Welcome back to the journal review. This is our walkthrough of the Journal of the American Academy of Dermatology for May twenty twenty-six, and we've got four pieces on the docket this episode — a health policy narrative review on indoor tanning legislation, a methodologic comment-and-response exchange on GPT-4 Vision in dermoscopy, and then the piece I suspect you'll want to sit with longest, the long-term multicenter randomized trial comparing surgical excision against topical five percent five-fluorouracil and photodynamic therapy for Bowen's disease. Let's get into it.

First up is a health policy and practice piece — a narrative review, not a primary study, so think of this as a synthesis and advocacy piece rather than something with its own methods and results section. The subject is state-level indoor tanning legislation and its effect on adolescent ultraviolet exposure in the United States. The clinical backdrop here is one you already know cold: early indoor tanning exposure raises melanoma risk substantially, on the order of a fifty-nine percent increase, and adolescents, especially white female teens, are the highest-risk users. The policy landscape as of twenty twenty-five is genuinely a patchwork — twenty-one states plus DC now have complete under-eighteen bans, twenty-three states allow tanning with parental consent, and six states have more nuanced carve-outs like requiring a parent to physically accompany the minor.

The review's central argument is built on behavioral surveillance data. Using the Youth Risk Behavior Surveillance System, indoor tanning among high schoolers fell dramatically over less than a decade, from roughly sixteen percent in two thousand nine down to about six percent by twenty seventeen, with the steepest decline in white female students — dropping from roughly thirty-seven percent to about ten percent. That timeline tracks the rollout of stricter access laws, though the authors are appropriately cautious that self-report surveys probably underestimate true tanning rates given growing social stigma around the behavior.

The more clinically useful finding is a direct comparison of ban types. States with complete under-eighteen bans showed a meaningfully lower prevalence of teen indoor tanning compared to states without bans — statistically significant and clinically substantial, roughly a fifty percent relative reduction. Parental consent laws, by contrast, showed essentially no effect on behavior at all — a null finding, meaning that if your state's law lets a parent simply sign off, the data suggest it isn't actually changing what teenagers do. State-specific natural experiments reinforce this: New Jersey saw tanning prevalence drop by roughly half among minors directly covered by its ban, and interestingly, older teens who were technically exempt from the law still showed a large decline too, which the authors interpret as a spillover or norm-shifting effect rather than pure legal deterrence. Minnesota saw an even larger drop, around seventy percent, in eleventh-grade white females within three years of its ban.

The review also touches on the commercial side — a twenty twenty-three economic analysis found reduced tanning-related search activity, falling salon revenue, and increased business closures in ban states, suggesting these laws don't just change individual behavior, they shrink the supply side of the industry. Where the piece is honest about limitations, it's around enforcement — facilities frequently fail to verify documentation, and teens can route around bans via home or gym tanning beds, so there's no illusion here that legislation alone solves the exposure problem. The authors also flag a real evidence gap: most of this data comes from white adolescent girls, while tanning marketing is increasingly targeting LGBTQ+ youth and other under-represented groups who haven't been well studied.

Practically, there's nothing here that changes your Mohs or surgical practice tomorrow, but it's a useful advocacy touchstone — if you're asked to comment publicly or lend your name to legislative advocacy in your own state, this review gives you the cleanest available argument: complete under-eighteen bans work, consent-based laws don't, and enforcement plus federal-level uniformity are the next logical asks.

Next, a linked pair of letters — a methodologic critique and its response — centered on a prior JAAD paper evaluating GPT-4 Vision, the multimodal large language model, on dermoscopic image interpretation. These are exactly what they sound like: a Letter to the Editor raising concerns, followed by the original authors' rebuttal, so there's no new data here, just a sharpened methodologic conversation worth knowing about if you're fielding questions about AI and dermoscopy.

The critique, from Ke and colleagues, raises five points. First, image standardization — the dataset pulled from public platforms may have had variable compression and resolution, the imaging device wasn't specified, and there's a legitimate concern that publicly available training and testing images skew toward lighter Fitzpatrick skin types, since white patients are overrepresented as research subjects, which could inflate or deflate model performance in ways that don't generalize to your actual patient population. Second, selection bias from sourcing images off social media rather than diverse clinical settings, compounded by small sample sizes for some diagnoses. Third, the absence of head-to-head comparison against other large language models like Gemini or Claude, which limits how much you can say about GPT-4 Vision specifically versus multimodal AI broadly. They also note the five comparator dermatologists' expertise levels weren't detailed, and finally, that diagnosing from a single image without any clinical context — age, sex, lesion location, history — is a meaningfully different task than real clinical decision-making, even though they acknowledge that's partly a deliberate privacy tradeoff.

The original authors' response is candid and largely concedes the substance. On image quality, they explain their priority was internal validity — human experts and the model graded the identical images, whatever their native quality, and they simply hadn't been given resolution, compression, or device metadata to control for. They acknowledge model performance might well differ, possibly favorably, on high-resolution standardized clinical images, and explicitly call for future work testing whether there's a quality threshold below which model accuracy degrades. On skin type and demographics, both the physicians and GPT-4 Vision were blinded to Fitzpatrick type and history, which they frame as symmetric blinding rather than a flaw, though they agree testing whether performance improves with demographic metadata is an important next study, particularly given AI's well-documented historical underperformance on skin of color. They agree cross-model comparison is worthwhile but was outside their scope, and they clarify their five-physician panel intentionally spanned four residents at different training levels plus one attending of over twenty years' experience specifically to reduce selection bias in the human comparator arm — and they note, almost in passing, that accuracy did track linearly with training level, which is reassuring construct validity for their design even if it wasn't the primary question.

