FDA Precedents for Removing or Narrowing Boxed Warnings and the Supporting Postmarketing Evidence

FDA precedents for removing or narrowing a boxed warning, and the evidence behind them

FDA has removed or narrowed boxed warnings many times. The evidence behind these changes falls into distinct types. A few removals rest on large randomized postmarketing requirement (PMR) safety trials built to answer the question that prompted the box. Varenicline and the ICS/LABA class are the clearest examples. Others rest on long-term pharmacoepidemiology programs (teriparatide), on new outcome-trial data that changed the size of the risk (canagliflozin, rosiglitazone), or on FDA's own class-wide labeling reassessments that found the box was not needed to manage a well-understood risk (bevacizumab, cabozantinib, ipilimumab). In the removal cases, the risk itself stayed in the label, usually moved to Warnings and Precautions. FDA was not saying the risk had gone away. Narrowing cases are different: some risks stayed in a shorter box.

The regulatory standard a removal has to meet

FDA labeling guidance says a boxed warning is ordinarily used when one of the following applies ¹⁹⁵:

• an adverse reaction is so serious relative to benefit (fatal, life-threatening or permanently disabling) that prescribers must weigh it;
• a serious reaction can be prevented or reduced by appropriate use (patient selection, monitoring, avoiding certain concomitant therapy);
• the drug was approved with restrictions to assure safe use.

The guidance also allows a box for a risk-benefit issue that is unique within a drug class ¹⁹⁴. The same guidance does not set separate criteria for removing a box. It does say that application holders must keep labeling accurate and must change it when new information makes it inaccurate or misleading ²¹⁰. A substantive boxed warning change must be listed under Recent Major Changes in Highlights for 12 months ¹⁹⁶. In practice, the precedents below show that removal arguments work best when they speak directly to these criteria. The strongest ones show that the risk is smaller than first thought, that it is manageable without a box, or that the box is inconsistent with how FDA labels comparable products.

Category 1: Removal supported by dedicated postmarketing randomized trials

Chantix (varenicline): neuropsychiatric events, removed December 2016

This is the best-documented precedent. FDA removed the boxed warning for serious neuropsychiatric adverse events on December 16, 2016. The risk stayed in Warnings and Precautions, which was revised to include frequencies from the postmarketing safety trial ⁶⁶²⁸.

The supporting evidence was PMR study A3051123 (EAGLES). This was a randomized, double-blind, active- and placebo-controlled trial of varenicline 1 mg twice daily, bupropion 150 mg twice daily, nicotine replacement therapy (NRT) and placebo. Patients received 12 weeks of treatment followed by 12 weeks off treatment ⁴⁴. The trial enrolled 8,144 patients at 140 centers in 16 countries. It was deliberately split into a non-psychiatric cohort and a psychiatric-history cohort of 4,074 patients with stable, confirmed DSM-IV diagnoses ⁴⁴.

• Primary endpoint: a composite of clinically significant neuropsychiatric events during treatment plus 30 days ²⁸²²⁸⁵. The trial was designed to estimate risk. It was not designed to exclude a prespecified margin or to assess rare outcomes such as suicidality ²⁸⁵¹⁷⁷.
• Non-psychiatric cohort: varenicline was not associated with an increased incidence of the composite endpoint ²⁸²²⁷⁹. Serious neuropsychiatric events occurred in 0.1% on varenicline versus 0.4% on placebo ²⁸⁶.
• Psychiatric cohort: event rates were higher in every arm. The risk difference for varenicline versus placebo was 1.59% (95% CI -0.42% to 3.59%) in one FDA summary and 2.7% (-0.05% to 5.4%) in the labeling summary ²⁸²²⁸⁶.
• Sensitivity analysis: an FDA analysis that excluded patients previously exposed to study drugs did not change the primary result ²⁶⁵⁶.

At the September 14, 2016 joint advisory committee meeting, members voted 10 to remove the box, 4 to modify it and 5 to retain it ⁷²²⁸. The committee concluded that observational studies added nothing beyond the randomized trial. It also stressed that psychiatric history was associated with greater risk, which should be communicated in labeling ⁵⁶. Earlier pooled analyses of Phase II to IV trials had found no increased risk, but FDA noted that those trials were not designed to collect or adjudicate these events ²⁶. That limitation is why a dedicated trial was needed.

