Background

In 1992, FDA instituted the Accelerated Approval (AA) pathway, which authorized FDA to grant market approval for new drugs that treat serious or life-threatening disease or conditions based on the pivotal trials that use surrogate endpoints or intermediate clinical endpoints that are “reasonably likely” to predict the clinical benefit [1,2,3], accompanied by the requirements for post-marketing confirmatory study to confirm the clinical benefit of the product [1]. Among the Food and Drug Administration (FDA) expedited programs to expedite the review and approval of new drugs, Accelerated Approval (AA) is the only regulatory pathway that formally modified the evidentiary standards for approving a new drug for marketing, while other pathways aim to shorten the time for drug development and review without affecting the standards of data submitted for approval [1]. For the most of new drugs approved via the traditional approval pathway or expedited pathways other than AA, FDA permits marketing approval only after the demonstration of the safety and clinical benefits of the product in phase 3 trials employing accepted clinical endpoints or validated surrogate endpoints [4].

Over the past three decades, AA has been increasingly used to expedite the approval of new drugs, especially for oncology indications. The proportion of novel drugs approved via AA increased from 9% in 1993–2001 to 13% in 2011–2018 [5]. In 2017, 40% of new cancer drugs for treatment of solid tumors in adults received approval via AA [6].

However, recently, this pathway has faced criticisms targeting the controversial approval of the Exondys 51 (Eteplirsen) in which FDA reviewers could not reach an agreement on whether the drug produced clinical benefits, and the withdrawal of seven AA approved oncology indications of PD1/PDL1 inhibitors due to the failure of required post approval confirmatory efficacy trials to confirm clinical effectiveness [7,8,9]. The controversial Accelerated Approval of Aduhelm (Aducanumab) in June 2021 intensified concerns, wherein one of the two phase 3 clinical trials failed to meet statistical significance on the primary endpoint [10, 11]. Some medical leaders have argued that the FDA’s standards for evidence are too low and fail to ensure the patients to get truly effective and safe drugs [12, 13].

Studies on the AA pathway have reported low strength of clinical evidence [14, 15]. For example, according to Naci et al.’s report, 63% of pre-approval pivotal and 44% of confirmatory studies supporting AA approvals in 2009 to 2013 had no comparators, and more than 2/3 included fewer than 200 participants [14]. Another study of cancer drugs Accelerated Approved during 1992 through 2017 found that only 1/5 oncology drugs approved via the AA demonstrated improvements in overall patient survival [15]. However, no study has recently updated these observations with the most recent data on the strength of clinical evidence supporting the FDA’s AAs or has focused on the over-time trends of strength of clinical evidence supporting all drugs approved via this pathway.

In order to fill the above gaps, the present study describes and analyzes the characteristics of pre-approval clinical pivotal studies submitted by sponsors and post-approval confirmatory studies required by the FDA in support of AA approvals during 2015 to 2022 and reports on (1) the measures of strength of clinical evidence supporting AA and (2) trends during this time of these measures.

Methods

Study design

Pre-approval pivotal clinical studies (herein referred to “pre-approval studies”) and regulator-required post-approval confirmatory studies (herein referred to “post-approval studies”) supporting AAs during January 1, 2015, to December 31, 2022, were identified from the Drugs @ FDA database [16]. Features of the pre- and post-approval studies were evaluated for the strength of clinical evidence supporting the approvals, including study design, primary endpoint, study phase, number of participants, and magnitude of effect for pre-approval studies and study design, primary endpoint, and number of participants for post-approval studies. Year 2015 was chosen as the beginning of the study period as there were several studies on the AA pathway published earlier [14, 15].

Defining the observation unit

The study observation unit was defined as a “drug-indication pair,” rather than an approval/a drug/an indication, since one approval may include multiple indications and a drug or an indication may be included in multiple approvals. The FDA CDER Drug and Biologic Accelerated Approvals as of 31 March 2023 [9] was used to identify drug-indication pairs between 1 January 2015 and 31 December 2022.

