A reported cancer survival link to COVID vaccines looked small and brief in 95,015 French patients

In October 2025 a paper in the journal Nature made a striking claim. Cancer patients who happened to get an mRNA COVID-19 vaccine around the time they started immunotherapy lived far longer than those who did not. In one group with lung cancer, median survival, the point by which half of patients had died, was nearly twice as long. The mRNA vaccines are the type that deliver genetic instructions for a viral protein.

A French team has now looked for the same effect in the national health records of patients who started that kind of treatment over more than two years, 95,015 people in all. Writing in Nature Medicine in September, they report that vaccinated patients did die at a slightly lower rate in the first months. But the gap was small, it was gone after a year, and a similar gap showed up with vaccines of other kinds.

Why a COVID vaccine was thought to help immunotherapy

The drugs in question release a brake on the immune system so that it attacks tumors. Immune checkpoint inhibitors (ICIs) extend survival in many patients with cancer, but not in all, and researchers have looked for ways to make more tumors respond.

The Nature paper, from the University of Texas MD Anderson Cancer Center and collaborators, proposed that mRNA vaccines against the coronavirus do this as a side effect. In mice, the vaccines set off a burst of an immune signal called type I interferon, which helped immune cells recognize tumors. The team then went looking in patient records.

They found what looked like a large effect. Among people with advanced non-small cell lung cancer, the commonest form of the disease, they identified 180 patients who received a COVID mRNA vaccine within 100 days of ICI initiation, meaning the start of immunotherapy, and 704 patients who were treated with ICI and did not receive a COVID vaccine. Median survival was 37.3 months in the vaccinated group and 20.6 months in the other. At three years, 55.7% of the vaccinated patients were alive, against 30.8%. A smaller melanoma group, with 43 vaccinated patients and 167 unvaccinated, showed a similar pattern.

The authors also checked two older vaccines. Patients who received a pneumonia or influenza vaccine within 100 days of initiating ICI saw no such gain. That detail matters for what came next.

Studies like this have a known weakness. Nobody assigned the vaccine at random. People who get vaccinated in the weeks around the start of cancer treatment may differ from those who do not, and a patient too ill to attend a vaccination appointment is also a patient more likely to die soon.

How the French study tested the survival link

The new study, led by Joe-Elie Salem of the Pitie-Salpetriere hospital group in Paris, set out to check the claim at national scale. As its authors put it, experimental data suggest that the vaccine could alter the effects of these drugs, but population evidence is limited.

It was a nationwide, retrospective observational cohort study using the French national health data system. Retrospective means that it looked back at records already collected. From that system, the researchers identified 95,015 adults initiating ICI therapy between December 2020 and March 2023. The start date was chosen as corresponding to first COVID-19 vaccine availability in France, so that every patient could in principle have been vaccinated.

They ran two separate comparisons, because the timing of the vaccine raises different problems.

The first looked at mRNA vaccination within 100 days before ICI initiation. Vaccinated and unvaccinated patients were statistically weighted, a way of adjusting the comparison so that the two groups matched on the characteristics the database records.

The second looked at vaccination within 100 days after ICI initiation among previously unvaccinated patients. This one is harder. A patient can only be vaccinated two months into treatment by surviving those two months, which hands the vaccinated group a head start. To remove it, the team used a method called target-trial emulation, which treats existing records as if they came from a randomized trial with the same starting line for everyone. The approach was chosen to address immortal-time bias, time-dependent vaccination and informative censoring, three ways in which the timing of events can distort such a comparison. Immortal-time bias means crediting a treatment with the time a patient had to survive in order to receive it.

The main outcome was death from any cause, tracked to the end of 2024. The secondary outcomes were cancer-specific and non-COVID-19 mortality, a check on whether any difference simply reflected fewer deaths from the infection itself.

A small early survival gap that faded within a year

Timing of mRNA vaccination	Early months	At six months	After 12 months
Within 100 days before treatment	Death rate about 10% lower at 1 to 3 months	About 4% lower, not a clear difference	No clear difference
Within 100 days after treatment began	Death rate about 16% lower at 3 months	About 14% lower	No clear difference

Pre-ICI mRNA vaccination was associated with lower early death rates, pre-ICI meaning before immunotherapy began. At one to three months the rate was about 10% lower than in unvaccinated patients, with a plausible range of 6% to 13%. By six months the figure was 4%, and its range reached zero.

Vaccination after treatment began showed a somewhat larger gap. The death rate was about 16% lower at three months, with a range of 12% to 20%, and about 14% lower at six months.

In both comparisons the association was gone after 12 months, and for vaccination before treatment it was already unclear at six.

Set against the Nature result, these numbers are different in kind. That study reported a survival gap still wide at three years, with 55.7% of vaccinated patients alive against 30.8%. The French data show a difference in death rates of a tenth to a sixth, lasting months.

The cause-specific results did not change the picture. Estimates for non-COVID-19 and cancer-specific mortality were similar to those for overall survival. Cancer-specific mortality means deaths from the cancer itself. So the early gap was not simply a matter of COVID-19 deaths prevented by the vaccine.

