Welcome back to the journal review. This episode covers the October twenty twenty-six issue of the Journal of the American Academy of Dermatology — four pieces this time, ranging from a short practice-reflection commentary to a couple of brief reports and a medico-legal case review. Let's get into it.

First up is a JAAD Game Changer piece, written by Robert Brodell, reflecting on the external validation of the Melanoma Institute Australia sentinel node metastasis risk prediction tool using the National Cancer Database. For those unfamiliar with the format, a Game Changer is not new primary research — it's a short retrospective narrative explaining how an article that previously ran in JAAD actually shifted practice, written by someone other than the original authors. The original study being discussed here is the Freeman et al. paper from 2023, which took the Melanoma Institute Australia nomogram — a tool that had already been developed and validated in an Australian cohort to estimate the probability of a positive sentinel node — and tested it against a large American dataset, the National Cancer Database, to see whether it held up outside its country of origin. The point Brodell is making is a conceptual one that's worth sitting with: for years, Breslow depth alone has functioned as the de facto gatekeeper for the sentinel lymph node biopsy conversation. This Game Changer's message is that relying on depth in isolation leaves predictive accuracy on the table. A multivariable nomogram — one that folds in the array of clinical and histologic variables the Melanoma Institute Australia tool uses, rather than Breslow depth by itself — produces a meaningfully better estimate of sentinel node positivity risk. Because this is a commentary rather than the primary paper, we don't have fresh effect sizes or concordance statistics to report here — the piece is making the case that the 2023 external validation already did the work, confirming the tool's accuracy generalizes to a large, heterogeneous American surgical population. The practical takeaway for you is really a reaffirmation rather than a new mandate: if you are still anchoring sentinel lymph node biopsy discussions primarily on depth, this is your reminder that a validated multivariable nomogram, now externally confirmed in a U.S. population, is the more defensible way to quantify risk and counsel patients on whether nodal staging is worthwh, and that ship sailed back in twenty twenty-three — this Game Changer is simply flagging that it was, in fact, a game changer.

Next is a brief report, and this one is genuine original data: lower artificial intelligence and clinician-derived dermoscopy scores in melanomas from CDKN2A pathogenic variant carriers compared to sporadic cases, out of Leiden in the Netherlands. The clinical observation motivating this is one a lot of you managing familial melanoma or FAMMM syndrome patients will recognize instinctively — melanomas arising in CDKN2A pathogenic variant carriers often just look less dramatic dermoscopically than their sporadic counterparts, which is a problem precisely because dermoscopy, now increasingly paired with AI-based scoring tools, is the backbone of surveillance in this high-risk population. The gap they're addressing is whether that subjective impression holds up when you actually run the lesions through both an AI malignancy classifier and standardized clinician-based dermoscopy algorithms.

Methodologically, they took one hundred eight pT1a melanomas from CDKN2A carriers diagnosed between twenty twenty and twenty twenty-four and compared them against a randomly selected set of one hundred nine sporadic pT1a melanomas — deliberately restricting to thin, stage-matched tumors so that stage itself wouldn't confound the comparison. Diagnoses came from expert dermatopathologists with ambiguous lesions excluded. Each lesion then got run through a previously validated open-source AI model generating a zero-to-one malignancy score, and separately scored by a blinded dermatologist using the revised seven-point checklist and the chaos-and-clues method, with a subset re-scored by three additional dermatologists to confirm inter-rater reliability, which came back good to excellent depending on the metric.

Here's the statistics teaching moment, because the analytic choice here is genuinely instructive. Because many patients contributed more than one melanoma — one hundred eight lesions came from only eighty patients — a standard linear regression or simple group comparison would treat each melanoma as an independent data point, when in fact lesions from the same patient are correlated with each other through shared genetics, skin phenotype, and surveillance behavior. Treating them as independent would artificially shrink your standard errors and inflate your confidence in the result. The authors instead used linear mixed-effects models with patient-level clustering, which add a random effect for each patient — essentially letting the model acknowledge that observations from the same person share some baseline tendency, and partitioning variance accordingly so the between-group comparison isn't fooled by that within-patient correlation. This is the same logic you'd apply with repeated measurements in any clustered design — multiple margins from the same Mohs case, multiple lesions from the same patient, serial time points from the same subject. Separately, to assess how well the AI score and the clinician scores agreed with each other, they used Spearman correlation with cluster bootstrap confidence intervals — Spearman because it's rank-based and doesn't assume the underlying scores are normally distributed or linearly related, and the bootstrap approach again respects that clustering structure rather than assuming independence.

