PURPOSE: Randomised controlled trials (RCTs) provide optimal evidence to assess the benefits of new treatments. However, clinicians routinely rely on cross-trial comparisons to assess competing treatments when head-to-head randomised comparisons are unavailable. We investigate the validity of cross-trial comparisons using individual patient data (IPD) where patients received the same treatment protocol. We also examine the extent to which statistical adjustment for baseline characteristics can account for inter-trial differences in outcomes. PATIENTS AND METHODS: We used pooled IPD of 378 women with advanced breast cancer assigned to oral cyclophosphamide, intravenous methotrexate and 5-fluorouracil (CMF) in the control arms of three first-line treatment RCTs (ANZ8101, ANZ8614 and ANZ0001) conducted between 1982 and 2001. The Kaplan-Meier method was used to compare progression-free survival (PFS) and overall survival (OS) across trials. Proportional hazard models were constructed to estimate the hazard rates across trials after adjustment for baseline characteristics. RESULTS: The distribution of baseline characteristics varied across trials. There was a statistically significant difference in survival among women treated with CMF in these trials (logrank p=0.009). The median OS were 17.7, 10.3 and 10.1 months for 0001, 8101 and 8614, respectively. The hazard ratios for survival, adjusted for baseline characteristics differences, were 1.44 (8614) and 1.45 (8101) compared to 0001 (p=0.03). PFS did not differ across trials (logrank p=0.38). CONCLUSIONS: Caution should be exercised when interpreting results from historical cross-trial comparisons even if the adjustment of baseline prognostic characteristics can be performed. Cross-trial comparisons have some role in hypothesis-generating, identifying and prioritising promising treatments for further investigation; however RCTs are still essential to guide sound clinical practice. Copyright (c) 2009 Elsevier Ltd. All rights reserved.
PURPOSE: Randomised controlled trials (RCTs) provide optimal evidence to assess the benefits of new treatments. However, clinicians routinely rely on cross-trial comparisons to assess competing treatments when head-to-head randomised comparisons are unavailable. We investigate the validity of cross-trial comparisons using individual patient data (IPD) where patients received the same treatment protocol. We also examine the extent to which statistical adjustment for baseline characteristics can account for inter-trial differences in outcomes. PATIENTS AND METHODS: We used pooled IPD of 378 women with advanced breast cancer assigned to oral cyclophosphamide, intravenous methotrexate and 5-fluorouracil (CMF) in the control arms of three first-line treatment RCTs (ANZ8101, ANZ8614 and ANZ0001) conducted between 1982 and 2001. The Kaplan-Meier method was used to compare progression-free survival (PFS) and overall survival (OS) across trials. Proportional hazard models were constructed to estimate the hazard rates across trials after adjustment for baseline characteristics. RESULTS: The distribution of baseline characteristics varied across trials. There was a statistically significant difference in survival among women treated with CMF in these trials (logrank p=0.009). The median OS were 17.7, 10.3 and 10.1 months for 0001, 8101 and 8614, respectively. The hazard ratios for survival, adjusted for baseline characteristics differences, were 1.44 (8614) and 1.45 (8101) compared to 0001 (p=0.03). PFS did not differ across trials (logrank p=0.38). CONCLUSIONS: Caution should be exercised when interpreting results from historical cross-trial comparisons even if the adjustment of baseline prognostic characteristics can be performed. Cross-trial comparisons have some role in hypothesis-generating, identifying and prioritising promising treatments for further investigation; however RCTs are still essential to guide sound clinical practice. Copyright (c) 2009 Elsevier Ltd. All rights reserved.