Welcome back to the podcast. This is our walkthrough of the September twenty twenty-six issue of the Journal of the American Academy of Dermatology, and we've got four pieces worth your time today — two brief reports, one full-length diagnostic accuracy study, and one multicenter pathology-utilization cohort. Let's get into it.

First up is a brief report titled "Clinical factors associated with major poor outcomes in kidney transplant recipients with cutaneous squamous cell carcinomas." This is a retrospective case series out of the University of Wisconsin. The background here is one you already live with clinically: solid organ transplant recipients carry a markedly elevated risk of cutaneous squamous cell carcinoma and of the bad outcomes that follow — metastasis and disease-specific death. What's less well characterized is the actual temporal and clinical pattern of how those bad outcomes unfold, which matters for how aggressively and how long we surveil these patients.

Methodologically, this is about as straightforward as retrospective work gets — an Institutional Review Board-approved chart review identifying sixteen kidney transplant recipients who had squamous cell carcinoma-related metastasis and or death after transplant. The authors don't spell out their rationale beyond the rarity of the event, but it's worth naming explicitly: because they enrolled based on already having had the outcome, this is a case series of the worst-case tail, not a comparative cohort. There's no denominator of transplant patients who did fine, so you can't calculate incidence or relative risk from this — the authors acknowledge exactly that limitation themselves. Think of this as hypothesis-generating pattern description, not a risk-factor analysis in the inferential sense, and there are no formal statistical comparisons to walk through here.

What did the pattern-finding show? These sixteen patients accumulated a total of a hundred and seventy-two squamous cell carcinomas after transplant, averaging almost eleven tumors per person, though the burden varied enormously by outcome group — patients who had metastasis alone averaged only two tumors, while those who died averaged around ten to twelve. Median time from transplant to first squamous cell carcinoma was six years, and about half developed their first tumor within the first five years. Annual tumor burden peaked around year fourteen post-transplant, and the majority of first tumors, metastatic events, and deaths all clustered within the first fifteen years after transplant. Here's the number worth remembering for surveillance planning: a substantial share of both metastases and deaths — roughly eight of thirteen metastatic events and six of fourteen deaths — occurred after only three or fewer documented tumors. In other words, catastrophic outcomes are not reliably preceded by a heavy tumor burden; low tumor count does not equal low risk. Head and neck tumors dominated both overall counts and metastatic risk. Aggressive histologic features — poor differentiation, perineural invasion, deep invasion beyond subcutaneous fat — were uncommon overall but clustered disproportionately in the patients who ultimately died rather than those with metastasis alone. And chemoprevention was used sparingly across this cohort — niacinamide, acitretin, and topical field agents were each used in only a small minority.

The honest limitations are small sample size driven by the rarity of the event, single-institution retrospective design, and no control group, so nothing here should be read as an adjusted risk factor. The practical takeaway is more about vigilance than about changing your treatment algorithm: this reinforces that transplant recipients need durable, long-horizon surveillance — well past the five-year mark and really out to fifteen years and beyond — and that a low cumulative squamous cell carcinoma count should not be reassuring in this population, since fatal outcomes can occur early and after very few lesions. It's also a nudge to be more proactive about chemoprevention discussions, given how underused they were here. This is interesting and worth internalizing for surveillance intensity, but it isn't a practice-changing dataset in the statistical sense — there's no comparator to hang a number on.

Second article: "Clinical outcomes and complications of large full-thickness skin grafts post-Mohs surgery," a single-center retrospective cohort study from University of California San Diego. This one does have real statistics, so let's spend a little more time on it.

The gap they're addressing: full-thickness skin grafts are a mainstay after Mohs when flaps aren't feasible, but most of the outcomes literature is built on smaller grafts. Larger grafts — the ones you reach for out of necessity rather than preference — are underdescribed. This group looked retrospectively at grafts twenty square centimeters or larger placed after Mohs surgery from 2010 to 2025, a fifteen-year single-center window. A retrospective design makes sense here for the obvious reason: these are relatively infrequent, size-selected reconstructions, and no one is going to randomize patients to graft size — the graft size is dictated by the defect. Their primary outcome was thirty-day graft necrosis, split into partial and total, with infection and hematoma as secondary outcomes.

