The recent article in the Journal by Mauri et al. ( 1 ) of a meta-analysis investigating the impact of aromatase inhibitors on survival provides an opportunity to explore some relevant issues in summarizing data from randomized trials. A synthetic approach, such as meta-analysis, should consider all randomized trials specifically designed to compare different strategies in a particular clinical scenario ( 2 ). This approach should exclude all studies not designed for a comparison (i.e., phase

more »

... randomized studies). It should also avoid any comparison with a small sample size or possible selection bias (e.g., for activity and safety) that could confound any survival result. It is unclear whether a meta-analysis approach could overcome these problems by use of individual patient data that were analyzed by pooling all single-patient survival data of each trial at an updated follow-up. From this perspective, the inclusion in the meta-analysis of randomized phase II studies that were not adequately powered for any survival analysis may introduce uninformative data into the analysis. Furthermore, a sort of melting pot of various clinical settings was explored by the authors of the metaanalysis ( 1 ), given that the primary objectives of trials that aimed to assess efficacy of hormonal therapy for advanced breast cancer as first-, second-, and third-line treatments were different. Indeed, because of the crossover design (which is relatively common in this strategy) and the impact on survival from second-line (and sometimes third-line) treatment in specific clinical situations, it has been suggested ( 3 ) that overall survival should not be considered as the primary endpoint in first-line hormonal therapy for advanced breast cancer. In a palliative setting such as metastatic disease, any advantage (whatever the outcome studied) should be balanced with the safety profile. For these reasons, specific outcomes for each drug class and/or generation should be followed by cumulative toxicity analyses ( 4 , 5 ). Although abstracts of unpublished trial results cannot provide definitive data, the first-line European Organization for Research and Treatment of Cancer phase III trial with the aromatase inactivator exemestane should have been included in the overall analysis to avoid publication bias in such a literature-based meta-analysis ( 6 , 7 ). Indeed, a previous meta-analysis that used survival as secondary endpoint did not find that a statistically significant survival benefit was associated with aromatase inactivator therapy ( 5 ). Finally, from a practical clinical perspective, the inclusion in metaanalyses of arms that use nonapproved drug dosages could interfere with the approval of that drug for use in standard practice. EMILIO BRIA PAOLO CARLINI FRANCESCO COGNETTI EDMONDO TERZOLI DIANA GIANNARELLI References (1) Mauri D , Pavlidis N , Polyzos NP , Ioannidis JPA . Survival with aromatase inhibitors and inactivators versus standard hormonal therapy in advanced breast cancer: meta-analysis . J Natl Cancer Inst 2006 ; 98 : 1285 -91 . (2) Pignon JP , Hill C . Meta-analyses of randomized clinical trials in oncology . Lancet Oncol 2001 ; 2 : 475 -82 . (3) Di Leo A , Bleiberg H . Overall survival is not a realistic endpoint for clinical trials of new drugs in advanced solid tumors: a critical assessment based on recently reported phase III trials in colorectal and breast cancer . J Clin Oncol 2003 ; 21 : 2045 -7 . (4) Carlini P , Bria E , Giannarelli D , Ferretti G , Felici A , Papaldo P , et al . New aromatase inhibitors as second-line endocrine therapy in postmenopausal patients with metastatic breast carcinoma: a pooled analysis of the randomized trials . Cancer 2005 ; 104 : 1335 -42 . (5) Ferretti G , Bria E , Giannarelli D , Felici A , Papaldo P , Fabi A , et al . Second-and thirdgeneration aromatase inhibitors as fi rst-line endocrine therapy in postmenopausal metastatic breast cancer patients: a pooled analysis of the randomised trials . Br J Cancer 2006 ; 94 : 1789 -96 . (6) Paridaens R , Therasse P , Dirix L , Beex L , Piccart M , Cameron D , et al . First line hormonal treatment (HT) for metastatic breast cancer (MBC) with exemestane (E) or tamoxifen (T) in postmenopausal patients (pts) -a randomized phase III trial of the EORTC Breast Group. CORRESPONDENCE Response Bria et al. assert that phase II trials should not be included in analyses of survival simply because these trials do not have survival as the primary objective but only as a secondary objective. However, this point goes against the fundamental concept and practice that a meta-analysis should be systematically inclusive. An inclusive approach provides the best chance to explore diversity (i.e., heterogeneity) in the trial results ( 1 ). We strongly believe that overall survival should be considered as a key endpoint, especially in meta-analysis of several clinical trials. Crossover is unavoidable: trials that deny patients other available treatments when they have had disease progression on the originally assigned treatment may be not only unrealistic but also occasionally unethical. In the intention-to-treat principle, survival is an outcome that makes most sense, and a meta-analysis of many trials is the best opportunity to assess this outcome. The notion that meta-analysis can only address the outcomes that were primary outcomes in the combined trials is a misunderstanding of the method. We fully agree with Bria et al. that efficacy should be balanced against safety. We did make this point in our paper, and we have previously voiced serious concerns about the quality of harms data from clinical trials and about the need for its improvement ( 2 ). As for aromatase inhibitors, we were fortunate to have a previously published meta-analysis of toxicity outcomes, which we did cite ( 3 ).