Lessons from the FDA’s Refusal to Review an mRNA Flu Vaccine
library(tidyverse)
library(patchwork)
library(knitr)
library(kableExtra)
# GHR color palette (matches website)
ghr_blue <- "#0284c7" # sky-600, primary
ghr_orange <- "#E69F00" # secondary/contrast
ghr_slate <- "#64748b" # slate-500, neutral
ghr_teal <- "#0d9488" # teal-600, accent
ghr_red <- "#dc2626" # red-600, for emphasis
theme_set(theme_minimal(base_size = 14))
set.seed(42)This note is a companion to the Experimental Designs chapter of Global Health Research in Practice, which covers randomized trial design, blinding, and control group selection. Here we use a real regulatory case to explore what “adequate and well-controlled” means in practice.
Update: On February 18, 2026, the FDA revised its stance and agreed to review Moderna’s application, as long as Moderna completes a post-marketing study in older adults.
On February 10, 2026, Moderna received a Refuse-to-File letter from the FDA for its Biologics License Application (BLA) of mRNA-1010, an influenza vaccine built on its mRNA platform. The agency didn’t reject the application after a full review. It declined to begin the review at all.
The reason wasn’t safety. It wasn’t efficacy. In its refusal letter, the FDA stated that the Phase 3 trial’s active comparator—the vaccine used as the comparison arm, standard-dose Fluarix—did not reflect the “best-available standard of care.”
By the company’s account, this came as a surprise. After enrolling more than 40,000 participants across 11 countries and demonstrating superiority over the comparator on the primary endpoint—no small feat—the FDA said the trial answered the wrong question.
In public statements, Moderna has said it discussed the protocol with the FDA in advance and believed the comparator choice was acceptable. The company’s president, Dr. Stephen Hoge, told the New York Times:
“This refusal to start a review is all confusing, to say the least. It is surprising, and we’re trying to understand what has changed.”
Whatever the source of that disconnect, the stated reason for refusal centers on a design decision that shapes every clinical trial: what do you compare your intervention against?
In its press release, Moderna reported that in April 2024 it submitted the Phase 3 protocol to the FDA’s Center for Biologics Evaluation and Research (CBER) for review. The company states that the agency provided written feedback including: “we agree it would be acceptable to use a licensed standard dose influenza vaccine as the comparator.” The FDA also recommended using vaccines preferentially recommended for older adults for participants over 65—but, the company says, raised no objections or clinical hold comments before the study began.
The refusal letter, however, states the decision was “consistent with FDA’s advice given to you prior to your study.” The details of those interactions aren’t fully public. But the takeaway is: in regulated trials, “design” includes not only statistical validity, but also regulatory alignment about what comparators will be considered adequate for the intended label and population.
Moderna’s trial (P304, NCT06602024) was a Phase 3, randomized, observer-blind, active-controlled study. Participants were randomized 1:1 to receive either mRNA-1010 or standard-dose Fluarix, a licensed inactivated influenza vaccine.
More than 40,000 adults aged 50 and older were enrolled across 11 countries.
The study met its pre-specified primary endpoint. mRNA-1010 demonstrated a relative vaccine efficacy (rVE) of 26.6% (95% CI: 16.7%–35.4%) against RT-PCR-confirmed influenza-like illness compared to Fluarix. Immune responses against influenza A strains were particularly strong.
On its face, this is exactly the kind of result sponsors seek before filing for approval: a large, randomized, active-controlled Phase 3 trial demonstrating superiority.
Yet regulatory review does not evaluate results in isolation. It evaluates whether the trial was designed to answer the question the agency believes must be answered.
In this case, the dispute turns on the choice of comparator.
In a randomized trial, the control group defines the counterfactual. The estimated effect is always relative to that comparison, not absolute.
Consider two scenarios for the same vaccine:
In Scenario A, the vaccine beats a licensed but potentially suboptimal comparator. In Scenario B, the relevant question for patients is whether mRNA-1010 is better than the best currently available option. These are different questions, and they lead to different clinical conclusions.
