Survival among metastatic colorectal patients in clinical trials compared to the real world.

673 Background: Randomized clinical trials play a central role in clinical research though only a small fraction of patients partake in clinical studies. Questions thus arise regarding the generalizability of clinical trial results to the remainder of the population. This study evaluated whether patient survival from randomized clinical trials in metastatic colorectal cancer reflects real world outcomes. Methods: A Pubmed search was used to identify randomized phase III clinical trials of first-line treatment for metastatic colorectal cancer published between 2005 and 2010. We excluded secondary or pooled analyses, second-line treatments, non-metastatic patients, non-English language, and non-randomized studies. Thirty-one clinical trials met these criteria, comprised of 79 distinct clinical trial arms. Overall survival among clinical trial patients was compared to metastatic colorectal cancer patients within the Surveillance, Epidemiology, and End Results (SEER) program. Within SEER, we restricted the analysis time-period and age of patients to match the enrollment period and age of patients within each individual clinical trial. Results: The clinical trials enrolled a total of 16,614 patients. Among all clinical trial arms the median survival ranged from 6.7-62 months, 1-year survival ranged from 30-97%, and 2-year survival ranged from 6-88%. Compared to SEER, the median survival was higher in 95% of the individual clinical trial arms by an average of 5.4 months (p<0.0001). The 1-year survival was higher in 94% of the clinical trial arms by an average of 16.7% (p<0.0001). The 2-year survival was higher in 71% of the clinical trial arms by an average of 7.2% (p<0.0001). Conclusions: This study found substantially improved survival among clinical trial participants compared to patients in the SEER database suggesting that survival estimates from clinical trials may not generalize to the “real world.” Potential patient factors such as differences in underlying comorbidity, performance status, disease burden, as well as variation in treatment could not be addressed in this study, though these factors likely explain some of the observed survival differences.