P5-14-01: Differences in Efficacy by Assessment Method: NCIC CTG Adjuvant Breast Cancer Trials MA.5, MA.12, MA.14, MA.21, MA.27 Meta-Analysis.

Abstract Background: Based on recent breast cancer literature, we hypothesized that there could be substantive differences in apparent efficacy estimates using a log-normal (LN) survival model rather than with standard Kaplan-Meier (K-M) or Cox model methods. While both Cox and LN survival analyses offer greater specification by individual patient characteristics, the LN model may more robustly estimate survival under model misspecification. Methods: We recently pooled data for 5 NCIC CTG primary breast cancer trials: MA.5, MA.12, MA.14, MA.21, and MA.27. The total patient count for patients who received at least 1 dose of trial therapy is 11,253. Compilation included definition of STEEP endpoints (C Hudis, JCO, 2008) and standardized factor categorizations. The primary endpoint is Breast Cancer Free Interval (BCFI) defined as the time from randomization until recurrence: first local invasive or DCIS; regional, or distant; contralateral invasive or DCIS; or death from breast cancer. We found substantive evidence of non-proportionality for 7 factors compiled for the meta-analyses. In this work, we fit multivariate Cox and LN models with these 7 factors, lymph node status and pathologic T status. We then compare BCFI efficacy estimates for patient and tumour characteristics at 1-, 3-, and 5-years obtained with K-M, Cox, and LN models. Results: There was evidence that the Cox assumption of proportional hazards was violated for 7 factors: age, menopausal status, hormone receptor status, anthracycline use, chemotherapy use, race, and ECOG performance status. Differences between models were intrinsically affected by timing and extent of non-proportionality; there was no consistent pattern. In particular, investigations to date indicate efficacy estimates with absolute differences between K-M, Cox and LN estimates which varied by time of assessment: at 1-year 0.0 to 6.7%, at 3-years 0.4 to 18.6%, and at 5-years 0.2 to 17.0%. BCFI estimates with the K-M were inconsistently closer to those with the LN or Cox model: for K-M to Cox at 1-year 0.4 to 5.2%, at 3-years 0.4 to 15%, at 5-years 0.4 to 14.3%; for K-M to LN at 1-year 0.0 to 6.7%, at 3-years 0.5 to 18.6%, at 5-years 0.2 to 17.0%; for Cox to LN at 1-year 0.8 to 1.8%, at 3-years 1.9 to 6.0%, at 5-years 0.6 to 5.7%. K-M and Cox models have step-wise adjustments at events for K-M and Cox, rather than smooth modeling with the LN. Discussion: Even with reasonably large population subgroups, there were substantive differences in apparent survival (0.0 to 18.6%) between K-M, Cox and LN model types. The magnitude of differences in survival estimates was large enough to be clinically relevant and warrant further consideration as we evaluate new therapies and prognostic/predictive factors. We will be statistically investigating framework robustness under differing levels of model misspecification. Citation Information: Cancer Res 2011;71(24 Suppl):Abstract nr P5-14-01.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.