Welcome back to the journal review. This is the March twenty twenty-six issue of the Journal of the American Academy of Dermatology, and we've got four pieces worth your time today — a commentary on AI regulation, a single-institution retrospective study on Merkel cell carcinoma recurrence, a full literature review updating cutaneous squamous cell carcinoma management, and a multi-cohort epidemiologic study looking at Merkel cell carcinoma by virus status and anatomic site across race and ethnicity. Let's get into it.

First up is a commentary — really a policy piece — titled "Current Landscape of AI Regulation in Dermatology," out of Duke. This isn't a study with methods and results; it's an argument, so I'll walk it through as the authors build it. Their starting premise is one you already live with: dermatology is a visually driven specialty, which makes it an obvious target for AI-augmented diagnostics, but the regulatory apparatus governing these tools hasn't kept pace with how fast they're proliferating. They lay out the three FDA pathways currently used for AI-augmented devices — the five ten k premarket notification, the de novo classification, and full Premarket Approval — and their core critique is that applying uniform scrutiny across such a heterogeneous and fast-moving product category doesn't make sense; they argue for a risk-stratified framework instead.

Where this gets practically relevant for you is the direct-to-consumer space. They cite a cross-sectional study of over forty dermatology-focused mobile apps and found only about one in seven had any peer-reviewed publication behind them, and of those, just a single one had gone through a multicenter diagnostic trial. Their concern, stated plainly, is that these apps can provide false reassurance and delay patients from getting to a board-certified dermatologist — which is exactly the failure mode you see downstream in clinic when a delayed melanoma walks in the door.

They contrast that with DermaSensor, which got de novo authorization after the DERM-SUCCESS trial — over a thousand participants, ninety-six percent sensitivity for skin cancer detection — and hold that up as the standard of rigor that should be expected, rather than an exception.

The more forward-looking part of the piece is about what current regulation simply doesn't address: model drift. Unlike a static diagnostic device, AI and machine learning tools keep retraining, and FDA oversight right now is essentially a snapshot at authorization, not a longitudinal surveillance system. The authors flag concept drift — deterioration in performance as a model meets new clinical contexts — as a real risk, particularly for rare melanoma variant discrimination or inflammatory disease mimickers, and warn that without dynamic oversight these tools could quietly reinforce existing disparities rather than closing gaps. Their asks are concrete: premarket "sandbox" testing in constructed clinical environments, prospective dermatologist-supervised trials stratified by Fitzpatrick skin type, lesion type, and clinical setting — primary care versus dermatology clinic versus teledermatology — and transparency around proprietary training datasets, which right now are a black box that blocks independent bias assessment.

For you, practically, there's nothing here that changes what you do in the Mohs suite tomorrow. But it's a useful primer if you're fielding patient questions about phone-based mole-checking apps, or if you're on any institutional committee evaluating AI-based triage tools — this piece gives you the vocabulary and the regulatory landscape to push back intelligently on vendor claims.

Second article: a brief report, single-institution retrospective study, "Identifying High-Risk Subtypes and Risk Factors for Recurrence in Merkel Cell Carcinoma." The gap they're addressing is a real one — NCCN guidance on MCC surveillance is thin compared to melanoma or squamous cell carcinoma, largely because of rarity, and there's very little data on how histologic subtype, specifically combined MCC, affects recurrence risk.

Methodologically, this is exactly the kind of question a retrospective single-center cohort is suited for — MCC is rare enough that you're not getting a prospective randomized answer to a recurrence-risk-factor question anytime soon, so pooling curative-intent cases over two decades at one institution is a pragmatic way to get numbers large enough to run multivariable analysis. They identified one hundred seventy-two patients treated with curative intent between 2003 and 2023, excluding stage four disease at baseline — a sensible exclusion since the question is about recurrence after attempted cure, not de novo metastatic behavior.

The cohort skewed as you'd expect — the vast majority over fifty, about two-thirds male, and about six in ten immunosuppressed or with hematologic malignancy at meaningfully elevated rates relative to the general population. Roughly one third of patients recurred, at a median of about eight months, which is a useful number to keep in mind for surveillance intensity in that first year. Distant recurrence was the single largest category, at about forty percent of recurrences, followed by local, in-transit, and nodal in roughly similar smaller proportions.

The signal that matters here is combined MCC — that's MCC with a concomitant non-neuroendocrine component, usually squamous cell carcinoma, SCC in situ, or basal cell carcinoma. It was rare in this cohort, but recurred in seventeen percent of patients who had it versus about two percent who didn't, a statistically significant difference, and on multivariable Cox regression it carried roughly a four-fold hazard for recurrence independent of other factors. Lymph node metastases and positive surgical margins were the other two independent predictors, with about a three-fold and five-fold hazard respectively — and that margin finding, a five-fold increase, is the kind of number that should reinforce your threshold for aggressive re-excision or Mohs referral when margins come back positive on MCC.

