Uniformly positive (>80%) HER2 expression maximizes sensitivity and specificity for prediction of response to trastuzumab in CALGB 9840.

Abstract

Abstract Abstract #6046 Purpose: The recent ACO/CAP guidelines revised the cut-point for HER2 expression by immunohistochemistry (IHC) from 10% up to 30%. Recently we showed that heterogeneity of expression is a function of the assessed biomarker and, for HER2, >80% area of expression was most prognostic and predictive. Here we test this hypothesis in CALGB 9840, a prospective cooperative group clinical trial of Paclitaxel (P) and Trastuzumab (T).
 Experimental Design: Tissue from 274 cases selected from cases with available slides from the TH arm were first scored by pathologists as 0-3+, but then again scored to assess percentage of tissue staining with prospectively chosen cut points of >10%, >30% or >80%. Response was assessed by modified RECIST criteria, including complete and partial response. Stable disease and progressive disease were considered non-response.
 Results: Amongst the 176 cases treated with P+T there were 90 (52%) responses (PR+CR), 3 cases were unevaluable for response. In the subset of cases treated with T, there was inconsistent differences in the >10% vs the >30% categories but the >80% category was uniformly statistically significantly predictive of response. Receiver Operator Characteristic (ROC) curves constructed to assess prediction of response showed areas under the curves (AUC) of 0.59 for IHC and 0.60 for FISH suggesting both are failed tests for defining an optimal cut-point. However, selecting only the cases scored as 3+ shows an AUC of 0.72 and the optimal cut-point is for the test is at 95% positive 3+ HER2 staining. This modification of the cut-point would have denied T to 6 of the 49 patients who responded to therapy in the 3+ group. However, there were also 41 (of 86) patients in the 0,1+ and 2+ groups that responded to therapy, raising the possibility that the response in those 6 patients may have been unrelated to T therapy.
 Conclusions: When using clinical outcome to assess the value of the current tests for prediction of response to T, ROC curves suggest that both the old and new HER2 tests fail at defining an optimal cut-point. However, using only strongly positive cases (3+), an acceptable test can be achieved which is maximized at 95% area of expression. This cut-point, while statistically rigorous, is limited by the fact that the patients were treated with both P and T. This result suggests a similar analysis is warranted on a larger cohort of T treated patients, perhaps in the adjuvant or neoadjuvant setting. The availability of other options for HER2 amplified cases (ie lapatinib) and the increased use of T in the adjuvant setting warrant optimization of tests for the best possible patient selection. Citation Information: Cancer Res 2009;69(2 Suppl):Abstract nr 6046.