Why other vaccines point to a healthy-vaccinee effect

The most telling check was a comparison that ought not to matter. The study’s registered plan included analyses of other adult vaccines as control exposures, meaning comparisons that should come up empty if the effect belongs to mRNA vaccines. If the mRNA vaccine were making tumors more sensitive to treatment, a different kind of vaccine given in the same window ought to show no such pattern.

The other vaccines showed one too. The authors’ conclusion reads: “Similar early patterns with non-mRNA COVID-19 and other adult vaccines suggest modest, transient associations or healthy-vaccinee effects, rather than a specific mRNA COVID-19 vaccine effect.”

The healthy vaccinee effect is a well-documented trap in vaccine research. As a Swedish team put it this year, the healthy vaccinee effect occurs when healthier persons are more likely to receive the vaccine than their less healthy peers. The vaccinated then do better for reasons unrelated to the vaccine.

That team gave a vivid demonstration in a Swedish population aged 65 and over, 245,696 people. Those who received the 2024-2025 COVID-19 vaccine died from any cause at less than half the rate of those who did not. A vaccine against a single infection would not be expected to do that. And removing individuals hospitalized with COVID-19 from this population did not change the difference. The Swedish authors concluded that frail older people had been less likely to be vaccinated.

Among cancer patients starting immunotherapy the same logic applies with force. In the first weeks, the patients least likely to book a vaccination are those who are deteriorating fastest. A short-lived gap in deaths, concentrated in the early months and appearing with other vaccines too, is the pattern such selection would be expected to produce.

One number from a second French analysis, described in the next section, shows how the vaccinated differ. In that study, 28.0% of patients received an influenza vaccination among those given the COVID-19 vaccine, against 15.4% in the comparison group. That suggests people who get one vaccine tend to get others, and to be in closer touch with routine care.

How it fits with other studies of vaccination and immunotherapy

The Nature Medicine paper is the second large French national analysis this year. Both found a far weaker link than the original report, though they differ on how long it lasts. In May, a French research group called EPI-PHARE published a study of 30,898 patients who started drugs of this type in 2021 and 2022. Each patient who had received an mRNA COVID-19 vaccine within three months of starting treatment was matched with a similar patient who had not, so half were vaccinated.

During follow-up, 67.3% of co-exposed patients died, compared with 70.5% of non-co-exposed patients, co-exposed meaning vaccinated in that window. Median survival was 21.6 months in the vaccine coexposed cohort compared with 18.5 months in the other group, a gap of about three months that the Salem study’s fading rates do not reproduce.

That group read its own result cautiously. Vaccination was associated with a modest improvement in survival, it wrote, far weaker than the Nature paper had reported. In its view, the size of the association does not appear sufficient, at this stage, to justify a change in the vaccination schedule around cancer treatment. Its authors added that the results “provide population-level evidence that may guide future studies”.

The Salem study goes a step further. It asks whether the modest gap is specific to mRNA vaccines at all, and its answer is that the data do not point that way.

Neither French study can say the Texas team’s biology is wrong. The mouse experiments stand on their own, and a real but small effect in people is still possible. What the national data show is that the dramatic survival difference seen in roughly a thousand patients at one center did not appear when nearly a hundred thousand patients with cancers of all kinds were counted.

The EPI-PHARE group named the next step. Randomized clinical trials will be necessary to confirm this effect, it wrote.

What the survival study cannot settle

It is observational. The weighting can only balance what the database records. An insurance database holds treatments and hospital stays. It does not hold how well a patient was on a given day, which is the thing most likely to decide both whether they were vaccinated and whether they survived the next three months. The study’s own plan included control analyses to explore residual confounding, confounding meaning the influence of such unmeasured differences.

All cancers together. The main results pool every tumor type treated with these drugs. The Nature study concerned lung cancer and melanoma. An effect confined to one cancer could be diluted in the overall figures.

A real effect could hide inside the bias. The healthy-vaccinee explanation fits the pattern, but the study cannot prove that none of the early gap is real. A small true benefit and a selection effect would look alike in these data.

Timing. The main analysis counted at least one mRNA dose in a 100-day window. Whether a dose given at a precisely chosen moment behaves differently is a question for a trial.

The ordinary purpose of the vaccine is a separate matter. The study asked only whether vaccination improves the results of immunotherapy. It did not examine protection against the infection. In the EPI-PHARE study, COVID-19-related hospitalization occurred in 0.6% of patients in the coexposed group compared with 2.0% in the other, though that comparison shares the same weakness.

Decisions about vaccination during cancer treatment depend on the individual patient and their treatment plan, which is assessed by their doctor.

Among 95,015 French patients starting immunotherapy, mRNA COVID-19 vaccination was linked to a death rate about 10% to 16% lower in the first months and no difference after a year, a pattern also seen in similar form with other vaccines and so consistent with a difference in who gets vaccinated more than with an effect on tumors, though only a randomized trial could settle it.