Clinical Trial Evidence Supporting US Food and Drug Administration Approval of Novel Cancer Therapies Between 2000 and 2016

Key Points Question What are the available data on cancer treatment outcomes for new cancer therapies approved by the US Food and Drug Administration? Findings In this comparative effectiveness study of 92 novel cancer therapies approved for 100 indications over 17 years, 44% of drug approvals were based on data from nonrandomized clinical trials. Randomized clinical trials typically reported that these drugs were associated with substantial tumor responses and delays in the time to progression or death, but the median absolute increase in overall survival was only 2 months. Meaning This study’s findings indicate that, at the time of drug approval, limited supporting data are available to decision-makers, and the increase in overall survival associated with new cancer drugs is typically small.


Introduction
Cancer research is characterized by the perceived urgency to develop novel drugs that may improve patients' survival and quality of life. Before patients have access to novel therapies, the available evidence on benefits and harms from clinical trials is assessed by authoritative institutions, such as the US Food and Drug Administration (FDA). Several regulatory programs have been established to expedite the development and approval of drugs for serious conditions, such as cancer. 1 These programs may allow patients to have earlier access to beneficial drugs; however, there is concern that these programs may increase uncertainty in clinical decision-making, as approvals based on these regulations often rely on evidence from fewer and smaller studies, surrogate outcomes, and studies that are more likely to be biased owing to a lack of randomization and adequate controls. [2][3][4][5][6] Previous analyses have described the evidence used to support FDA approval of cancer therapies for periods before 2013, 4,7-10 with a focus on certain types of cancer [11][12][13][14][15] or on the use of certain end points in clinical trials aimed at drug approval. 6, 10 The objective of our study was to systematically investigate the available data on treatment outcomes for all cancer drugs approved by the FDA for the first time between 2000 and 2016. We described the regulatory characteristics and supporting clinical trials and calculated the treatment outcomes of overall survival (OS), progressionfree survival (PFS), and tumor response.

Methods Database
This comparative effectiveness study was performed as part of the Comparative Effectiveness of Innovative Treatments in Cancer (CEIT-Cancer) project. Full details regarding the database and the processes used for data identification, selection, extraction, and handling have been described elsewhere. 16,17 This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline when applicable. 18 We used only published information and aggregated clinical trial-level data. The University of Basel and University Hospital Basel, Switzerland, do not require institutional review board approval for this type of research as the data used were anonymized and not subject to the health regulations of Switzerland (as confirmed by the regulatory team at the Department of Clinical Research, University of Basel).
In brief, we identified all novel drugs and therapeutic biologic therapies (ie, new molecular entities or novel biologic drugs approved for the first indication) that received FDA approval as treatment for any malignant disease for the first time between January 2000 and December 2016.
In this study, we did not consider any additional indications for drugs that received later approval for those indications. We excluded supportive care or imaging drugs that did not produce anticancer activity. We obtained the corresponding FDA approval documents from drugs@FDA, 19 a publicly available database containing information on drug and biologic products approved for human use in the US, and we searched the documents for randomized clinical trials (RCTs) in which the novel drug was compared with some type of active control, placebo, or standard of care and for single-arm clinical trials that may have provided data on treatment benefits. Dose-comparison clinical trials, in which all patients received the novel drug at different doses without the use of any control arm, were considered together with the single-arm clinical trials, merging all doses. We included all RCTs that were explicitly labeled as pivotal and any other RCT that was conducted in the target population and that compared the novel drug with a control that did not contain the novel drug.
In addition, we included single-arm clinical trials that were explicitly described as pivotal or that we inferred were pivotal and essential for drug approval (eg, based on statements such as, "The clinical review of efficacy was primarily based on an analysis of clinical trial 101-09"). 20 We extracted characteristics of the drugs, indications, clinical and regulatory details, and features of all eligible clinical trials. For RCTs, we extracted the reported treatment outcomes for OS, PFS, and tumor response. All steps and extractions were conducted by 2 independent reviewers (A.L. and either A.A., B.K., B.S., F.N., F.S., H.E., J.M.-L., T.S., or T.V.P.; A.L. and A.K.H. for extractions of line of treatment, type of control, and approval pathway). Any disagreement was resolved by consensus or by a third reviewer (A.M.S. or B.K.). Only information on sample size, clinical trial phase, and blinding was extracted by 1 reviewer (A.L.) alone.

