Background
The FDA established the Accelerated Approval Program (AAP) in 1992.1,2 The AAP was codified in 2012 under Section 901 of the FDA Safety and Innovation Act, amending the Federal Food, Drug, and Cosmetic Act (FD&C Act), which contains the rules of drug approval processes and FDA procedures.3
The AAP was instituted to fill an unmet medical need by allowing earlier approval of drugs that treat serious conditions that typically do not have highly effective treatment options currently available based on surrogate endpoints.4 A surrogate endpoint is used in clinical trials on the assumption that it is a predictor of clinical benefit. It can be a laboratory measurement, a radiographic image, a physical sign, or another measure considered to predict clinical benefit adequately.4 The FDA decides whether to accept a proposed surrogate or intermediate clinical endpoint based on scientific evidence supporting the endpoint. Using a surrogate endpoint is especially advantageous when clinical outcomes take a long time to study.5 For example, under the AAP, the FDA can approve a treatment for cancer based on evidence that the drug shrinks the tumor, because such shrinkage is a predictor of real clinical benefit.
Manufacturers of AAP drugs are required to produce data establishing that a surrogate endpoint is “reasonably likely” to predict a drug’s intended clinical benefit. Postmarketing confirmatory studies must be conducted to verify the clinical benefit of the product. The FDA may withdraw or change the labeled indication of an AAP drug if the postmarketing confirmatory trials fail to verify the clinical benefit or do not demonstrate a sufficient clinical benefit to justify the risks associated with the drug (eg, a significantly smaller magnitude or duration of benefit than anticipated based on the observed effect on a surrogate).5
Approving therapies via surrogate clinical endpoints enables patients suffering from life-threatening diseases to receive potential life-changing therapies much sooner than with traditional approval based on patient survival. As the FDA AAP regulations specifically recognize, it is exactly the fatal diseases where patients and physicians are generally willing to accept greater level of risks and uncertainty.
Several recent papers have reported on the success and limitations of the AAP.6–9 However, a population-level assessment of the clinical benefit obtained via the AAP is urgently needed before addressing policy action to alter the program.
The objective of the current work is to estimate the clinical benefit of the FDA AAP by modeling the number of life years gained (LYG) by patients with cancer using drugs with accelerated approvals (AAs) relative to the standard of care (SoC). The focus on oncology was driven by its frequency in the AAP, accounting for 66% of all approvals since 1992 and 80% of approvals since 2006,10 and interest for policymakers as indicated by multiple papers recently published.6–8 From a modeling perspective, narrowing down the analysis to the oncology therapeutic area alleviates some of the computational burden; established methods used to model cancer therapies are relatively simple. Furthermore, the high frequency of AAP utilization for oncology treatments provides several data points to enable a robust analysis for informing policymakers on this important topic.
Methods
Modeling Approach
A model (Figure 1A) was developed to calculate the incremental LYG accumulated by cohorts of patients receiving AA products rather than the SoC they would have received in an unobserved counterfactual scenario with no AAP in place. The model uses a partitioned survival model with 2 health states: alive and dead. A simplified illustration of the flow of patient cohorts and accumulation of overall survival (OS) benefit is presented in Figure 1B and is described in the following sections.
Time horizon, uptake in observed and counterfactual scenario, and illustration of accumulating life year gain of an indication. (A) Uptake of a new therapy is assumed to be the same in both the observed and unobserved counterfactual scenarios (see light purple shaded area after RA). Therefore, the model takes into account patients who start therapy between the AA and RA date. There is no new cohort after RA, but cohorts starting before the RA date will have some follow-up time (see panel B). In the counterfactual scenario, patients who received the new therapy between the AA and RA date would receive the SoC. The model sets a hypothetical RA date for products that have not yet converted to the AA date plus the average time to approval for the rest of the products (36.8 months).
