When the US Food and Drug Administration (FDA) approves a drug while a serious risk is still uncertain, the law sets an order of tools. Before FDA can require the company to run a postmarketing safety study, it must decide that its own systems cannot answer the question. Those systems are the FDA Adverse Event Reporting System (FAERS) and the Active Risk Identification and Analysis (ARIA) system in the Sentinel Initiative [1][8].

A research letter published online in JAMA Internal Medicine on 16 November 2026, by a team from Emory and Yale, shows how that order plays out [1][2]. The authors reviewed the 448 novel drugs and biologics FDA approved from 2016 to 2025. At first approval, 303 (67.6%) had at least one serious risk that FDA said needed further assessment, 865 risks in all [1]. Adding 96 risks identified when indications were extended gives 961 serious risks. FDA judged FAERS sufficient to evaluate none of them, and ARIA sufficient for 10 (1.0%) [1]. Almost everything else became a postmarketing requirement (PMR) for the sponsor to fulfil.

This article sets out what the study found, why FDA's systems fall short, how sponsor-run studies are progressing, and what this means for safety teams. It is analysis, not legal or regulatory advice.

Key takeaways

  • Two thirds of novel therapeutics approved in 2016–2025 (303 of 448) had at least one serious risk needing further assessment at approval [1].
  • Of 961 serious risks at original and supplemental approval, FAERS was judged sufficient for 0 and ARIA for 10. The six ARIA studies that resulted were all still ongoing in June 2026, with no regulatory outcomes reported [1].
  • Only 178 of the 951 risks where ARIA was insufficient had a published ARIA Sufficiency Memorandum. In those memos, the most common problem was that ARIA could not identify or measure the outcome (151, 84.8%) [1].
  • FDA issued 900 safety-related PMRs at original approval and 63 at supplemental approval. New clinical trials made up 6.4% and 9.5% of them [1].
  • For sponsors, the message is to assume that a serious risk flagged at approval will become their own study. Plan and resource it before the action letter.

The study

The authors used Drugs@FDA to identify novel therapeutics approved from 2016 to 2025 and their supplemental approvals for new indications or populations [1]. They read the approval letters to find serious risks needing further assessment and FDA's decisions on whether FAERS and ARIA were sufficient. They used ARIA Sufficiency Memoranda, where available, for FDA's reasons [1]. For risks that ARIA could handle, they looked up the matching ARIA studies on the Sentinel website. For the rest, they classified the safety-related PMRs by study design [1]. The study is cross-sectional, uses only public documents, and follows the STROBE reporting guideline [1].

Table 1. Characteristics of the therapeutics studied (adapted from the paper's Table [1]).
CharacteristicOriginal approval, n = 448, No. (%)Supplemental approval, n = 131, No. (%)
Drug / biologic314 (70.1) / 134 (29.9)86 (65.6) / 45 (34.4)
Cancer and haematology indication154 (34.4)60 (45.8)
Infectious disease55 (12.3)18 (13.7)
Neurology and psychology54 (12.0)14 (10.7)
Autoimmune, musculoskeletal, dermatology52 (11.6)16 (12.2)
Cardiovascular, diabetes, hyperlipidaemia22 (4.9)5 (3.8)
Other indications111 (24.8)18 (13.7)
Orphan designation229 (51.1)62 (47.3)
Any expedited pathway315 (70.3)84 (64.1)
Priority review / fast track / breakthrough / accelerated approval281 (62.7) / 164 (36.6) / 144 (32.1) / 82 (18.3)80 (61.1) / 11 (8.4) / 21 (16.0) / 12 (9.2)
At least one serious risk needing further assessment303 (67.6)37 (28.2)
Boxed warning105 (23.4)6 (4.6)
Both a serious risk and a boxed warning74 (16.5)2 (1.5)
Neither114 (25.4)90 (68.7)

Source: research letter, Table [1]. Approval years are spread across 2016–2025 (20 to 59 original approvals a year). The paper's table footnotes are not reproduced. For supplemental approvals, the paper's percentages for the risk rows use the 131 therapeutics with a supplemental approval as the denominator.

Conceptual flow from a serious risk identified at approval, through FAERS and ARIA sufficiency checks, to a sponsor postmarketing requirement
Figure 1. Conceptual schematic, not data. A simplified view of the section 505(o)(3) sequence, in which FDA considers FAERS and then ARIA before requiring a sponsor study. Sources: FD&C Act section 505(o)(3) [8]; FDA, Assessment in Support of the Sentinel System, 2022 [4].

What the authors found

Serious risks are common at approval. At original approval, 303 of 448 therapeutics (67.6%) had at least one serious risk needing further assessment, 865 unique risks in total. The most common type was neoplasm risk (96, 11.1%), which includes carcinogenic potential and second primary cancers [1]. At supplemental approval, 37 of 131 therapeutics (28.2%) had at least one such risk, 96 in total. Here infection was the most common (18, 18.8%) [1].

