Association Between Survival Benefits of Anticancer Drugs and Probability of US Food and Drug Administration Approval: A Systematic Review.

Key Points

Introduction

Improving patient survival is a fundamental requirement for novel anticancer drugs [ 1 , 2 ]. The American Society of Clinical Oncology (ASCO) articulated that relative improvements in median overall survival (OS) by at least 20% are necessary for clinically meaningful effect [ 3 ]. Similarly, the European Society for Medical Oncology (ESMO) used OS to evaluate clinical benefits on a standardized scale for the magnitude of clinical benefit (ESMO-MCBS) [ 4 , 5 ]. The World Health Organization (WHO) requires at least a 4- to 6-month increase in OS for essential medicine [ 6 ]. Although these attempts clarified the expected effect size from medical societies or public health, there have been criticisms about approved anticancer drugs showing only marginal survival benefits [ 7 ].

In the new drug approval process, the United States (US) Food and Drug Administration (FDA) individually evaluates the clinical data for the investigational drug, considering the specific therapeutic context. Likely attributable to these individual-based reviews, the survival benefits of pivotal trials for FDA-approved drugs are broadly distributed. Some trials showed marginal survival benefits, whereas others showed significant improvement in OS [ 8 – 12 ]. Moreover, several drugs that received accelerated approval recently failed to demonstrate survival benefits in the post-marketing studies [ 13 , 14 ]. Although anticancer drug development is characterized by the urgency of the need for new drugs, the tradeoff of the quality of evidence has been questioned [ 15 – 17 ]. The bar for regulatory approval for anticancer drugs should be maintained high to ensure their positive impact on public health. Therefore, it is critical to understand the threshold of survival benefit sufficient to grant regulatory approval.

Several previous studies explored the performance of FDA-approved drugs through systematic reviews [ 3 , 18 – 20 ]. They consistently reported that a significant fraction of the FDA-approved anticancer drugs did not exceed the OS benefit expected by the ASCO or the ESMO, and the regulatory approval relied on the surrogate endpoints, e.g., progression-free survival (PFS). However, these studies collected data from only FDA-approved drugs; thus, the perspective of comparing FDA-approved and non-approved drugs has remained intact. This study was designed to collect data on both FDA-approved and non-approved drugs and explore the boundary between them.

Materials and Methods

This study collected data from publicly available databases and journals; therefore, institutional review board approval was not required per 45 Code of Federal Regulations (CFR) §46. This study was conducted with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement [ 21 , 22 ]. The study protocol was not prospectively registered in PROSPERO or any other public registry. Figure 1 shows the data collection flow, and Online Resource Table S1 details the data source for individual studies (see the electronic supplementary material).

Eligibility Criteria

Trial eligibility was determined in accordance with the PRISMA statement. We included randomized phase 3 trials of novel anticancer agents that utilized OS as a primary or co-primary endpoint. To ensure data robustness for quantitative analysis, included trials were required to provide published clinical results, specifically reporting the hazard ratio (HR) for OS.

Several exclusion criteria were applied to ensure the analysis focused strictly on FDA regulatory outcomes for new molecular entities. First, we excluded trials that were not intended for a formal New Drug Application (NDA) or Biologics License Application (BLA), such as those focused on biomarker development, surgical techniques, or diagnostic procedures. Studies investigating new combination regimens consisting solely of previously approved agents were also excluded, as they were not directed toward the initial approval of a new therapeutic agent.

Second, to ensure the relevance of the clinical evidence to the FDA’s review process, trials conducted exclusively outside the US (e.g., regional trials limited to China or other specific territories) were excluded. Third, we excluded trials with non-inferiority designs for OS, focusing instead on superiority trials for OS. Fourth, trials using surrogate time-to-event endpoints—such as PFS, event-free survival (EFS), or disease-free survival (DFS)—as the sole primary endpoint were excluded; all included trials were required to utilize OS as a primary or co-primary endpoint. Finally, we excluded trials that did not report a 95% confidence interval (CI) for the OS HR, as this information was essential for our quantitative analysis.

Information Sources, Search Strategies, and Selection Process

We identified relevant clinical trials through a comprehensive search of ClinicalTrials.gov. Detailed search parameters are summarized in Online Resource Table S2 . The search was conducted for phase 3 investigational studies initiated between January 1, 1990, and January 1, 2021. This timeframe was selected to encompass major breakthroughs in oncological drug mechanisms, including molecularly targeted therapies and immune checkpoint inhibitors (ICIs), which have significantly improved survival for patients with cancer [ 23 ]. To ensure a broad initial capture, the search criteria included all study statuses (e.g., active, completed, terminated) and used the keyword “survival” within the outcome measures.