(B) In the example, the intervention receives AA in January 2018 and RA in 2021 January. Three cohorts of patients start using the drug: the first cohort in January 2018, the interim cohort in January 2019, and the last cohort in December 2020. OS curves for the intervention (orange curve) and the SoC comparator (blue curve) determine the OS gain (shaded blue area between the curves). The OS tracked over time for both the intervention (AA scenario) and the comparator (counterfactual). OS is multiplied by the uptake (ie, number of patients who started treatment with the intervention in a given month). The time window for accumulating benefit depends on RA date (determines the last cohort starting the intervention, last cohort in the diagram) and the end of the model time horizon (cuts the accumulation of benefits). The last cohort only accumulates part of the OS benefit, as the model time ends on December 31, 2022.
Abbreviations: AA, accelerated approval; OS, overall survival; RA, regular approval; SoC, standard of care; St Comp, fitted overall survival curve of the comparator; St Interv, fitted overall survival curve of the intervention/AA treatment.
Citation: Journal of the National Comprehensive Cancer Network 22, 6; 10.6004/jnccn.2024.7010
Patient Cohorts and Time Horizon
The analysis considers cohorts newly starting an AA treatment who would not have received that therapy without the AAP. Ideally, these cohorts are those initiating the AA treatment between the AA date and when regular approval (RA) would have occurred in the counterfactual scenario with no AA. As the latter date is unknown, it is approximated with the observed time of RA in the base case analysis, and alternative dates are explored in scenario analyses by assuming that the RA would have happened earlier or later than in reality. Patients starting therapy after RA date are excluded, reflecting an implicit assumption that they could have received the treatment based on the RA in the counterfactual scenario too. For products that have not yet been converted, the RA date in the counterfactual scenario is set to AA date plus 36.8 months (ie, the calculated average time between AA and RA for all other oncology products).
Each cohort is tracked from model entry until December 31, 2022, in the base case, irrespective of the time of RA. This corresponds to the implicit assumption that the treatment course of patients receiving SoC is not disrupted due to the availability of a new drug on the market. Cohort sizes are based on observed product uptake.
Estimation of Survival Gain for Patients
For the counterfactual scenario, zero uptake is assumed before RA, and patients receive SoC instead. For cohorts entering each calendar month, the model tracks OS with the treatment and with SoC. The proportion alive is multiplied by the cohort size each month, thereby accumulating the months lived. Finally, the model calculates the difference between the life years accumulated with the new FDA AAP therapy versus those accumulated with the SoC (Figure 1B).
Clinical trials usually report the median OS obtained from Kaplan-Meier survival estimators. However, because the objective was to accumulate LYG for each cohort, an area under the curve (AUC) approach (blue-colored areas in Figure 1B) is more appropriate than assigning the median OS gain to all patients. Because the model estimates OS for 69 active and control treatment pairs, digitizing OS curves and fitting a distribution was simplified: the AUC was estimated by fitting an exponential curve through the median OS. When OS data were available but medians were not reached, other milestones (eg, OS estimates at 24 months follow-up) were used for fitting the exponential distribution. When a milestone was only available for SoC or the novel therapy, a published OS hazard ratio (HR) comparing the AA therapy versus SoC was used to estimate OS. For products where crossover-adjusted HR was available, it was used, but the reported gain may underestimate real OS gain due to lack of crossover adjustment in the reported results.
In an alternative calculation scenario (“lump sum”), median OS difference between AA product and SoC either directly reported or estimated using the exponential distribution (as described earlier) was assigned to each cohort. This method was only applied in a scenario analysis and only to cancers with <36 months’ median OS for SoC or AA therapy.
Given multiple sources for OS data, median OS values from the same source for both treatment and SoC were selected, with a preference for the more mature data from the RA trial, and in some cases for the mature AA trial to be aligned closely with the original AA indication, resulting in the use of a mix of AA and RA trial results (see Table S1 in the supplementary materials, available online with this article).
A list of assumptions appears in Table 1, along with an assessment of their impact on the results. Some assumptions were tested in scenario analyses. The model is implemented in Microsoft Excel with Visual Basic for Applications.
List of Main Assumptions
Data
Oncology Therapy Approvals
All oncology therapy AAs between January 1, 2006, and June 30, 2021, were reviewed regardless of their final approval status. A 15-year timeframe was selected to include most cancer therapies without returning to the chemotherapy era.