The authors frame combined MCC as biologically distinct — typically Merkel cell polyomavirus-negative and UV-driven, which tracks with a more aggressive natural history, and they cite prior data showing combined MCC has nearly double the metastasis rate and shorter survival compared to pure MCC, consistent with what they found here.

Limitations are the ones you'd expect and the authors are upfront about them: retrospective, single institution, modest sample size, and even the terminology for "combined MCC" isn't standardized across the literature, which limits cross-study comparison. This isn't practice-changing in the sense of altering your surgical technique, but it is clinically actionable as a risk-stratification signal: if a Mohs or excision specimen comes back showing combined MCC-SCC or MCC-basal cell carcinoma histology, that's a patient who probably warrants closer surveillance intervals, similar to how you'd treat a positive nodal or margin status.

Third, and the meatiest piece this month: a full literature review and update, "Management of Cutaneous Squamous Cell Carcinoma," out of a multi-institutional group including Memorial Sloan Kettering and Columbia. This is a synthesis piece, not a primary study, so I'll walk it through as a state-of-the-evidence update rather than hunting for a methods section.

They start where you'd expect — most cSCC is low risk and cured with surgery alone, cure rates over ninety-five percent with wide local excision or Mohs, but a subset behaves badly, and NCCN's low, high, and very-high-risk stratification is the backbone for triage. Useful nuance they highlight: for staging and prognosis specifically, Brigham and Women's Hospital T-staging generally outperforms AJCC eighth edition, with better specificity and positive predictive value for nodal metastasis and disease-specific death — worth remembering when you're trying to have an actual prognostic conversation with a patient rather than just a treatment-planning conversation. They also flag that even BWH T1 tumors with multiple minor risk factors stacked together — moderate differentiation, larger diameter, subcutaneous fat invasion — behave worse than a "clean" T1, and that lymphovascular invasion should be treated as a very-high-risk feature on par with perineural invasion. There's also a newer web-based predictive tool, the RISC model, that reportedly outperforms both BWH and AJCC eight for five-year local recurrence, metastasis, and disease-specific death risk — worth bookmarking if you haven't already incorporated it into your risk conversations.

On imaging, the honest state of the evidence is that there's no firm consensus on baseline imaging thresholds, but the data trend is compelling: imaging changes management in roughly a third of cases in older data, and more recent series push that to about half of high-risk cases. The authors' own practice pattern — which they state explicitly — is to image BWH T2b or T3 tumors, recurrent tumors, cases with extensive perineural invasion, or when there's pain, paresthesia, or suspicion of bony, orbital, or nodal involvement. They're appropriately balanced about the tradeoffs — cost, false positives, incidental findings, and radiation exposure, citing a sobering projection that if current CT utilization patterns continue, radiation-induced cancer could account for a meaningful slice of future cancer diagnoses — which is really an argument for preferentially using ultrasound or MRI where anatomically appropriate.

On treatment, the low-risk algorithm is one you know well — wide local excision with four to six millimeter margins, curettage and electrodesiccation reserved for small, well-defined, low-risk lesions, and their caution that tissue rearrangement or flap reconstruction shouldn't proceed until clear histologic margins are confirmed. For high and very-high-risk disease, the review is unambiguous that Mohs or other peripheral and deep en face margin assessment techniques outperform wide local excision — citing data showing wide excision carries a higher three-year cumulative incidence of recurrence, metastasis, and disease-specific death compared to Mohs, particularly for BWH T2b and T3 tumors. That's a data point worth having ready when you're justifying Mohs appropriateness for a high-stage SCC to a payer or a referring physician.

The systemic therapy landscape gets a meaningful update here too: cemiplimab, the anti-PD-1 agent already approved for locally advanced or metastatic cSCC, has now also received FDA approval in the adjuvant setting for high-risk patients after surgery and radiation, based on improved disease-free survival and reduced locoregional and distant recurrence. That's a genuinely practice-relevant shift — it means your postoperative high-risk patients now have a systemic option to discuss with medical oncology that didn't exist as an approved pathway before. Adjuvant radiation also gets a nod — in BWH T2b/T3 tumors with negative margins, adjuvant radiotherapy roughly halved local and locoregional recurrence risk in the cited data, reinforcing a role for radiation oncology referral even after clear Mohs margins when other high-risk features stack up, like extensive perineural invasion involving larger nerves.

Sentinel lymph node biopsy remains genuinely unsettled — positivity rates reported anywhere from a fraction of a percent up to forty percent depending on the series, no prospective outcome data, and one retrospective study found no relationship between SLNB status and relapse-free or overall survival. The authors' pragmatic suggestion is to borrow the melanoma ten-percent risk threshold and consider SLNB for BWH T2b and AJCC T3-and-above tumors, but they're clear this is consensus-adjacent extrapolation, not high-level evidence.

