# JAAD Journal Watch — July 2026

Welcome back to Journal Watch, where we go straight to the data from the Journal of the American Academy of Dermatology. This month we're covering four items: a methodologic letter on artificial intelligence in evidence synthesis, a cohort study on skin cancer risk in systemic sclerosis patients with organ transplantation, a critical comment on Delphi guidelines for topical five-fluorouracil and calcipotriene, and an original cohort study comparing melanoma outcomes between a Veterans Affairs center and a tertiary care center with shared physician staffing. Let's get into it.

First up: "Artificial Intelligence in Evidence Synthesis: Benefits, Limitations, and the Importance of Transparent Reporting," by Kahn, Tepper, and Marmon. This is a letter responding to a prior editorial by Elston on classifying review types. The clinical question here isn't a patient outcome question — it's a methodologic one: how should dermatology handle the growing, often undisclosed, use of artificial intelligence tools in systematic reviews and other evidence synthesis?

The authors lay out a taxonomy of tools. General-purpose large language models — think ChatGPT, Claude, Gemini — can assist with drafting, editing, summarization, and screening, but the authors flag that these produce stochastic outputs and can fabricate or misstate citations. Retrieval-augmented research assistants, such as Elicit or Consensus, link outputs to specific retrieved papers, which improves transparency but does not eliminate misinterpretation or incomplete retrieval. Then there are full systematic review management platforms like Covidence, DistillerSR, and EPPI-Reviewer, and dedicated screening or extraction tools like Abstrackr, SWIFT-ActiveScreener, and RobotReviewer — these are described as more auditable and workflow-specific, but still dependent on human-supplied search strategies and verification.

The key data point cited: when ChatGPT-4o was tasked with applying the Cochrane Risk of Bias 2.0 tool to one hundred randomized trials, it rated none of them as overall high risk of bias, and showed poor agreement with human reviewers — a citation to Kuitunen and colleagues. The authors note that whether large language models reliably incorporate other quality-relevant factors — industry funding, heterogeneity, sample size — has not been systematically evaluated.

On reporting frameworks, the letter distinguishes PRISMA 2020, which is the baseline but doesn't address AI use at all; PRISMA-AI, which applies only when AI is the subject being evaluated, such as a diagnostic algorithm under review; and PRISMA-trAIce, a fourteen-item checklist proposed specifically for disclosing AI used as a methodological tool during review conduct — covering tool identity, version, prompts, human-AI interaction, and performance evaluation. The authors are explicit that PRISMA-trAIce remains early-stage, without broad validation or consensus endorsement.

Takeaway for practice: if you are authoring or evaluating systematic reviews in dermatology — including anything informing surgical or oncologic guidelines — transparent disclosure of AI tool use should be treated as a baseline expectation, not an afterthought, given documented poor agreement between large language models and human reviewers on risk-of-bias judgments. The main limitation of this piece is that it's a letter, not a primary study — it synthesizes existing evidence and proposes a framework rather than generating new data.

Second article: "Assessing Skin Cancer Risk in Systemic Sclerosis Patients with Organ Transplantation: A Cohort Study," by Kanwar, Anshelevich, Sakunchotpanit, Fettel, and Nambudiri. Clinical question: does organ transplantation modify skin cancer risk specifically in patients with systemic sclerosis, a population already known to carry elevated malignancy risk?

This is a retrospective cohort from the Mass General Brigham registry, spanning 1979 to 2024. Investigators queried for systemic sclerosis with kidney, bone marrow, or lung transplantation, identifying 178 exposed and 374 unexposed patients initially; after chart review and two-to-one matching by age, sex, and race, the final analytic cohort was 48 systemic sclerosis patients with transplant and 45 without.

Key findings: transplantation was associated with a relative risk of five point three one for overall skin cancer — ninety-five percent confidence interval one point six seven to sixteen point nine one — p equals zero point zero zero four seven. For keratinocyte carcinoma specifically, relative risk was four point six nine, ninety-five percent confidence interval one point four five to fifteen point one two, p equals zero point zero zero nine seven. Looking only at skin cancers arising after transplantation, relative risk was three point seven five, p equals zero point zero three zero six, driven by squamous cell carcinoma with a relative risk of four point six nine, p equals zero point zero three eight four.