Data collection was initiated in May 2021, identifying 3463 trials. The structured dataset from ClinicalTrials.gov, including all available fields (e.g., National Clinical Trial [NCT] number, study design, and outcome measures), was used to screen each trial against our pre-specified eligibility criteria. During this identification phase, 2209 trials without posted results and 116 non-randomized controlled trials were excluded.

For the remaining 1138 candidate trials, we excluded 161 non-oncology studies. For the 977 oncology-related trials, targeted searches were performed in PubMed using the NCT identifier to locate corresponding peer-reviewed publications. We excluded 267 trials for which no full-text publication was available. The study included all relevant papers published through August 2021. Upon retrieving the full-text articles, for the remaining 710 trials, we performed a rigorous manual review. This process led to the exclusion of 131 non-therapeutic or non-US trials, 21 non-inferiority trials, 332 trials not powered for OS, and 37 trials with insufficient statistical data (e.g., missing 95% CI for OS HR). Consequently, 189 eligible individual trials comprising 208 arm-pairs were identified for inclusion.

The entire data collection and screening process was conducted by one investigator (KI) and independently verified by a second investigator (SO). Any disagreements regarding study eligibility or data categorization were resolved through thorough discussion until a consensus was reached. No automated data extraction tools were utilized, and no additional information was sought from trial investigators.

Determination of Regulatory Status and Unit of Analysis

For each identified arm-pairs, the target indication—including cancer type and line of therapy—was meticulously verified. We accessed the Drugs@FDA database to retrieve the United States Prescribing Information (USPI) and FDA Approval Packages, including the Summary Review, for the respective agents. By cross-referencing these regulatory documents with trial characteristics, we confirmed whether each specific trial served as a pivotal study for the approved indication. A trial was categorized as "FDA-approved" only if the approval strictly matched the study population and the line of therapy investigated in the trial; otherwise, it was classified as "non-approved." This rigorous adjudication ensured that approval status was accurately assigned at the trial-and-indication level, avoiding misclassification of trials involving drugs already approved for other indications.

Regarding the unit of analysis, for multi-arm trials involving three or more groups, each treatment–control arm-pair was treated as a distinct analytical unit. For example, if a trial compared both “new drug A + drug B vs. control” and “new drug A monotherapy vs. control,” these were analyzed as two separate arm-pairs. In such cases, the regulatory status was assigned according to the specific regimen; we distinguished whether approval was granted for the combination therapy, the monotherapy, or neither. This approach ensured that the approval status accurately reflected the regulatory outcome generated by each specific clinical comparison.

Finally, we identified Breakthrough Therapy, Fast Track, and Orphan Drug designations consulting the FDA Orphan Drug Database and official communications from the respective pharmaceutical companies.

Data Items

The literature in Online Resource Table S1 was used to extract study characteristics, i.e., sample size, study duration, type of study sponsorship, year of study start, cancer type, treatment line, modality of the study drug, age of control drug in the market, type of primary endpoint, study blind setting, and study arm design. We estimated the age of the control drug/regimen in the market by the duration of the first year of reporting the control regimen/drug and the year when the cited literature was published. In addition, we collected primary study results, including HR for OS and PFS, absolute extension of median OS and median PFS, the p value for the statistical tests, and median OS of the control arm. Nearly all data points were available for extraction; however, in instances where specific data were missing from the published reports, they were coded as “not available” (NA).

Statistical Analysis for Effect Measure and Assessment of Risk of Bias

All statistical analyses were performed using Stata statistical software (StataCorp, 2022; Stata Statistical Software Release 17. College Station, TX: StataCorp LP). All tests were two-sided, and a p value of < 0.05 was considered statistically significant. Data analysis was performed from January 2023 to December 2023 by the first author, KI, and then verified by the second author, SO. Descriptive statistics were used to characterize the FDA-approved and non-approved groups. Pearson’s chi-square test was used for categorical variables, and the t test was employed for continuous variables. To evaluate the potential impact of publication bias, we performed a funnel plot analysis and conducted a sensitivity analysis using the non-parametric trim-and-fill method. This approach estimates the number of potentially missing studies required to achieve funnel plot symmetry and calculates an adjusted pooled effect size.

Using a random-effects restricted maximum likelihood model, we estimated the pooled mean effect size in HR for OS by the FDA approval status, regulatory designation, and study characteristics. The heterogeneity was described using the I 2 statistics.