There's no practice-changing takeaway from this exchange — no clinical numbers to act on — but it's a good snapshot of where the methodologic scrutiny on dermatologic AI studies currently sits, and it's a useful mental checklist if you're asked to peer review or interpret similar AI-dermoscopy papers going forward: image provenance, skin type distribution, comparator model breadth, and blinding to clinical metadata are now the standard critique points.

Now the main event: a multicenter randomized controlled trial reporting long-term results comparing surgical excision, topical five percent five-fluorouracil, and methylaminolevulinate photodynamic therapy — MAL-PDT — for Bowen's disease, which of course is squamous cell carcinoma in situ.

The background here is genuinely a gap worth addressing. Guidelines have never had solid head-to-head randomized data comparing these three modalities, and the oft-quoted three to five percent lifetime progression risk from Bowen's disease to invasive squamous cell carcinoma traces back to two studies from nineteen sixty-one and nineteen eighty that didn't even specify treatment modality or follow-up duration. So this trial's stated purpose is to fill that gap with modern, prospective, long-term data — this is the long-term follow-up extension of a trial the same group already published one-year results on.

Methodologically, this is a multicenter, non-inferiority randomized controlled trial — two hundred fifty patients with biopsy-confirmed primary Bowen's disease, lesions between four and forty millimeters, randomized one-to-one-to-one to surgical excision, five-fluorouracil, or MAL-PDT, stratified by center and lesion size. The non-inferiority design makes complete sense here and the authors say so explicitly: since excision is assumed to be the most effective option, the clinically relevant question isn't "is topical therapy better," it's "is topical therapy an acceptable tradeoff," given that non-invasive options offer real advantages — better cosmesis, feasibility for multiple lesions, treatment where surgery is anatomically awkward, and for five-fluorouracil specifically, at-home dosing and lower cost. Their pre-specified non-inferiority margin was a generous twenty-two percentage points, justified by the reasoning that Bowen's disease is low-risk, slow-growing, and easy to retreat if it recurs — so some sacrifice in primary clearance is clinically tolerable in exchange for those benefits. What makes this particular paper valuable is that it's not a new trial — it's a pre-planned three-to-five-year long-term follow-up of that same one-year cohort, which is exactly the right design choice for answering the question that actually matters for a chronic low-grade neoplasm: does early treatment success hold up over time, or does the recurrence curve keep climbing?

The results are, frankly, reassuring and rest on a single elegant observation: almost everything happened in year one. Of the twenty-seven total treatment failures in the whole study, only one additional failure occurred during the entire long-term follow-up window — a single MAL-PDT recurrence at five years. No new failures at all occurred in the excision or five-fluorouracil arms beyond year one. Translating that into the headline numbers, four-year cumulative tumor-free survival was about ninety-eight percent for excision, eighty-six percent for five-fluorouracil, and eighty-three percent for MAL-PDT. Statistically, five-fluorouracil met the pre-specified non-inferiority bar relative to excision — that twenty-two-point margin held. MAL-PDT did not clear that bar; non-inferiority could not be concluded for photodynamic therapy. And critically for anyone counting risk of missed invasive disease: nobody in any treatment arm developed invasive squamous cell carcinoma in the treated field during long-term follow-up — that's the number I'd actually quote a patient who asks about their long-term cancer risk after topical therapy.

On limitations, the authors are upfront: loss to follow-up ran at just over twenty-three percent, though the majority of that was death unrelated to Bowen's disease in an elderly population with a median age of seventy-six, which is more attrition-by-mortality than true loss-to-follow-up bias. There were also fourteen treatment crossovers baked into the original randomization that required both modified intention-to-treat and per-protocol analyses — reassuringly, both approaches gave similar results, which strengthens confidence that the findings aren't an artifact of protocol deviation. The wide confidence intervals around the four-year survival estimates, particularly for MAL-PDT, also reflect that the failure counts driving these percentages are still modest in absolute terms.

So what do you actually do with this. The practice-changing takeaway is that five-fluorouracil can be confidently offered as a legitimate first-line alternative to excision for appropriately selected Bowen's disease — small to midsized lesions, four to forty millimeters, in patients where the cosmetic, cost, or multiplicity advantages matter — because the durability data now extend convincingly out to four years with a non-inferiority conclusion that holds up over time, not just at one year. The interesting-but-not-quite-as-clean signal is around MAL-PDT: it remains a reasonable option and clearly effective in the majority of patients, but the trial could not statistically confirm it as non-inferior to surgery, so if you're counseling a patient toward PDT specifically because they want to avoid surgery, it's worth setting expectations that recurrence risk sits meaningfully higher than after excision or five-fluorouracil, even if any recurrence remains readily salvageable. And probably the most reassuring message for your broader practice pattern: the near-total absence of late recurrences and the zero rate of progression to invasive disease across all three arms supports relaxing anxiety about long-interval surveillance for treated Bowen's disease — if a lesion hasn't failed by one year, it is very unlikely to fail later, and it is very unlikely to become invasive squamous cell carcinoma in the treated field regardless of which of these three modalities was used.

That wraps our four articles for this May issue — a policy review reinforcing that hard bans, not consent laws, are what actually change adolescent tanning behavior; a sharp methodologic exchange on the current evidentiary weak points of AI-assisted dermoscopy; and a genuinely practice-relevant long-term trial confirming that five-fluorouracil holds its own against surgery for Bowen's disease over years, not just months. Thanks for listening, and I'll see you next issue.