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 picked out four pieces that are particularly relevant to your practice — an editorial perspective on artificial intelligence, a brief report on immunotherapy and skin cancer chemoprevention, a workforce demographics study specifically about Mohs and dermatopathology fellowships, and a JAAD Reviews highlight on disability and skin cancer disparities. Let's get into it.

First up is a clinician's perspective piece by Warren Heymann, titled "Artificial Intelligence in Dermatology: A Matter of Trust." This isn't an original study — it's an editorial that surveys four other papers in the same issue dealing with AI, tied together with the author's own commentary, so I'll walk through it the way he does, hitting each study briefly and then his synthesis.

Heymann opens by noting that his own three-year-old prediction about chatbots becoming integral to practice has already come true faster than expected, and then he uses four papers to make the point that the technology is outrunning our regulatory and trust infrastructure. The first paper he cites found that only about fifteen percent of direct-to-consumer AI dermatology applications are backed by any peer-reviewed publication, which is the core patient-safety concern — these tools are reaching patients without validation, and the call is for prospective, dermatologist-supervised trials before that gap closes.

The second study is more encouraging: a group tested multimodal large language models, meaning models that can take both text and images and produce a structured diagnosis with reasoning, against a set of over three hundred JAMA Dermatology clinical challenge cases. The best model, GPT-o1, hit about eighty-four percent accuracy on its own. But the more clinically interesting number is what happened to the dermatologists — their baseline accuracy was under sixty percent, and it jumped to roughly eighty percent once they had AI assistance, a statistically significant and clinically meaningful improvement. This is the augmentation model working as intended — the AI didn't replace the clinician, it made the clinician better.

Third, a Chinese group built a deep-learning tool to quantify vitiligo repigmentation more precisely than standard scoring tools like the vitiligo area scoring index, and used it to show topical crisaborole was noninferior to halometasone for repigmentation — more a methods validation than a clinical practice shift for you directly, but worth knowing the tool exists.

The fourth study is the cautionary tale, and probably the one most relevant to how you should calibrate your own trust in these tools. Researchers fed ChatGPT-4o several hundred dermoscopic images from a well-validated Harvard dataset — melanoma, benign nevi, seborrheic keratoses, vascular lesions — and simply asked it to identify the lesion and call it benign or malignant. It was bad. Melanoma accuracy was under fifty percent, benign nevi around fifty percent, seborrheic keratosis under thirty percent, vascular lesions around sixteen percent — all significantly worse than the greater-than-seventy-percent accuracy dermatologists typically achieve with dermoscopy. The message is that consumer-facing general-purpose chatbots are nowhere near ready for unsupervised image-based diagnosis, even as the text-based, structured clinical-reasoning tools are looking genuinely additive.

Heymann closes by citing a survey of Australasian dermatologists in which the dominant barrier to AI adoption — for about seven in ten respondents — wasn't accuracy or access, it was trust, or uncertainty about trust. His own take, and I think it's the right one for us, is that these tools are going to become as unremarkable as GPS navigation, but only if we use them the way you'd want a trainee to use any adjunct — verify, maintain skepticism, and go back to the primary literature when something doesn't sit right. Nothing here is practice-changing today, but the trajectory is worth watching, particularly the diagnostic-augmentation data, since that's the use case most likely to show up in your workflow first.

Second, a brief report — a single-center retrospective cohort study out of Ohio State asking whether PD-1 and PD-L1 inhibitor therapy, meaning programmed cell death protein 1 and its ligand, has a chemoprophylactic effect against new non-melanoma skin cancers. The background here is straightforward: checkpoint inhibitors have transformed treatment of advanced cutaneous squamous cell carcinoma and other skin cancers, but almost nothing has looked at what these drugs do to a patient's future risk of developing new non-melanoma tumors, despite the obvious biological plausibility that unleashing antitumor immunity might suppress field cancerization as well.