The distribution of the observed effect sizes in OS and PFS was visualized in the histogram. The probability of FDA approval was derived by the number of FDA-approved drugs divided by the total number of drugs in the specific effect size range. Logistic regression analysis was performed to fit the probability of FDA approval over the observed effect size in OS. Importantly, to account for potential intra-trial correlation in multi-arm designs, a sensitivity analysis was performed exclusively for the logistic regression using generalized estimating equations (GEE) with trial-level clustering. The boundary of FDA-approved and non-approved drugs was defined as a range of HR for OS that varied the probability of FDA approval from 20% to 80%. To further characterize the boundary, we reviewed the Oncologic Drug Advisory Committee (ODAC) incidence rate for the drugs in and outside the boundary and calculated the odds ratio. Finally, the summary receiver operating characteristic (ROC) curve and the area under the curve (AUC) were utilized to compare the regression models with HR for OS and other survival endpoints: absolute extension of median OS, HR for PFS, and absolute extension of median PFS.

Results

Characteristics of the Included Studies

Following a comprehensive screening of 3463 phase 3 studies, a total of 189 oncology studies comprising 208 arm-pairs and 158,250 participants were identified as eligible for analysis (Fig. 1 ). Approval status for each arm-pair was adjudicated by cross-referencing with official FDA documents. Specifically, we verified whether each trial was cited as pivotal evidence in the USPI or the FDA Summary Basis of Approval (Approval Package). Through this meticulous classification, 79 arm-pairs (38.0%) were categorized as supporting FDA approval, while 129 (62.0%) did not.

Tables 1 and 2 summarize the study characteristics and cancer types stratified by FDA approval status. Pharmaceutical companies sponsored the vast majority of the included studies (197 arm-pairs, 94.7%). The mean study duration was 5.6 years, and the average sample size was 761 patients. Although our search of ClinicalTrials.gov encompassed trials initiated from January 1990 to January 2021, those meeting all eligibility criteria were ultimately initiated between 1999 and 2018, with 93.7% starting after 2005. This distribution is primarily attributable to the limited adoption of NCT identifiers in peer-reviewed publications prior to the mid-2000s, which hindered the identification of earlier trials. Furthermore, because this study focused on large-scale phase 3 trials with OS as a primary endpoint, there was a substantial time lag (mean 5.6 years) between trial initiation and primary publication. Consequently, while the earliest trial in our cohort was initiated in 1999, the corresponding publication window spanned from December 2006 to August 2021.

The dataset spanned 27 cancer types, including 19 carcinomas, one sarcoma, and seven hematological malignancies. Treatment lines were balanced, with 47.6% ( n = 99) in the first-line setting and 47.6% ( n = 99) in the second-line or later. The most common malignancies were non-small cell lung cancer ( n = 41, 19.7%), followed by prostate cancer ( n = 27, 13.0%), and gastric cancer, melanoma, and pancreatic cancer ( n = 14 each, 6.7%).

The temporal evolution of therapeutic mechanisms is illustrated in Online Resource Fig. S1 . Following the rise of various molecularly targeted therapies in the 2000s, ICIs have predominated since 2013, driven by the intensive development of anti-programmed cell death 1 (anti-PD-1)/programmed cell death ligand 1 (PD-L1) and anti-cytotoxic T-lymphocyte-associated antigen 4 (anti-CTLA-4) antibodies.

There was no significant difference in the characteristics between the FDA-approved group and non-approved group for sample size, study duration, study sponsorship, cancer types, treatment line, modality of study drug, age of control drug in market, median OS of control drug, primary endpoint, study arm design, Fast Track designation, and Orphan Drug designation. On the other hand, the year of study start ( p = 0.017), study blind setting ( p = 0.029), and Breakthrough Therapy designation ( p < 0.001) showed significant differences between FDA-approved and non-approved drugs.

The funnel plot showed symmetric distribution and no significant publication bias for the FDA-approved drugs (Egger test, p = 0.727; Begg-Mazumdar test, p = 0.403, see Online Resource Fig. S2 ). However, the FDA non-approved drugs exhibited an asymmetrical distribution and showed significance (Egger test, p < 0.001; Begg-Mazumdar test, p < 0.001). The sensitivity analysis using the trim-and-fill method estimated 19 missing arm-pairs for the FDA-approved group and 40 for the non-approved group (see Online Resource Table S3 and Fig. S3). After imputation, the pooled OS HR shifted from 0.703 to 0.667 for the approved group and from 0.948 to 0.900 for the non-approved group. Despite these shifts, the relative disparity in HR between the two groups remained stable, changing by only 0.012 (unadjusted difference 0.245 vs. adjusted difference 0.233).

