What is the probability that new cancer treatments are better than standard treatments?

6120 Background: How often new treatments tested in phase III randomized controlled trials (RCT) are superior to standard treatments is not known. To accurately answer this question three factors have to be considered: publication rate, quality of trials and the choice of the adequate comparator intervention. Methods: All consecutive RCTs conducted by 6 National Cancer Institute sponsored cooperative groups from 1955 to 2000 were reviewed. Data on primary outcomes from published and unpublished trials were analyzed. The results were assessed for the possible effects of bias, random error and choice of comparator intervention. Results: We abstracted data from 444 trials, involving 566 comparisons, enrolling more than 150,000 patients. Hazard ratio (HR) for overall survival favored experimental treatments 0.96 [99%CI (0.94, 0.98), p<0.0001]. Treatment related mortality (TRM) was worse with innovative treatments (HR 1.45 [99%CI (1.24, 1.69)], p<0.0001). In absolute difference this amounts to <0.4% in survival and 0.38% in TRM. Although the distribution of successes was on average similar between experimental and standard treatments, we found that one in 14 trials lead to discovery of a treatment that improved survival by 50%. The findings were not affected by publication bias, methodological quality, treatment type, disease, or comparator. Conclusions: In clinical trials of new cancer drugs, experimental treatments in cancer which have progressed to testing in RCTs are, on average, as likely to be inferior as to be superior to standard treatments. Individual treatments may be more or less successful but this cannot be predicted and can only be known after a trial is conducted. This pattern of treatment success is not accidental, but is directly related to moral principle for conduct of clinical trials known as equipoise or uncertainty principle (BMJ2005;331:1295). Such uncertainty makes it easier for patients to decide whether to participate in such trials, and for researchers to justify the clinical trial system, which has led to advances in treatment of cancer. No significant financial relationships to disclose.

Evaluation of publicly-sponsored lung cancer trials in US: Are experimental treatments better than the control ones?

7157 Background: Lung cancer is one of the leading causes of cancer deaths in the US. A considerable number of these patients are treated in randomized clinical trials (RCTs). However, how often experimental lung treatments are superior to control treatments is not known. To accurately assess this, three factors have to be taken into consideration: publication rate, methodological quality of trials and the choice of the comparator intervention. Methods: All phase 3 RCTs that were completed to date by 3 National Cancer Institute cooperative groups (ECOG, NCCTG and RTOG) were eligible for the analysis. We identified 50 RCTs enrolling 11,631 lung cancer patients. The methodological quality of the trials was assessed for possible effects of bias and random error on the outcomes of the trials. The possible impact of the choice of a comparator intervention was also assessed. To evaluate the outcomes, we extracted data on survival (OS), disease free survival (DFS), response rate (RR) and treatment-related mortality (TRM). In addition, the final investigators’ preference about experimental or control interventions was used to assess whether experimental treatments were better than controls. Results: In terms of OS, DFS and RR experimental treatments were as likely as control treatments to be successful—Hazard ratio [HR]= 0.96 (99%CI 0.92–1.01)], [HR = 0.98 (99%CI 0.98–1.17)] and [RR = 1.11 (99%CI 0.84–1.46)], respectively. TRM was worse with experimental treatments [HR = 1.61 (99% CI 1.10–2.36)]. Investigators’ preferences for experimental vs. control treatments were 30% and 70%, respectively (p < 0.001). The quality of trials was high. We did not find any evidence that the methodological quality of trials/choice of comparator influenced the results. Conclusions: We found that there is no clear pattern that predicts which treatment will be better. In aggregate, there is about an equal chance for experimental and standard treatments to result in successful outcomes or that the outcomes may not differ between two types of the treatments, although TRM is slightly higher with experimental arms. This is a welcome finding because if one intervention (experimental or control) was consistently better, maintaining randomization will be difficult. No significant financial relationships to disclose.