# Hypothetical data for the comparator visualization
comparator_df <- tibble(
scenario = factor(c(
"Scenario A:\nvs. Standard-Dose",
"Scenario A:\nvs. Standard-Dose",
"Scenario B:\nvs. High-Dose",
"Scenario B:\nvs. High-Dose"
), levels = c("Scenario A:\nvs. Standard-Dose", "Scenario B:\nvs. High-Dose")),
arm = factor(c("Comparator", "mRNA-1010", "Comparator", "mRNA-1010"),
levels = c("Comparator", "mRNA-1010")),
efficacy = c(55, 70, 65, 70),
lower = c(50, 64, 60, 63),
upper = c(60, 76, 70, 77),
color = c(ghr_orange, ghr_blue, ghr_orange, ghr_blue)
)
# Panel A: Absolute efficacy estimates
p_absolute <- ggplot(comparator_df, aes(x = arm, y = efficacy, color = color)) +
geom_pointrange(aes(ymin = lower, ymax = upper),
size = 1.2, linewidth = 1, fatten = 3) +
scale_color_identity() +
facet_wrap(~scenario, scales = "free_x") +
geom_hline(yintercept = 0, color = "gray80") +
labs(
title = "Same Treatment, Different Comparators",
subtitle = "Hypothetical absolute vaccine efficacy (%) against influenza",
x = NULL,
y = "Absolute vaccine efficacy (%)"
) +
coord_cartesian(ylim = c(40, 85)) +
theme(strip.text = element_text(face = "bold", size = 11))
# Panel B: Relative vaccine efficacy
rve_df <- tibble(
scenario = factor(c("vs. Standard-Dose", "vs. High-Dose"),
levels = c("vs. Standard-Dose", "vs. High-Dose")),
rve = c(26.6, 7.1),
lower = c(16.7, -8.5),
upper = c(35.4, 21.0)
)
p_relative <- ggplot(rve_df, aes(x = scenario, y = rve)) +
geom_hline(yintercept = 0, linetype = "dashed", color = ghr_red, linewidth = 0.8) +
geom_pointrange(aes(ymin = lower, ymax = upper),
size = 1.2, linewidth = 1, fatten = 3, color = ghr_blue) +
annotate("text", x = 2.3, y = 0, label = "No difference",
color = ghr_red, fontface = "italic", hjust = 0, size = 3.5) +
labs(
title = "Relative Vaccine Efficacy (rVE)",
subtitle = "The same vaccine looks very different depending on the comparator",
x = NULL,
y = "Relative vaccine efficacy (%)"
) +
coord_cartesian(ylim = c(-15, 40), xlim = c(0.5, 2.8)) +
theme(plot.title = element_text(face = "bold"))
p_absolute / plot_spacer() / p_relative +
plot_layout(heights = c(4, 0.5, 3)) +
plot_annotation(
caption = "Note: Values for Scenario B (vs. High-Dose) are hypothetical, for illustration only.",
theme = theme(plot.caption = element_text(color = "gray50", size = 9))
)
A randomized trial estimates performance relative to a specific comparator, and the choice in comparator defines the question being answered. The FDA’s role is to decide whether that question satisfies the legal standard for approval, what the statute calls “adequate and well-controlled” evidence.
At the center of the refusal is the FDA’s requirement that approval decisions rest on “adequate and well-controlled” investigations. That phrase is not rhetorical. It is defined in regulation.
Under 21 CFR 314.126, the FDA specifies what qualifies as an adequate and well-controlled study and recognizes five types of controls as valid:
control_types <- tibble(
`Control Type` = c(
"1. Placebo concurrent control",
"2. Dose-comparison concurrent control",
"3. No treatment concurrent control",
"4. Active treatment concurrent control",
"5. Historical control"
),
Description = c(
"Participants randomized to the experimental treatment or an inert placebo",
"Participants randomized to different doses of the same treatment (including possibly zero dose)",
"Participants randomized to treatment or no treatment (no placebo given, often unblinded)",
"Participants randomized to experimental treatment or a known effective therapy",
"The experimental group is compared to a prior group treated under similar conditions"
),
`When Appropriate` = c(
"When no proven treatment exists, or when withholding treatment is ethical and the condition is not life-threatening",
"When the dose-response relationship itself is the question, or when placebo control is unethical",
"When the outcome is objective and unlikely to be influenced by knowledge of treatment assignment",
"When an effective treatment exists and withholding it would be unethical; when the goal is to show the new treatment is at least as good as (or better than) the standard",
"When the natural history is highly predictable and consistent; rarely sufficient as the sole evidence for approval"
),
`Global Health Example` = c(
"Early HIV vaccine trials (before effective vaccines existed)",
"Dose-finding studies for artemisinin-based combination therapies for malaria",
"Trials of surgical vs. non-surgical interventions where blinding is impractical",
"New TB regimens compared to standard isoniazid/rifampicin therapy",
"Emergency-use evaluations during Ebola outbreaks when randomization was debated"
)
)
kable(control_types,
caption = "The five types of controls recognized by the FDA under 21 CFR 314.126") %>%
kable_styling(full_width = TRUE) %>%
row_spec(4, background = "#E69F0030")| Control Type | Description | When Appropriate | Global Health Example |
|---|---|---|---|
| 1. Placebo concurrent control | Participants randomized to the experimental treatment or an inert placebo | When no proven treatment exists, or when withholding treatment is ethical and the condition is not life-threatening | Early HIV vaccine trials (before effective vaccines existed) |
| 2. Dose-comparison concurrent control | Participants randomized to different doses of the same treatment (including possibly zero dose) | When the dose-response relationship itself is the question, or when placebo control is unethical | Dose-finding studies for artemisinin-based combination therapies for malaria |
