Correlation between changes in 2- or 3-year disease-free survival (DFS) and 5-year overall survival (OS) in adjuvant breast cancer trials from 1966–2006

581 Background: Although disease-free survival (DFS) is accepted as a valid endpoint in adjuvant breast cancer trials, improvement in 2- or 3-year DFS has never been formally established as an adequate surrogate for 5- or 10-year overall survival (OS). We set out to establish if changes in 2- or 3-year DFS can be used to accurately predict changes in 5- or 10-year OS. Method: We conducted a systematic Medline search for phase III randomized adjuvant breast cancer trials published between 1966–2006 with >100 patients per arm with data for 2- and 3-year DFS as well as 5- or 10-year OS. Only trials of systemic therapies (e.g. chemotherapy, hormonal therapy) were included. We excluded studies investigating the effects of surgery, radiotherapy and neoadjuvant treatment. A univariate regression model weighted by trial sample size was constructed to determine if changes in 2-year DFS between treatment arms within trials were predictive of changes in 5- year OS. Computations of the correlation coefficient, proportion of variation, predicted estimates and 95% prediction interval were undertaken. Results: 126 studies containing 149 treatment arms met the inclusion criteria. Median sample size per trial was 533 and median follow up time was 81 months. Only 26 trials provided 10-year OS data thus association with 10-year OS was not attempted. Results were similar between analyses using either 2- or 3-year DFS, hence only 2-year DFS statistics are reported. 2-year DFS was a significant predictor of 5-year OS regardless of what other covariates were included (p<0.001). For every 1% increase in difference between treatment arms in 2-year DFS, the estimated difference in 5-year OS increased by 0.52%. The proportion of variation explained (R2) ranged from 0.38 to 0.49, with a wide prediction interval. Indeed, if a future trial accrued 1,000 patients and the 2-year DFS in the experimental arm was better than the control by 10%, the 95% prediction interval would still range from -0.2% to 11%. Conclusion: There is a statistically significant correlation, of moderate strength, between changes in 2-year DFS between treatment arms and changes in 5-year OS but the wide prediction intervals mean that the correlation is not strong enough to be used as a surrogate. No significant financial relationships to disclose.