Statistically, they used two tools worth unpacking. Fisher's exact test was used for the categorical comparisons — this is the test you reach for instead of a standard chi-square when your subgroup cell counts are small, which they clearly were once you stratify by graft size band and complication type. Fisher's exact calculates the exact probability of the observed table of counts arising by chance, rather than relying on the chi-square's large-sample approximation, so it's more trustworthy when some cells have single-digit counts. The second tool was logistic regression, used to generate the odds ratios you'll hear about in a second. Logistic regression models the probability of a binary outcome — necrosis yes or no — as a function of one or more predictors simultaneously, which lets you estimate, say, the effect of graft size while accounting for diabetes or anticoagulant use rather than looking at each variable in isolation. The odds ratio it spits out tells you how much the odds of the outcome multiply for a given predictor, holding the others roughly constant.

On to results. A hundred and ten cases, mean age seventy-seven, most grafts harvested from the abdomen or the supraclavicular area, with a median graft size of about twenty-seven square centimeters. Thirty-day complication rates overall: any necrosis in about one in six grafts, partial necrosis in about one in eleven, total necrosis in about one in fourteen, infection in about one in eight, and hematoma in about one in sixteen. The single most clinically important finding is a complication threshold at forty square centimeters. Grafts at or above that size had roughly an eight-fold increase in odds of any necrosis and a roughly sixteen-fold increase in odds of total necrosis compared with grafts in the twenty-to-thirty range — and looking at the raw rates makes it concrete: any necrosis went from about one in nine grafts in the smallest size band to nearly one in two grafts at forty square centimeters and above. That's not a subtle signal — that's a number you'd use to counsel a patient and change your postoperative plan.

Timing mattered too. Hematomas tended to show up around days ten to thirteen, infections around days six to thirteen, and necrosis later — partial around day fifteen, total out at three to four weeks. And critically, of the grafts that developed hematoma or infection, more than half went on to progress to necrosis, suggesting these early complications are often a warning sign rather than a separate, self-limited problem. Donor site mattered independent of size: abdominal donor grafts had a necrosis rate several-fold higher than supraclavicular grafts, even within the same size band, which the authors speculate — and flag as their interpretation — may relate to abdominal skin being thicker and more metabolically demanding to revascularize. Recipient site also stratified risk, with trunk and lower extremity recipient sites showing the highest necrosis rates and scalp and upper extremity the lowest. Diabetes independently predicted total necrosis with a roughly fourteen-fold increase in odds. Anticoagulation shifted the necrosis pattern from partial toward total, though that effect washed out when necrosis was pooled as one category. Age, interestingly, was not predictive.

Limitations are the expected ones for single-center retrospective work — no randomization of graft size or donor site, so selection effects around why a given surgeon chose abdomen versus supraclavicular, or why a certain patient got a bigger graft, can't be fully disentangled from the outcome. But the practical takeaway here edges toward practice-changing: for grafts forty square centimeters and above, especially on the trunk or lower extremity, sourced from the abdomen, or in diabetic or anticoagulated patients, this supports building in earlier postoperative follow-up — inside that first one-to-two week window — specifically to catch hematoma or infection before it progresses to necrosis, rather than waiting for a standard follow-up interval.

Third article, and this is your full original study for the episode: "Diagnostic accuracy of two-photon fluorescence microscopy in Mohs surgery of basal cell carcinomas," out of University of Rochester, MIT, and Beth Israel Deaconess. This is a diagnostic accuracy study evaluating a slide-free imaging technology against the gold standard of frozen section analysis.

