Welcome back to the journal review. This is the October 2026 issue of the Journal of the American Academy of Dermatology, and we've got four pieces to get through this time — a correction notice, an international highlights roundup, a Game Changer commentary on a melanoma follow-up cohort study, and a letter responding to an AI dermatology paper. Let's get into it.

First up is a correction notice, so this is just a quick housekeeping item rather than a full article. It concerns a policy-based narrative review that ran back in May of 2026 looking at state-level indoor tanning legislation and its effect on adolescent behavior and UV exposure. The authors went back and re-reviewed the underlying state legislation and found they'd misclassified several states in their original table. The corrected version now has Montana, New Mexico, Alaska, Colorado, and South Dakota moved into the "no statewide governing law" category, while Connecticut, Oklahoma, Virginia, and West Virginia get reclassified as states with a complete under-eighteen ban. The corrected bottom line: as of 2025, twenty-four states plus DC have complete under-eighteen bans on indoor tanning, seventeen states allow tanning with parental consent, four have mixed or conditional rules like requiring a physician's note or parental accompaniment, and five states have no governing law at all. Nothing earth-shaking clinically here, but if you've cited that review's table in a patient education handout or an advocacy letter to your state legislature, it's worth swapping in the corrected version so you're not misstating your own state's law to a patient or lawmaker.

Next, the Highlights from JAAD International piece, which is a curated summary rather than original research, so think of this as the editor pointing you toward two related survey studies on occupational and recreational UV exposure. The first looked at Boston firefighters attending workplace skin cancer screenings. Three-quarters had no regular dermatologist — which, if you think about the ladder-truck-and-turnout-gear sun exposure profile plus chronic heat and soot exposure, is a striking care gap. The encouraging part: firefighters who were repeat attendees at these workplace screenings reported more confidence doing skin self-exams and scored higher on an objective knowledge quiz, so repeated contact seems to build durable skin cancer literacy, at least on paper. The second study surveyed California surfers about sun protection and skin cancer history. Nearly one in five reported a biopsy-confirmed skin cancer, and that climbed to over three-quarters in surfers fifty and older — which tells you this population has been accumulating meaningful chronic UV dose for decades. Yet only about a fifth applied sunscreen at every session, and ironically the youngest surfers were the least consistent users, which is exactly the group you'd want to intervene with earliest. Strengths across both studies include decent response rates and, in the firefighter study, an objective rather than self-reported knowledge measure; the surfer study gets credit for stratifying by age and Fitzpatrick type. Main limitations are the usual self-report biases on diagnosis history and selection bias in who shows up to answer a survey at a screening event or on a beach. The editorial's own candid takeaway, which I'd echo, is that confidence and knowledge gains are a reasonable surrogate endpoint but we don't yet know whether workplace or recreational-setting outreach programs actually move the needle on earlier diagnosis or outcomes — that's the next study that needs to be done. For your practice, this is less "change what you do" and more "here's supporting rationale" if you're building or justifying an occupational outreach screening program, say partnering with a local fire department or surf shop — the uptake and knowledge data support that kind of meeting-patients-where-they-are approach, even though outcome data isn't there yet.

Now the substantive one — a JAAD Game Changer commentary, written by Jane Grant-Kels, revisiting a population-based cohort study originally published in JAAD in 2024 by Huang and colleagues, titled "Association between dermatology follow-up and melanoma survival." This is a retrospective population-based cohort study, and the clinical gap it's addressing is one you've no doubt felt in your own practice: there is no consensus, guideline-level answer on how often or for how long we should actually be seeing melanoma patients for surveillance after diagnosis. Everyone does annual-ish skin checks, but the evidence base for that habit has been thin.

The design: they identified adults diagnosed with primary invasive melanoma between 2010 and 2013 and followed them through the end of 2018, so this is a population-based registry-linked cohort with meaningful follow-up, on the order of five to eight years depending on diagnosis date. A retrospective cohort is really the only feasible design to answer this question — nobody is going to randomize melanoma patients to annual dermatology follow-up versus none, that would be an ethically fraught trial to recruit for and run. So this falls into the realm of using real-world adherence variation as a natural quasi-experiment: some patients were adherent to annual visits, many weren't, and the investigators used that variation to estimate an association with survival. The commentary doesn't spell out the exact statistical model, but for this kind of time-to-event question with patients followed for different lengths of time and with adherence measured as a time-varying exposure, you'd typically reach for a Cox proportional hazards model — comparing the instantaneous risk of melanoma death between more-adherent and less-adherent patients at each point in follow-up, which handles the fact that not everyone has the same amount of observation time and that adherence itself can change over the course of follow-up. That's a more appropriate tool here than a simple comparison of raw death rates, which would ignore timing entirely and bias you if adherent and nonadherent patients had systematically different follow-up durations.

