In this study, the correlation of the efficacy data between small and large clinical trials was investigated to assess the reliability of the efficacy data of small clinical trials. As mentioned above, linear regression analysis indicated a significant linear trend in the median OS, median PFS, and ORR data between small and large clinical trials; correlation analysis showed that the correlation coefficients between both clinical trials were high for all endpoints, and the correlation coefficients of median OS and ORR were >0.85, the threshold used for the assessment of the validation of surrogate endpoints [15]. Thus, our results show that median OS, median PFS, and ORR data of small clinical trials properly reflect that of large clinical trials. Additionally, it is noteworthy that even when the sample size of the small clinical trial is limited (approximately 30–50), the median OS, median PFS, and ORR still accurately reflect the data from large clinical trials. This finding supports the use of the accelerated approval pathway, which is based on promising data obtained from small populations in the early phase. These data suggest that pharmaceutical companies can be confident in the efficacy data of small clinical trials and should not expect more efficacious data in future large clinical trials.
It has been reported that phase III trials of several specific compounds failed despite predecessor trials showing promising results [16, 17]. However, significant differences in the median OS, PFS, and ORR between comparable phase II and III clinical trials of anticancer drugs listed by the FDA were not observed, and their phase III trials failed to show the significant benefit of investigative products, compared with comparators [16]. In the report from Lara and Redman [17], there was a discrepancy in the concomitant medication setting, which potentially affected the efficacy data of the phase II and III clinical trials. In our investigation, we evaluated clinical trials with a more comparable study design, including the combinatory treatment setting, which potentially impacts the efficacy data based on pairing criteria. This could have caused the differing results of our investigation.
Furthermore, Vreman et al. [6] studied the efficacy data between phase II and III clinical trials of anticancer drugs and reported no consistent differences in the median OS, PFS, and ORR data, similarly to the outcome of our investigation. However, while their investigation included only phase II and III clinical trials, our investigation included earlier phase clinical trials, including phase I and I/II. Therefore, the mean sample size of the small clinical trials in our analysis was 40.1, which was smaller than that of the investigation by Vreman et al., which reported a mean sample size of 85. Nevertheless, a correlation between small and large clinical trials was observed. Moreover, Vreman et al. reported only on compounds that were submitted to the European Medicines Agency for approval and may have therefore been biased. The efficacy data of these phase III clinical trials tended to reflect that of phase II clinical trials or produce more efficacious data. In our investigation, the efficacy data of development failure cases were included, and almost half of the samples in the median OS dataset were those of clinical trials of development failure cases. Nevertheless, a significant linear trend and strong correlation were observed in the median OS of small and large trials. Additionally, in a study by Zia et al. [7], a 12.9% higher ORR was observed in phase II than in phase III clinical trials. Meanwhile, our study demonstrated a 1.6% higher ORR in small clinical trials than in large clinical trials. One possible reason for this discrepancy is that we included only eight pairs of clinical trials of development failure cases in the ORR dataset and Zia et al. could have included more compounds whose development was terminated due to the unexpected and inefficacious results observed in phase III clinical trials, compared to those in phase II clinical trials. Another possible cause for the discrepancy is that we chose a more homogeneous patient population. Approximately 27% (13 pairs) of small and large clinical trials included in the ORR dataset enrolled cancer patients with specific antigen or gene mutation expression. Meanwhile, Zia et al. [7] searched for clinical trials investigating chemotherapeutic regimens from 1998 to 2003, and it is the possible that these clinical trials investigated less specific and broader patient populations regardless of specific antigen or gene mutation expression. Therefore, it is possible that they compared the efficacy data of clinical trials that enrolled a more heterogeneous patient population with more inter-clinical trial variability and differences in patient characteristics. This could result in an efficacy gap between phase II and phase III clinical trials.
It is well-known that large clinical trials comparing the investigative product with the standard of care or placebo provide insights into the efficacy and safety of the new treatment option based on the high amount of patient data. However, development of anticancer drugs is active and fast owing to the advanced technology and quickly changing standard of care [18]. Therefore, the need for faster development of anticancer drugs has been growing, and conducting time-consuming, large, randomized phase III clinical trials is no longer ideal for anticancer drug development because it is possible that the standard of care will change before a large clinical trial can be completed, making its findings obsolete. In addition, phase III studies are costly, and a placebo-controlled clinical trial is even more expensive than studies without a control group [11, 19]. Thus, a limited research and development budget provides a challenge to the further development of health technology. As mentioned above, there is an ethical challenge regarding patient enrollment in the comparator arm of randomized, confirmatory clinical trials. In our study, more than 85% of large clinical trials were randomized and a total of 9,150 patients were enrolled in the comparator arm of a placebo-controlled or active-controlled trial. Moreover, our database included 27 large clinical trials that targeted patients who received the treatment previously and experienced disease progression. Therefore, it is expected that a number of patients were hoping to receive new treatment and participated in the clinical trials but were enrolled in the standard of care or placebo arm of the clinical trial previously. We believe that the accelerated approval pathway can help eliminate the challenges of large clinical trials by gaining patients earlier access to novel drugs and reducing the time, cost, and ethical dilemma of confirmatory clinical trials.
Finally, we extracted the data of compounds whose clinical trials in ClinicalTrials.gov were terminated as development failure cases. However, there are also compounds in phase III clinical trials registered as “completed” whose development was stopped or paused by the pharmaceutical company due to unexpected efficacy and safety results in clinical trials. These cases were not evaluated in our datasets as development failure cases.