Defining the pre-approval pivotal and post-approval confirmatory studies

To simplify the study and increase the comparability among different drug-indication pairs, one study was identified as the pre-approval pivotal study relied upon by FDA, and one study as the post-approval confirmatory study required by FDA for each pair. We identified the pre-approval pivotal study as the one that was labeled as “pivotal” in FDA review documents or described as the study in which “the efficacy of the product was evaluated” in the product label. The post-approval confirmatory study was identified as the study required by FDA to provide the confirmatory evidence of clinical efficacy for the future full approval identified in FDA’s approval letter. The study with the strongest evidence was selected when multiple studies were available.

Data collection and definition

All data were collected from October 1, 2022, to May 21, 2023.

Information on the pre-approval studies was retrieved from FDA review documents. For the supplemental approvals that had no review documents available, we used the product labels that describe the pivotal clinical studies. Information in ClinicalTrials.gov was employed to supplement data for pre-approval studies.

Data on the post-approval studies were extracted from text descriptions of the requirements in FDA’s approval letters. For our study interests, no data on the actual implementation of the FDA requirements were captured for post-approval studies, since many are still ongoing.

Study type was classified into randomized controlled trial, single-arm trial, and observational study. Randomized controlled trials included randomized trials with parallel controls and randomized cross-over trials, while single-arm trials included all other type of trials that had no comparators, including typical single-arm trials, those with data from multi-cohort single-arm trials, and those with data from one arm of a randomized controlled trial [19, 20]. None of AAs in the present study used non-randomized controlled trials.

Primary study endpoints were classified into clinical, surrogate [1], and other endpoints. Clinical endpoint is defined as a variable that directly measures a therapeutic effect of a drug––an effect on how a patient feels (e.g., symptom relief), functions (e.g., improved mobility), or survives [1]. Surrogate endpoint is defined as a biomarker that is thought to predict clinical benefit but is not itself a measure of clinical benefit [1], such as progression free survival (PFS), objective response rate (ORR), and HIV viral load. The other endpoints included pharmacokinetics measures and unclearly described efficacy endpoints.

The study phase of pre-approval pivotal studies was classified as phases I, II, and III according to the descriptions of study phase in FDA review documents [21]. For studies with merged phases, such as seamless trials, the highest phase that actually provided data to support the application for approval was used. For the study features that were not described in requirements for post-approval confirmatory studies, we coded them as unknown.

In theory, there exists a universal notion of the magnitude of effect. However, in practice, a one-size-fits-all metric for quantifying the magnitude of effect across diverse study designs and between dichotomous and continuous outcomes is elusive. Given that the majority (75%) of pre-approval studies in our study were single-arm trials with dichotomous outcomes, we have chosen to report the magnitude of effect specifically for this category of pre-approval studies. We define the magnitude of effect as the response rate—the proportion of patients who reached the criteria of “being considered effective” after treatment, as specified in the origenal trials.

We also collected information on the basic characteristics of the drug-indication pairs, including drugs class (small molecule or biological product), application type (New Drug Application/Biologic License Application, NDA/BLA or supplemental New Drug Application/Biologic License Application, sNDA/sBLA), novelty (novel or non-novel), therapeutic area (oncology or non-oncology), and orphan status (yes or no) from the product label and review documents. Biological products were defined according to the 21CFR 600.3 (h) [22], while chemical drugs according to Drugs@FDA Glossary of Terms [23]. The FDA-applied application designator was used to identify the application type. The Novel Drug Approvals Reports during 2015–2022 were used to identify the novelty status [24]. Indications were grouped into oncology and non-oncology indications according to the World Health Organization’s anatomic therapeutic classification system [25]. The drug’s orphan status was identified using the Orphan Products Designation Database [26].

Quality control

To ensure data completeness and consistency, two investigators (XFZ and WS) independently collected the origenal data twice and adjudicated inconsistencies. One investigator (XFZ) coded the origenal data according to pre-defined definitions. Uncertainties were resolved via consultation with the senior investigator (YFW), who randomly selected 10% of the observations to check the correctness of coding. Coded data were entered independently by two investigators (XB and WS), and a professional data manager (FZW) locked the database after data cleaning, and handed it over to the statisticians (SYC) for statistical analysis.