Pooled Effect Size by Subgroup Meta-Analysis

To explore the difference in average survival benefit by the regulatory status and the study characteristics, we used meta-subgroup analysis based on a random-effects restricted maximum likelihood model (Fig. 2 ). Overall pooled mean HR for OS in total 208 arm-pairs was 0.85 (95% CI 0.83–0.87) with 70.12% I 2 , indicating large amounts of heterogeneity in the dataset. The clearest separation was observed between FDA-approved and non-approved subgroups ( p < 0.001). The combined risk reduction of death associated with any condition across all FDA-approved drugs was 30% compared with the control drugs (HR 0.70 [95% CI 0.68–0.73]). In contrast, the FDA non-approved drugs showed only a 5% risk reduction in OS (HR 0.95 [95% CI 0.93–0.97]). Followed by FDA approval status, Breakthrough Therapy designation showed a significant separation (HR 0.70 [95% CI 0.66–0.74] with Breakthrough Therapy designation; HR 0.87 [95% CI 0.85–0.90] without Breakthrough Therapy designation; p < 0.001). Conversely, Fast Track and Orphan Drug designations did not significantly differ in the combined risk reduction of death ( p = 0.83 and p = 0.13, respectively). The pooled effects size by study characteristics are detailed in Online Resource Fig. S4 and Fig. S5.

Boundary Between FDA-Approved Drugs and Non-approved Drugs

The histogram in Fig. 3 a shows the distribution of HR for OS by FDA approval status, which indicates that each group has a distinct peak of the distribution. The FDA-approved group peaked at around HR 0.65–0.70 for OS, whereas the non-approved group peaked at around HR 0.95–1.0 for OS. The two groups’ distributions overlap from HR 0.70 to HR 0.90 for OS. Within this overlap, there were 37 FDA-approved and 35 non-approved arm-pairs. The largest overlap region was observed between HR 0.80 and 0.85 for OS.

Figure 3 b shows an overlay of a scatter plot for the actual probability of FDA approval over the HR for OS and a logistic regression curve using HR for OS to predict the probability of FDA approval. The logistic regression model showed a reverse sigmoidal curve with sharp sensitivity of FDA approval to the HR for OS (coefficient − 24.56, p < 0.001).

This model projected the probability of FDA approval at 20% at HR 0.86 and 80% at HR 0.74. Therefore, the range of HR for OS at 0.74–0.86 was identified as a boundary between FDA-approved and non-approved drugs based on the predefined criteria. This boundary was consistent with the overlap area observed in the histogram. Drugs in the boundary area showed a significantly higher odds ratio for the incidence that the FDA consults with the ODAC, compared with those outside the boundary area (odds ratio 22.3, p = 0.0001, see Online Resource Table S4 and Table S5).

To account for potential intra-trial correlation in multi-arm designs, we performed a sensitivity analysis using GEE with trial-level clustering. Among the 208 arm-pairs analyzed, 189 unique trials were utilized as the clustering units. The results demonstrated that the regression coefficients remained highly stable, with − 24.56 in the original logistic regression model and − 24.50 in the GEE model (both p < 0.001). The GEE-based findings were consistent with those of the primary analysis; the predicted probabilities of FDA approval plotted against the HR for OS demonstrated a nearly identical overlap between the two models (see Online Resource Table S6 and Fig. S6 for details).

The same histogram was depicted for other survival endpoints: extension of median OS, HR for PFS, and extension of median PFS. However, contrary to HR for OS, no clear separation was observed between FDA-approved and non-approved groups (see Online Resource Fig. S7). Similarly, the ROC curve showed the most extensive AUC in the predictive model using HR for OS (AUC 0.97 [95% CI 0.94–0.99]) among the survival endpoints (Fig. 4 ). The extension of median OS (AUC 0.92 [95% CI 0.89–0.96]), HR for PFS (AUC 0.81 [95% CI 0.75–0.87]), and the extension of median PFS (AUC 0.72 [95% CI 0.64–0.80]) followed, respectively.

Discussion

This cross-sectional study characterized the registrational phase 3 studies for anticancer drugs with measurements of OS over the past 20 years and reported a potential boundary between FDA-approved and non-approved drugs in survival benefit. Our findings underscore the importance of HR for OS in the regulatory decision and introduce a quantitative perspective for the regulatory bar for the new anticancer drugs in the US. In recent years, it has been criticized that the survival benefits of several FDA-approved drugs were lower than expected, resulting in concerns that the regulatory bar might have been lowered. However, the threshold identified by this study revealed that at least an HR of 0.80 for OS was required to maintain a 50% probability of FDA approval. Remarkably, this level was in line with the expectations of the ASCO. The finding could be a valuable reference for drug development in oncology.

