1.17, 2.66)
 DFS, Multivariate F-S 1.57 (1.04, 2.38)
 OS, Univariate Cox 1.95 (1.25, 3.07)
 OS, Univariate F-S 1.55 (0.98, 2.45)
 OS, Multivariate Cox 1.66 (1.04, 2.66)
 OS, Multivariate F-S 1.29 (0.81, 2.08)
  • Node negative patients in the study (n=130):
 RFS, Univariate Cox 2.22 (1.22, 4.05)
 RFS, Univariate F-S 1.99 (1.09, 3.63)
 RFS, Multivariate Cox 1.98 (1.07, 3.68)
 RFS, Multivariate F-S 1.73 (0.92, 3.25)
 DFS, Univariate Cox 2.41 (1.39, 4.18)
 DFS, Univariate F-S 2.12 (1.22, 3.68)
 DFS, Multivariate Cox 2.03 (1.15, 3.59)
 DFS, Multivariate F-S 1.77 (0.99, 3.16)
 OS, Univariate Cox 2.90 (1.47, 5.72)
 OS, Univariate F-S 2.35 (1.21, 4.58)
 OS, Multivariate Cox 2.40 (1.19, 4.84)
 OS, Multivariate F-S 2.01 (1.02, 3.99)
Jansen, 200772Context: Validation of the H/I signature by measuring expression levels normalized to a different set of control genes respect to Ma et al 2006.61High HOXB13-to-IL17BR ratio expression levels associate with both tumor aggressiveness and tamoxifen therapy failure.
Aims: to evaluate if the HOXB13-to-IL17BR expression ratio predicts response to tamoxifen, and/or cancer intrinsic aggressivenessThe ratio was significantly associated with DFS and PFS in the specific subsets of patients.
End points: disease-free survival (DFS), progression free survival (PFS), post-relapse survival (PRS), and overall survival (OS);In multivariate analysis, the ratio was associated with a shorter DFS for node-negative patients only.
  • Subsets of patients used from the total population (n=1252) for specific end points:
 DFS ER+, node negative primary breast cancer, no adjuvant therapy (N = 468);
 PFS in ER+ primary breast cancer, whose recurrence was treated with first-line tamoxifen (N = 193)
  • Two-gene ratio as dichotomized or continuous variable
  • Non-parametric methods:
 Spearman rank correlations for ER status, PR status, age;
 Kruskal - Wallis exact test for grade, size, LN status;
 Mann-Whitney U test for menopausal status;
  • Univariate and multivariate Cox regression analysis to compute the hazard ratio
  • In multivariate analysis the model for OS and DFS included age, menopausal status, tumor size, lymph node status, grade, and log ER and log PgR mRNA levels
  • The model for PRS and PFS included age, menopausal status, DFS, site of relapse, and log ER and log PgR mRNA levels
  • All 1252 patients, associations with clinical factors, HOXB13:
 Grade, P < 0.001
 ER status, P < 0.001
 PR status, P < 0.001
  • All 1252 patients, associations with clinical factors, I17RB (both 3′ and 5′):
 Age, P < 0.001;
 Menopausal status, P < 0.005 and P < 0.026
 Tumor size P < 0.035 and P < 0.098
 LN status, P < 0.001
 Grade, P < 0.001
 ER status, P < 0.001
 PR status, P < 0.001
  • Multivariate analysis, ER+, node negative, no adjuvant therapy (N = 468):
 Continuous ratio with 3′ I17RB, DFS HR = 1.04; 95% CI = 1.01 to 1.08; P = 0.015
 Continuous ratio with 5′ I17RB, DFS HR = 1.05; 95% CI = 1.02 to 1.08; P = 0.004
 Dichotomized ratio with 3′ I17RB, DFS HR = 1.74; 95% CI = 1.17 to 2.59; P = 0.006
 Dichotomized ratio with 5′ I17RB, DFS HR = 1.61; 95% CI = 1.08 to 2.41; P = 0.019
  • Multivariate analysis, relapsing ER+, tamoxifen treated (N = 193):
 Continuous ratio with 3′ I17RB, PFS HR = 1.07; 95% CI = 1.04 to 1.11; P < 0.001
 Continuous ratio with 5′ I17RB, PFS HR = 1.05; 95% CI = 1.02 to 1.08; P = 0.004
 Optimal dichotomized ratio with 3′ I17RB, PFS HR = 2.97; 95% CI = 1.82 to 4.86; P < 0.001;
 Standard dichotomized ratio with 3′ I17RB, PFS HR = 1.95; 95% CI = 1.39 to 2.73; P < 0.001;
 Optimal dichotomized ratio with 5′ I17RB, DFS HR = 3.31; 95% CI = 2.05 to 5.34; P < 0.001;
Standard dichotomized ratio with 5′ I17RB, DFS HR = 2.12; 95% CI = 1.52 to 2.97; P < 0.001;
Jerevall, 200763Context: Validation of the H/I ratio signature. Expression levels were normalized to b-actin.The ratio or HOXB13 alone can predict the benefit of endocrine therapy, with a high ratio or a high expression rendering patients less likely to respond.