The clinical problem: Mohs achieves excellent margin control, but it's rate-limited by the time and staffing cost of frozen section processing. Two-photon fluorescence microscopy is a slide-free technique that stains fresh tissue with fluorescent analogs of hematoxylin and eosin and generates virtual H-and-E images essentially in real time, using near-infrared light that also happens to penetrate blood and debris better than confocal alternatives. The specific gap the authors identify is twofold: nobody had tested how accurately this technology evaluates actual surgical margins, as opposed to small biopsy specimens, and prior slide-free studies had been confounded by comparing images taken from different depths or tilts within the tissue — meaning apparent disagreement could just reflect the two techniques literally looking at different tissue planes, not a real limitation of the technology.

Their methodological fix is the most elegant part of this paper. After routine frozen section processing was completed for clinical purposes on Mohs specimens from basal cell carcinoma cases, they cut one standard four-micron slide as the frozen section gold standard, and immediately below it, a fifty-micron-thick section for two-photon imaging — designed to keep the two modalities looking at nearly the same tissue plane. That co-registration step directly addresses the confound they identified; it's the kind of design choice that turns a previously unreliable comparison into a fair one. A hundred thirty-three specimens from seventy-eight tumors across sixty-nine patients made it through quality control, with twelve excluded for tissue plane mismatch or cryosectioning damage.

Two fellowship-trained Mohs surgeons who had never seen this imaging modality before got a brief orientation and a small training set, then read the two-photon images blinded and in random order, recording margin status. Separately, a dermatopathologist and another Mohs surgeon built a consensus frozen-section reading as the reference standard, adjudicating disagreements and using permanent sections as a tiebreaker when needed.

Two statistical concepts are worth teaching here. First, sensitivity and specificity — sensitivity is the proportion of truly positive margins the test correctly calls positive, specificity is the proportion of truly negative margins it correctly calls negative — and both are only as good as the reference standard you're comparing against, which is exactly why the co-registration protocol mattered so much. Second, the authors report a Cohen's kappa of point-seven-four between the two masked readers, a measure of interrater agreement that corrects for the agreement you'd expect by pure chance alone; a kappa in that range is generally interpreted as substantial agreement, which matters because it tells you the technology's readability doesn't depend heavily on which surgeon is looking at it.

Now the numbers. Unadjusted, pooling both readers, average sensitivity for detecting basal cell carcinoma was about ninety-four percent and specificity about eighty-two percent — with real spread between the two readers, one running quite sensitive and less specific, the other more balanced. Here's the key analytic move: the authors went back and classified every discordant read into categories, distinguishing true discordances — where the technology itself simply missed or over-called pathology that was genuinely visible on both modalities — from discordances that were artifacts of the comparison process itself, like a deeper two-photon plane picking up tumor that a shallower frozen section plane missed, or disagreement over how to classify a detached tissue fragment created during cryosectioning. Their argument, which is reasonable, is that these artifact categories wouldn't exist if two-photon imaging were used as the primary margin tool rather than compared after the fact against a separately cut slide. Excluding those artifact-driven discordances, sensitivity rose to ninety-five percent and specificity to ninety-four percent — and the authors frame this as the truer measure of the technology's intrinsic diagnostic accuracy.

The stated limitation is important and self-imposed: this evaluated agreement using specimens obtained after the surgery had already concluded, not live intraoperative decision-making, so real-world workflow integration, turnaround time, and how surgeons perform under actual time pressure remain untested. For your practice, I'd call this genuinely promising but not yet actionable. The accuracy numbers, once you account for the comparison artifacts, are in the same neighborhood as what we'd accept from frozen section itself, and the minimal-training finding is notable — but this is still an ex vivo validation study, not a clinical deployment trial. Watch for a prospective intraoperative study before this changes how you think about your frozen section lab.

