Neoadjuvant as Future for Drug Development in Breast Cancer—Response Free

We thank Buyse and colleagues (1) for their letter and agree with much of their perspective, most importantly the need to continue to test new regimens in the neoadjuvant setting to further elucidate the relationship between pathologic complete response (pCR) and event-free survival (EFS).

However, we disagree with the comment that there is “no justification for equating” individual-level surrogacy with trial-level surrogacy in the FDA-led meta-analysis. For HER2⁺ breast cancer, the meta-analysis “trial-level” analysis of six trial results actually depends upon patient-level data (Fig. S1 “bubble plot” in Supplementary Appendix; ref. 2). Indeed, one can derive the distributions of the 6 EFS HRs in these trials from the Kaplan–Meier curves describing the relationship between EFS and pCR in the patient-level analysis and the treatment-specific pCR rates within the six trials; the resulting means of these distributions closely mirror the respective bubbles in the plot.

Evaluating the data at the “trial level” introduces variability (always present in clinical trials) and loses important information regarding the relationship between pCR and EFS. In the extreme case, if all six trials had the same pCR OR, there would be no information about the slope of the regression line or the correlation between OR and HR. Indeed, the actual plot shows less than 10% treatment difference in improvements in pCR rates in a cluster of five of the six trials. The lone outlier is NOAH, which therefore determines the slope of the regression line. NOAH was predicted to land where it did by the patient-level analysis, but regardless, there is no power to show statistical significance for the slope of a line through two points.

Although the trial-level analysis reduces the information from the patient-level analysis, it provides an important service. Impressive HRs of EFS by pCR could lead to the erroneous conclusion that statistical significance in pCR will lead to significance in EFS. But the HR of 0.39 for HER2⁺ disease applies only to a treatment that changes the pCR rate from 0% in the control group to 100% in the intervention group (3). NeoALTTO's improvement in pCR rate moves only 19.2% of the patients from the non-pCR curve to the pCR curve, which implies a more realistic but less impressive HR of 0.83.

We agree that the accuracy of this prediction may be a coincidence; one example is hardly conclusive. Our point was simply that the NeoALTTO/ALTTO results were not inconsistent with the FDA's patient-level meta-analysis and therefore were not cause for reconsidering the utility of taking a neoadjuvant approach to drug development. It is also important to recognize that a new therapy may have an entirely different relationship between pCR and EFS. The need for validating or suitably modifying the FDA's patient-level analysis will be essential for any particular therapy, which is precisely what the FDA's guidance requires. Randomized neoadjuvant trials incorporating both pCR and EFS endpoints in the same trial, preferably using an adaptive design based on sample size, provide the best path forward in early-stage breast cancer.

See the original Letter to the Editor, p. 268