FDA's systems almost never sufficed. Across all 961 risks, FAERS was never judged sufficient. ARIA was judged sufficient for 10 risks (1.0%) in six therapeutics: six cardiovascular, three neoplasm and one gastrointestinal [1]. Each of the six therapeutics had one matching ARIA study. As of June 2026 all six were still ongoing, and none had reported a regulatory outcome [1]. At the product level, ARIA covered at least one risk for 4 of the 303 original approvals with serious risks (1.3%) and 2 of the 37 supplemental ones (5.4%) [1].

The reasons are mostly about the outcome. Of the 951 risks where ARIA was insufficient, only 178 (18.7%) had an ARIA Sufficiency Memorandum explaining why [1]. In those 178, the reasons cited were: the outcome could not be identified or measured (151, 84.8%); limits of the analytic tools (76, 42.7%); covariates could not be captured (74, 41.6%); and the study population could not be identified (67, 37.6%) [1]. A memo can give more than one reason.

Table 2. Key results of the study [1].
MeasureResult
Novel therapeutics approved, 2016–2025448
With a supplemental approval for a new indication or population131 (29.2%)
Original approvals with ≥1 serious risk needing further assessment303 of 448 (67.6%); 865 risks
Most common risk at original approvalNeoplasm, 96 of 865 (11.1%)
Supplemental approvals with ≥1 serious risk37 of 131 (28.2%); 96 risks
Most common risk at supplemental approvalInfection, 18 of 96 (18.8%)
FAERS judged sufficient0 of 961 risks
ARIA judged sufficient10 of 961 (1.0%): 6 cardiovascular, 3 neoplasm, 1 gastrointestinal; 6 therapeutics
ARIA studies for those risks6, all ongoing in June 2026; no regulatory outcomes reported
ARIA-insufficient risks with a published ARIA memo178 of 951 (18.7%)
Reasons in memos (n = 178)Outcome 151 (84.8%); analytic tools 76 (42.7%); covariates 74 (41.6%); population 67 (37.6%)
Safety-related PMRs at original approval900 for 303 therapeutics
Safety-related PMRs at supplemental approval63 for 37 therapeutics
Bar chart showing that of 961 serious risks, FAERS was sufficient for none, ARIA for 10, and ARIA insufficient for 951, of which 178 had an ARIA memo
Figure 2. FDA's sufficiency decisions for the 961 serious risks identified at original (865) or supplemental (96) approval of novel therapeutics, 2016–2025. The darker part of the bottom bar shows the 178 risks with a published ARIA Sufficiency Memorandum. Source: research letter, Results [1].
Bar chart of the reasons ARIA was judged insufficient among 178 risks with an ARIA memo
Figure 3. Reasons for ARIA insufficiency among the 178 risks with an ARIA Sufficiency Memorandum. A memo can cite several reasons, so shares add to more than 100%. Source: research letter, Results [1].

What FDA required instead

FDA issued 900 safety-related PMRs for the 303 therapeutics with serious risks at original approval [1]. The most common were nonhuman studies (276, 30.7%), new clinical pharmacology studies (194, 21.6%) and new prospective observational studies (136, 15.1%). New clinical trials made up 58 (6.4%) [1]. At supplemental approval, FDA issued 63 PMRs for 37 therapeutics. These were mostly new retrospective observational studies (18, 28.6%), new prospective observational studies (13, 20.6%) and nonhuman studies (13, 20.6%), with 6 (9.5%) new clinical trials [1].

The authors use this mix to qualify their own headline. Just over half of the requirements at original approval (52.3% in the paper) were nonhuman or clinical pharmacology studies, which address questions FAERS and ARIA could never answer [1]. At supplemental approval, 49.2% were observational studies. That suggests ARIA has more room to contribute later in a product's life, once there is real-world use to study [1].

Stacked bars of safety-related PMR study types at original approval (900) and supplemental approval (63)
Figure 4. Safety-related PMR study types at original (n = 900) and supplemental (n = 63) approval, 2016–2025. Only categories with numbers stated in the paper's text are shown. The rest (236 at original and 13 at supplemental approval) are grouped as "other / not itemised in text". Retrospective observational studies are not itemised for original approval, and clinical pharmacology studies are not itemised for supplemental approval. Source: research letter, Results [1].

How the result fits earlier evidence

The direction matches earlier work that counted ARIA decisions rather than products.

  • FDA's own 2022 Sentinel assessment found that from 2017 to 2021, ARIA was sufficient for 11 of 145 observational PMR candidates (7.6%), and for none of 92 pregnancy questions [4].
  • A 2023 analysis by FDA and Sentinel authors counted 133 safety concerns evaluated in ARIA and 197 insufficiency determinations over 2016–2021. Pregnancy and fetal outcomes, cancers and death were the hardest areas, and thromboembolic events the easiest [5].
  • A 2026 review by several of the same authors found ARIA insufficient for 165 of 187 safety concerns raised in premarket review (88.2%) in 2016–2021 [3].