Data concerning the AA indications were extracted from the FDA Center for Drug Evaluation and Research (CDER) AA Database as of December 31, 2021,11 and from the FDA Oncology AA Database compiled from tables of ongoing, verified, and withdrawn AA indications.11
Data related to AA and RA dates and exact indications for each indication/therapy pairs were extracted up until June 10, 2022.
The final dataset (see Supplementary Figure S1 for inclusion/exclusion criteria) contained 130 oncology indications with 57 verified (converted), 55 ongoing, and 18 withdrawn AAs as of June 10, 2022.
Procedural Data
For AAs, the postmarketing requirements are set by the FDA (presumably following discussion between the sponsor and the FDA) along with an expected submission date of confirmatory trial results. To determine whether a submission was late, the expected submissions dates for confirmatory trial results were collected from the Project Confirm website12 for ongoing trials; for verified (converted) and withdrawn indications, they were collected from FDA letters written to the sponsor at the time of the AA.13 Information concerning breakthrough and orphan designations of the therapies were extracted.14,15
Clinical Data
The primary sources of the efficacy data were the “clinical studies” sections of FDA labels published in their current format since 2007. These sources include relevant clinical trial results at the time of the drug’s initial AA and/or RA.
For OS data, trial results published on ClinicalTrials.gov and peer-reviewed publications were collected. If unavailable, the OS results were extracted from other sources, such as various oncology conference abstracts (eg, ASCO, ESMO), published health technology assessments, or, in rare cases, real-world evidence.
OS results were collected from the trials supporting AA and RA, covering multiple data timepoints; the most mature reported trial OS data were used. OS data were available for 69 of 130 indications. OS data were available for 50 of 57 converted AAs (Figure 2). Among withdrawn indications, OS was available for 12 of 18, and 7 of 55 ongoing indications contributed data to the model.
AA database by conversion status and OS data availability.
Abbreviations: AA, accelerated approval; OS, overall survival.
Citation: Journal of the National Comprehensive Cancer Network 22, 6; 10.6004/jnccn.2024.7010
Importantly, for ethical and other practical reasons, RA trials are often conducted in an indication that differs from the original AA indication.16 Even within this group, the RA trial may report results for the population in the AA indication. This is not only possible when the AA and RA indications are identical. In some cases, the RA was provided for a narrower population than the AA, yet data were available to represent the entire AA population because the RA trial covered a broader population than the RA label. In other instances, the AA indication was broader than the RA indication, but the RA trial publications reported results for the AA label’s subgroup. When OS data were only available in an indication that differed from the original AA, these were used to calculate OS. The relationship between the 69 AA and RA indications/trial populations was categorized as (1) having very good overlap between AA indication and trial results used for the calculation (51%), either because (i) mature AA trial results were used for confirmation, (ii) only the original AA trial results were available, or (iii) the RA trial reported on a population very similar to that in the AA trial; (2) a “line change down” (ie, the confirmatory trial was conducted in an earlier line of therapy [26%]); or (3) other changes (23%), such as from monotherapy indication to combination therapy, combined with a line change down; a broadening or narrowing of the patient population; a “biomarker change”; or a “line change up” (see Supplementary Table S2).
Based on the information collected regarding the expected submission dates and an assumed 90-day time frame needed for an FDA revision, 40 of 55 ongoing indications (84%) were categorized as “not yet due” as of June 10, 2022. Therefore, these indications were not expected to have any OS data at the time of writing. Of 50 converted AA indications, in 3 cases 2 drugs in combination received AA for a given indication. To avoid double-counting, only one of the drugs was used in the model to calculate the benefits.
Market Data
Market data concerning uptake were obtained to estimate the number of patients benefiting after the drug received AA. Number of patients newly starting a therapy each month after AA date was extracted from the Ipsos Oncology Uptake Tool (see Supplementary Figure S2), which includes data on oncology indications and their treatments spanning back to 2004. Uptake was extracted for AA indications (and not the possibly different RA indications), specific to tumor type, line of therapy, biomarkers, and combination versus monotherapy.
Of 69 indications with OS data, 3 had no uptake information due to very small populations in the approved indications; for 2 indications, data collection started after the RA date. For AAs based on trials including multiple tumor types, Ipsos data only cover the major cancer sites (lung and colon). For therapies without confirmation, uptake was only taken into account for 36.8 months (ie the average time from AA to RA for the 57 converted products).