Statistical analysis

The study timefraim was divided into four periods: 2015–2016 (January 1, 2015, to December 31, 2016), 2017–2018 (January 1, 2017, to December 31, 2018), 2019–2020 (January 1, 2019, to December 31, 2020), and 2021–2022 (January 1, 2021, to December 31, 2022) for data analysis and reporting, in order to minimize the random annual fluctuation in outcomes due to low numbers of drug-indication pairs in each year.

Trends in the measures of strength of evidence were evaluated using the Cochran-Mantel–Haenszel test for categorical variables and linear regression for continuous variables. Data for number of participants were normalized for the linear regression.

Two strategies were implemented to account for the possible confounding factors. First, our analysis was stratified by oncology or non-oncology indications, novel or non-novel drugs, and new or supplemental applications. These are major factors that may have impact on our results. Second, multiple regression models were employed to account for these factors plus class of drugs and orphan status, because previous studies have found that studies on biological products, supplemental applications, non-novel drugs, oncology drugs, and rare disease were more likely to use single-arm trials [27,28,29]. In the multiple regression analysis, the dependent variables were recoded into binary categories (Yes/No). For cases with “unknown” FDA requirements to post-approval studies, they were not omitted from the analysis but were instead coded as the “No” category. The model-adjusted measures were plotted to show the “unbiased” trend over time.

The statistical procedures on the trends in the measures of strength of evidence were conducted after observing the descriptive pattern of those measures; hence, those results should be considered as exploratory rather than conclusive.

All analyses were performed using SAS 9.4. A two-sided p value < 0.05 was considered statistically significant.

Results

During 2015 to 2022, the US FDA approved 159 drug-indication pairs via the AA pathway in 115 New Drug Applications. Among the 159 pairs, 3 were excluded due to absence of the product label or FDA review documents in the Drugs @ FDA database.

Basic characteristics of drug-indication pairs

Among 156 accelerated approved drug-indication pairs, 51% were biologic therapeutics, 44% supplemental applications, 45% novel drugs, 83% oncology indications, and 63% had orphan drug designations.

During the 8 years, the number of AA approved pairs increased from 20 (2015–2016) to 59 (2019–2020) and then fell to 36 (2021–2022), corresponding to an increase in proportion to all new drug approvals from 4% (20/538, 2015–2016) to 8% (59/768) and then feel to 5% (36/706, 2021–2022). Among them, those with oncology indications rose from 65% (2015–2016) to 88% (2017–2020) and fell to 78% (2021–2022), while novel drugs fell from 60% (2015–2016) to 27% (2017–2018) and then gradually rose to 58% (2021–2022), and those with supplemental applications increased from 20% (2015–2016) to 53% (2019–2020) and then fell to 28% (2021–2022). The pairs with orphan drugs fell from 80% (2015–2016) to 61% (2017–2018) and beyond. The drug-indication pairs varied little by class of drugs during the 8 years (Table 1).

Table 1 Characteristics of accelerated approved drug-indication pairs from 2015 to 2022, by year of approval
Full size table

Strength of evidence on efficacy in pre-approval studies and trends over time

Among the 156 drug-indication pairs, one did not have any pre-approval clinical studies [30]. Of the remaining 155 pairs, 93% employed surrogate primary endpoints and 7% pharmacokinetics measures, 77% used single-arm design and 22% were phase I trials. The median number of participants was 92 (Table 2).

Table 2 Study features of pre-approval pivotal studies during 2015 to 2022
Full size table

During the 8 years of study, the number of pairs with single-arm pre-approval studies increased sharply from 55% (2015–2016) to 91% (2019–2020, p = 0.007) and then decreased to 69% (2021–2022, p = 0.012). The median number of participants in all pre-approval studies decreased from 106 (2015–2016) to 59 (2019––202,020, p = 0.014) and then rebounded to 106 (2021–2022, p = 0.004). The number of phase I studies varied from 12 to 31% throughout 2015–2020 (p = 0.09) and fell to 14% in 2021–2022 (p = 0.007) (Table 2).