Aims: to evaluate if the HOXB13-to-IL17BR expression ratio can predict the benefit of 5 years versus 2 years of tamoxifen treatment of postmenopausal patients;Lower expression of IL17BR, but not HOXB13, was correlated to several factors related to poor prognosis, IL17BR might be an independent prognostic factor in breast cancer.
End points: Recurrence-free survival (RFS), defined as the time from diagnosis to local, regional, or distant recurrence or death due to breast cancer; OS, defined as the time elapsed from diagnosis to the date of death due to breast cancer;
  • The relationships between grouped variables were analyzed with the chi-squared test, or the chi-squared test for trend when required;
  • Spearman's rank order correlation was used when comparing HOXB13 and IL17BR levels;
  • Survival curves were calculated using the Kaplan-Meier method and analyzed with the log rank test.
  • Multivariate analysis of recurrence and mortality rates was performed with Cox proportional hazard model;
  • A Cox model was also used to test for the interaction between the two-gene ratio and the benefit from prolonged duration of tamoxifen treatment;
  • The following variables were included: HOXB13:IL17BR, treatment (5 vs. 2 years), lymph node status (N+ vs. N-), tumor size ( > 20 mm vs. > 20 mm), PgR status (PgR+ vs. PgR), and an interaction variable (HOXB13:IL17BR × treatment);
  • The ratio is significantly correlated to:
 tumor size, P = 0.003
 ER, P < 0.001
 PR, P < 0.001
 HER2, P = 0.003
 NHG, P < 0.001
 Ploidy, P < 0.001
 S-phase, P = 0.005
 ER, HER2, S-phase and NHG correlations are mostly due to IL17BT
 PR and ploidy correlation have contribution from both genes
  • Post-menopausal ER+ patients, low ratio was associated with a benefit from a prolonged tamoxifen treatment (P = 0.021; in KM analysis for RFS), mainly due to the low expression of HOXB13 genes (P = 0.010, in KM analysis for RFS)
  • Postmenopausal ER+ patients (n=179), multivariate Cox proportional hazard model analysis:
 Recurrence Rate (5y vs 2y), low ratio: 0.39 (CI 95% = 0.17–0.91), P value = 0.030
 Test for interaction: P value = 0.035
 Recurrence Rate (5y vs 2y), HOXB13: 0.37 (CI 95% = 0.17–0.83), P value = 0.015
 Test for interaction: P value = 0.018
  • Postmenopausal ER+, node negative, patients (n=134), multivariate Cox proportional hazard model analysis:
 Recurrence Rate (5y vs 2y), low ratio: 0.27 (CI 95% = 0.10–0.72), P value = 0.0087
 Test for interaction: P value = 0.014
 Recurrence Rate (5y vs 2y), HOXB13: 0.30 (CI 95% = 0.12–0.73), P value = 0.0083
Test for interaction: P value = 0.0081
Ma, 200464Context: This study used the two-gene ratio signature for the development and preliminary validation of the of the two-gene ratio signature prognostic value in patients with early stage breast cancer treated with tamoxifen. The training set was analyzed by microarray (n=60) and by RT-PCR (n=59); the validation set (n=20) was evaluated by RT-PCR (3′ probes used)2-gene ratio predicts tumor recurrence in the setting of tamoxifen therapy.
End points: DFS was calculated from the date of diagnosis.
  • Two-sample t test;
  • Receiver Operating Characteristic (ROC) analysis; Statistical test of significance of ROC curves by the DeLong method
  • Univariate and multivariate logistic regression; All predictors (tumor size, PGR, ERBB2, 2 gene-ratio) used as continuous variables
  • Kaplan-Meier survival analysis and log-rank test
  • Microarray data, training set, two-genes ratio, Univariate logistic regression:
 OR = 10.17 (95% CI 2.9–35.6, p Value = 0.0003)
  • Microarray data, training set, two-genes ratio, Multivariate logistic regression:
 OR = 7.3 (95% CI 2.1–26.3, p Value = 0.002
 This outperforms the other variables
  • Classification results in the validation set (RT-PCR data):
 16/20 correctly classified
 Correct prediction probability = 0.01 (95% CI = 56%–94%), by exact binomial test;
 9/10 disease free patients correctly classified
 7/10 relapse patients correctly classified
  • Kaplan-Meier survival analysis, log-rank test, RT-PCR on the training set: P value = 0.0000058
  • Kaplan-Meier survival analysis, log-rank test, RT-PCR on the validation set: P value = 0.002
Ma, 200661Context: Development of the HOXB13:IL17BR two-gene index, through validation of the HOXB13:IL17BR two-gene ratio signature.The HOXB13:IL17BR index was only significant in node-negative patients.
