Estimating benefit from dose dense adjuvant chemotherapy for early breast cancer in the PANTHER randomized phase 3 trial.

Abstract

521 Background: Adjuvant dose dense chemotherapy for early breast cancer improves patient outcomes across all subgroups compared with standard interval treatment. However, the effect of dose dense treatment based on individual patient risk and over time, and the performance of the PREDICT model at this setting are not well-studied. Methods: This is an exploratory analysis from the phase 3 PANTHER trial (NCT00798070) that compared adjuvant sequential epirubicin/cyclophosphamide (EC) and docetaxel (D) administered in either tailored dose dense (tDD EC/D, every 2 weeks) or standard interval schedule (FEC/D, every 3 weeks) to patients with high-risk resected early breast cancer. Dose tailoring in the experimental arm was performed according to a predefined algorithm based on hematologic and non-hematologic toxicity. We evaluated the performance of PREDICT in terms of calibration and discrimination in the entire cohort and subgroups of interest, conducted a subpopulation treatment effect pattern plot (STEPP) analysis and observed the relative treatment effect over time. Results: The intention-to-treat population consists of n=2003 randomized patients. Median follow-up was 10.3 years (interquartile range, 9.1 – 10.5 years). PREDICT consistently underestimated 5-year (absolute difference 1.4%) and 10-year overall survival (absolute difference 6.6%) in the entire population and across all subgroups, but mostly so in the highest risk patients where the difference between observed and predicted 10-year survival reached 20.4% for patients with tumors larger than 5 cm. Discriminatory accuracy was lower for 10-year than for 5-year survival (AUC 0.728 and 0.786, respectively), although the difference in model performance between the two treatment groups was small. In STEPP analysis, absolute benefit from tDD EC/D in the primary endpoint of breast cancer relapse free survival was stable across risk-defined overlapping subpopulations and ranged from 3.8% in the lowest risk patients to 3.6% in the highest risk ones. There was no evidence of differential treatment effect through time (pinteraction >0.1 for all time-to-event endpoints). Conclusions: In this largely node-positive population, PREDICT consistently overestimated risk for death in all patient subgroups. We could not reliably identify any subgroup not benefiting from dose dense treatment, which should be the standard of care for primary resected high-risk breast cancer. Clinical trial information: NCT00798070 .