The new study's 1.0% is lower than these figures because its denominator is broader. It counts every serious risk in an approval letter, including the many that need nonhuman or pharmacology studies, which ARIA could never handle [1].

Table 3. FDA's postmarket tools for a risk identified at approval.
ToolWho runs itStrengthMain limitation for risks found at approvalSource
FAERSFDADetects and describes signals from spontaneous reportsCannot estimate incidence; variable data quality; judged sufficient for 0 of 961 risks[1][9]
Sentinel ARIAFDA, with Sentinel data partnersRapid analyses across large claims-based networks with standard toolsOutcomes, covariates, populations or designs often not captured; sufficient for 10 of 961 risks[1][4][5]
505(o)(3) PMRSponsor, required by FDAPurpose-built and enforceableTakes years; delays and incomplete public reporting[1][6][7]

When the job passes to the sponsor

FDA's annual performance report shows how PMRs progress once they are set [6]. On 30 September 2024 there were 1,636 open PMRs for NDAs and BLAs, 30% of them behind schedule. Of these, 885 were section 505(o)(3) safety PMRs: 667 on schedule and 218 off schedule, almost all delayed [6]. Of the 750 annual status reports expected in FY2024, 78% arrived on time and 13% did not arrive [6]. Of the PMRs FDA set in FY2018, 35% had been fulfilled and 24% were delayed six years later [6].

Table 4. Open postmarketing requirements on 30 September 2024 (FDA FY2024 report).
MeasureNDABLATotal
Open PMRs, all authorities1,1964401,636
Open PMRs off schedule415 (35%)74 (17%)489 (30%)
Open 505(o)(3) safety PMRs on schedule478189667
Open 505(o)(3) safety PMRs off schedule17048218
Annual status reports received on time in FY2024430 of 537 (80%)153 of 213 (72%)583 of 750 (78%)

Source: FDA, Report on the Performance of Drugs and Biologics Firms in Conducting Postmarketing Requirements and Commitments, Fiscal Year 2024, Tables 2–5 [6]; notice in the Federal Register, 24 April 2026 [11]. The 505(o)(3) rows add FDA's status columns in Tables 4 and 5.

Stacked bars showing the status of FDA postmarketing requirements on 30 September 2024 by fiscal year of establishment, FY2018 to FY2024
Figure 5. Status on 30 September 2024 of PMRs set in each fiscal year from FY2018 to FY2024, as a share of the PMRs set that year. All PMR authorities are included, not only 505(o)(3). FDA cautions against reading these snapshots as a year-on-year trend. Source: FDA FY2024 PMR/PMC performance report, Table 8 [6].

Earlier work also found gaps in public reporting. Among new drugs and biologics approved in 2009–2012, 47 of 65 completed required prospective studies and trials (72.3%) had reported results or been published, a median of 47 months after approval [7].

Why ARIA falls short, in the authors' view

The authors point to limited clinical data for risk evaluation and the clinical complexity of the risks themselves [1]. Long-term cancer risks and pregnancy complications need detailed clinical characterisation and long follow-up. FAERS cannot provide these, and ARIA often lacks them [1]. ARIA did work for a few acute, clinically well-defined outcomes, such as cardiovascular events and lymphoma [1]. Their proposed fixes are to:

  • link ARIA to broader clinical data
  • improve outcome ascertainment with advanced pharmacoepidemiological methods and artificial intelligence
  • be more transparent about sufficiency decisions [1].

They also note that FDA increasingly accepts a single pivotal trial for approval, so more products may reach the market with serious risks still to be assessed [1].

Limitations and funding

The authors list three limitations: reliance on public documents; the infrequent availability of ARIA Sufficiency Memoranda (only 178 of 951 insufficient risks had one); and possible under-counting or misclassification of risks where FDA sources disagree [1]. They stress that the findings do not mean FDA-approved therapeutics are unsafe, or that FDA's overall surveillance system is not robust [1]. Two further points are worth adding. The study counts FDA's determinations, not the quality of the PMRs that followed. And the ARIA studies it found are too recent to show whether they changed any labels.

The work was funded by Arnold Ventures through grants to Yale and Emory. The funder had no role in the design, conduct, analysis, manuscript or decision to submit [1]. Several authors also report grants from FDA, among other funders, for work outside this study [1].