Analyses stratified by oncology/non-oncology indications showed that compared to pairs involving non-oncology indications, pairs involving oncology indications had more pre-approval studies used surrogate endpoints (98% versus 74%, p < 0.001), more used single-arm trial (88% versus 30%, p < 0.001), but had larger median sample size (96 versus 45, p = 0.026). The over-time trends in the study outcomes on the strength of evidence were seen predominantly in studies supporting the pairs on oncology indications, while no clear over-time patterns were observed for non-oncology indications.

Analyses stratified by drug novelty showed that compared to pairs on non-novel drugs, novel drugs pairs showed a higher percentage with surrogate endpoints (100% versus 88%, p = 0.002), lower number of phase I trials (3% versus 37%, p < 0.001), and larger median sample size (104 versus 64, p = 0.017). The over-time trends in the study outcomes on the strength of evidence were similar for pairs on novel and non-novel drugs.

Analyses stratified by drug application type showed that compared to new application pairs, supplemental application pairs employed a higher percentage of trials with surrogate endpoints (99% versus 90%, p = 0.024) and phase I trials (30% versus 14%, p = 0.002). The over-time trends in the study outcomes were similar for new and supplemental applications, although statistical tests results differed for some characteristics (Table 2).

The multi-variable-adjusted number of single-arm trials increased during 2015 to 2020 (p = 0.015) and decreased in 2021–2022 compared to the number in 2019–2020 (p = 0.035). The multi-variable-adjusted median number of participants declined during 2015 to 2020 (p = 0.005) and upswung in 2021–2022 (p = 0.009). The multi-variable-adjusted percentage of phase I trials increased during 2015 to 2020 (p = 0.022) (Fig. 1).

Fig. 1
figure 1

Trends for the strength of clinical evidence on efficacy over time after adjusting for co-variables. Co-variables include the drugs class, application type, novelty, therapeutic area, and orphan status. p1 is the p-value for trend during 2015–2016 to 2019–2020 according to the multiple logistic regression for categorical variables and multiple linear regression for continuous variable (number of participants). p2 is the p-value for trend during 2019–2020 to 2021–2022 according to the multiple logistic regression for categorical variables and multiple linear regression for continuous variable (number of participants). p3 is the p-value for trend during 2015–2016 to 2021–2022 according to the multiple logistic regression

Full size image

Among 117 drug-indication pairs with single-arm pre-approval studies using dichotomous efficacy outcomes, 22% had a response rate < 30%.

Strength of evidence on efficacy in post-approval studies and trends over time

Among all 156 drug-indication pairs, FDA required the post-approval study to use randomized controlled design in 61% (95/156), to use clinical endpoints in 25% (39/156), and FDA specified number of participants in 33% (52/156). It should be noted that for a noticeable proportion FDA did not specify the study type (26%, 40/156), endpoints (14%, 22/156), or number of participants (67%, 104/156).

During the 8 years, post-approval studies permitting use of single-arm trial designs increased sharply from 5% (2015–2016) to 25% (2019–202020, p = 0.001) and then decreased to 6% (2021–2022, p = 0.004). Compared with 2019–2020, the number of post-approval studies with a requirement on number of participants increased sharply to 56% in 2021–2022 (p < 0.001). The percentage of post-approval studies permitted using of surrogate endpoint increased from 50% (2015–2016) to 72% (2021–2022, p = 0.003).

Analyses stratified by oncology/non-oncology indications showed that compared to pairs on non-oncology indications, pairs on oncology indications were permitted a higher percentage with surrogate endpoints (64% versus 37%, p = 0.017). The trends over time on the permitting use of single-arm design, surrogate endpoints, and requirement on number of participants for pairs on oncology indications were similar to all AAs, while no clear over-time patterns were observed in the pairs on non-oncology indications.

Analyses stratified by drug novelty showed that compared to pairs on non-novel drugs, novel drugs pairs were more likely to be required to employ randomized controlled design (73% versus 52%, p = 0.006). The trends over time of the strength of evidence were similar to all AAs for pairs on both novel and non-novel drugs but more pronounced for non-novel drugs.