The sex-stratified analysis is the standout finding: female patients with systemic sclerosis and transplant had a relative risk of three point seven for overall skin cancer, p equals zero point zero two nine eight, and three point three three for keratinocyte carcinoma, p equals zero point zero four six eight, compared to female systemic sclerosis patients without transplant. Males showed no significant difference — relative risk nine point two six for keratinocyte carcinoma, but with a wide, non-significant confidence interval spanning zero point five five to one hundred fifty-five, p equals zero point one two. Subgroup analysis by transplant type — kidney, lung, or bone marrow — did not yield significant associations, likely reflecting sample size constraints.

Descriptively, transplant patients averaged two point two four skin cancers each versus one point zero among non-transplant patients, and immunosuppressive regimens were similar between transplant patients with and without skin cancer — steroids in one hundred percent of both groups, tacrolimus in eighty-one point three percent versus seventy-one point four percent, mycophenolate mofetil in forty-three point eight percent versus fifty-seven point one percent — suggesting drug regimen alone doesn't explain the skin cancer difference.

Limitations are significant and acknowledged directly: retrospective design, single health system, and a small cohort that precluded robust multivariable adjustment for smoking or specific immunosuppressive exposures. Takeaway: this supports targeted dermatologic surveillance for systemic sclerosis patients undergoing transplantation, with particular attention to female patients, who showed a disproportionate risk increase relative to males — a pattern the authors speculate may relate to loss of baseline sex-related immune surveillance advantages, though this is not directly tested in the data.

Third: "Comment on 'Topical Combined 5-Fluorouracil and Calcipotriene for Actinic Keratosis and Superficial Keratinocyte Carcinoma: Modified Delphi Expert Panel Guidelines from ITSCC,'" by Dai and Cao. This is a critical letter responding to the ITSCC Delphi consensus guidelines on combination five-fluorouracil and calcipotriene, referred to as five-FU-C, raising three methodologic and safety concerns relevant to anyone using this regimen for actinic keratosis or superficial keratinocyte carcinoma in lieu of surgery.

First concern: endpoint selection. The Delphi guidelines cite an eighty-six point seven percent lesion count reduction, but the authors point out this doesn't address complete clearance. They cite a systematic review by Mohney and colleagues showing only twenty-seven percent of patients achieved complete facial clearance at week eight — substantially lower than typical complete clearance rates reported for FDA-approved five-fluorouracil monotherapy. The authors argue complete clearance, not proportional lesion reduction, is the metric that correlates with chemoprevention and reduced malignant transformation risk, and that anchoring on lesion reduction risks overstating curative potential.

Second concern: pretreatment histologic confirmation before using five-FU-C for squamous cell carcinoma in situ. The authors cite a 2024 systematic review of three thousand forty-nine squamous cell carcinoma cases showing an upstaging rate of twenty-two point seven four percent when biopsy was compared to definitive surgical specimens — with punch biopsy discordance at twenty-eight point four percent and shave biopsy discordance at seventeen point six percent. Even in a lower-risk scenario — shave biopsy of routine squamous cell carcinoma in situ — a 2024 cohort of six hundred sixty-five patients showed one case, zero point two percent, upstaged to invasive disease. Small percentage, but the authors note the clinical stakes for that individual patient are substantial: topical therapy instead of surgery could mean delayed definitive treatment.

Third concern: relying on clinical complete response as the efficacy endpoint may mask subclinical residual tumor. They cite Patel and colleagues' pilot study showing histologic clearance of only eighty-three point three percent at day seven and eighty-seven point five percent at day fourteen — meaning twelve to seventeen percent of clinically cleared lesions still harbored residual disease on histology.

The authors' proposed refinements: base treatment duration recommendations on complete clearance rather than lesion-count reduction; require pretreatment biopsy for high-risk lesions — greater than ten millimeters, located on ears, nose, lips, or eyelids, or with an indurated base; and consider post-treatment histologic or confocal confirmation for squamous cell carcinoma in situ and superficial basal cell carcinoma in future trials, with a risk-stratified approach given practical limitations of confocal access and patient willingness to undergo rebiopsy. For Mohs surgeons, the actionable point is that biopsy-based staging discordance rates in the twenty to twenty-nine percent range for punch biopsies are directly relevant when counseling patients about topical alternatives to surgical clearance, particularly for lesions in high-risk anatomic locations.

Fourth and final article: "Differential Melanoma Outcomes in a Regional VA Center and a Tertiary Care Center with Shared Physician Staffing: A Cohort Study," by Chen, Fernandez, Tang, and colleagues, senior author Maria Wei. Clinical question: when veteran and non-veteran melanoma patients are cared for by an overlapping pool of physicians in the same city — controlling for provider-level variation — do outcomes still differ by institution, and if so, why?