The strengths of this study are the inclusion of both FDA-approved and non-approved drugs without selection by clinical or regulatory outcomes. This approach contributed to the comprehensiveness of this study. Our study showed a 38.0% (79 of 208 arm-pairs) success rate of FDA approval for the drugs in which the pivotal phase 3 trials were initiated from 1999 to 2018. The success rate is consistent with several previous reports. Wong et al. (2019) reported a 35.5% success rate from phase 3 to approval in oncology based on the 1236 phase transitions from 2000 to 2015 [ 24 ]. Similarly, Hay et al. (2014) reported a 37.0% success rate for 383 phase transitions from 2003 to 2011 [ 25 ]. Biotechnology innovation organizations reported a consistent success rate in oncology from 2006 to 2015 (33.0%) and a slightly higher rate from 2011 to 2020 (43.9%) [ 26 , 27 ]. These previous studies comprehensively collected the clinical trials without selection. Therefore, the consistency of our report supports the comprehensiveness of our dataset. Consequently, the dataset enabled us to compare the FDA-approved and non-approved drugs for the study characteristics and calculate the probability of success in the FDA approval.

Boundary Between FDA-Approved and Non-approved Drugs

Our dataset suggested that the probability of FDA approval is sharply sensitive to the HR for OS. Survival benefit is primarily expected for anticancer drugs; thus, it seems reasonable that such a sensitivity is observed. The range of HR at 0.74–0.86 for OS was identified as the boundary to separate the FDA-approved and non-approved drugs, exhibiting a 20–80% probability of FDA approval. From the logistic regression model, a 50% likelihood of FDA approval was derived at an OS HR of 0.80, which is consistent with the ASCO’s recommended criteria of at least a 20% risk reduction in death compared to available standard care (i.e., HR = 0.80). Although the ASCO working groups did not intend to set the standards for regulatory approval, the identified boundary showed a notable alignment. Importantly, however, this identified range should be interpreted as a descriptive pattern within the current evidence rather than as a definitive or prescriptive regulatory threshold.

In the ROC analysis, the regression model with HR for OS showed greater predictability to the FDA approval over the other models using an extension of median OS, HR for PFS, and an extension of median PFS. Two aspects of study design may partly explain this tendency. First, all studies in our dataset employed OS as a primary or co-primary endpoint. Therefore, the regulatory outcome may have shown higher sensitivity with OS than PFS. Second, the hypothesized HR generally determines the sample size rather than the expected absolute extension of median OS or PFS. Therefore, the HR has more influence than the absolute median extension of survival endpoints. These two aspects may determine the order of sensitivity, as shown in the ROC curve. Further interpretation of this trend implies that the FDA may make decisions based on OS over PFS and the result of the HR over median extension.

In August 2025, the FDA released a draft guidance outlining approaches for assessing OS in oncology clinical trials, underscoring the critical importance of accurate OS measurement as it reflects both the efficacy and safety profiles of anticancer drugs. The guidance designates OS as the gold standard endpoint in oncology, given that life extension in the context of life-threatening diseases possesses clear inherent value. Our study’s observed strong correlation between survival benefits and the likelihood of FDA approval aligns with the FDA’s emphasis on OS as a pivotal criterion in the regulatory evaluation of new anticancer drugs.

Drugs in the Boundary Area

The drugs outside of the boundary area were considered relatively easy to predict the FDA approval status based on the OS results because the drugs with less than HR 0.74 showed clear separation on the survival curve; conversely, the drugs with HR 0.87 or more for OS showed marginal or no separation to the control drugs. On the other hand, the drugs in the boundary area from HR 0.74 to 0.86 resulted in variable regulatory outcomes. As the FDA’s guidance regarding benefit–risk assessment for new drug and biological products clarifies, the FDA should have considered therapeutic context, e.g., cancer type, stage of disease, available treatment in the target population, and made regulatory decisions [ 28 ]. If the FDA needs to be more evident in its regulatory decision, it can convene the ODAC to hear the experts’ opinions. Our dataset includes six drugs for which the FDA consulted with ODAC. Among them, five drugs are in the boundary area in HR for OS: quizartinib for acute myeloid leukemia (HR 0.76 [95% CI 058–0.98]), sipuleucel-T for prostate cancer (HR 0.78 [95% CI 0.61–0.98]), pembrolizumab for gastric cancer (HR 0.82 [95% CI 0.66–1.03]), atezolizumab plus platinum chemotherapy for urothelial carcinoma (HR 0.83 [95% CI 0.69–1.00]), and atezolizumab plus nab-paclitaxel for triple-negative breast cancer (HR 0.86 [95% CI 0.72–1.02]) (see Online Resource Table S4). The odds ratio for ODAC consultation was 22.3 with the drugs in the boundary area compared with the drugs outside the boundary area (see Online Resource Table S5). Suppose a drug with the effect size in the boundary is filed for an NDA. In that case, the investigators and the sponsor should anticipate the possibility of the FDA’s consultation with ODAC. In that situation, further supportive data from other studies may be required to reinforce the uncompelling survival benefit in OS.