| 3. No treatment concurrent control | Participants randomized to treatment or no treatment (no placebo given, often unblinded) | When the outcome is objective and unlikely to be influenced by knowledge of treatment assignment | Trials of surgical vs. non-surgical interventions where blinding is impractical |
| 4. Active treatment concurrent control | Participants randomized to experimental treatment or a known effective therapy | When an effective treatment exists and withholding it would be unethical; when the goal is to show the new treatment is at least as good as (or better than) the standard | New TB regimens compared to standard isoniazid/rifampicin therapy |
| 5. Historical control | The experimental group is compared to a prior group treated under similar conditions | When the natural history is highly predictable and consistent; rarely sufficient as the sole evidence for approval | Emergency-use evaluations during Ebola outbreaks when randomization was debated |
Moderna’s trial was an active treatment concurrent control study: mRNA-1010 vs Fluarix. The regulation says the control should be a “known effective therapy”, but it doesn’t define whether that means any licensed option or the best available standard for the population studied.
Fluarix is licensed. But for adults 65 and older, U.S. guidance preferentially recommends enhanced influenza vaccines: high-dose formulations (like Fluzone High-Dose), adjuvanted vaccines (like Fluad), or recombinant vaccines (like Flublok). Standard-dose vaccines like Fluarix are licensed and widely used, but they are not the best available option for older adults in the United States.
Moderna argued that its approach followed a familiar regulatory pathway. In its response to the refusal letter, the company noted that enhanced vaccines such as Fluzone High-Dose and Fluad were themselves approved on the basis of superiority over standard-dose comparators. In other words, the design it used mirrors the designs that established the very products now considered the preferred standard.
The dispute turns on interpretation. When multiple effective therapies exist, which one defines the standard against which a new product must be judged?
This tension—licensed versus best available, local versus global standard—is not unique to influenza vaccines. It recurs wherever comparator choice determines what a trial can claim.
In the mid- to late 1990s, short-course AZT regimens to prevent mother-to-child HIV transmission were tested against placebo in several sub-Saharan African countries, even though a longer AZT regimen had already been shown effective in the United States (ACTG 076). The ethical controversy centered on whether trial comparators should reflect the “best proven intervention” globally or the local standard of care in the host country. Marcia Angell and others argued in the New England Journal of Medicine that ethical standards should not vary by geography. Defenders countered that the long-course regimen was not feasible in many low-resource settings and that placebo-controlled trials were necessary to determine whether shorter, affordable regimens could provide benefit. The debate contributed to revisions of the Declaration of Helsinki and to ongoing clarification of standards governing placebo use.
Malaria prevention trials face similar questions. When testing a new bed net or chemoprevention strategy, should the comparator reflect WHO-recommended standard practice, what is actually implemented in the study setting, or a more intensive strategy used elsewhere?
The Declaration of Helsinki (2013, Paragraph 33) states that new interventions should be tested against the “best proven intervention.” Yet the interpretation of what counts as “best proven” has long been debated in light of geography, feasibility, cost, and implementation context.
The Moderna case is less ethically charged, but structurally similar. The question is not whether the comparator Fluarix is a known effective therapy—it is. The question is whether superiority over Fluarix answers the question regulators believe needs answering. And because the trial spanned 11 countries and was under review in multiple jurisdictions, a comparator that one regulator rejects may meet the evidentiary expectations of others.
“Adequate and well-controlled” is a specific regulatory standard, not just a synonym for “good.” The FDA recognizes five types of controls under 21 CFR 314.126. An active-controlled trial must compare against a “known effective therapy”—but what that means in practice is a judgment call, and that judgment can differ between the sponsor and the regulator.
The choice of comparator determines what a trial can prove. Not just whether the new treatment works, but whether it works better than the current best option. Superiority over a weaker comparator is a real finding—but it may not answer the question that matters to patients, clinicians, or regulators.
In global health, “standard of care” is context-dependent. A trial designed for global licensure may use a comparator that satisfies regulators in some markets but not others. This reflects real variation in what’s available, what’s affordable, and what’s recommended across settings.
The next time you’re reading a clinical trial or designing one, stop and ask: what is this being compared against, and why? The answer shapes everything that follows.