Welcome to this month's journal review — we're covering the March 2026 issue of JAMA Dermatology, and today we've got two pieces: a research letter out of Memorial Sloan Kettering looking at antibiotics and immune-related skin toxicity during checkpoint inhibitor therapy, and an editorial on a companion study examining Agent Orange exposure and acral melanoma risk in US veterans. Let's get into it.

First up is a research letter — so think of this as a compact, hypothesis-driven retrospective cohort study rather than a full-length original article. The question they're asking is whether antibiotic exposure during immune checkpoint blockade increases the risk of immune-related cutaneous adverse events, what the field shorthand calls ircAEs. The background here is familiar territory for anyone managing these patients: checkpoint inhibitors have transformed oncology outcomes, but immune-related adverse events are common, and skin is the most frequently affected organ. What's less well worked out is the underlying mechanism. There's already a body of work linking the gut microbiome to both efficacy and to immune-related colitis with these agents, and separately the skin microbiome is increasingly implicated in various dermatologic conditions. So the authors' logical leap is: if microbiome perturbation from antibiotics can drive gut toxicity, might the same phenomenon drive skin toxicity through disruption of the cutaneous microbiome?

To test this, they pulled a retrospective cohort of patients with solid tumors treated with checkpoint blockade at Memorial Sloan Kettering between 2016 and 2021 — just under ten thousand patients. Methodologically, the interesting design choice is how they handled antibiotic exposure. Rather than a simple yes-or-no exposure classification, they treated antibiotic exposure as a time-varying covariate in a Cox proportional hazards model, with a two-month exposure window following each antibiotic course. The stated rationale is that this window approximates expected microbiome reconstitution time after antibiotic perturbation. This is also, worth noting, the correct way to handle a covariate that changes during follow-up — using antibiotic exposure as a fixed baseline variable here would risk immortal time bias, since patients who happen to survive longer on therapy have more opportunity to receive antibiotics. So the time-varying approach is doing real methodological work, not just being fancy. They also stratified antibiotics by their microbiome-perturbation potential and by spectrum of activity, which lets them test a dose-response type question rather than a blunt exposed-versus-unexposed comparison. The primary outcome — an oncodermatologist-confirmed ircAE — was identified by diagnostic code query and then validated by manual chart review, including photographs and pathology when available, which adds a layer of diagnostic rigor you don't always see in claims-based studies.

On to results. About seven percent of the cohort developed an ircAE during their first checkpoint regimen, with pruritus and maculopapular rash being the dominant phenotypes, together accounting for roughly eight in ten cases. The headline finding: antibiotic exposure during checkpoint therapy was associated with a significant sixty percent relative increase in ircAE risk. When they broke it down by microbiome-perturbation potential, both low and medium-to-high perturbation antibiotics carried increased risk, though interestingly the low-perturbation group showed an even larger point estimate, almost a three-fold increase, albeit with a wide confidence interval reflecting a smaller sample. When they broke it down by spectrum, broad-spectrum antibiotics were clearly and significantly associated with increased risk, roughly a two-fold increase, while narrow-spectrum antibiotics showed no significant association at all — though that estimate was quite imprecise, likely because relatively few patients received narrow-spectrum agents alone, so I'd be cautious reading too much into that null finding rather than treating it as a true negative.

In their discussion, the authors frame this as biologically plausible but explicitly not proof of causality — this is a retrospective association study, and they're appropriately careful about that language. They tie it back to the antibiotic-colitis literature as a mechanistic analog and speculate about dysbiosis promoting cutaneous inflammation either through cross-reactive antigens or altered local immune tone, but that part remains speculative. On limitations, they flag likely underreporting of milder ircAEs, and acknowledge that as a retrospective chart-based study, residual confounding from preexisting conditions or concurrent medications wasn't uniformly captured. One nice piece of internal validation: to make sure they weren't just capturing straightforward antibiotic drug eruptions rather than true ircAEs, they looked at timing — classic antibiotic drug eruptions cluster around seven to fourteen days after exposure, whereas the ircAEs in this cohort occurred considerably later, which supports that they're capturing a genuine immune-related phenomenon rather than a misclassified hypersensitivity reaction.

Where does this leave us practically? This is hypothesis-generating, not practice-changing in the sense of altering how you manage an established ircAE. But it is a reasonable nudge toward antibiotic stewardship in patients on checkpoint blockade — if a patient doesn't clearly need broad-spectrum coverage, this data adds one more reason to favor a narrower or more judicious choice, particularly in patients you're worried about for cutaneous toxicity risk. It's also a useful thing to keep in your differential reasoning: a new rash presenting weeks after an antibiotic course in a checkpoint patient shouldn't automatically be filed as a drug eruption. The authors themselves call for prospective study of the skin microbiome's role, and that's really the right way to characterize where this data sits — provocative correlation, awaiting mechanistic and prospective confirmation.