EAGLES also included a Zyban (bupropion) arm ⁴⁴. The Zyban labeling from the supplement decided June 1, 2016 still carried a boxed warning for serious neuropsychiatric reactions in patients taking bupropion for smoking cessation ²⁶⁷. FDA removed that part of the box on May 4, 2017 (NDA 020711/S-047, after EAGLES), and released the smoking-cessation REMS the same day; the antidepressant-class boxed warning on suicidal thoughts and behaviors remains (FDA approval letter, May 4, 2017). Earlier (2013) bupropion labels had cross-referenced the smoking-cessation neuropsychiatric risk to the Boxed Warning ²⁵⁷. The retrieved record does not fully document the decision rationale for bupropion on its own.

ICS/LABA combinations: asthma-related death, class-wide removal December 2017

On December 20, 2017, FDA removed the asthma-related death boxed warning from inhaled corticosteroid/long-acting beta-agonist (ICS/LABA) combination products, including Advair Diskus and Symbicort, as a class labeling change ⁴²⁴⁷.

The box originally came from the SMART trial, which studied salmeterol added to usual therapy without requiring an ICS. That trial found 13 asthma-related deaths in 13,176 patients on salmeterol versus 3 in 13,179 on placebo (relative risk 4.37; 95% CI 1.25 to 15.34) ³⁰¹³¹⁶. FDA regarded this as a class effect of LABA monotherapy ³⁰¹.

After advisory committee discussions in 2008 and 2010, FDA required a harmonized PMR program. The trials were similarly designed and shared adjudication, steering and data monitoring committees so that rare events could be pooled ³¹²³⁰⁶. Four trials were completed. The fifth (Foradil) was terminated after that product left the U.S. market ⁴⁵.

• Design: 26-week, randomized, double-blind noninferiority trials of ICS/LABA versus ICS alone. Three enrolled adults and adolescents and one enrolled children aged 4 to 11. The primary endpoint was a composite of asthma-related hospitalization, intubation or death ³⁰²³⁰¹.
• Noninferiority margins: 2.0 for adults and adolescents, 2.7 for children. Every trial met its noninferiority objective ³⁰²²⁷⁴.
• FDA meta-analysis: hazard ratio 1.10 (95% CI 0.85 to 1.44) in adults and adolescents, based on 17,537 ICS/LABA and 17,552 ICS patients, and 1.29 (0.73 to 2.27) in children ³⁰⁵³⁰¹.
• Event counts in adults and adolescents: 116 composite events on ICS/LABA versus 105 on ICS, including 2 versus 0 deaths ³⁰¹. The pediatric trial had no asthma-related deaths or intubations ³⁰¹.

FDA's epidemiology and pulmonary divisions concluded that the four trials and the meta-analysis did not support keeping the box ⁴⁵³¹⁰. Warnings and Precautions was revised to keep and emphasize the asthma-related death risk of LABA monotherapy without an ICS ²⁹⁸. FDA also acknowledged that the trials could not rule out every level of risk and deferred conclusions on race and sex subgroups ³⁰²²⁹⁷.

Category 2: Removal supported by long-term pharmacoepidemiology

Forteo (teriparatide): osteosarcoma, removed November 2020

FDA removed the osteosarcoma boxed warning on November 16, 2020 ²²⁹³⁵⁶. The evidence was a required five-part postmarketing surveillance program ³⁵¹³⁵⁶:

FDA judged the registry underpowered against its 1.7 million person-year target. It still concluded that the registry results did not suggest a risk ³⁴⁹. The multidisciplinary review found no evidence of increased osteosarcoma risk with up to two years of treatment ³⁴⁸³⁴⁹. FDA's Medical Policy and Program Review Council had agreed in April 2019 that the box was no longer necessary ²²⁹.

