Inferring the Effects of Cancer Treatment: Divergent Results From Early Breast Cancer Trialists' Collaborative Group Meta-Analyses of Randomized Trials and Observational Data From SEER Registries.
Henson KE., Jagsi R., Cutter D., McGale P., Taylor C., Darby SC.
PURPOSE: To compare the effect of breast cancer radiotherapy as estimated from observational data with findings from randomized trials. MATERIALS AND METHODS: Rate ratios were obtained for selected end points among 13,932 women randomly assigned to receive radiotherapy or not in trials contributing to recent meta-analyses by the Early Breast Cancer Trialists' Collaborative Group. Estimates of the same quantities were derived for 393,840 women registered with breast cancer in the US SEER registries between 1973 and 2008. RESULTS: In the randomized trials, radiotherapy after breast-conserving surgery reduced mortality from both breast cancer (rate ratio, 0.82; 95% CI, 0.75 to 0.90) and all causes (rate ratio, 0.92; 95% CI, 0.86 to 0.99). Reductions of similar magnitude were seen in the trials of radiotherapy after mastectomy in node-positive disease (rate ratios, breast cancer 0.84; 95% CI, 0.76 to 0.94; all causes, 0.89; 95% CI, 0.81 to 0.97). In the observational data, radiotherapy after breast-conserving surgery was associated with much larger mortality reductions (rate ratios, breast cancer, 0.64; 95% CI, 0.62 to 0.66; all causes, 0.63; 95% CI, 0.62 to 0.65), whereas radiotherapy after mastectomy in node-positive disease was associated with substantial increases in mortality (rate ratios, breast cancer, 1.34; 95% CI, 1.31 to 1.37; all causes, 1.23; 95% CI, 1.22 to 1.25). Detailed adjustment of the observational data for potential confounders did not reduce the divergence from the randomized data. CONCLUSION: This study of mortality after radiotherapy for breast cancer found strikingly divergent results between the Early Breast Cancer Trialists' Collaborative Group meta-analyses of randomized data and the SEER observational data, even when efforts had been made to remove confounding and selection biases. Nonrandomized comparisons are liable to provide misleading estimates of treatment effects. Therefore, they need careful justification every time they are used.