Overall, between January 2006 and December 2021, approximately 3.3 million people started therapies that received AA. Between the AA and RA date of each intervention, that number narrows to 1.1 million people, and ultimately, 910,757 patients were estimated to receive the 69 modeled interventions that had corresponding OS data (see Figure 2 and Supplementary Figure S2).
Results
The base-case analysis includes all oncology AA indications with analyzable OS data regardless of their conversion status (69 as of June 10, 2022; see Supplementary Table S2). The model uses the latest available OS data (mostly from RA trials and, in some cases, AA trials; see Supplementary Table S1), and for the counterfactual scenario it assumes that RA would occur at the same time as it originally happened.
It was estimated that by end of 2022, 263,378 life years were accrued overall for 910,757 patients with cancer who initiated an AA therapy before the RA date. The total is composed of total net life years, including those lost for patients with triple-negative breast cancer on AA therapies that eventually proved inferior to SoC (−3,027) and the large gains for patients in other indications, with the highest number of LYG (73,667) in non–small cell lung cancer (Figure 3 and Supplementary Table S3).
Annual and cumulative LYG, by cancer type.
Abbreviations: CML, chronic myeloid leukemia; LYG, life years gained; MM, multiple myeloma; NSCLC, non–small cell lung cancer; RCC, renal cell carcinoma.
aAnnual numbers are decreasing as newer indications become biomarker-specific.
Citation: Journal of the National Comprehensive Cancer Network 22, 6; 10.6004/jnccn.2024.7010
Scenario analyses assuming that all products have worse or better OS results, varied according to the 50% confidence limits of the HR, showed that results would range between 147,917 to 367,834 LYG. Allowing accumulation of benefits until 2026 and 2030 increased the number of LYG by 45.2% and 82.0%, respectively (Table 2). If RA were 3 or 6 months later (or to similar effect, the start of uptake following RA would be delayed) in the counterfactual scenario, the number of patients benefitting from AA therapy increased by 12.5% or 24.9% and the number of LYG increased by 21.3% or 41.4%, respectively. If RA in the counterfactual situation occurred 3 months earlier, LYG and number of patients would drop by 18.4% and 11.9%, respectively. A scenario combining the more favorable HRs with ≥6 months of use and time horizon extended to 2026 and a scenario combining the upper limit of OS HRs with a limited time horizon for benefit accumulation are also presented.
Scenario and Subgroup Results
Examining AA by approval status indicates that most LYG are brought about by fully approved products (Table 2). The 12 withdrawn indications have on average a small benefit. Overall, 5 products are included with negative LYG. Ongoing assessments that have survival data in the public domain contribute 5% of the results (38,429 LYG) impacting only approximately 4% (13,212 LYG) of patients, partly due to the limited time for uptake. Relative to SoC, the subgroups of 41 AAs with orphan designation and 26 with breakthrough therapy designation resulted in the highest life years gained. The 24 hematology indications comprise 20 orphan indications, and therefore these results are relatively similar.
A subgroup analysis was conducted including only advanced cancers with a median life expectancy of <4 years (“fast-progressing cancers”), because the exponential distribution assumption is more tenable in these indications. Running this analysis increases the proportional LYG to 16.2% (212,074 LYG). When median OS is immediately assigned to patients (“lump sum”), LYG increased to 454,916 (35.2%).
Discussion
Interpreting Results
In the most comprehensive analysis to date, we estimate that AA of cancer products between 2006 and 2021 contributed approximately 263,000 additional life years in approximately 911,000 patients who received these therapies before RA. Extending the time horizon to 2026 would increase LYG to approximately 382,000 years.
A wide range of cancer drugs were approved in the FDA AAP in the selected time frame, including a small group of chemotherapies, several targeted therapies, antibody–drug conjugates, and immunotherapies across 44 different cancer types.
This analysis incorporated cancer therapies with an AA regardless of the final outcome, including all withdrawn indications. It calculated the benefit for patients newly starting therapy in the AA indication between AA and RA, and for yet unapproved products, for a limited amount of time.
