Treatment Efficacy Score—continuous residual cancer burden-based metric to compare neoadjuvant chemotherapy efficacy between randomized trial arms in breast cancer trials

Background: Difference in pathologic complete response (pCR) rate after neoadjuvant chemotherapy does not capture the impact of treatment on downstaging of residual cancer in the experimental arm. We developed a method to compare the entire distribution of residual cancer burden (RCB) values between clinical trial arms to better quantify the differences in cytotoxic efficacy of treatments. Patients and methods: The Treatment Efficacy Score (TES) reflects the area between the weighted cumulative distribution functions of RCB values from two trial arms. TES is based on a modified Kolmogorov–Smirnov test with added weight function to capture the importance of high RCB values and uses the area under the difference between two distribution functions as a statistical metric. The higher the TES the greater the shift to lower RCB values in the experimental arm. We developed TES from the durvalumab + olaparib arm (n = 72) and corresponding controls (n = 282) of the I-SPY2 trial. The 11 other experimental arms and control cohorts (n = 947) were used as validation sets to assess the performance of TES. We compared TES to Kolmogorov–Smirnov, Mann–Whitney, and Fisher's exact tests to identify trial arms with higher cytotoxic efficacy and assessed associations with trial arm level survival differences. Significance was assessed with a permutation test. Results: In the validation set, TES identified arms with a higher pCR rate but was more accurate to identify regimens as less effective if treatment did not reduce the frequency of high RCB values, even if the pCR rate improved. The correlation between TES and survival was higher than the correlation between the pCR rate difference and survival. Conclusions: TES quantifies the difference between the entire distribution of pathologic responses observed in trial arms and could serve as a better early surrogate to predict trial arm level survival differences than pCR rate difference alone.