Statistical Analysis
We used descriptive statistics to analyze drugs, indications, clinical and regulatory details, clinical trial characteristics, and treatment outcomes. We used only RCTs for the analysis of treatment outcomes because single-arm clinical trials and dose-comparison clinical trials do not provide comparative treatment outcomes for experimental drugs. In five 3-arm RCTs that evaluated the experimental treatment using 2 different doses, we selected the comparison with the later-approved dose. Three studies were considered twice because each was pertinent to 2 indications for the same drug.
We combined treatment outcomes from all RCTs in meta-analyses using random-effects models. 21 We described the statistical heterogeneity using the I 2 statistic. 22 Tumor response rates were presented as unadjusted relative risks (RRs). Odds ratios (ORs) were also reported to address potential differences between measures when events were frequent or rare. We used a continuity correction of 0.5 to account for cases of 0 events. The increase in OS and PFS per study was calculated as the difference between the median OS or PFS of the experimental vs control arms for all indications for which data on the median OS or PFS were available.
Analyses were conducted overall and stratified by cancer type (solid tumors vs hematological cancers), orphan status (with vs without; based on the Orphan Drug Act, 23 which provides incentives for pharmaceutical manufacturers to develop drugs to treat rare diseases), and approval pathway (nonaccelerated vs accelerated; accelerated pathway based on the FDA Accelerated Approval Program, 24 which enables earlier approval of drugs that treat serious diseases and address an unmet medical need). We compared the OS and PFS effect sizes by calculating the ratio of hazard ratios (HRs). We also conducted several sensitivity and subset analyses (eTable 1 and eTable 2 in the Supplement).

Results
We identified 92 novel cancer drugs approved between January 2000 and December 2016 for 100 indications (7 drugs with multiple indications) (

Treatment Outcomes
Across all 54 RCTs with reported treatment outcomes for OS, the combined risk of death associated with any condition across all novel cancer treatments was lower by a mean of 23% compared with the control (HR, 0.77; 95% CI, 0.73-0.81; I 2 = 46%), with a median survival increase of 2.40 months (IQR, 1.25-3.89 months; range, −2.10 to 11.80 months) (Figure 1 and Table 3 (Figure 2 and Table 3  trials. In contrast, drugs for solid tumors, drugs without orphan indications, and drugs without accelerated approval pathways entered the market with evidence that was more frequently obtained from RCTs and larger patient samples.  Relative treatment outcomes were better for surrogate outcomes (PFS and tumor response) than for OS. There was moderate to high statistical heterogeneity across treatment effect sizes, but the overall range of effect sizes was similar across the various subsets. The only drug that indicated absolute survival increases of more than 6 months was olaratumab. In this case, the clinical trial reported a survival improvement of almost 1 year, while PFS was prolonged by only 2.5 months. 25 Notably, a confirmatory clinical trial did not substantiate the OS benefit, 26 and olaratumab was subsequently withdrawn from the market. 27

Limitations
This study has several limitations. First, our analysis was restricted to data presented to the FDA and reported in approval packages. There may be other studies that have evaluated the drugs for these indications. We assumed that a manufacturer would present the most favorable supporting evidence to the FDA. Moreover, our sample not only included clinical trials that were explicitly labeled as pivotal, it included any RCT of the same target population. Thus, because it is more likely that positive results were submitted for approval, there is a low risk that the results from the included clinical trials underestimated the actual benefits of the approved drugs.
approvals, there are substantial concerns about bias in one-half of all randomized approval studies.
Although major sources of bias resulting from a lack of blinding or randomization are reflected in our data, we did not perform a detailed assessment of each study. This limitation is important, as suboptimal controls would produce overestimation of the benefits of the experimental drug. 3

Conclusions
Overall, the data from 17 years of studies conducted for the approval of novel cancer drugs indicate that patients and clinicians typically have limited information available when a novel cancer treatment enters the market; data from RCTs are available for only one-half of indications. This lack of data is even more problematic for patients with hematological cancers. Although these novel therapies may have substantial consequences for tumor size or other markers of tumor response, they were associated with prolonging the life of patients by a median of only 73 days. Our findings suggest that these novel drugs should be used cautiously without the expectation that they will markedly extend survival. Moreover, additional clinical trials performed after a drug is licensed can offer insights on the exact benefit the drug may confer. Many of these drugs were approved to address an unmet medical need. We believe this need still exists.