Extensive data collection covered each AA in detail, including letters of approval, FDA labels, populations and outcomes of trials cited for AA, and the confirmatory trials, including their long-term follow-up. A standardized methodology was applied throughout to be consistent across products. The impact of assumptions was considered in Table 1.
Behind the standard methods, each approved indication has its unique story. The indications covered rare and fast-progressing late-line cancers, such as persistent, recurrent, or metastatic cervical cancer with a median OS of 16 months with 6 months of improvement in life-expectancy for approximately 3,600 patients, and therapies for slow-progressing cancers, such as the tyrosine kinase inhibitors for approximately 40,000 patients with chronic myeloid leukemia. All 5 indications with point estimates for the OS HR >1 were included, reducing the total estimated number of LYG. For completeness, indications that were omitted due to missing OS data were tabulated (Supplementary Table S4). Although OS data were not identified in the published literature for 13 of the converted or withdrawn therapies, their extent of use was established. Approximately 46,800 patients used the 7 products that were converted, but had no OS data available, mostly because the studies were single-arm trials. Overall, 7,750 patients used the 6 therapies that were eventually withdrawn and had no OS data, representing only approximately 0.7% of the total population that used AAs.
Strengths and Limitations
Benefits of therapies were quantified in terms of LYG, the ultimate outcome of interest. The LYG were accumulated continuously for the cohorts, up to a cutoff date—a more accurate approach than assigning the median gain compared with SoC to all patients. Furthermore, if available, the OS data from the confirmatory trial were selected for the population closest to that in the AA.
Identifying OS data required extensive research and relied on the fact that in almost all cases, the FDA’s letters of approval identified the trial that would serve as the confirmation of the AA. Decisions were well documented (Supplementary Table S1), and the analysis is reproducible based on that documentation. Actual uptake based on chart reviews for very specific indications were used instead of relying on prescription data, which are not linked to an exact indication. Furthermore, some relatively prevalent indications lack market share data (Supplementary Figure S3), and for the not-yet-converted indications, the uptake was only included for approximately 3 years rather than their full duration, truncating the time frame for indications with very long trials/time to conversion. Therefore, the number of patients benefitting from the program was underestimated. Furthermore, we assumed no additional time to a peak uptake after the RA, even though it is likely that uptake after RA is not immediate. Lastly, the counterfactual assumes that RA occurs at the same time as following an AA, which may be conservative given that RA trials are ongoing at time of AA and their assessment may be faster due to familiarity with the therapies.
The model has several limitations. The exponential distribution may not be the optimal fit to cancer survival data, but use of other distributions would require additional data points or assumptions. For fast-progressing cancers, the exponential distribution is a reasonable approximation; for slow-progressing cancers with long median survival, the exponential may overestimate or underestimate LYG depending on the time course of the hazards. The model time horizon is limited, and therefore the LYG estimate is truncated. Modeling OS by digitizing survival curves and finding the best fitting distribution was beyond the scope of this study. LYG is a patient-relevant metric that can be easily operationalized. Its disadvantage is that it does not incorporate health-related quality of life impact of therapies. Modeling health-related quality of life associated with conditions and with adverse events of therapies was beyond the scope of this study. Because new therapies replacing chemotherapies have favorable safety profiles, the impact of the latter may be small. However, quality of life with advanced oncology is often severely impaired, and therefore the impact in terms of quality-adjusted life years is likely smaller.
The Ipsos data had gaps for some indications and small samples for some late-line cancers (see Supplementary Figure S3). For several products, some months had zero uptake, with positive values in the months before and after, likely biasing the estimation downward. Using OS data for the RA indication that is in an earlier line compared with the AA indication in some instances may bias the LYG results upwards. Finally, the model did not include costs associated with new therapies, and therefore did not consider the full opportunity cost of AAs, leaving the question of value for money for future research.
Conclusions
Policy discussions about AAP are incomplete without assessing its impact on survival. To date, LYG due to AAP have not been estimated, presumably due to the lack of databases covering outcomes related to AAs. Any major change to the program should consider the impact on patient outcomes for groups with a high unmet need.
