Abstract No. 
2014 San Antonio Breast Cancer Symposium
December 9-13

I-SPY 2 qualifying biomarker evaluation (QBE): The challenge and opportunity for interrogating predicted pathways in an adaptive design biomarker rich trial

Yau C, Wolf D, Sanil A, van 't Veer L, Petricoin EF, Buxton M, Gray J, DeMichele A, Hogarth M, Hylton N, Perlmutter J, Paoloni M, Symmans F, Yee D, Berry D

I-SPY 2, a multicenter phase 2 neoadjuvant trial in high-risk breast cancer, uses adaptive randomization within biomarker subtypes to evaluate novel agents added to standard chemotherapy. In addition to efficiently evaluating agent/signature pairs, I-SPY 2 is a biomarker rich trial, where samples are profiled for gene expression, protein levels, and mutation status. Biomarkers are classified as established, qualifying, or exploratory. Established biomarkers are those used clinically (HR/HER2 status) or FDA cleared (MammaPrint), and,used for adaptive randomization to generate the 10 signatures from which a drug can graduate. Qualifying biomarkers (QB) represent evidence-based, biologic pathway markers (e.g.cell line predictors, known drug targets). QB analyses must be pre-specified and performed under CLIA. Exploratory markers are for discovery and may allow integration of data from different technologies.

The QBE goal is to (1) evaluate biomarkers related to an agent’s mechanism of action to identify promising candidates for testing/patient selection in future trials, and (2) create a resource to elucidate biological mechanisms of response.

The wealth of biomarker data is both a boon and a challenge. Our small size limits the generalizability of our findings. There are multiple genes in each pathway measured on multiple platforms, creating the problem of multiplicity, which is compounded by the evaluation of multiple proposals. Biomarkers may correlate with HR/HER2/MP subtypes. The adaptive randomization may increase the prevalence of biomarker positive subsets and bias our findings. These challenges limit definitive conclusions, so our statistics are descriptive rather than inferential, and are intended to avoid adding to the false positive biomarker literature.

Methods: Three filters are applied: 1-The difference in biomarker performance in the experimental vs control arm (biomarker x treatment interaction) is evaluated using a logistic model under a pre-specified analysis plan 2-Biomarkers with a treatment interaction are dichotomized. The QB-High group is added to the graduating subtype to define a novel signature and the treatment effect in this group is evaluated 3-If the treatment effect is comparable to the graduating signature, and the prevalence is increased, the I-SPY 2 Bayesian model is modified to include the QB to assess the novel signature.

QBE to date: Veliparib in combination with carboplatin (V/C) and neratinib (N) are the first two agents to graduate from I-SPY 2. For V/C, we have completed initial evaluation for 5 biomarker proposals, including BRCA1/2 germline mutations and expression signatures associated with DNA repair deficiencies. For N, 6 biomarker proposals, including HER family protein signaling markers, have been assessed. Evaluation of the best candidates from these initial analyses in the I-SPY 2 Bayesian framework is ongoing. Mutational analyses are pending.

Conclusions: We have developed a rigorous approach for QB analysis. A small number of QB warrant further assessment. However, I-SPY 2 QB require validation, and should be considered preliminary efforts to effectively screen QB candidates for evaluation in ongoing and future trials.

View original