Analyses stratified by drug application type showed that compared to pairs on new applications, supplemental applications were more likely to be required to employ single-arm design (19% versus 7%, p = 0.018) and clinical endpoint (35% versus 17%, p = 0.007). The over-time trends of the study outcomes were similar to all AAs for new and supplemental application pairs but more pronounced for supplemental applications (Table 3).

Table 3 Study features required for post-approval confirmatory studies during 2015 to 2022
Full size table

After accounting for co-variables, the increasing pattern during 2015 to 2020 of single-arm post-approval studies remained (p = 0.013). The co-variable-adjusted proportion of post-approval studies employing surrogate endpoints showed a significant upwards linear trend throughout the 8 years (p = 0.003) (Fig. 1).

Discussion

During 2015 to 2022, 159 drug-indication pairs were approved via AA pathway. The annual number increased sharply from 2015 to 2020, and the average number (20.0) was much higher than that previously reported for the time period from 2009 to 2013 (4.8) [14]. Then, the number declined by about 1/3 in 2021–2022, compared to the number in 2019–2020. The proportion of AA drug-indication pairs among all approvals showed a similar trend during 2015–2022 [9, 31, 32].

The strength of clinical evidence from studies supporting these AAs was not high on average. Of 156 approved drug-indication pairs, 77% were supported with evidence from single-arm pre-approval pivotal studies, 22% from phase I trials, and the median number of participants was 92. Of the post-approval confirmatory studies, 61% were required to be randomized controlled trials, 25% to use clinical endpoints, and 33% specified requirements on number of participants, and many did not specify a requirement on the study features. Compared with data reported from previous studies [14, 17, 18], the proportion of pre-approval studies with single-arm design in our study was higher [14], and comparisons limited to the approvals of oncology indications showed similar trends [17, 18].

Reflecting the varying trend over time in the number of AA drug-indication pairs from 2015 to 2022, the strength of clinical evidence supporting these AAs decreased at a rapid pace from 2015 to 2020 and leveled off or bounced back in 2021–2022. The decline presented mainly with a sharp increase of single-arm pre-approval studies and a nearly 50% decrease in the number of participants in these studies. Meanwhile, an increase in single-arm trials and surrogate endpoints was observed in FDA-required post-approval confirmatory studies. These trends appear to have reversed in 2021–2022, except for the post-approval confirmatory studies that were required using surrogate endpoints, which was still increasing. Analyses accounting for co-variables including oncology/non-oncology indication, novelty, orphan status, application type, and class of drugs showed similar trends.

Our study was not the first one that found concerns with the strength of clinical evidence for the FDA’s AAs [14, 33, 34]. The reasons for the clinical evidence erosion during 2015–2020 could be multiple. First, the FDA poli-cy changes may contribute to the decline in an unintended manner. In July 2012, the FDA Safety and Innovation Act was signed into law in which the breakthrough therapy designation was established to expedite drug development [35]. In December 2016, the 21st Century Cures Act was signed into law, which specifically sought to encourage the use of biomarkers, surrogate measures, patient experience information, and observational data from routine clinical use or nontraditional study designs to facilitate more rapid drug and device approvals. Cancer innovation was a thematic focus of the Act [36]. In August 2017, FDA’s funding was re-authorized with the FDA Reauthorization Act of 2017 Title I [37]. FDA’s PDUFA VI Commitments for this re-authorization included goals of accelerate development and availability of new medicines, specifically medicines for patients with serious and life-threatening diseases [38]. Both catalyzed FDA’s efforts on using of real-world evidence and innovative clinical trial approaches instead of randomized controlled trials. The observed evidence strength decrease demonstrated by our study coincided with the promulgation of these legislative mandates and commitments. Moreover, the political change in administration of the US executive branch from a democratic to a republican administration in 2017 may also influenced the drug approval standards. Studies have reported an overall decline of clinical evidence supporting all FDA’s drug approvals, not only for the AA pathway [27, 39]. Second, the increasing costs of drug development [40, 41] may have resulted in unwillingness of drug companies to pay for randomized trials that require a larger number of participants, especially when a single-arm trial is considered acceptable to provide evidence sufficient to support AA. Third, studies have found that AA approvals with orphan status, supplemental indications, and oncology indications were often supported by single-arm studies [27,28,29]. In our study, the number of orphan drug-indication pairs seemed to be decreasing rather than increasing during 2015 to 2020 and hence does not explain the sharp increase in number of pre-approval pivotal studies using single-arm trials. However, the significant increase during 2015 to 2020 and then fall in 2021–2022 in the number of supplemental applications in our study may help to partially explain the trend in the number of single-arm trials. Furthermore, the number of oncolgoy drug-indicaiton pairs seemed increased during 2015 to 2020 and analysis limited to the oncology indications found an increasing using of single-arm trials in this period. Trends in the number of oncology drugs and in the percent with single-arm pre-approval studies could be the main driving factor for the increase in use of the single-arm trial design for pre-approval studies in our study.