The results were consistent and significant across every measure. CDKN2A carriers were younger, with a median age of fifty-four versus sixty-four in the sporadic group, though Breslow thickness, skin type, histologic subtype, and anatomic location were all comparable between the two cohorts — which matters, because it argues against the simplest explanation that these are just earlier, thinner tumors caught sooner. Despite that matching, the mean AI malignancy score was substantially lower in the carrier group, zero point three-two versus zero point five-two in the sporadic group, a statistically significant and clinically meaningful gap. The clinician-based scores told the same story: the seven-point checklist averaged two point seven in carriers versus three point nine in sporadic cases, the chaos score was lower, and the clues count was lower too, all statistically significant. AI and clinician scores correlated only moderately with each other, in the range of zero point five to zero point six, meaning they're capturing overlapping but not identical information. Perhaps the most clinically sobering number here: thirteen percent of carrier melanomas scored a flat zero on the seven-point checklist, compared to three percent of sporadic melanomas, and twenty-one percent showed no chaos at all on dermoscopy, versus nine percent in the sporadic group. That is, roughly one in five to one in eight genetically confirmed melanomas in this high-risk population would look reassuringly boring by standard feature-based criteria.

The authors are appropriately cautious in their discussion. Part of this could reflect surveillance-related enrichment — these are patients seen frequently enough that inconspicuous lesions get caught before they develop classic features — though they note many sporadic patients were also under regular follow-up, and a meaningful fraction of carrier melanomas actually prompted unscheduled visits, which pushes back against pure surveillance-interval artifact. The matched Breslow thickness and skin type across groups further suggests this is a real phenotypic signal, not simply a staging artifact. The limitation that matters most for your practice is the one they state directly: feature-based algorithms, whether human or AI, risk providing false reassurance in this population, and the AI model used here wasn't specifically trained on CDKN2A-associated melanomas, so its threshold may simply be miscalibrated for this group.

Practically, this is not yet a change-your-algorithm moment — the authors themselves suggest the fix is either lowering AI malignancy thresholds for known high-risk hereditary patients or enriching training datasets with CDKN2A-associated lesions, neither of which exists yet in deployed tools. What is actionable today is the mindset: in CDKN2A carriers under surveillance, a low AI score or a reassuringly low feature-count on dermoscopy should not be weighted the same way it would be in a sporadic lesion. Clinical context — lesion evolution, patient history, genotype — needs to keep driving your biopsy threshold in this population, independent of what the score says.

Third is a retrospective review, not an experimental study, and it's one worth your full attention because it's about you specifically: medico-legal claims involving dermatologic biopsy procedures, a review of United States and Canadian court cases from nineteen sixty-nine through twenty twenty-five. The clinical problem is straightforward — skin biopsies are ubiquitous in dermatologic practice and a recognized driver of malpractice litigation, yet the actual patterns behind those claims are not well characterized. The authors searched the LexisNexis legal database for civil cases involving a board-certified dermatologist in which a skin biopsy was a central component of the allegation. This is a sensible design choice for this question — malpractice litigation isn't captured in any clinical registry, so a legal-database search of published court decisions is really the only way to systematically characterize case patterns, even though, as the authors acknowledge, it misses the large volume of claims resolved through confidential settlement before ever reaching a published opinion.

They identified thirty-nine cases spanning sixteen U.S. states and two Canadian provinces, with New York, California, and Louisiana the most frequent jurisdictions. The overwhelming majority, about nine in ten, arose from outpatient private dermatology clinics rather than hospital or institutional settings, and plaintiffs were nearly evenly split between men and women. A key structural finding: fewer than half of the cases, about forty-four percent, actually involved a biopsy that was performed — meaning the majority of claims centered on the dermatologist's failure to biopsy a lesion in the first place, which was in fact the single most common allegation overall, accounting for roughly thirty-seven percent of all allegations. Where biopsies were performed, excisional biopsy was the most common type, and head and neck was by far the most frequent site, consistent with where concerning lesions tend to cluster and where consequences of missed diagnosis tend to be most visible.

Beyond failure to biopsy, the next largest allegation category was delayed diagnosis after a biopsy was performed, at roughly thirty percent, followed by misdiagnosis due to histopathologic misinterpretation, at close to one in five cases. Procedural allegations like improper technique, unnecessary biopsy, inadequate supervision of ancillary staff, and failure to obtain informed consent were each individually uncommon but collectively notable. Of the thirty-one cases that reached a final verdict, about two-thirds favored the defendant — including fifteen that were dismissed before trial — while roughly sixteen percent favored the plaintiff and the remainder were remanded. Interestingly, among the smaller set of cases without a final verdict, the pattern flipped, with most favoring the plaintiff, though the authors don't make much of that given the small numbers. Looking specifically at the five plaintiff-favored final verdicts, a clear theme emerges: every one of them involved either delayed biopsy performance, failure to communicate results back to the patient, or misinterpretation of histopathology — not exotic technical errors, but breakdowns in timeliness and communication. The case examples are illustrative rather than statistical: a persistent lip lesion treated topically for over a year before biopsy eventually revealed squamous cell carcinoma requiring multiple reconstructive surgeries, and a delayed biopsy and referral for a suspicious acral pigmented lesion progressed to metastatic melanoma. One case resulted in license suspension tied specifically to substandard histopathologic interpretation.