Methodologically, they identified seventy-eight patients treated with a PD-1 or PD-L1 inhibitor for any indication between 2016 and 2020 who had at least one cutaneous squamous cell carcinoma within a two-year window before or after starting immunotherapy, pulled from infusion records at Ohio State. This is a within-patient, pre-post design — each patient serves as their own control, comparing their two-year tumor burden before immunotherapy against their two-year burden after. That's a sensible choice here because it controls for a lot of patient-level confounding — field cancerization risk, sun exposure history, skin type — that would otherwise plague a between-patient comparison; the authors don't spell this out explicitly, but that's almost certainly why a paired pre-post design was the natural fit for a retrospective single-center cohort like this.

The results were fairly striking. Across the cohort, three hundred total non-melanoma skin cancers were captured, and nearly three-quarters of them — about seventy-two percent — occurred in the two years before immunotherapy started, versus only about twenty-eight percent in the two years after. That's a statistically significant drop, and patients developed on average one fewer non-melanoma skin cancer in the two years following treatment initiation. When they broke it down by tumor type, the effect was almost entirely a squamous cell carcinoma story — a significant reduction, again roughly one fewer tumor per patient on average — while basal cell carcinoma incidence didn't budge at all, with essentially identical numbers before and after. There was no difference in effect based on which specific checkpoint inhibitor was used, and no relationship to whether the primary tumor actually responded to treatment.

The authors frame this as consistent with known biology — cemiplimab and other checkpoint inhibitors show much better response rates in squamous cell carcinoma than basal cell carcinoma, plausibly because squamous tumors are more immunogenic, so it makes sense that any chemopreventive halo effect would track the same pattern. But they're appropriately cautious about their own limitations. The big one is selection bias: because having a squamous cell carcinoma was actually part of the inclusion criteria, this cohort was inherently enriched for squamous-prone patients, which could easily explain why the squamous effect is so much more prominent than the basal cell finding — so the basal cell null result in particular should be read with real caution rather than as reassurance that checkpoint inhibitors simply don't affect basal cell risk. Other limitations are the ones you'd expect from a retrospective single-center design: overwhelmingly white, male cohort, a short observation window, and loss to follow-up.

Practically, this is interesting, not practice-changing. It's hypothesis-generating for a potential chemopreventive role of checkpoint inhibitors in high-risk squamous disease, but nobody is recommending you start counseling patients that immunotherapy will reduce their future skin cancer burden, and certainly nobody is suggesting checkpoint inhibitors as an alternative to nicotinamide or acitretin for standard chemoprevention given the risk profile of immunotherapy itself. File this away as a data point suggesting our high-risk squamous cell carcinoma patients on checkpoint inhibitors may need somewhat less anxious surveillance for new squamous tumors during treatment, while their basal cell risk should be monitored just as before.

Third, and probably the one that hits closest to home professionally — a cross-sectional analysis of demographic trends among fellows in Micrographic Surgery and Dermatologic Oncology and in Dermatopathology, covering academic years 2010 through 2024. The gap being addressed is that dermatology remains one of the least diverse specialties in medicine, prior work has shown some improvement in fellowship diversity, particularly for gender, but nobody had systematically tracked whether that progress has actually reached the procedural subspecialty level — your level — over a full fifteen-year span.

Methodologically this was straightforward and pragmatic: the authors pulled publicly available ACGME Data Resource Books, that's the Accreditation Council for Graduate Medical Education, for both fellowship types across each academic year, and looked at trends in gender, race and ethnicity, age, and medical school or graduate status, using chi-square tests and ANOVA to assess change over time. Using the ACGME data resource books makes sense as basically the only source of longitudinal, standardized fellow-level demographic data across this many programs and years — you're not going to get a comparably complete dataset any other way without surveying every program director individually.

What did they find. Fellowship class sizes stayed essentially flat over the fifteen years in both specialties. In Micrographic Surgery and Dermatologic Oncology, male fellows generally outnumbered female fellows throughout, though the female proportion did rise from about one quarter to about four in ten — that trend was not statistically significant. In dermatopathology, women were often the majority, hovering in the forty-to-fifty percent range throughout, also without significant change over time — in other words, dermatopathology started from a better gender balance and basically stayed there, while Mohs made some numeric gains that didn't reach significance. On race and ethnicity, white fellows remained the majority across both fellowships, and in Mohs specifically that share actually declined significantly, from around seventy-four percent to about sixty-seven percent — but critically, that decline was not matched by a corresponding rise among underrepresented minority groups. Black and Hispanic fellows stayed under five percent in both fellowships across the entire study period, essentially flat. Asian representation increased modestly. Mean age was stable throughout, in the mid-thirties for both fellowships, and U.S. allopathic graduates remained the dominant pathway into both, with osteopathic and international medical graduates a persistent minority.