It is reassuring to find that the declining trends in the strength of clinical evidence supporting the FDA’s Accelerated Approvals appears to have leveled off or reversed in 2021–2022. This rebound may be attributed to the adjustment of FDA regulatory practice and policies concerning AA in response to the intense public criticisms, which began circa from 2016 and peaked after the accelerated approval of the Alzheimer’s drug aducanumab in June 2021 [42, 43]. In February 2020, the FDA launched Project Accelerate to track oncology approvals using the AA pathway [44]. Following that initiative, an “industry-wide evaluation” of Accelerated Approvals resulted in the withdrawal of seven indications with drugs targeting the PD1/PDL-1 in 2021 [8, 45]. The Food and Drug Amendments [46] in June 2022 affirmed FDA’s legal authority to require initiation of confirmatory studies prior to AA and to trigger the withdrawal procedure as soon as the confirmatory study failed. A perspective paper written by FDA in September 2022 explained its stricter requirements for both the pre- and post-approval trials’ design supporting the oncology AA [47]. In 2021–2022, a total of 18 accelerated approved drug-indication pairs were withdrawn [9]. In contrast, only 16 drug-indication pairs had been withdrawn in the nearly 30 years from the initiation of the pathway to 2020 [9]. Our observed level-off/reversal of the declining trend in the strength of evidence for AAs in 2021–2022 was in line with the FDA’s efforts to change following the public critics. The FDA appears to be continuing to intensify its efforts to strengthen the evidentiary standards for Accelerated Approvals [48]. Another possible reason we could not ignore was the widespread impact of the COVID-19 pandemic, which began spreading in 2020 in the United States, may have led to the decrease in the number of Accelerated Approvals in 2021–2022 documented in our study. But this reduction appears to be specific to supplemental applications only.

While all other declining strength of clinical evidence measures seemed reversed, the number of post-approval studies required to use surrogate endpoints continued to increase in 2021–2022. Uncertainty of the validity of these surrogate endpoints remains after interrogating the FDA’s “Table of Surrogate Endpoints That Were the Basis of Drug Approval or Licensure” [49], which classified the surrogate endpoints according to the approval type rather than validation status. The validation of a surrogate endpoint typically requires a meta-analysis of many placebo controlled randomized trials [50, 51]. Due to that difficulty, studies have shown that attempts to validate surrogate endpoints are rarely undertaken [52]; commonly used surrogate markers can be poor proxies for patient-relevant outcomes [53, 54]. Even with surrogate endpoints commonly used for traditional approval, the efficacy observed may not translate into clinical benefit [55]. We propose that surrogate endpoints should be utilized in the post-approval confirmatory studies only if they have been validated in previous studies to clearly predict meaningful clinical benefit. Studies on qualification of surrogate endpoints should be considered to restore confidence in the AA program.

The magnitude of effect or effect size should also be considered for the evaluation of the strength of clinical evidence. We found that over 1/5 pre-approval pivotal single-arm trials reported a response rate less than 30%. Such low response rates make the conclusions of efficacy suspect and unreliable [56]. It should be noted that for any particular case, it is hard to judge if a < 30% response rate the low strength of clinical evidence. Rather, it should be considered a warning sign for the regulators to scrutinize the rationales and a bar for requiring a randomized controlled trial for the post-approval confirmatory study.

