Use of Extrapolation in New Drug Approvals by the US Food and Drug Administration

Key Points Question How often does the US Food and Drug Administration extrapolate the findings of a drug’s pivotal trial data to its approved indication, in terms of disease severity, subtype, or use of concomitant medication? Findings In this cross-sectional study of 105 new drugs approved from 2015 to 2017, extrapolation beyond pivotal trial participants to other patient populations was noted 23 times in 21 drug approvals. Extrapolation of disease severity was most common, followed by disease subtype and concomitant medication use. Meaning The findings of this study suggest that extrapolation by the US Food and Drug Administration beyond pivotal trial data to final indications is common, highlighting the need for close postapproval monitoring.


Introduction
The US Food and Drug Administration (FDA) approves a new prescription drug if its benefits are judged to outweigh its risks in pivotal clinical trials 1 ; the FDA and manufacturers then define the approved indications for the drug based on evidence from the trials. The indication is registered in drug labeling to guide prescribing and promotional statements. It can also help establish the contours of insurance coverage. 2 However, pivotal trials leading to FDA approval are necessarily limited in the populations they cover. 3 Although attention has been given to limitations in the demographic diversity of the trial population, 4,5 less attention has been paid to whether a narrowly defined clinical trial population is then used as the basis for extrapolation of trial findings to a broader population of patients than was represented in those trials. Controversially, the FDA initially approved aducanumab in 2021 for all patients with Alzheimer disease, even though it was studied only in those with mild disease. 6 Extrapolation from pivotal trials to a broader approved indication may be appropriate when there is no basis for expecting a different outcome in the wider postapproval population. 7 However, extrapolation can also lead to different clinical outcomes in practice than were observed in the pivotal trials. Such differences were seen on a large scale with the approval of spironolactone for heart failure based on the Randomized Aldactone Evaluation Study when, after publication of the trial, spironolactone prescription rates in routine practice increased by a factor of 5 to include a substantially older population than was included in the initial trial. Inclusion of these individuals led to a 3-fold increase in mortality associated with hyperkalemia in patients given the drug, far higher than was seen in the original clinical trials. 8,9 This increase was due in part to differing populations being prescribed spironolactone in practice, as well as concomitant medication use and eligibility criteria compared with those used in the original trial. 9 To our knowledge, no systematic analysis of such extrapolation in FDA approvals has been published; however, several meta-analyses and trials discussed herein have reported differences in outcomes when a drug is used in patients with disease severity, disease subtypes, or concomitant medication use that differed from those studied in the pivotal trials. In a study of patients with rheumatoid arthritis included in trials of tumor necrosis factor inhibitors compared with patients receiving the drug in routine practice, the latter individuals had substantially lower disease severity than those eligible for the trials and experienced less clinical benefit. 10 Use of corticosteroids occurred in 44% to 69% of the trial participants, compared with 29% to 54% of those seen in routine care. 11 This higher concurrent corticosteroid use during the trial could have also contributed to the differing outcomes seen in the 2 settings. Differences in disease subtype can also influence outcomes. 12 A study of treatments for non-small cell lung cancer found that improved accuracy of histologic categorization would result in more appropriate prescribing and outcomes in routine care. 13 Yet extrapolation of findings from a pivotal trial that predominantly included 1 histologic subtype to all histologic subtypes of non-small cell lung cancer could influence patients' response to that treatment. This pattern has been observed across many diseases. [14][15][16] To determine how frequently extrapolation occurs from pivotal trial populations to final FDA-approved indications, we analyzed a cohort of FDA drug approvals to determine how 3 key clinical characteristics of patients in the pivotal trials were represented in the approved indications: disease severity, disease subtype, and concomitant medication use.

Data Sources and Extraction
We identified 113 new molecular entity and biologic drug approvals after pivotal trials from 2015 to 2017, using the FDA website, 17 and reviewed the approved indications on the original drug labeling.
We then used Section 7 of the FDA Summary Reviews to identify the pivotal trials for each medication (ie, those submitted to the FDA to support the drug's approval). Additional resources included the because of our inability to analyze data from all pivotal trials, as well as 1 drug approved based on animal studies alone.