The authors' discussion is candid: diversity, equity, and inclusion efforts have expanded broadly in dermatology, but this data suggests that impact has not meaningfully trickled down to fellowship-level representation, particularly for Black and Hispanic trainees. They raise structural explanations — smaller applicant pools at the subspecialty level, entrenched selection practices, cumulative disparities compounding from earlier training stages, and reduced access to mentorship for underrepresented, international, osteopathic, and female applicants navigating an already competitive process. The limitations are the ones inherent to any registry-based study like this: it relies on aggregate publicly reported data rather than individual-level records, gender was captured as binary with no accounting for nonbinary or gender-fluid identities, and because there's no applicant-pool denominator, the study can't actually distinguish whether the disparity originates in who applies versus who gets selected.

For you, I'd call this important context rather than practice-changing in the clinical sense — there's no procedure or protocol to change here. But if you're involved in fellowship selection, mentorship, or program leadership, this is a direct, current data point that the diversity conversation in Mohs and dermatopathology specifically has stalled for Black and Hispanic trainees even while overall specialty-level DEI initiatives have expanded, and that's worth carrying into any conversation about recruitment or mentorship pipeline efforts at your own institution.

Fourth and last, a "Highlights from JAAD Reviews" piece by Shari Lipner, titled "Invisible Barriers, Visible Consequences: Skin Cancer Disparities in Intellectual Disability." This is a commentary highlighting a scoping review, not a primary study, so there's no methods or results section to walk through — it's a synthesis, and I'll treat it that way.

The piece spotlights a scoping review by Lapolla and colleagues in JAAD Reviews looking specifically at skin cancer outcomes among people with intellectual disability, and situates it alongside a two-part continuing medical education series on disability in dermatology more broadly, covering cognitive, mobility, hearing, and vision impairments. The core finding being highlighted is that people with intellectual disability are less likely to receive routine total body skin exams, less likely to engage in sun-protective behavior or perform self-skin exams, and when melanoma is eventually diagnosed, it tends to be caught at a more advanced stage. The scoping review is explicit that this isn't a biological risk story — it's structural: communication barriers, dependence on caregivers to notice and report changes, symptoms getting misattributed to the underlying disability rather than worked up as a new skin finding, and generally limited access to preventive education.

The companion CME series adds a practical framework — screening all patients for cognitive impairment even when it isn't obvious on a routine visit, understanding a patient's goals and capacity to participate in shared decision-making, and concrete accommodations like longer appointment slots, caregiver-inclusive visits, visual aids, breaking exams into steps, and procedural flexibility. Lipner's synthesis frames the disparity as predictable and, importantly, modifiable, and calls for prospective work testing tailored surveillance protocols and caregiver education models.

There's no data here to round or caveat in the usual sense — this is a call to awareness and a practice-workflow prompt rather than a trial result. The practical takeaway for a Mohs surgeon or dermatologic oncologist is concrete even without new efficacy data: patients with intellectual disability in your practice or your referral base are a population at elevated risk for delayed melanoma diagnosis for reasons that have nothing to do with tumor biology, and simple accommodations — longer visits, caregiver involvement, lower threshold for a full skin exam even when the presenting complaint is unrelated — are low-cost interventions you can implement today, well ahead of any forthcoming trial data.

That wraps up this month's four articles — a sobering but useful look at where AI diagnostic tools actually stand today, a hypothesis-generating signal on checkpoint inhibitors and squamous cell carcinoma chemoprevention, a hard look at where our own subspecialty's diversity efforts have and haven't worked, and a reminder that some of the most fixable disparities in melanoma outcomes are structural rather than biological. Thanks for listening, and I'll see you next month.