The authors are candid about limitations — this is a small sample accumulated across more than five decades, so they couldn't identify any clear temporal trend, meaning these failure patterns, delay in biopsy and diagnostic error, have simply persisted unchanged over time rather than improving or worsening. Publicly available court records miss confidential settlements entirely, likely the majority of real claims, and the small Canadian sample limits cross-border comparison given differing tort systems. There's no inferential statistics to dissect here — this is a descriptive accounting, and it should be read as such. For your practice, the takeaway is concrete and genuinely actionable, even if it's not a change to your surgical technique: the liability exposure clusters overwhelmingly around delay and communication rather than technical biopsy performance itself. Robust systems for tracking pending pathology results, ensuring patients are actually notified of findings, timely re-biopsy when margins or clinical suspicion remain unresolved, and clear documentation of informed consent for biopsy-related scarring — these are the levers that actually show up in the plaintiff-favored cases, more than refining your punch or shave technique ever will.

Last is an ecologic study: sex-based disparities in melanoma outcomes across U.S. counties, drawing on the Surveillance, Epidemiology, and End Results program and the National Program of Cancer Registries. The background here is well established at the individual level — men have consistently higher melanoma incidence and mortality than women — but what hadn't been quantified is whether that disparity is uniform geographically or concentrated in particular places, which matters for where you'd target outreach and early-detection resources.

The design is explicitly ecologic, meaning the unit of analysis is the county, not the individual patient — they pulled five-year averaged incidence and mortality data from two hundred eight urban counties covering over one hundred eighty million people, stratified by sex, and calculated a mortality-to-incidence ratio, or MIR, for men and women in each county, then took the ratio of those two ratios as their headline disparity metric. It's worth pausing on why MIR is a useful construct here rather than simply comparing mortality rates directly: mortality rate alone conflates how often a cancer occurs with how often it kills, while MIR normalizes mortality against incidence, functioning as a rough population-level proxy for case fatality. If men simply got diagnosed with melanoma more often but died at the same proportional rate once diagnosed, you'd expect similar MIRs between sexes — the fact that the MIR itself differs by sex tells you the disparity isn't just about incidence, it's about something worse happening between diagnosis and death in men specifically. This is also a good moment to flag the core vulnerability of any ecologic design, which the authors appropriately discuss in their limitations: associations found at the county level don't necessarily hold at the individual level, a principle sometimes called the ecological fallacy. A county with a high male-to-female MIR might reflect genuine sex-specific biological or care-access differences, or it might simply reflect that county's particular demographic composition, without telling you anything definitive about any individual man's risk.

On to the numbers. Mean melanoma incidence was about thirty-one per one hundred thousand in men versus about twenty per one hundred thousand in women — already a substantial gap. But the MIR gap tells the more interesting story: the mean MIR was zero point one-one in men versus zero point zero-eight in women, both differences statistically significant, and the average male-to-female MIR ratio across counties was one point four-three — meaning that, relative to how often melanoma occurs, men are dying from it at roughly one and a half times the rate of women. Some counties showed far starker gaps, with several exceeding a ratio of two, meaning men in those counties were more than twice as likely to die relative to incidence. The counties with the single highest male MIRs were Spokane County, Washington, and Jackson County, Oregon, while the largest male-to-female disparities clustered in Multnomah County, Oregon, Denton County, Texas, and Placer County, California — notably concentrated in the western United States, which the authors speculate could reflect regional sun exposure patterns, though they're careful to frame this as speculative rather than confirmed.

The discussion ties this back to known individual-level biology — differences in tumor microenvironment and immune response by sex have been described elsewhere — while also raising access-to-care and regional detection practices as plausible contributors to the geographic clustering. The limitations are substantial and explicitly acknowledged: this is aggregated county data with no ability to adjust for age, race, or socioeconomic status, counties with fewer than three cases for a given variable were censored entirely, there's no individual-level staging or treatment data, and detection bias from variable regional reporting practices can't be excluded. So in terms of practice impact, this is hypothesis-generating population health data, not something that changes how you manage an individual patient in clinic. Where it may be genuinely useful is at the level of program design — if you're involved in community screening initiatives or outreach efforts, particularly in the western U.S. counties flagged here, this gives you a data-driven argument for why male-targeted early-detection messaging deserves specific emphasis rather than assuming general awareness campaigns close the gap equally for both sexes.

That wraps this episode. To summarize the throughline: reinforcement that multivariable risk tools outperform Breslow depth alone for sentinel node counseling, a genuine caution flag about trusting AI and dermoscopy scores at face value in CDKN2A carriers, a sobering but actionable reminder that biopsy-related liability is driven by delay and communication failures far more than technique, and population-level evidence that melanoma mortality disparities between men and women are real, significant, and geographically uneven. Thanks for listening, and I'll see you next issue.