Supervised risk predictor of breast cancer based on intrinsic subtypes

Abstract

PURPOSE To improve on current standards for breast cancer prognosis and prediction of chemotherapy benefit by developing a risk model that incorporates the gene expression-based "intrinsic" subtypes luminal A, luminal B, HER2-enriched, and basal-like. METHODS A 50-gene subtype predictor was developed using microarray and quantitative reverse transcriptase polymerase chain reaction data from 189 prototype samples. Test sets from 761 patients (no systemic therapy) were evaluated for prognosis, and 133 patients were evaluated for prediction of pathologic complete response (pCR) to a taxane and anthracycline regimen.

Results: The intrinsic subtypes as discrete entities showed prognostic significance (P = 2.26E-12) and remained significant in multivariable analyses that incorporated standard parameters (estrogen receptor status, histologic grade, tumor size, and node status). A prognostic model for node-negative breast cancer was built using intrinsic subtype and clinical information. The C-index estimate for the combined model (subtype and tumor size) was a significant improvement on either the clinicopathologic model or subtype model alone. The intrinsic subtype model predicted neoadjuvant chemotherapy efficacy with a negative predictive value for pCR of 97%. CONCLUSION Diagnosis by intrinsic subtype adds significant prognostic and predictive information to standard parameters for patients with breast cancer. The prognostic properties of the continuous risk score will be of value for the management of node-negative breast cancers. The subtypes and risk score can also be used to assess the likelihood of efficacy from neoadjuvant chemotherapy.

Fig 1.

PAM50 intrinsic subtype prognosis for relapse-free survival (RFS). (A) Outcome predictions according to the four tumor subtypes in a test set of 710 node-negative, no systemic adjuvant therapy patients. (B) Outcome by subtype in the subset of patients with estrogen receptor (ER) –positive disease from Figure 1A. (C) Outcome by subtype in patients with ER-negative disease. (D) Outcome by subtype in HER2clin-positive patients.

Fig 2.

Risk of relapse (ROR) predictions using a test set of node-negative, no systemic therapy patients. (A) C-index analyses of four different Cox models using a test set of node-negative, untreated patients. (B) ROR-C (tumor size and subtype model) scores stratified by subtype. (C) Kaplan-Meier plots of the test set using cut points determined in training. (D) Analysis of the ROR-C model versus probability of survival shows a linear relationship (with the dashed lines showing the 95% CIs). ER, estrogen receptor; RFS, relapse-free survival.

Fig 3.

Relationship between risk of relapse (ROR) score and paclitaxel, fluorouracil, doxorubicin, and cyclophosphamide neoadjuvant response. (A) Receiver operating characteristic curve analysis of ROR-S (subtype only model) versus pathologic complete response (pCR) in the Hess et al test set. (B) ROR-S score versus probability of pCR. PPV, positive predictive value; NPV, negative predictive value; AUC, area under the curve.

Fig A1.

Hierarchical clustering and SigClust analysis of microarray data using 1,906 “intrinsic” genes and 189 samples. The SigClust algorithm statistically identifies significant/unique groups by testing the null hypothesis that a group of samples is from a single cluster, where a cluster is characterized as a multivariable normal distribution. SigClust was run at each node of the dendrogram beginning at the root and stopping when the test was no longer significant (P > .001). Statistical selection using SigClust identified nine significant groups, including the previously identified subtypes designated as luminal A (dark blue), luminal B (light blue), HER2-enriched (pink), normal-like (green), and basal-like (red).

Fig A2.

Focused heatmap of Classification by Nearest Centroids (ClaNC) selected genes for each subtype. The ClaNC algorithm was optimized to select 10 genes per class for a total of 50 genes. The 10 genes for each class are shown as red/green according to their expression in a class. Black indicates that gene was not selected for the given class.

Fig A3.

Heatmap of the centroid models of subtype. The centroids were constructed using the Classification by Nearest Centroids selected genes and calculated as described for the Prediction Analysis of Microarray algorithm. The expression values are shown as red/green according to their relative expression level.

Fig A4.

Developing a continuous risk score based on subtypes and clinical variables. (A) A family of smoothing functions illustrates the general linear relationship between the risk of relapse (ROR) score and relapse-free survival at 5 years. (B) Significance of the slopes in the smoothed lines is plotted for each bandwidth across the range of scores. Blue indicates a significant positive slope, purple indicates nonsignificant slope, and green indicates too few data for inference.

Fig A5.

Risk classification for training set of untreated patients using a combined model of intrinsic subtypes and clinical variables. (A) Risk of relapse based on the combined model with low-risk scores less than 29 (black), moderate-risk scores between 29 and 53 (green), and high-risk scores ≥ 53 (red). (B) Kaplan-Meier plot and significance of the risk score shown for 141 training set cases. LumA, luminal A; LumB, luminal B.

Fig A6.

Analysis of an old-aged formalin-fixed, paraffin-embedded patient cohort. (A) The combined risk of relapse (ROR)-C model predicting high, medium, and low risk of relapse in 279 patients from the University of British Columbia. (B) Receiver operating characteristic curve analysis of quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) data for ESR1 versus immunohistochemistry (IHC) data. (C) Receiver operating characteristic curve analysis for qRT-PCR data for HER2/ERBB2 versus HER2clin-positive (IHC and/or fluorescent in situ hybridization). PPV, positive predictive value; NPV, negative predictive value; AUC, area under the curve.