For locally advanced or unresectable disease, neoadjuvant PD-1 inhibition — cemiplimab or pembrolizumab — produced a response in roughly half of patients with advanced but resectable disease in the cited study, and there are ongoing trials specifically addressing the thorny transplant population, combining checkpoint inhibition with modified immunosuppression, given the real risk of graft rejection with these agents.

Bottom line for your practice: the biggest practice-relevant updates here are the adjuvant cemiplimab approval, the reinforced data favoring Mohs over wide excision for high-risk and very-high-risk disease, and the BWH-over-AJCC8 prognostic preference. The SLNB question and the newer RISC prediction tool are interesting and worth watching but not yet things that should override your current risk-stratification approach.

Last piece: an original multi-cohort study, "Population Differences in Merkel Cell Carcinoma by Virus Status and Anatomic Site," combining University of Washington institutional data, SEER, NCDB, and a global pooled literature analysis. The background question is one that's been murky in the MCC literature for a while — unlike melanoma, where Black and Hispanic patients clearly have worse outcomes, MCC outcome data by race and ethnicity has been inconsistent, and this group's hypothesis is that the explanation runs through tumor biology rather than access or bias alone: specifically, that ultraviolet-protective melanin might shift where Black and Hispanic patients develop their tumors, which in turn shifts virus status, which in turn shifts prognosis.

Methodologically, this is a smart, layered design, and it's worth appreciating why they built it this way rather than relying on any single dataset. The University of Washington cohort gives them a curated, deeply annotated repository with actual virus-status testing and long-term follow-up — good for survival modeling, but a single center, so limited generalizability and modest numbers. SEER and NCDB give them the opposite tradeoff: enormous population-level numbers — over six thousand and sixteen thousand patients respectively with usable data — with race, ethnicity, and anatomic site, but no virus status data at all, since that's not something cancer registries capture. So they layer in a third piece, a systematic pooled analysis of the published literature — thirty-nine cohorts and over thirteen hundred cases for the site-versus-virus-status question, and fifty-eight cohorts across fifteen countries for the geographic virus-positivity question — to bridge the gap between what SEER and NCDB can tell them about race and site, and what only smaller published series can tell them about virus status. That triangulation strategy is really the only way to answer this question given that no single existing dataset contains race, site, virus status, and outcome all together.

On results: virus-positive MCC showed significantly better survival than virus-negative MCC, consistent with prior biology-driven expectations, and virus-positive tumors were significantly more likely to arise on UV-protected skin. Critically, Black and Hispanic patients' tumors were significantly more likely to present on UV-protected anatomic sites compared to White patients. And in the global pooled analysis, ambient ultraviolet radiation had a much bigger association with virus-negative MCC incidence than with virus-positive incidence — supporting the idea that virus-negative disease is the UV-driven subtype and virus-positive disease arises largely independent of UV exposure.

Put together, the chain of inference is: Black and Hispanic patients get relatively more UV-protected-site tumors, UV-protected-site tumors are more often virus-positive, and virus-positive tumors carry better survival — which offers a plausible biological explanation for why race-based MCC outcome data has looked inconsistent or even favorable for Hispanic patients in some series, in contrast to the clearly worse melanoma outcomes seen in the same populations.

Limitations are honestly stated and matter for how much weight to put on this: virus assays weren't standardized across the pooled literature studies — different labs used different PCR and immunohistochemistry approaches, which they tried to reconcile with a preference hierarchy, but that's an imperfect fix. Patient migration history is unknown, which matters for a UV-exposure-driven hypothesis. The global incidence data is incomplete for many regions. And there's inherent registry selection bias in both SEER and NCDB, particularly around who gets biopsied, staged, and reported.

Practically, this doesn't change what you do in the room with an individual MCC patient today — you're not ordering a virus status test to decide margins or adjuvant therapy based on this paper. But conceptually, it's an important reframe: it suggests race-based outcome disparities in MCC may be substantially explained by a virus-and-anatomic-site pathway rather than access-to-care disparities alone, which has implications for how future MCC prognostic models and NCCN guidance might eventually incorporate site and viral status stratified by population. Interesting and mechanistically important — not yet actionable at the bedside, but worth watching as virus-status testing potentially becomes more routine in MCC workup.

That's the March issue. To recap the practice-relevant threads: adjuvant cemiplimab is now a real option for high-risk cSCC patients after surgery and radiation, Mohs continues to outperform wide excision for high and very-high-risk cSCC with real survival-relevant numbers behind it, combined-histology MCC deserves the same surveillance intensity as node-positive or margin-positive disease, and keep an eye on the regulatory conversation around AI diagnostics as more of these tools start showing up in referral patterns. Thanks for listening, and I'll see you next month.