What this means for safety teams

  • Plan for the sponsor study, not the FDA query. With FAERS sufficient for no risk and ARIA for 1%, a serious risk named in the approval letter should be assumed to become a PMR [1]. Build protocol concepts, feasibility work and budget into the launch plan.
  • Expect nonhuman and pharmacology work, not only epidemiology. Over half of original-approval PMRs were nonhuman or clinical pharmacology studies [1]. Bring in nonclinical and clinical pharmacology teams, not only epidemiologists, when negotiating PMRs.
  • Design around the outcome. The outcome was the main reason ARIA failed in 84.8% of memos [1]. If a risk depends on clinical detail, such as laboratory values, histology or pregnancy outcomes, plan linked health records, registries, chart validation or a trial from the start.
  • Read any ARIA memo as a design brief. When FDA publishes one, it names the gaps and proposes PMR wording [1][3]. A study that does not close those gaps will struggle to fulfil the requirement.
  • Revisit at each new indication. Supplemental approvals carried fewer serious risks but more observational PMRs [1]. Once real-world use exists, it is worth testing with FDA whether ARIA, or a sponsor study using existing data, can answer the question.
  • Protect the schedule and report on time. FDA measures PMR status against the original milestones even when it has accepted an extension [6]. Treat milestones and annual status reports as compliance obligations.
  • Keep FAERS in its lane. Use spontaneous reports to detect and describe signals, not to estimate risk [1][9].
  • Align with the EU risk management plan. EU marketing authorisation holders must carry out imposed post-authorisation safety studies, and PRAC assesses their protocols and results [10]. A single global study can sometimes meet both a US PMR and an EU PASS, but only if it is designed for both regulators' questions from the outset.

What is established and what is not

Established: the study's counts of products, serious risks, FAERS and ARIA sufficiency decisions, reasons in ARIA memos, and PMR study types [1]; the legal sequence under section 505(o)(3) [4][8]; and FDA's published data on PMR status [6].

Open questions:

  • The other 773 decisions. No reasons are published for most ARIA insufficiency determinations, so the reasons in Figure 3 may not represent the whole set [1].
  • Outcome of the six ARIA studies. None had reported a regulatory outcome by June 2026 [1].
  • Trend over time. The text we reviewed does not give sufficiency by approval year. With only 10 sufficient risks, any trend would be hard to read anyway.
  • What the PMRs delivered. The study does not follow PMRs to completion, so it cannot say how many changed labels.

The finding is stark but consistent with nearly a decade of data. Active surveillance based mainly on claims can answer a narrow set of acute, well-coded questions. The serious risks flagged at approval remain, in practice, the sponsor's to resolve.

References

  1. Elshafie S, Chen RZ, Taherifard E, Ramachandran R, Ross JS, Wallach JD. Evaluation of Premarket-Identified Serious Risks for Drugs and Biologics Using FDA Surveillance Systems. JAMA Intern Med. Published online 16 November 2026. doi:10.1001/jamainternmed.2026.5774. PMID 43037980
  2. JAMA Network media release, 16 November 2026
  3. Taherifard E, Ross JS, Ramachandran R, Wallach JD. US Food and Drug Administration Sentinel Initiative and Postmarket Drug Safety Surveillance: A Review. JAMA Health Forum 2026;7(4):e260754
  4. FDA. Assessment in Support of the Sentinel System. September 2022
  5. Maro JC, Nguyen MD, Kolonoski J, et al. Six Years of the US Food and Drug Administration's Postmarket Active Risk Identification and Analysis System in the Sentinel Initiative: Implications for Real World Evidence Generation. Clin Pharmacol Ther 2023;114(4):815–824
  6. FDA. Report on the Performance of Drugs and Biologics Firms in Conducting Postmarketing Requirements and Commitments, Fiscal Year 2024
  7. Wallach JD, Egilman AC, Dhruva SS, et al. Postmarket studies required by the US Food and Drug Administration for new drugs and biologics approved between 2009 and 2012: cross sectional analysis. BMJ 2018;361:k2031
  8. Federal Food, Drug, and Cosmetic Act, section 505(o)(3), 21 U.S.C. 355(o)(3)
  9. Potter E, Reyes M, Naples J, Dal Pan G. FDA Adverse Event Reporting System (FAERS) Essentials: A Guide to Understanding, Applying, and Interpreting Adverse Event Data Reported to FAERS. Clin Pharmacol Ther 2025;118(3):567–582
  10. European Medicines Agency. Post-authorisation safety studies (PASS)
  11. Federal Register. Report on the Performance of Drug and Biologics Firms in Conducting Postmarketing Requirements and Commitments; Availability. 91 FR 22155, 24 April 2026

Safety Strategy · Pharmacovigilance and safety risk management · Vaud, Switzerland


This article is for information only and is not medical or regulatory advice.

Related services: signal management and benefit-risk assessment and risk management (EU RMP, US REMS, aRMMs).

Follow Safety Strategy on LinkedIn (opens in a new tab) for new articles and updates. Back to all news