Limitations

This study has several limitations. First, our analysis focused exclusively on registrational trials powered for OS as a primary or co-primary endpoint. In modern oncology, however, regulatory approvals frequently rely on other clinically meaningful endpoints, such as PFS, objective response rate (ORR), and duration of response (DOR) [ 29 ]. For instance, registrational trials for targeted therapies directed at actionable genomic alterations often employ ORR or PFS as primary endpoints, as these measures may better reflect the therapeutic benefit in specific molecularly defined populations. Consequently, our findings are specific to OS-driven clinical trials and may not be directly generalizable to drugs approved based on surrogate endpoints. Second, our study included clinical studies initiated from 1999 to 2018. Over the last 2 decades, the oncology landscape has evolved significantly with the introduction of diverse therapeutic modalities, including molecularly targeted therapies, ICIs and antibody–drug conjugates. The emergence of these agents is notably time dependent; for instance, molecularly targeted drugs became prevalent in the 2000s and 2010s, whereas ICIs have predominated since the mid-2010s, as illustrated in the supplemental time-course chart (Online Resource Fig. S1 ). These modalities are characterized by distinct survival patterns; in particular, ICIs often exhibit non-proportional hazards with delayed clinical benefit or a "long-tail" plateau in survival curves, representing long-term survivors [ 30 ]. Future modalities, such as bispecific antibodies and cell therapies, may introduce further complexities in survival kinetics. Consequently, a traditional analysis based on the HR for OS may not fully capture the nuanced benefits provided by these diverse therapeutic mechanisms. While our findings provide a robust descriptive pattern of historical regulatory decisions, they should be interpreted with caution when applied to agents that demonstrate unconventional survival benefits. Third, the observed high I 2 value (70.12%) indicates substantial heterogeneity, which may reduce the certainty and interpretability of the pooled effect sizes. This heterogeneity likely stems from the diversity of anticancer agents and tumor types included. Furthermore, inherent imbalances in study characteristics between the FDA-approved and non-approved groups may limit the interpretation of the results. Fourth, while data extraction was performed by a single investigator, all entries were verified by a second investigator to ensure accuracy. Despite this rigorous verification, this process may still carry a minor risk of bias compared to larger, multi-reviewer systematic reviews. Fifth, it should be noted that our dataset included several comparisons derived from the same clinical trials or the same drug development programs. Although a sensitivity analysis using GEE with trial-level clustering confirmed the stability of the regression coefficients, the potential for intra-study or intra-drug correlation should remain a consideration when interpreting the broader patterns of FDA approval. Sixth, the FDA non-approved group exhibited an asymmetrical distribution in the funnel plot, suggesting potential publication bias. Although a sensitivity analysis using the trim-and-fill method indicated that the relative relationship between survival benefits and FDA approval status remained stable, the presence of unidentified or unpublished trials could still influence our findings. As this is a literature-based retrospective analysis, the possibility that missing data from negative or discontinued trials may have affected the observed results should be considered. Finally, as the therapeutic landscape continues to shift with new modalities, the observed relationship between OS benefit and approval may evolve. Therefore, our results should be viewed as a snapshot of the current regulatory era.

Conclusion

The probability of FDA approval was significantly associated with the HR for OS, with an HR range of 0.74–0.86 serving as a descriptive boundary between FDA-approved and non-approved drugs. Notably, an HR of 0.80 corresponded to a 50% likelihood of FDA approval. Our findings suggest that prioritizing the investigational agents with the potential to meet this efficacy benchmark could optimize drug development strategies.

Given the discussed limitations of our analysis, these findings should be considered a descriptive pattern derived from 2 decades of historical data rather than a formal regulatory threshold. The findings are contingent upon the accuracy of published reports and potential publication bias. Furthermore, as our analysis focused on OS-powered trials, the findings may not be generalizable to drugs approved based on surrogate endpoints.

Nevertheless, given that regulators evaluate broader facets—including safety, disease severity, and unmet medical needs—future research integrating these multi-faceted considerations will further refine our understanding of approval boundaries. Ultimately, these evidence-based benchmarks serve as a vital tool for stakeholders to streamline the delivery of truly impactful therapies to patients with cancer.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary file1 (PDF 1639 KB)

Acknowledgements

This work was presented as a preprint on Research Square. Koji Ishizuka, Shunsuke Ono. Boundary Between FDA Approved and Non-Approved Anticancer Drugs in Survival Benefit: A Meta-analysis, 14 October 2024, PREPRINT (Version 1) 10.21203/rs.3.rs-4883359/v1 . Aside from the authors, no other persons, institutions, or entities provided substantial contributions to this study.

Funding

Open Access funding provided by The University of Tokyo. This study was funded by the Ministry of Education, Culture, Sports, Science and Technology (MEXT): Shunsuke Ono KAKEN-HI: 25K10043. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The open-access publication fee was covered by the University of Tokyo.