Data Analysis
We studied extrapolation of findings from trial participants to approved indications across 3 domains: disease severity, subtypes of disease treated, and concomitant medication use. We compared a drug's pivotal trial characteristics with the FDA-approved indications. Extrapolation was defined as the granting of an indication for use in a broader population than was included in the pivotal trials on the basis of disease severity, disease subtype, or concomitant medication use. Underrepresentation or skew toward 1 end of a spectrum of disease severity, disease subtype, or concomitant medication use was also considered extrapolation. More than 1 category of extrapolation could be related to each approved indication.
Instances of extrapolation were systematically identified by one of us (D.F.) and analyzed by the team thereafter. In addition, one of us (A.S.K.) independently analyzed 20% (n = 21) of the 105 approvals. The approvals were selected randomly by assigning a number from 1 to 105 and using a random number generator to select the sample. The findings from the sample analysis were discussed to ensure consistency and reproducibility.
To assess disease severity, we examined the scales or criteria used to classify, include, or exclude patients from a trial based on disease activity or severity. For example, the trials for secukinumab to treat psoriasis and other conditions included only patients with a modified investigators' global assessment score of 3 to 4, indicating moderate to severe plaque psoriasis. This limited population was reflected in the language of the approved indication. If the indication had not specified moderate to severe, we would have considered the indication to be based on extrapolation because the trial did not include patients with lower disease severity. Similarly, extrapolation was designated if the trial population was heavily skewed toward one side of the spectrum of disease severity but was not indicated on the formal indication.
To assess disease subtypes, we determined the range of variations or subtypes for a given condition and compared those included in the pivotal trials with those in the approved indication. For example, patients in the secukinumab trial were included only if they had plaque-type psoriasis; the indication reflects this, as the drug is indicated for use only in patients with this diagnosis. If, however, the indication had not specified plaque psoriasis, we would have considered this to be extrapolation because the trial did not include patients with any other psoriasis subtype. If trials were heavily skewed toward one subtype of disease that was not reflected in the official indication, that would also represent extrapolation. Heavily skewed, in most cases, was defined as 5% or less of the study population.
To assess concomitant medication use, we determined whether trial participants were receiving additional relevant medications and whether that concomitant therapy was reflected in the approved indication. For example, extrapolation would be designated if patients in a trial of secukinumab were using a topical corticosteroid along with the study drug, but the combination was not included in the indication. Investigators, however, prohibited the use of topical corticosteroids during the trial, which aligns with the approved indication for secukinumab. Data analysis was performed using Microsoft Excel, Version 16.58, (Microsoft Corp). Extrapolation of indications based on the use or nonuse of concomitant medications was the least common (n = 3). For example, patients in the pivotal trial for ivabradine for heart failure were also receiving a variety of other treatments to manage this condition. However, the approved indication refers only to β-blockers, which were being used by 89% of patients given the study drug;

JAMA Network Open | Health Policy
the label does not refer to concurrent use of angiotensin-converting enzyme inhibitors, which were being used by 79% of patients receiving the study drug.

Discussion
Using data from 105 novel FDA drug approvals from 2015 to 2017, this cross-sectional study sought to define instances of clinically meaningful extrapolation of indications beyond the characteristics of the pivotal trials on which drug approval was based. With this information, we found that  extrapolation occurred in 20% of new drug approvals. We noted that such extrapolation is common and describe its prevalence in approved indications of many new drugs.
Some of the extrapolations we identified may be clinically plausible, but others could have implications for the effectiveness and safety of drugs when used in accordance with the FDA-approved indication. In some cases, the FDA acknowledges the presence of extrapolation in its review materials. For example, in the approval of edaravone for ALS, for which the pivotal trial showed varying efficacy based on disease severity, the FDA concluded that the disease has a variable course and it may not be possible to identify a stage at which treatment benefit may or may not be achieved. However, only trials that excluded patients with more severe disease found statistically

Limitations
This study has limitations. The analysis was limited to 3 years of new drug approvals and was based on publicly accessible data. We therefore did not have insight into any nonpublished confidential data or discussions between the FDA and manufacturers and were unable to calculate descriptive statistics that were not included in the sources reviewed. Similarly, we were unable to identify whether all subtypes of a disease being treated were included in pivotal trials if the accessible data did not include the breakdown within participants. This level of access, however, simulates the access clinicians would have to guide their prescribing should they explore the data behind the approval.
Another limitation of this study is that, owing to the variability between trials and therapy targets, we could not apply a consistent definition for extrapolation across all the drugs. In addition, this study was not designed to investigate the clinical outcomes associated with the extrapolations identified, which should be the subject of future clinical trials and observational studies.

Conclusions
Extrapolation of a drug's indications beyond information available from the pivotal trials on which approval was based is often necessary in the approval of new medications because preapproval trials cannot possibly cover all patient subpopulations, age categories, and comorbidities. However, extrapolation of clinically limited data to generate broad official indications can sometimes extend past the bounds of what is plausible given the characteristics of patients studied in preapproval trials, potentially influencing the drug's effectiveness and safety when used in routine practice. Such problems of nonrepresentativeness have been described in terms of patient age, sex, and race and ethnicity. To our knowledge, this is the first study to document the frequency of such extrapolation in terms of the clinical features of the disease being treated such as severity, subtype, and concomitant medication use. Although such extrapolation may be justified at the time of regulatory approval, our findings also point to the importance of follow-up research to confirm the expected outcomes. It may be beneficial to incorporate formal postapproval surveillance, using both prospective trials and wellconducted observational studies, into the rollout of novel therapies for which such extrapolation has occurred to ensure better ascertainment of real-world effectiveness and safety. When such clinical extrapolation is necessary at the time of approval, its details should be made clear in the labeled indications for physicians and patients. Until that time, it would be useful for physicians to recognize that the FDA-approved indication alone may be insufficient information on which to decide whether a given medication will benefit a patient who may differ meaningfully from those studied in the clinical trials on which approval was based.