The removal came with related changes. An osteosarcoma precaution was kept for patients at higher baseline risk (open epiphyses, Paget's disease, prior skeletal radiation) ³⁶¹. The lifetime two-year limit was relaxed so that longer use "should only be considered" in patients who remain at high fracture risk ³⁵⁷³⁶¹. This is the main precedent for removing a box that was based on a nonclinical signal, using a mix of case series, a registry and claims-based cohort studies.

Category 3: Removal or narrowing after new outcome-trial data

Invokana and Invokamet (canagliflozin): lower-limb amputation, removed August 2020

FDA approved removal of the amputation boxed warning in August 2020 ²³⁷³³². The original box rested on CANVAS (hazard ratio 2.12; 95% CI 1.34 to 3.38) and CANVAS-R (hazard ratio 1.80; 95% CI 1.10 to 2.93) ³¹⁸³³⁰.

CREDENCE studied a higher-risk diabetic kidney disease population. It showed a much smaller difference: 3.2% versus 2.9% (12.3 versus 11.2 per 1,000 patient-years), with an on-study hazard ratio of 1.1 (95% CI 0.8 to 1.6) ³¹⁸. FDA reviewers said the smaller gap likely reflected risk-mitigation measures used in CREDENCE ³²⁹. FDA's August 26, 2020 Drug Safety Communication gave the public rationale: newly identified heart- and kidney-related benefits significantly enhanced the benefit of canagliflozin, and the amputation risk, while still increased, was lower than previously described when patients were monitored. Lower-limb amputation stayed as Warnings and Precautions section 5.1, with instructions to assess risk factors, monitor for infection or ulcers, and stop the drug if these occur ³³³.

Avandia (rosiglitazone): REMS and box narrowed after the RECORD readjudication, 2013 to 2014

By 2011, the Avandia box covered congestive heart failure and myocardial infarction. The drug was available only through a restricted access program ¹⁵³. On November 25, 2013, FDA used section 505(o)(4) to require safety labeling changes reflecting ischemic cardiovascular risk "as assessed by the readjudicated results" of the RECORD trial ¹⁵⁵.

The resulting supplement, approved May 7, 2014, removed the access-program language from the box and the patient-population restrictions from Indications and Usage. The same action approved a modified REMS: the restricted prescriber, pharmacy and patient requirements were dropped, and the remaining REMS consisted of health-care-provider training. FDA eliminated the REMS entirely on December 16, 2015 (NDA 021071/S-050 approval letter). The remaining boxed warning addressed congestive heart failure only ³⁶²³⁶³. The retrieved records do not report the readjudication's numerical findings.

Category 4: Removal through FDA class-wide labeling reassessment

In several recent cases, FDA itself requested removal after comparing labeling across a drug class. The sponsor submitted no new safety data.

• Avastin (bevacizumab), June 2019: the box for GI perforation, wound-healing complications and hemorrhage was removed, and that content moved into Highlights and Warnings and Precautions ¹³¹. FDA found that VEGF-inhibition risks were similar across the class. Postmarketing experience with other VEGF inhibitors that had no boxed warning showed no unacceptable toxicity ³¹⁷³. Genentech submitted no new clinical or safety data ⁷³.
• Cometriq (cabozantinib), January 2020: the box for perforations, fistulas and hemorrhage was removed because comparable VEGFR inhibitors had no box and there was no evidence that patients on those drugs faced greater risk ³². The sponsor also submitted a global safety database search (1,214 surgical or procedural cases, narrowed to 186) and a literature review ²⁸⁹²⁹¹. FDA found the wound-healing data too limited to set a drug-specific restart interval ²⁹¹. The label adopted class-based withholding guidance instead: at least three weeks before elective surgery and two weeks after major surgery ²⁹².
• Yervoy (ipilimumab), June 2020: the box for immune-mediated adverse reactions was removed after a reassessment of PD-1/PD-L1 labeling, none of which carried such a box ²³⁶²⁴¹. The "evidence" was accumulated clinical experience. FDA concluded that oncologists could now recognize and manage these toxicities, and that Warnings and Precautions was adequate. No new data were submitted ³³⁴³³⁵.
• Older labeling clean-ups: Erbitux (2007) cut its boxed warnings down to the most critical elements ⁷. Doxil (2015) dropped myelosuppression, hepatic dose reduction and accidental substitution from its box because they did not reach boxed-warning severity ⁸³. Fludarabine (2024) moved its box content to Warnings and Precautions to align with current labeling practice, based on a review of existing safety data and published literature ²²⁸.