Declarations

Conflict of Interest

KI is an employee of Daiichi Sankyo, Inc. SO is an Editorial Board member of Pharmaceutical Medicine . SO was not involved in the selection of peer reviewers for the manuscript nor any of the subsequent editorial decisions.

Availability of Data and Material

The datasets generated and/or analyzed during the current study are available from the corresponding author (shun-ono@mol.f.u-tokyo.ac.jp) on reasonable request.

Ethical Approval

Not applicable. This study is a retrospective analysis of publicly available, de-identified clinical trial data and does not involve the participation of human subjects. Therefore, institutional review board approval was not required.

Consent to Participate

Not applicable. This study is a retrospective analysis of publicly available, aggregated trial-level data and did not involve direct interaction with human participants. Therefore, informed consent was not required.

Consent for Publication

Not applicable. This article does not contain any individual person’s data or identifiable images.

Code Availability

Not applicable. Standard statistical software (Stata) was used for the analyses, and no custom-made software applications or original codes were developed for this study.

Author Contributions

All authors contributed to this study’s concept, design, and contents. They had full access to the study’s data and took responsibility for its integrity and accuracy. KI conducted the statistical analysis and drafted the manuscript. SO supervised all study processes. All authors have read and approve the final manuscript, and agree to be accountable for the work.

References

1. Ladanie A, Schmitt AM, Speich B, et al. Clinical trial evidence supporting US Food and Drug Administration approval of novel cancer therapies between 2000 and 2016. JAMA Netw Open. 2020. 10.1001/jamanetworkopen.2020.24406.

2. Royle KL, Meads D, Visser-Rogers JK, et al. How is overall survival assessed in randomized clinical trials in cancer, and are subsequent treatment lines considered? A systematic review. Trials. 2023;24(1):708. 10.1186/s13063-023-07730-1.

3. Ellis LM, Bernstein DS, Voest EE, et al. American Society of Clinical Oncology perspective: raising the bar for clinical trials by defining clinically meaningful outcomes. J Clin Oncol. 2014;32(12):1277–80. 10.1200/JCO.2013.53.8009.

4. Cherny NI, Sullivan R, Dafni U, et al. A standardised, generic, validated approach to stratify the magnitude of clinical benefit that can be anticipated from anticancer therapies: the European Society for Medical Oncology Magnitude of Clinical Benefit Scale (ESMO-MCBS). Ann Oncol. 2015;26(8):1547–73. 10.1093/annonc/mdv249.

5. Cherny NI, Dafni U, Bogaerts J, et al. ESMO-magnitude of clinical benefit scale version 1.1. Ann Oncol. 2017;28(10):2340–66. 10.1093/annonc/mdx310.

6. World Health Organization. Executive summary, the selection and use of essential medicines 2019: report of the 22nd WHO Expert Committee on the Selection and Use of Essential Medicines. https://www.who.int/publications/i/item/WHO-MVP-EMP-IAU-2019.05 .

7. Fojo T, Mailankody S, Lo A. Unintended consequences of expensive cancer therapeutics—the pursuit of marginal indications and a me-too mentality that stifles innovation and creativity: the John Conley Lecture. JAMA Otolaryngol Head Neck Surg. 2014;140(12):1225–36. 10.1001/jamaoto.2014.1570.

8. Senan S, Wang LH, Vansteenkiste J, et al. PROCLAIM: randomized Phase III trial of Pemetrexed-Cisplatin or Etoposide-Cisplatin plus thoracic radiation therapy followed by consolidation chemotherapy in locally advanced nonsquamous non-small-cell lung cancer. J Clin Oncol. 2016;34(9):953–62. 10.1200/JCO.2015.64.8824.

9. Garon EB, Ciuleanu TE, Arrieta O, et al. Ramucirumab plus docetaxel versus placebo plus docetaxel for second-line treatment of stage IV non-small-cell lung cancer after disease progression on platinum-based therapy (REVEL): a multicentre, double-blind, randomized phase 3 trial. Lancet. 2014;384(9944):665–73. 10.1016/S0140-6736(14)60845-X.

10. Saura C, Oliveira M, Feng YH, et al. Neratinib plus capecitabine versus lapatinib plus capecitabine in HER2-positive metastatic breast cancer previously treated with ≥ 2 HER2-Directed regimens: phase III NALA trial. J Clin Oncol. 2020;38(27):3138–49. 10.1200/JCO.20.00147.

11. Thatcher N, Hirsch FR, Luft AV, et al. Necitumumab plus gemcitabine and cisplatin versus gemcitabine and cisplatin alone as first-line therapy in patients with stage IV squamous non-small-cell lung cancer (SQUIRE): an open-label, randomised, controlled phase 3 trial. Lancet Oncol. 2015;16(7):763–74. 10.1016/S1470-2045(15)00021-2.