Last article: "Use of ancillary studies in pediatric spitzoid neoplasms," a multicenter retrospective cohort study anchored at University of Iowa, spanning fifteen institutions. The clinical problem is a familiar headache — distinguishing benign atypical spitzoid neoplasms from spitzoid melanoma on overlapping clinicopathologic grounds, and the question here isn't about prognosis directly but about how ancillary immunohistochemistry and molecular testing are actually being used in practice, to inform optimization of that workup.

Methodologically, this is a retrospective chart review across fifteen institutions covering pediatric patients diagnosed with either diagnosis between 2010 and 2021, following STROBE reporting guidelines, which is simply a checklist standard for transparent reporting of observational studies. Statistically, they used Fisher's exact test for categorical comparisons — same rationale as before, appropriate for the frequently small subgroup counts here — and a new one worth teaching: the Kruskal-Wallis test for numeric variables, such as the number of stains ordered per lesion. Kruskal-Wallis is the nonparametric cousin of a one-way analysis of variance — instead of comparing group means, which assumes something like a normal distribution, it ranks all the observations across groups and compares the sum of ranks. That's the right tool when your data are skewed or count-based, as "number of stains ordered" clearly is, with most lesions getting zero to a handful and a few outliers getting many more.

Across four hundred atypical spitzoid neoplasms and thirty-eight spitzoid melanomas, just over half had at least one immunohistochemical stain ordered, most commonly Melan-A and Ki-67. Molecular testing was ordered in about a third of cases overall, with fluorescence in situ hybridization, or FISH, the most common modality. Here's the finding worth sitting with: no individual immunohistochemical stain — including p16, which gets a lot of attention as a discriminator — showed a statistically significant difference in positivity between the benign and malignant groups. PRAME was only used in the final two years of the study window and was negative in all nine cases tested, all of which were ultimately called benign — too few cases to draw a real conclusion, but consistent with its emerging role. Molecular testing told a different story: abnormal FISH results were strongly and significantly associated with a spitzoid melanoma diagnosis — roughly eight in ten spitzoid melanomas had abnormal FISH compared with only about one in four of the benign lesions. TERT promoter testing, done in a smaller subset, was negative across the board in the benign group and wasn't tested at all in the melanoma group in this cohort — the authors read the uniformly negative results in benign lesions as offering some reassurance about a negative TERT result, consistent with prior literature linking TERT promoter alterations to aggressive behavior. Next-generation sequencing was used in only three cases total, reflecting that this cohort predates its widespread clinical availability.

The authors are appropriately cautious in discussion: individual stains didn't discriminate statistically, but they may still have clinical value as a triage step to decide who gets sent for molecular testing rather than as standalone diagnostic arbiters. And there's an important honest caveat about the FISH finding specifically — abnormal molecular results likely influenced the final diagnosis in real time, meaning some of that strong association between abnormal FISH and a melanoma label could be circular, since the molecular result may have helped drive the diagnostic label itself rather than being an independent predictor of it. Small subgroup sizes for PRAME, next-generation sequencing, and TERT further limit how far any of these individual findings can be pushed.

For your practice, the takeaway is more conceptual than procedural: don't lean on any single immunohistochemical stain, p16 included, to resolve an equivocal pediatric spitzoid lesion — treat IHC as a screening layer that helps triage toward molecular testing rather than a stand-alone answer. An abnormal FISH result should raise your index of suspicion and prompt more aggressive workup and multidisciplinary discussion, and given current availability, next-generation sequencing should generally be favored as the first-line molecular test going forward, more so than this older cohort reflects. This isn't a study that changes your surgical margins or technique, but it's valuable background for how you counsel families and coordinate with pathology when one of these ambiguous pediatric lesions crosses your desk.

That's the September lineup — a sobering reminder to keep surveilling transplant patients indefinitely regardless of tumor count, a genuinely useful complication threshold for large full-thickness grafts that should change your follow-up scheduling, a promising but still ex vivo slide-free imaging technology to keep on your radar, and a clear-eyed look at how ancillary testing is and isn't helping in pediatric spitzoid diagnosis. Thanks for listening, and we'll see you next month.