Abstract P6-06-03: Differing patterns of treatment effect in a trial assessing sequential docetaxel as adjuvant chemotherapy for early breast cancer (TACT)

Abstract

Abstract Background : Plausible underlying treatment effects were investigated in exploratory analysis of the translational component of TACT, a large adjuvant chemotherapy trial (Lancet 2009:373, 1681). Some breast cancer patients are cured by primary therapy and are therefore not at risk of relapse, potentially confounding treatment comparisons using Cox regression, exploratory comparison of outcomes can be undertaken using parametric cure models (PCM) which estimate the proportion of patients cured and the pattern of events in patients not cured. Methods : TACT compared sequential anthracycline-taxane (FEC-D) adjuvant chemotherapy with an anthracycline regimen of similar duration. As part of TransTACT ER Allred scores were assessed centrally on two cores per patient by two pathologists (CG and SP). PCM was implemented using the CUREREGR command in Stata (Buxton A 2007), a lognormal model for time to relapse being fitted for the category of patients who were not cured. The primary endpoint for the analyses was relapse free interval. Results : 2892 patients had Estrogen Receptor (ER) assessed centrally, 1851 tumors were ER positive and 1041 were ER negative, median follow up was 8 years. The proportion of patients cured was related to established prognostic characteristics such as nodal status. In unselected ER+ patients evidence of cure was weak but it could not be ruled out in some subgroups e.g. in node negative ER+ patients the estimate of the proportion cured was approximately 70% (p = ns). In unselected ER- patients the pattern of relapse implied by PCM was found to differ significantly between taxane treated and control patients (p<0.0001), despite the fact there was no overall evidence of a treatment effect (Hazard ratio 0.98, p = 0.71 by Cox Regression). In taxane treated patients the estimate of the proportion of patients cured was 65% (95%CI:60-69), and the estimated median time to relapse in those not cured was 1.9 years (95%CI:1.7-2.1), in the control group the estimates were 45% (95%CI:29-62) cured and in patents not cured median time to relapse 4.2 years (95%CI:2.1-8.4). Conclusions : The concept of cure is of particular interest in tumour subtypes, such as ER negative tumours, in which event rates are high in early years (1-4) but low subsequently. In ER- patients PCM suggested the model-defined cure rate was about 20% higher in docetaxel treated patients, however this was counterbalanced by worse prognosis in the patients without model defined cure. EBCTCG results examining the taxane effect on recurrence were not directly available to support this finding but breast cancer mortality curves in more than 35,000 women with follow up to 8 years (Lancet 379, 432 Webappendix, p15) showed patterns in ER- patients that were consistent with a higher cure rate in taxane treated patients, in addition there was little evidence of cure in unselected ER+ patients. In this study PCM identified differences in the treatment effect between the randomised groups despite the fact there was no evidence of an overall difference based on the hazard ratio by Cox regression, suggesting PCM may have a useful role in exploratory analysis to look for treatment effects in situations where cure is a possibility. Citation Information: Cancer Res 2013;73(24 Suppl): Abstract nr P6-06-03.