The US Food and Drug Administration’s use of regular approval for cancer drugs based on single-arm studies: implications for subsequent evidence generation

Abstract

for patients with BRAF V600E mutation in metastatic non-small-cell lung cancer (NSCLC). The approval was made on the basis of a 63% response rate (RR) in a single-arm study. Dabrafenib-trametinib is the second recent regular approval in NSCLC made based on RR-a 50% or greater shrinkage in tumor on serial computed tomography scans. It follows crizotinib (Xalkori; Pfizer Inc., NY, New York) used for patients with ROS1 proto-oncogene receptor tyrosine kinase (ROS1) rearranged metastatic NSCLC [1]. In both cases, what is notable is not that the FDA demonstrated flexibility in making highly active and promising drugs quickly available on the basis of small, single-arm data, but that the agency granted full or regular-not accelerated-marketing authorization for these drugs. Regular approval omits the need for further post-marketing efficacy testing requirements (though not further safety commitments). Here, we summarize the FDA's use of regular approval on the basis of single-arm data, describing potential unintended consequences. Regular versus accelerated approval Currently, there are two forms of marketing authorization granted by the US FDA: accelerated and regular approval. Accelerated approval is given to agents that are expected to provide significant benefit over existing therapies for serious conditions [2] based on their improvement in surrogate markers of disease considered 'reasonably likely' to predict clinical benefit. However, continued approval is contingent on the completion of post-marketing efficacy studies, which provide clearer estimates of the magnitude of benefit on clinical end points. If such studies are not carried out, or if they are negative, the FDA may revoke marketing authorization-as occurred with bevacizumab in breast cancer. In contrast, regular approval is made for drugs with improvement in overall survival, quality or life, or an established surro-gate for these end points [3]. In such cases, drugs can only be removed from the market for unanticipated safety concerns, and the FDA does not mandate post-marketing trials to confirm efficacy. In recent years the FDA has utilized regular approval based on the RR from single-arm studies. In Table 1, we list cancer drugs that received regular approval based on RR since 2008. We believe granting regular approval based on surrogate end points in single-arm trials raises four concerns for public policy. Are these surrogates truly established? In order to grant BRAF and ROS1 inhibitors regular approval, the FDA must consider RR in untreated metastatic NSCLC as an established surrogate end point. To make this claim, a surrogate validation study must be conducted. Using all randomized trials carried out in that indication, a validation study asks whether a change in the surrogate correlates with a change in survival or quality of life. Guidance from the German Institute for Quality and Efficiency in Health Care advocates for the 95% confidence interval of correlation coefficient to be >0.85 for a surrogate to be considered strong and suitable for regulatory purposes. Before designating RR as an established surrogate, the FDA carried out exactly such an analysis in NSCLC. The FDA concluded that 'there was no association between overall survival and objective response rate (R(2) ¼ 0.09; 95% confidence interval , 0-0.33)' [4]. Thus, there is concern that the FDA's own analysis and interpretation does not justify RR in this setting as an established end point. In several other examples in Table 1, the empirical evidence supporting these surrogates is limited or weak [5]. Some may contend that randomized trials for some of these indications may be impossible due to the rarity of molecular subgroups. Yet, though both ROS1 and BRAF represent rare mutations in a common cancer, researchers estimate that, given the 1.5 million new cases of lung cancer annually, tens of thousands of cases globally may be due to such drivers [6]. Without confirmatory studies, the efficacy of a therapy cannot be known Although it is appealing to believe that drugs with very high RRs must confer large benefits to patients regarding survival, quality of life, or both, empirical evidence suggests caution is warranted. In an evaluation of 181 studies, Zia et al. found RRs in phase II trials were on average 12.9% larger than RRs in randomized trials for the same therapies [7]. Furthermore, even drugs with very high (>60%) RRs have revealed unanticipated surprises in randomized testing. Iodine 131 Tositumumab (Bexaar; Corixa Corp., Seattle, Washington) was approved on the basis of a RR of 63% in follicular lymphoma [8]. After a decade of being on the market, the drug failed to show it could improve outcomes in two randomized trials, and the manufacturer voluntarily withdrew it from market [8]. Thus, a high RR is not guarantee of benefit and does not eliminate equipoise around new drugs. In NSCLC, randomized trials of crizotinib for EML4-ALK rearranged cancers have also yielded surprising results. Randomized trials with crossover to crizotinib have shown that though the drug improves progression-free survival, it does not