Category 5: Narrowing to a specific population or condition of use

• Promacta (eltrombopag), February 2014: the general hepatotoxicity box was narrowed to hepatic decompensation in chronic hepatitis C patients receiving interferon and ribavirin. This took ITP patients out of the box, while liver enzyme elevations stayed in Warnings and Precautions ^2512033. The sponsor pointed to accumulated clinical-trial and postmarketing experience. FDA found no new or emerging safety issues and concluded the REMS communication plan was no longer needed ²⁰³³⁴¹.
• Addyi (flibanserin), 2019: the box was kept but revised. The absolute alcohol prohibition became a timing-based restriction: alcohol must be stopped at least 2 hours before the bedtime dose ¹²¹⁰⁸. The evidence came from postmarketing alcohol-interaction studies. In one, 64 women drank about two drinks 2, 4 or 6 hours before dosing and had hypotension rates similar to controls ¹⁶⁸¹⁷⁰. A concurrent-dosing PMR in 96 women showed missing orthostatic measurements that rose with alcohol dose. FDA read this as evidence that the concurrent-use risk was real ¹⁶⁰¹⁶⁴.
• Soriatane (acitretin), 2003: non-teratogenicity content was removed from the box and the pregnancy box was kept. The record says this was not based on new "improved" safety data ⁶³.
• Menopausal estrogen products: in 2019, Divigel's box was revised to state that the WHI estrogen-alone findings came from daily oral 0.625 mg conjugated estrogens only, and that their relevance to other doses and routes is unknown ⁹⁵. Labeling revisions dated February 12, 2026 went further. The cardiovascular disorders, breast cancer and probable dementia components were removed from the boxes of Cenestin, Enjuvia and Divigel, leaving only the endometrial cancer box for unopposed estrogen ^125126127. Bijuva's box was removed entirely ¹³⁰. Those risks remain in Warnings and Precautions and Contraindications ¹²⁵¹³⁰. The change was not yet uniform: EstroGel's May 2026 label still carried the broad box ¹³². The retrieved labeling does not state FDA's evidentiary rationale ¹²⁵¹³⁰.

Summary of precedents

Practical takeaways for sponsors

1. Match the evidence to the original signal. Removals based on randomized trials (Chantix, ICS/LABA) used trials designed around the exact endpoint that prompted the box. They used adjudicated composite endpoints and, for ICS/LABA, prespecified noninferiority margins ³⁰²²⁸⁵. Pooled analyses of older trials that did not collect the events systematically were given little weight ²⁶.
2. Rare-event risks can be addressed with epidemiology. For osteosarcoma, FDA accepted a set of case series, a registry and claims cohorts even though the registry was underpowered, because every source pointed the same way ³⁴⁹.
3. Class consistency is a real argument. When comparable products carry similar risks with no box and postmarketing experience shows no excess harm, FDA has removed boxes on its own initiative ^3132334.
4. Expect the risk to stay in the label. In the removal cases, the underlying risk stayed in Warnings and Precautions, often with new detail, monitoring instructions or trial data ^28298333361. Where FDA only narrowed the box, part of the risk stayed in the box itself (Avandia congestive heart failure; Addyi alcohol timing).
5. Plan for REMS and restrictions at the same time. Box changes often came with REMS release or relaxed restrictions (Avandia, Promacta) ³⁶²³³.
6. Advisory committees help but are not required. Chantix went to a joint advisory committee vote ⁷². The retrieved record for the 2017 ICS/LABA decision shows no advisory committee vote on removal ³⁰⁶.

Topics that deserve a closer look include the RECORD readjudication results behind the Avandia change, FDA's stated rationale for the 2026 menopausal hormone therapy labeling, and whether the remaining estrogen products have since followed the same path.