Welcome back to the journal review. This month we're working through the March 2026 issue of the Journal of the American Academy of Dermatology, and I've pulled three pieces for you — one pharmacovigilance study out of the Brief Reports section, and two entries from the Ethics Journal Club that are worth sitting with even though neither one is going to change how you hold a blade tomorrow.

Let's start with the pharmacovigilance analysis on relatlimab. This is a retrospective comparative pharmacovigilance study, meaning the authors went into the FDA's adverse event reporting system, FAERS, and mined it for signal rather than running a trial. The clinical backdrop here is that in 2022 the FDA approved nivolumab plus relatlimab — that's the first LAG-3 inhibitor, lymphocyte activation gene 3, paired with a PD-1 inhibitor — as a combination regimen for metastatic melanoma. We already have a mature combination in nivolumab plus ipilimumab, the CTLA-4 pairing, and we know its immune-related adverse event profile intimately because we've been managing it for over a decade. What we don't have a great handle on yet is how the relatlimab combination's toxicity profile actually compares in a real-world reporting population, since the pivotal trial data is efficacy-focused and trial populations are curated in ways that real-world prescribing populations are not.

So the authors queried FAERS from 2022 through 2025 for adverse event reports tied to each regimen when used for melanoma, excluding reports where patients were on multiple concurrent medications, where the reported event wasn't melanoma-related, or where the report was simply "treatment failure" rather than a discrete adverse event. That's a sensible design choice for this kind of question — FAERS is a spontaneous reporting database, so you can't calculate true incidence rates or do time-to-event analysis, but you can compare the composition and severity of what does get reported between two drugs approved for the same indication, which is exactly what they did using chi-square tests and t-tests.

They ended up with 121 adverse event reports for the nivolumab-relatlimab combination and 779 for nivolumab-ipilimumab. A few things jumped out. Patients reported on relatlimab combination therapy were, on average, about a decade older than those on the ipilimumab combination, roughly 70 versus 60, a statistically significant difference. Gender distribution was essentially the same across both groups. Reaction severity, though, diverged sharply — about 95 percent of ipilimumab combination reports were classified as serious, compared to about 71 percent for the relatlimab combination, a significant and clinically meaningful gap. The relatlimab group also had a lower proportion of reports containing at least one immune-related adverse event — about two-thirds versus more than three-quarters for ipilimumab — and notably fewer associated hospitalizations, around 27 percent versus 45 percent. Mortality, interestingly, was essentially identical between the two, around 9 percent in each, so the safety advantage for relatlimab shows up in morbidity and severity, not in reported deaths.

Where it gets clinically textured is in the organ-system breakdown. Relatlimab combination therapy showed a higher proportion of cardiac, neurologic, and musculoskeletal immune-related events — cardiac driven almost entirely by myocarditis, neurologic driven by myasthenia gravis and encephalitis, musculoskeletal by myositis, myalgia, and arthralgia — all significant differences. Conversely, gastrointestinal and endocrine immune-related events were significantly less common with relatlimab than with ipilimumab, which tracks with what we already know about ipilimumab's predilection for colitis, hepatitis, and endocrinopathies like hypophysitis and thyroiditis. For us specifically, the dermatologic signal is the interesting null result: there was no significant difference in cutaneous immune-related adverse events or infusion reactions between the two regimens, and pulmonary toxicity was also similar.

The authors' discussion frames this as consistent with the proposed mechanism — LAG-3 and PD-1 co-blockade synergistically restoring T-cell cytotoxic function through a somewhat different immunologic route than CTLA-4 blockade, which may explain the shifted toxicity spectrum toward cardiac and neurologic events rather than gastrointestinal and endocrine ones. They note this pattern of higher cardiotoxicity with relatlimab and higher GI and endocrine toxicity with ipilimumab replicates prior work, though their finding of higher neurologic and musculoskeletal events with relatlimab, plus the significantly lower hospitalization rate for relatlimab, appears to be new relative to earlier reports.

Limitations are the ones you'd expect from any FAERS study, and the authors are upfront about them: a modest sample size, comparison limited to just these two regimens rather than the full melanoma immunotherapy landscape, no standardized adverse event grading since these are voluntary spontaneous reports, and no data on whether events actually led to treatment discontinuation. There's also the inherent reporting bias of a passive surveillance system — a newer drug like the relatlimab combination may simply be under-reported relative to a combination that's been on the market and in wide use far longer.