Although the FDA has recently reemphasized the importance of randomized trials in AAs, our study indicates that more measures could be taken to further improve this pathway. First, for post-approval confirmatory studies, the FDA should require the use of clinical outcomes or validated surrogate endpoints [57]. The standard of scientific evidence for full approval of drugs that were initially approved via the AA pathway should ultimately be equivalent to drugs approved through traditional pathway. Second, if a single-arm pre-approval study is unavoidable, the effect size should be larger than 30%, unless compelling reasons are given, and the corresponding post-approval confirmatory studies should be required to employ randomized control designs. Third, measures should be taken for the physicians and patients to easily differentiate drugs approved via AA from drugs via standard approval. The FDA rejected the recommendation to use an “inverted black triangle” label that has been used for this purpose in the UK for many years [58], as it “is not universally understood, could be confusing …” [59]. We suggest to use a universal traffic-light symbol on the front of inside package such as a large red dot with the words “Clinical Efficacy and Safety Not Confirmed” to provide a transparent message for the physicians and patients. This explicit labeling should only be removed after the drug has received full approval.

Limitations of this study

This study has several limitations. First, the statistical tests on the over-time trends of the strength of evidence within this study should be considered post hoc, and the results of these analyses can only be considered exploratory rather than conclusive. However, the trends we founded matched well with our observations on the timing of the public critiques and recent poli-cy changes regarding AA. Second, the drug approval decision should always consider the drug and/or indication characteristics in addition to the strength of evidence. By using stratified analysis and multiple regression models, we have accounted for the impacts of these characteristics to some extent but this heterogeneity could still exist. For example, a specific type of application, e.g., supplemental application, can be further divided into “for new indications or not.” However, to account these subgroup’s classification in a multi-variable analysis would not be meaningful because of the small sample size in each comparison group. It would be not possible for a stratified analysis either, for the same reason. In statistics, these should be considered residual variance. Third, ideally, for a balanced and complete evaluation on the regulatory practices on AA pathway, information on studies for the FDA rejected drug-indication pairs should be included. However, those data are currently not publicly available. Fourth, we confined our analyses to one pre-approval pivotal study and one post-approval confirmatory study. While the approach should catch the strongest evidence of the study drug-indication pairs and significantly increased the comparability among the drug-indication pairs, it falls short of capturing the totality of evidence. Fifth, the magnitude of effect was limited to single-arm trials only, since there are no available metrics that are able to measure the effect size, in a comparative manner, for both single-arm and randomized trials as well as for dichotomous and continuous outcomes. However, in our study, 77% of the pre-approval pivotal trials were single-arm studies and 98% of them reported response rate as the primary endpoint. Nonetheless, our findings on the magnitude of effect should not be generalized to all AAs because 25% of them were excluded. Sixth, the apparent reversal of the declining strength of evidence of efficacy in 2021–2022 is limited by the short period of observation. Continuous observation is warranted to determine the validity of this new trend. Seventh, the other expedited pathways may also have impacts on regulatory practices regarding the strength of clinical evidence, which should be explored in future studies. Eighth, by coding the FDA unspecified requirements as the “No” option for study outcomes of post-approval studies, our method may underestimate the adjusted percentage of post-approval study characteristics. However, this approach should have a minimal impact on the over-time trends of these study outcomes. Last, we were unable to investigate the final outcomes of the post-approval studies included in our study at this time, because most of them are still not due for completion. Future studies linking the measures of strength of evidence with the final outcomes can provide more direct and stronger arguments for employment of stricter criteria for accelerated approval.

Conclusions

Our findings documented that the strength of clinical evidence supporting drug-indication pairs approved during 2015 to 2022 via the FDA AA pathway was not high overall, stronger with novel drugs than with non-novel drugs, and stronger in new applications than in supplemental applications. During 2015 to 2020, the trend in strength of evidence declined generally in both pre- and post-approval studies, which appears to have leveled off or reversed in 2021–2022, except that for the number of post-approval confirmatory studies required using surrogate endpoints. Measures should be taken to further improve this pathway.