12. Tabernero J, Yoshino T, Cohn AL, et al. Ramucirumab versus placebo in combination with second-line FOLFIRI in patients with metastatic colorectal carcinoma that progressed during or after first-line therapy with bevacizumab, oxaliplatin, and a fluoropyrimidine (RAISE): a randomised, double-blind, multicentre, phase 3 study. Lancet Oncol. 2015;16(5):499–508. 10.1016/S1470-2045(15)70127-0.

13. Beaver JA, Pazdur R. “Dangling” accelerated approvals in oncology. N Engl J Med. 2021;384(18):e68. 10.1056/NEJMp2104846.

14. Lemery S, Pazdur R. Approvals in 2021: dangling accelerated approvals, drug dosing, new approvals and beyond. Nat Rev Clin Oncol. 2022;19(4):217–8. 10.1038/s41571-022-00605-5.

15. Schnog JB, Samson MJ, Gans ROB, et al. An urgent call to raise the bar in oncology. Br J Cancer. 2021;125(11):1477–85. 10.1038/s41416-021-01495-7.

16. Niraula S, Nugent Z. New cancer drug approvals from the perspective of a universal healthcare system: analyses of the Pan-Canadian oncology drug review recommendations. J Natl Compr Canc Netw. 2018;16(12):1460–6. 10.6004/jnccn.2018.7084.

17. Gyawali B, Hey SP, Kesselheim AS. Assessment of the clinical benefit of cancer drugs receiving accelerated approval. JAMA Intern Med. 2019;179(7):906–13. 10.1001/jamainternmed.2019.0462.

18. Molto Valiente C, Tibau A, Ocana Fernandez A, et al. Comparison of the European Society for Medical Oncology Magnitude of Clinical Benefit Scale (ESMO-MCBS) in clinical trials supporting US Food and Drug Administration (FDA) approval of orphan vs. non-orphan drugs. ESMO Abstr. 2017;28(5):V642. 10.1093/annonc/mdx440.063.

19. Tibau A, Molto C, Borrell M, et al. Magnitude of clinical benefit of cancer drugs approved by the US Food and Drug Administration based on single-arm trials. JAMA Oncol. 2018;4(11):1610–1. 10.1001/jamaoncol.2018.4300.

20. Kumar H, Fojo T, Mailankody S. An appraisal of clinically meaningful outcomes guidelines for oncology clinical trials. JAMA Oncol. 2016;2(9):1238–40. 10.1001/jamaoncol.2016.0931.

21. Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. 10.1136/bmj.n71.

22. Page MJ, Moher D, Bossuyt PM, et al. PRISMA 2020 explanation and elaboration: updated guidance and exemplars for reporting systematic reviews. BMJ. 2021;372:n160. 10.1136/bmj.n160.

23. Paik PK, Zhang J, West HJ, et al. Squamous non-small cell lung cancer: current and emerging treatment options. Clin Lung Cancer. 2025. 10.1016/j.cllc.2025.10.001.

24. Wong CH, Siah KW, Lo AW. Estimation of clinical trial success rates and related parameters. Biostatistics. 2019;20(2):273–86. 10.1093/biostatistics/kxx069.

25. Hay M, Thomas DW, Craighead JL, et al. Clinical development success rates for investigational drugs. Nat Biotechnol. 2014;32(1):40–51. 10.1038/nbt.2786.

26. Thomas DW, Burns J, Audette J, et al. Clinical development success rates 2006–2015. BIO Industry Analysis; 2016. https://share.google/Ft21d1g3vnD4GH2vZ . Accessed 10 Feb 2026.

27. Thomas DW, Chancellor D, Micklus A, et al. Clinical development success rates and contributing factors 2011–2020. BIO/QLS Advisors/Informa UK Ltd. White paper; 2021. https://go.bio.org/rs/490-EHZ-999/images/ClinicalDevelopmentSuccessRates2011_2020.pdf . Accessed 10 Feb 2026.

28. U.S. Department of Health and Human Services Food and Drug Administration. Benefit-risk assessment for new drug and biological products guidance for industry. 2023. https://www.fda.gov/media/152544/download . Accessed 10 Feb 2026.

29. Saad ED, Buyse M. Statistical controversies in clinical research: endpoints other than overall survival are vital for regulatory approval of anticancer agents. Ann Oncol. 2016;27(3):373–8. 10.1093/annonc/mdv562.

30. Ferrara R, Pilotto S, Caccese M, et al. Do immune checkpoint inhibitors need new studies methodology? J Thorac Dis. 2018;10(13):S564–1580. 10.21037/jtd.2018.01.131.

Associated Data

Supplementary Materials

Supplementary file1 (PDF 1639 KB)