So what do you actually do with this. It's not practice-changing in the sense of altering your biopsy or excision technique, but it is useful counseling data. If you're fielding questions from a melanoma patient or their oncologist about relative tolerability between these two checkpoint combinations, this adds real-world weight to the idea that the relatlimab combination carries a somewhat more favorable overall toxicity and hospitalization profile, with the important caveat that you should have a lower threshold for cardiac and neuromuscular symptom review in these patients rather than assuming the ipilimumab-era toxicity checklist transfers directly. The reassuring point for dermatology specifically is that cutaneous immune-related events look similar between the two, so your clinical vigilance for lichenoid or bullous eruptions and other cutaneous immunotoxicity doesn't need to shift based on which combination the patient is on.

Now let's turn to the ethics side of the issue, both pieces this month coming from the Ethics Journal Club's "Dear Dr Dermatoethicist" format — these are advice-column-style ethics commentaries, not studies, so there's no methods or results to walk through, just a scenario, the ethical reasoning, and a practical resolution.

The first is titled "Invisible Hands," addressing artificial intelligence use by medical students working with dermatologists. The scenario: a mentor receives a suspiciously polished, suspiciously fast clinical review draft from a medical student and suspects undisclosed AI use. The piece opens with some framing data — a 2024 survey found about three-quarters of researchers now use AI tools in their work, including machine translation and chatbot tools, and a separate analysis of manuscripts before and after ChatGPT's late 2022 launch found the proportion of abstracts containing detectable AI-generated text roughly doubled, a statistically significant jump. The core tension the authors lay out is that suspicion based on writing speed or polish alone isn't evidence, and unfounded accusations can genuinely damage a mentorship relationship — but silence has its own costs. Undisclosed AI use compromises the mentor's autonomy to make informed decisions about the research, threatens academic justice if it constitutes an unfair and unacknowledged advantage in something like a residency or fellowship application, and creates real liability exposure if AI-generated inaccuracies make their way into a publication that then gets cited or acted on clinically. Their practical resolution is refreshingly concrete: don't accuse, ask. Open with a neutral, open-ended question like "walk me through how you developed this," which either prompts voluntary disclosure or simply resolves the suspicion. If AI use is confirmed, the response should hinge on degree — using it to polish a draft the student intellectually authored is a different ethical category than having it generate the substantive content — and the relationship going forward should depend on the student's honesty and willingness to engage, not on the mere fact that a tool was used. The broader institutional recommendation is proactive: document concerns, have the conversation, loop in the research ethics office if it doesn't resolve, and push your program toward explicit AI use guidelines before this comes up again, because it will.

The second piece tackles a scenario a lot of you have probably lived through directly: a dermatologist is asked by a primary care department to teach their residents, attendings, and advanced practice providers how to perform skin biopsies. The ethical tension here is genuinely two-sided. On one hand, dermatology access is a real problem, particularly in rural, inner-city, and under-resourced areas, and long wait times for a suspicious lesion carry their own morbidity and mortality risk — so expanding the base of clinicians who can biopsy has a real beneficence argument behind it. On the other hand, the authors cite prior work showing dermatologists require a significantly lower number needed to biopsy to catch one skin cancer than primary care providers do, meaning less selective biopsying by less experienced clinicians translates into more unnecessary procedures, higher costs, and avoidable patient morbidity — the maleficence side of the ledger. There's also a pointed observation that procedural reimbursement is lucrative, which creates a financial incentive for primary care providers or newly trained advanced practice providers to acquire and use this skill regardless of whether it's actually improving triage accuracy.

The authors' resolution is that procedural training divorced from diagnostic training is the actual hazard — teaching someone to cut without teaching them what to cut is where the ethical exposure lives. Their recommendation is a curriculum that leads with clinical and dermoscopic lesion recognition before any biopsy technique is introduced, ideally paired with dermatology department rotations so trainees see the full spectrum of benign versus malignant presentations and learn how to handle procedural complications and interpret the resulting dermatopathology report correctly, so an ambiguous or discordant result doesn't get either dismissed or over-acted upon. They also flag an informed consent dimension specific to this situation — patients should know the training level and experience of whoever is about to biopsy them — and cite older survey data showing patients generally have more confidence in a dermatologist's skin assessment than a primary care provider's and prefer direct dermatology access when possible. Their closing point is that the dermatologist invited to build this curriculum has real leverage and real responsibility in shaping it, and that periodic proficiency assessment isn't optional if the program is going to avoid trading an access problem for a new safety problem.

Neither ethics piece is going to change a technique or a treatment algorithm, but both are the kind of thing worth having a considered position on before you're the one standing in the room being asked the question — whether that's a mentor sitting across from a suspiciously fast manuscript draft, or a primary care chair asking you to build their procedural curriculum. That's the issue for this month. Thanks for listening, and I'll see you next time.