Aim: to evaluate if the two-gene expression index (HOXB13:IL17BR) could predict outcome in patients treated and untreated with tamoxifen monotherapyHigher HOXB13:IL17BR index was associated with a higher risk of relapse.
End points: Relapse-free survival (RFS), defined as the time from initial diagnosis to any recurrence. Optimal threshold for dichotomization was identified and applied in the analysisTwo-gene index was a significant predictor of clinical outcome in ER+, node-negative, patients irrespective of tamoxifen therapy.
  • Spearman rank correlation was used to assess association between the HOXB13:IL17BR index and the other prognostic factors
  • Cox proportional hazards regression and Kaplan-Meier analyses were used to examine the associations between gene expression indices and RFS
  • Using untreated patients, cut-point selection was tuned to obtained the smallest log-rank P value using values between the 10th and 90th percentile; Cut-point selection was obtained from a training set and validated on a test set; Such cut-off was then used in treated patients
  • Interaction of HOXB13:IL17BR index with other clinical factors in ER+ patients:
 Node status, r = 0.13, P value = 0.0015
 HER2, r = 0.17, P value < 0.0001
 S-phase, r = 0.15, P value = 0.0004
 ER, r = -0.23, P value < 0.0001
 PR, r = -0.26, P value < 0.0001
  • Univariate Cox regression analysis:
 All patients (n=852), Interquartile HOXB13:IL17BR HR = 1.47 p = 9e-06
 Node positive (n = 239), Interquartile HOXB13:IL17BR HR = 1.19 p = 0.1
 Node negative (n = 613), Interquartile HOXB13:IL17BR HR = 1.64, p = 2e-05
  • Kaplan-Meier survival analysis results, ER+, node negative patients:
 Untreated patients, Test set (n=103), P value 0.0012
 Treated patients, (n=122), P value =0.026
  • Multivariate Cox Regression Analysis; ER+ node negative, untreated test set and tamoxifen treated patients (n = 225), dichotomized HOXB13:IL17BR index (high versus low):
 HR = 3.9 (95% CI = 1.5 to 10.3) p value = 0.007
  • Two-gene index on a continuous scale, 5-year recurrence risk for untreated patients:
 Index of -2.0 = 15% (95% CI, 9.8% to 20.5%)
 Index of +2.0 = 36% (95% CI, 26.5% to 45.2%)
Reid, 200569Context: Validation of the H/I ratio signature using an independent cohort of 58 ER+ patients treated with tamoxifen. Re-analysis of the original microarray data and of an independent data setAlthough the proposed predictive model is very appealing the use of the two-gene ratio signature in an independent population yielded statistically non-significant results.
End points: DFSThe authors failed to confirm he association of the ratio with response to tamoxifen on their cohort (which is however different in terms of clinical characteristics from the original Ma, 200464 cohort). The authors also failed to classify patients using Discriminant Linear Analysis on two published data sets, including the Ma, 200464 original series.
  • RT-PCR measurement of HOXB13 and IL17BR gene expression (5′ probes used)
  • Univariate logistic regression
  • Area Under the receiver-operating-Characteristic curve (AUC)
  • Two-sample t test and Mann-Whitney test
  • Univariate logistic regression: odds ratio:
 HOXB13 OR = 1.04, 95% CI = 0.92 to 1.16, P = 0.54
 IL17BR OR = 0.69, 95% CI = 0.40 to 1.20, P = 0.18
 HOXB13/IL17BR OR = 1.30, 95% CI = 0.88 to 1.93, P = 0.18
  • Similar results by the other methods
  • About Ma 2004 validation data:
 9/10 disease free patients were correctly classified; Correct prediction probability 0.02 (95% CI = 52%–99%)
 7/10 relapse patients correctly classified; Correct prediction probability = 0.34 (95% CI = 35%–93%)

ER = estrogen receptor; OS = overall survival; DFS= disease free survival; HR= hazard ratio; CI= confidence interval; RFS = recurrence free survival; RT-PCR= reverse transcriptase polymerase chain reaction; RNA= ribonucleic acid; HER-2 = human epidermal growth factor receptor 2; PFS = progression LN = lymph node; mRNA= messenger ribonucleic acid; NHG= Nottingham histologic grade; ROC= receiver operating characteristic; PGR= progesterone receptor gene; AUC= area under the receiver operating characteristic curve

From: Appendix I: Evidence Tables

Cover of Impact of Gene Expression Profiling Tests on Breast Cancer Outcomes
Impact of Gene Expression Profiling Tests on Breast Cancer Outcomes.
Evidence Reports/Technology Assessments, No. 160.
Marchionni L, Wilson RF, Marinopoulos SS, et al.

NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.