We have provided an assessment of the AAP as it played out between 2006 and 2022 in cancer-related indications. Our results are a snapshot in time, specific to the time frame examined, and may not be generalizable for the future; the oncology therapies approved now include cell and gene therapies with curative potential, and the program itself may be changing.
Acknowledgments
The authors would like to thank Júlia Palotás and Walter Morris for their assistance with data extraction and editorial help; J. Jaime Caro for comments; and Tereza Lanitis for programming.
References
- 1.↑
Code of Federal Regulations. Biologics. 21 CFR §601.41 Approval based on a surrogate endpoint or on an effect on a clinical endpoint other than survival or irreversible morbidity. Accessed March 31, 2023. Available at: https://www.ecfr.gov/current/title-21/section-601.41
- 2.↑
Code of Federal Regulations. Drugs for human use. 21 CFR §314.510 Approval based on a surrogate endpoint or on an effect on a clinical endpoint other than survival or irreversible morbidity. Accessed March 31, 2023. Available at: https://www.ecfr.gov/current/title-21/section-314.510
- 3.↑
U.S. Food & Drug Administration. Accelerated approval. Accessed January 1, 2024. Available at: https://www.fda.gov/patients/fast-track-breakthrough-therapy-accelerated-approval-priority-review/accelerated-approval
- 4.↑
U.S. Food & Drug Administration. Accelerated approval program. Accessed January 1, 2024. Available at: https://www.fda.gov/drugs/nda-and-bla-approvals/accelerated-approval-program
- 5.↑
U.S. Food & Drug Administration. FDA facts: biomarkers and surrogate endpoints. Accessed January 1, 2024. Available at: https://www.fda.gov/about-fda/innovation-fda/fda-facts-biomarkers-and-surrogate-endpoints
- 6.↑
Beaver JA, Howie LJ, Pelosof L, et al. A 25-year experience of US Food and Drug Administration accelerated approval of malignant hematology and oncology drugs and biologics: a review. JAMA Oncol 2018;4:849–856.
- 7.↑
Gyawali B, Hey SP, Kesselheim AS. Assessment of the clinical benefit of cancer drugs receiving accelerated approval. JAMA Intern Med 2019;179:906–913.
- 8.↑
Cherla A, Naci H, Kesselheim AS, et al. Assessment of coverage in England of cancer drugs qualifying for US Food and Drug Administration accelerated approval. JAMA Intern Med 2021;181:490–498.
- 9.↑
Kaltenboeck A, Mehlman A, Pearson SD. Strengthening the accelerated approval pathway: an analysis of potential policy reforms and their impact on uncertainty, access, innovation, and costs. Accessed January 1, 2024. Available at: https://icer.org/wp-content/uploads/2021/04/Strengthening-the-Accelerated-Approval-Pathway-_-ICER-White-Paper-_-April-2021.pdf
- 10.↑
Subbiah V, Wirth LJ, Kurzrock R, et al. Accelerated approvals hit the target in precision oncology. Nat Med 2022;28:1976–1979.
- 11.↑
U.S. Food & Drug Administration. CDER drug and biologic accelerated approvals based on a surrogate endpoint as of December 31, 2023 . Accessed January 1, 2024 . Available at: https://www.fda.gov/media/151146/download?attachment
- 12.↑
U.S. Food & Drug Administration. Ongoing | Cancer accelerated approvals. Accessed January 1, 2024. Available at: https://www.fda.gov/drugs/resources-information-approved-drugs/ongoing-cancer-accelerated-approvals
- 13.↑
U.S. Food & Drug Administration. Drugs@FDA: FDA-approved drugs. Accessed January 1, 2024. Available at: https://www.accessdata.fda.gov/scripts/cder/daf/index.cfm
- 14.↑
U.S. Food & Drug Administration. Search orphan drug designations and approvals. Accessed January 1, 2024. Available at: https://www.accessdata.fda.gov/scripts/opdlisting/oopd/
- 15.↑
U.S. Food & Drug Administration. Breakthrough therapy approvals. Accessed January 1, 2024. Available at: https://www.fda.gov/drugs/nda-and-bla-approvals/breakthrough-therapy-approvals