The key results: adherence to annual dermatology visits was low — only about 28% of patients were actually compliant with seeing their dermatologist every year as recommended. Predictors of better adherence included younger age, female sex, higher income, better access to dermatologic care, higher stage disease at diagnosis — stage II or III — a prior history of keratinocyte carcinoma, fewer comorbidities, and having had any outpatient visit in the year before their melanoma diagnosis. In other words, the patients most plugged into the healthcare system to begin with were the ones who stayed plugged in. And the headline finding: greater adherence to annual dermatology follow-up was associated with reduced melanoma-specific mortality compared with patients who were less adherent — a statistically significant and clinically meaningful association.

As with any retrospective registry analysis like this, the obvious caveat — which the Game Changer commentary doesn't belabor but which you should keep in mind — is confounding by indication and by general health-seeking behavior. Patients adherent enough to show up every year are probably different in a dozen unmeasured ways from patients who don't, healthier at baseline, more health-literate, better resourced, more engaged with their own care overall, and some of that survival benefit could reflect that underlying propensity rather than a causal effect of the visits themselves. The authors can statistically adjust for the predictors they measured, but that's not the same as a randomized comparison.

The practical takeaway here is squarely practice-affirming rather than practice-changing in the sense of giving you a brand-new protocol — Grant-Kels's own lesson-learned line sums it up nicely: continuing to see melanoma patients annually, even years out from diagnosis, is the right thing to do, and now there's population-level outcome data, not just consensus opinion, behind that habit. What is actionable, though, is the adherence data — knowing that older patients, those with lower income or more comorbidities, and those without a recent outpatient visit history are the ones most likely to fall out of your surveillance pipeline gives you a concrete population to target with extra recall effort, reminder calls, or patient navigation resources.

Last up is a letter to the editor — a response to a prior paper by Nouyed and colleagues on text-dominant decision-making by large multimodal models in dermatology clinical challenges. This is correspondence rather than original research, so there's no new methods or results section to walk through; it's the authors extending a discussion. The original Nouyed paper apparently found that across four contemporary large multimodal models, diagnostic accuracy on JAAD Case Reports clinical challenges was about the same whether or not the model actually had access to the clinical image, and that when these models did generate image-anchored explanations, those explanations mostly just paraphrased the text vignette rather than reflecting genuine image analysis.

This letter's authors, who work in neuroradiology and dermatology at a French academic center, write in to say this finding isn't a dermatology-specific quirk — they've seen the identical pattern in their own neuroradiology work, where GPT-4o and Gemini 1.5 Pro got no real accuracy boost from images over clinical context alone, while actual board-certified neuroradiologists did benefit from having the images. They extend the point into histopathology, citing a kidney biopsy study: giving models a vignette plus a biopsy image with only a minimal caption didn't improve accuracy over text alone — for one model, accuracy with a minimal-caption image was numerically worse than text alone — but once the model was given a detailed histopathologic description alongside the image, accuracy jumped substantially, up into the 80% range for one of the models tested. Image-only analysis in that same study showed the model recognized normal histology most of the time but caught disease-specific abnormalities in under half of cases, with most of its errors traceable to training-data bias rather than genuine visual misread.

The authors then raise what they call an expertise paradox from a separate multireader neuroradiology study: AI assistance boosted accuracy more for less experienced readers — residents in adjacent specialties, then radiology residents — and boosted accuracy least for the actual subspecialist neuroradiologists, even though the neuroradiologists had the highest baseline accuracy to begin with and arguably used the tool better. The implication these authors draw, and propose as a research direction for dermatology specifically, is that these models are reasoning predominantly off text, which means the quality of the clinician's own written description of what they're looking at — lesion morphology, distribution, histopathologic detail — is probably the single biggest lever on how useful the AI assistance actually turns out to be. Their suggested next step is for dermatology studies to explicitly test standardized vignette text against clinician-generated text as a moderating variable, stratified by training level, essentially asking whether AI assistance helps junior dermatologists more than senior ones, the same pattern seen in their radiology data.

There's no new primary data generated in dermatology here, so take this as a conceptual contribution rather than anything actionable in your Mohs or oncology practice today. But it's a genuinely useful framing if you're evaluating any AI diagnostic support tool being pitched to your practice or health system: the message is that these tools may be amplifying your own descriptive and semiologic skill rather than substituting for it, so vague documentation in, vague AI output out. Worth keeping in your back pocket as these tools inevitably get pitched to dermatology groups over the next few years.

That wraps up this October 2026 issue — a legislative correction worth updating in your patient materials, encouraging but outcome-agnostic data on meeting high-UV-exposure populations where they work and play, solid supportive evidence for continuing annual melanoma surveillance indefinitely with some useful adherence-risk-factor data to act on, and a thoughtful caution about what multimodal AI models are actually doing under the hood when you hand them a clinical image. Thanks for listening, and I'll see you next issue.