This is a retrospective cohort spanning January 1987 to May 2023, comparing the San Francisco VA Health Care System, six hundred thirty-eight patients, six hundred twelve male, to the University of California San Francisco tertiary center, nine thousand six hundred eighty-two patients, five thousand five hundred sixty-four male. Because of the overwhelming male predominance at the VA — ninety-six point two percent versus fifty-seven point five percent at the tertiary center, p less than zero point zero zero zero one — the primary analysis focused on male patients.

Baseline differences were substantial: VA male patients were older at diagnosis, mean seventy point nine versus fifty-nine point four years, p less than zero point zero zero zero one; lived farther from the treating hospital; resided in lower median household income zip codes; and had a higher burden of prior cancers, HIV, or immunosuppression. Critically, fifty-four point one percent of VA male patients presented with melanoma in situ compared to eighteen point six percent at the tertiary center, p less than zero point zero zero zero one, while tertiary center patients were more likely to present with regional disease, nineteen point three percent versus six point six percent, or distant disease, seven point one percent versus five point eight percent.

Despite these baseline differences, overall melanoma-specific survival excluding melanoma in situ showed no significant difference between institutions: five-year survival eighty point six percent tertiary versus seventy-eight point four percent VA, ten-year seventy-three point nine percent versus seventy-five point one percent, p equals zero point four five.

The more clinically important finding emerged on stratification by SEER stage. For localized disease, VA male patients had significantly better melanoma-specific survival than tertiary center patients, p equals zero point zero zero three. But for regional disease, tertiary center patients did significantly better, p equals zero point zero zero zero two, and for distant disease, tertiary center patients again did better, p equals zero point zero two. Consistent with this, a greater percentage of VA male patients with localized disease had Breslow depth under one millimeter — sixty-nine point three percent versus forty-seven point one percent at the tertiary center.

On multivariable Cox modeling, SEER stage at diagnosis was the strongest predictor of melanoma-specific mortality at both institutions. Within the tertiary center specifically, insurance type mattered significantly: compared to privately insured patients, those with Medicare had a hazard ratio of one point four six, ninety-five percent confidence interval one point two two to one point seven five; Medicaid, one point six seven, ninety-five percent confidence interval one point two eight to two point one eight; and Military or VA insurance, one point nine two, ninety-five percent confidence interval one point one five to three point two zero. A history of other cancers was associated with increased mortality at the tertiary center, hazard ratio one point three four, p equals zero point zero three, but not at the VA, hazard ratio one point one eight, p equals zero point seven one. At the VA, age forty to sixty carried over three times the mortality risk compared to over sixty, hazard ratio three point zero four, p equals zero point zero zero four — a pattern not seen at the tertiary center.

When the authors isolated veterans at both sites — comparing VA-insured patients treated at the tertiary center, n equals fifty-four, to patients treated at the VA itself, n equals six hundred twelve — tertiary-center veterans were younger, mean fifty-nine point three versus seventy point nine years, presented with more regional and distant stage disease, eighteen point nine percent each versus six point six percent and five point eight percent at the VA, and were far more likely to die from melanoma, thirty-one point five percent versus nine point six percent, p less than zero point zero zero zero one. Household income by zip code did not differ significantly between these two veteran subgroups, suggesting the divergence reflects something other than socioeconomic status alone.

The authors' interpretation is that differences in preventive screening intensity and selective referral patterns likely contribute to the VA's early-stage survival advantage, while insurance type — both within and between institutions — is independently associated with worse outcomes for advanced-stage disease, pointing to a need for care standardization regardless of treatment site. The main limitation flagged by the authors is the small number of VA-insured patients treated at the tertiary center, fifty-four, which constrains precision in that specific comparison. For Mohs surgeons and dermatologic oncologists working within multi-site systems, this study is a useful reminder that shared physician staffing does not equalize outcomes — institutional screening infrastructure and insurance-linked access to advanced therapies remain independent drivers of melanoma-specific mortality, particularly for regional and distant stage disease.

That wraps up this month's review. Four papers spanning methodology, autoimmune-associated skin cancer risk, topical field therapy safety considerations, and systemic outcome disparities in melanoma care — each reinforcing that the numbers behind our guidelines and referral patterns deserve as much scrutiny as the techniques we use in clinic. Thanks for listening, and we'll see you next month.