CAR-T's New Boxed Warning: Reading the Evidence on Secondary T-Cell Cancers

In January 2024 the FDA required a boxed warning for secondary T-cell cancers on all six approved CAR-T cell products, spanning Kymriah, Yescarta, Tecartus, Breyanzi, Abecma, and Carvykti. The agency acted after receiving 22 reports of T-cell malignancies among CAR-T recipients as of December 31, 2023, three of which carried the engineered CAR gene inside the tumor itself. The label documents an observed association and requires lifelong monitoring. It does not prove the therapies cause these cancers, and it is not a verdict that CAR-T is unsafe.

From a November signal to a January label

The sequence matters because it shows how a regulator escalates. In November 2023 the FDA announced it was investigating reports of T-cell malignancies, including CAR-positive lymphoma, after treatment with BCMA-directed and CD19-directed autologous CAR-T therapies. By January 19, 2024, the agency had sent safety labeling change letters to manufacturers requiring a boxed warning across the entire class. In a New England Journal of Medicine perspective published that February, Nicole Verdun and Peter Marks of the FDA's Center for Biologics Evaluation and Research described the underlying data: 22 cases known to the agency by the end of 2023, drawn from adverse-event reports and the medical literature.

Of those 22, fourteen had adequate data for closer analysis. All fourteen arose within two years of infusion, and roughly half appeared within the first year. That temporal clustering is what pushed the signal past the threshold for a class-wide action. A boxed warning is the FDA's most prominent labeling tool, and applying it to a therapy class that has been transformative for relapsed leukemias, lymphomas, and multiple myeloma was not a routine step.

What the three transgene-positive cases actually show

The most consequential detail is the smallest number. In three of the reported cases, genetic sequencing detected the CAR transgene within the malignant T-cell clone. That is a mechanistic red flag. CAR-T manufacturing uses a viral vector to insert the receptor gene into a patient's T cells, and any integrating vector carries a theoretical risk of insertional mutagenesis, where the insertion lands near a gene that drives uncontrolled growth. Finding the transgene inside the cancer is consistent with the product contributing to that specific tumor.

Consistent with is not the same as caused. Two separate reports published in NEJM in June 2024 reached opposite conclusions on individual cases. One found no CAR gene in the T-cell lymphoma and concluded the therapy did not directly cause it. The other found the CAR gene alongside other pre-existing genetic changes and concluded the treatment probably contributed. Both can be true, because these are different patients with different tumors. The honest read is that insertional mutagenesis is plausible in a small subset and unproven as a general mechanism.

What pharmacovigilance can and cannot establish

Much of the public analysis of this signal leans on disproportionality analysis of spontaneous reporting databases. A 2024 study pooled second-primary-malignancy reports from the FDA's FAERS system and the WHO's VigiBase, extracting 310 cases from FAERS and 297 from VigiBase between 2017 and 2023. T-cell lymphoma emerged with a strong reporting odds ratio, the strongest disproportionality signal among the second cancers examined, alongside weaker signals for myelodysplastic syndrome and acute myeloid leukemia.

A reporting odds ratio measures whether an event is reported more often than expected for a given drug relative to all other drugs in the database. It is a signal-detection tool, not an incidence rate. These databases are subject to reporting bias, cannot supply a denominator of everyone treated, and often lack complete treatment histories. That last gap is decisive here. Nearly every CAR-T recipient has received prior lymphodepleting and cytotoxic chemotherapy, which itself raises the long-term risk of secondary cancers, and most carry blood cancers with an accumulated burden of mutations. A high ROR flags where to look. It cannot separate the vector's contribution from the chemotherapy, the underlying disease, or the simple fact that patients who live longer have more time to develop a second cancer.

Rarity and the denominator problem

Context restores proportion. More than 34,000 people have received CAR-T therapy, and the National Cancer Institute has emphasized that second cancers of any type remain rare in this population. The largest single-center cohort followed more than 700 treated patients over nine years and identified 25 second cancers in total, only one of which was a T-cell lymphoma. Against a base of tens of thousands of recipients, 22 reported T-cell malignancies describe a low-frequency event, which is exactly the regime where causation is hardest to pin down and where a precautionary label makes sense.

This is educational content, not medical advice, and no one facing a cancer diagnosis should weigh CAR-T against its alternatives from a blog post rather than with their own oncology team. The regulatory point is narrower. A boxed warning describes what the label legally must disclose, an observed and serious potential risk warranting monitoring. It is not an endorsement, not a safety all-clear, and not a statement that benefits no longer outweigh harms. For the approved indications, where CAR-T often follows the failure of every other option, the FDA has continued to state that the benefits still exceed the known risks.

The appraiser's takeaway

Three things are established. The temporal pattern is real, the class-wide warning is a defensible response to a clustered signal, and at least a few cases carry direct molecular evidence of product involvement. Three things are not. General causation is unproven, disproportionality signals cannot deliver an incidence rate, and the contribution of prior chemotherapy and underlying disease cannot yet be subtracted out. Prospective registries with complete treatment histories and long follow-up, not spontaneous reports, are what will eventually convert this signal into a rate.