Now to the second piece, which is an editorial — a commentary written to accompany and contextualize a nested case-control study by Hwang and colleagues on risk factors for acral melanoma in US veterans, rather than an original research report itself. So what we're getting here is an expert's guided tour through that companion study's findings and their historical and methodological context, not a standalone data set.

The editorialist opens with a fairly detailed history lesson that's actually clinically relevant background: Agent Orange was one of several herbicide mixtures used by the US military in Vietnam between 1962 and 1971, containing a phenoxy herbicide combined with the dioxin compound TCDD, which is classified as a known human carcinogen. Since the Vietnam War, there's been an ongoing formal review process — the National Academies of Sciences, Engineering, and Medicine has periodically evaluated the epidemiologic evidence connecting Agent Orange exposure to various health outcomes, sorting associations into categories from "sufficient evidence" down to "inadequate or insufficient evidence." As of the most recent 2018 update, several cancers — soft tissue sarcoma, non-Hodgkin lymphoma, chronic lymphocytic leukemia, Hodgkin lymphoma — had sufficient evidence of association, but melanoma and nonmelanoma skin cancer remained in the "inadequate or insufficient" bucket, essentially meaning the existing studies weren't strong or consistent enough to draw a conclusion either way.

This is the gap the companion study steps into. Hwang and colleagues ran a nested case-control study using Department of Veterans Affairs health system data, including the VA Cancer Registry, comparing an acral melanoma case group against two separate matched control groups — one made up of patients with nonacral cutaneous melanoma, matched four-to-one, and a second group of veterans without any melanoma diagnosis at all. Using two distinct control groups is a nice design choice because it helps distinguish whether any signal is specific to acral melanoma as opposed to melanoma risk generally, or just reflects general health-care-seeking or demographic differences. They also used natural language processing on pathology reports to help validate melanoma subtype classification, which the editorialist specifically calls out as a methodological strength.

The key finding: Agent Orange exposure status was associated with a statistically significant, and consistently elevated, adjusted odds ratio for acral melanoma against both control groups, with an effect size in the range of about a thirty percent increase in odds. The editorialist is careful to immediately contextualize the real-world magnitude of that number, and this is worth sitting with: acral melanoma is genuinely rare, on the order of two cases per million person-years in the US. So even if this association is fully causal, the authors themselves calculate it would translate into something like zero-point-four to zero-point-eight additional acral melanoma cases per million veterans per year. Scaled up across the entire estimated Vietnam-era veteran population — upper estimates run to about four-point-three million people — the absolute number of attributable cases would be exceedingly small. That's a really important distinction the editorial draws explicitly: statistically significant, but of limited absolute population impact.

The editorialist also spends real time on the exposure measurement problem, and this is a good methodological teaching point regardless of specialty. Agent Orange exposure status in this study is a presumptive yes-or-no flag based on whether a veteran successfully filed a VA disability claim tied to documented service in a time and place where Agent Orange was used — it is not a direct, individualized measure of actual chemical exposure or dose. That's a real source of exposure misclassification. But the editorialist makes a useful point about the direction of that bias: this kind of non-differential misclassification would typically be expected to pull odds ratios toward the null, meaning the true effect, if the association is real, may actually be larger than what was observed rather than smaller. There's also acknowledged uncertainty about temporality — whether the presumed Agent Orange-exposing service actually preceded the melanoma diagnosis in every case — though the editorial notes the authors provide reassuring data on that front. And because so little is understood about acral melanoma etiology generally, the potential for residual confounding by unmeasured factors remains open.

The editorial also places this in the context of the sparse existing literature — mainly the Air Force Health Study of Operation Ranch Hand personnel with direct herbicide handling exposure and measured serum TCDD levels, which produced suggestive but statistically imprecise associations with melanoma in two separate reports, findings the 2018 NASEM review explicitly flagged as needing more comprehensive follow-up analysis that, as of this writing, still hasn't been done.

So what's the practical takeaway here for those of us seeing these patients in clinic? This is interesting and hypothesis-generating, not practice-changing. It doesn't change screening intervals or biopsy thresholds for Vietnam-era veterans. But it is a reasonable piece of history to have in the back of your mind when you're evaluating an older male veteran with a pigmented acral lesion — remembering that acral melanoma is thought to arise through mechanisms largely independent of the classic UV-driven, fair-skin, nevus-count risk profile, and that exposure history, including military service history, might eventually earn a place in that risk conversation as this literature matures. The editorial's real contribution is underscoring how understudied acral melanoma etiology remains altogether — chemical exposures like this one, but also non-UV factors like mechanical stress on weight-bearing surfaces, which the editorialist explicitly flags as another underexplored lead deserving more rigorous study.

That wraps up this month's review — a hypothesis-generating signal linking antibiotic stewardship to immune-related skin toxicity in checkpoint inhibitor patients, and a thoughtful editorial reminder that the Agent Orange–melanoma question, particularly for the rare and mechanistically distinct acral subtype, is still very much an open one. Thanks for listening, and we'll see you next month.