The Impact of Prophylactic Axillary Node Dissection on Breast Cancer Survival?A Bayesian Meta-Analysis

To determine whether preoperative doxorubicin and cyclophosphamide (AC) permits more lumpectomies to be performed and decreases the incidence of positive nodes in women with primary breast cancer. Women (n = 1,523) were randomized to National Surgical Adjuvant Breast and Bowel Project (NSABP) B-18; 759 eligible patients received postoperative AC and 747, preoperative AC. The clinical size of breast and axillary tumors was determined before each of four cycles of AC and before surgery. Tumor response to preoperative therapy was clinically complete (cCR), partial (cPR), stable (cSD), or progressive disease (cPD). Tissue from patients with a cCR was evaluated for a pathologic complete response (pCR). Breast tumor size was reduced in 80% of patients after preoperative therapy; 36% had a cCR. Tumor size and clinical nodal status were independent predictors of cCR. Twenty-six percent of women with a cCR had a pCR. Clinical nodal response occurred in 89% of node-positive patients: 73% had a cCR and 44% of those had a pCR. There was a 37% increase in the incidence of pathologically negative nodes. Before randomization, lumpectomy was proposed for 86% of women with tumors or = 5.1 cm. Clinical tumor size and nodal status influenced the physician's decision. Overall, 12% more lumpectomies were performed in the preoperative group; in women with tumors > or = 5.1 cm, there was a 175% increase. Preoperative therapy reduced the size of most breast tumors and decreased the incidence of positive nodes. The greatest increase in lumpectomy after preoperative therapy occurred in women with tumors > or = 5 cm, since women with tumors less than 5 cm were already lumpectomy candidates. Preoperative therapy should be considered for the initial management of breast tumors judged too large for lumpectomy.

Meta-analysis is the process of combining study results that can be used to draw conclusions about therapeutic effectiveness or to plan new studies. We review important design and statistical issues of this process. The design issues include protocol development, objectives, literature search, publication bias, measures of study outcomes, and quality of the data. The statistical issues include consistency (homogeneity) of study outcomes, and techniques for pooling results from several studies. Guidelines are provided to assess the quality of meta-analyses based on our discussion of the design and statistical issues. Limitations and areas for further development of this approach are discussed; researchers should come to a general agreement on how to conduct meta-analysis. As an explicit strategy for summarizing results, meta-analysis may help clinicians and researchers better understand the findings of clinical studies.

In 1971 we began a randomized trial to compare alternative local and regional treatments of breast cancer, all of which employ breast removal. Life-table estimates were obtained for 1665 women enrolled in the study for a mean of 126 months. There were no significant differences among three groups of patients with clinically negative axillary nodes, with respect to disease-free survival, distant-disease--free survival, or overall survival (about 57 per cent) at 10 years. The patients were treated by radical mastectomy, total ("simple") mastectomy without axillary dissection but with regional irradiation, or total mastectomy without irradiation plus axillary dissection only if nodes were subsequently positive. Similarly, no differences were observed between patients with clinically positive nodes treated by radical mastectomy or by total mastectomy without axillary dissection but with regional irradiation. Survival at 10 years was about 38 per cent in both groups. Our findings indicate that the location of a breast tumor does not influence the prognosis and that irradiation of internal mammary nodes in patients with inner-quadrant lesions does not improve survival. The data also demonstrate that the results obtained at five years accurately predict the outcome at 10 years. We conclude that the variations of local and regional treatment used in this study are not important in determining survival of patients with breast cancer.