Welcome to this June twenty twenty-six review from the Journal of the American Academy of Dermatology. We've got four items on the docket this time — three letters to the editor engaging critically with recent papers, and one brief original report modeling the economics of melanoma surveillance. Let's get into it.

First up is a letter to the editor responding to the Choi et al study on anatomic site and prognosis in early-stage acral lentiginous melanoma, a two hundred eighty-one patient retrospective cohort that you may already be familiar with. The original paper described a metastatic dichotomy — upper-limb acral lentiginous melanoma tending toward distant, systemic spread, while lower-limb disease tended toward lymphatic routes — and on multivariate Cox regression, upper-limb location came out as an independent predictor of worse systemic metastasis-free survival, with a hazard ratio north of four and a half. That's a big number, and the letter writers, a group from Zhejiang and Hangzhou, push back on how we should interpret it. Their argument, essentially, is that anatomic location itself has no plausible direct biological mechanism for driving metastatic behavior — it's almost certainly a proxy for something else, namely the distinct mutational landscape that varies by acral subsite. They marshal some genomic literature to make the point: NRAS mutations are roughly ten times more common in sole melanomas than in palm melanomas, subungual melanomas — which make up a good chunk of the upper-limb cohort — are frequently driven by GNAQ and KIT alterations rather than the BRAF mutations more typical of other acral sites, and KIT-mutant melanomas have separately been linked to more aggressive features like ulceration and greater Breslow depth. Put together, their contention is that the striking hazard ratio for upper-limb location is probably capturing a more aggressive underlying molecular subtype rather than any intrinsic property of hand anatomy. This is a commentary, not a new dataset, so there's no methods or results section to walk through — the entire contribution is this conceptual critique, and their ask is straightforward: future prospective, multi-institutional acral melanoma studies should incorporate molecular profiling alongside anatomic and clinical variables, so we can figure out whether we're risk-stratifying by geography or by genotype. It's a good reminder for anyone reading site-based prognostic data in acral melanoma to hold the anatomic variable loosely until the molecular correlates are actually measured.

Second, a brief report — this one an actual original analysis rather than a letter — estimating the cost of melanoma surveillance under the twenty twenty-five National Comprehensive Cancer Network guidelines. The background here is simple: NCCN gives stage-based surveillance recommendations for melanoma survivors, but nobody had really priced out what adherence to those recommendations costs, especially given how wide the guideline-permitted ranges are for visit frequency and imaging. The authors built a cost model using twenty twenty-five Medicare hospital outpatient reimbursement rates, stratifying by stage and modeling two scenarios for advanced disease — a base scenario of routine clinical follow-up and biopsies, and a maximal scenario that layers in the full complement of NCCN-permissible imaging, meaning cross-sectional brain MRI and regional nodal ultrasound at the guideline's upper frequency. Methodologically, this is a modeling exercise built on published utilization assumptions — biopsy rates per visit, the proportion of encounters with multiple biopsies, and so on — rather than a chart review of actual billed encounters, and that's a reasonable choice here because you can't observe "guideline-concordant cost" directly in claims data when practice patterns already vary; you have to construct the boundaries of the guideline yourself and see what falls out. The findings: five-year base surveillance costs run from roughly nine hundred to about seventeen hundred dollars for early-stage, local melanoma, and from about fourteen hundred to thirty-seven hundred dollars for advanced disease under base follow-up. But if you layer in every NCCN-permissible imaging study for advanced melanoma, the five-year cost balloons to somewhere between about eleven and a half thousand and thirty-four thousand dollars — a genuinely enormous range, and one that's clinically meaningful because it tells you the guidelines themselves are loose enough to produce wildly different resource utilization depending on how aggressively a clinician chooses to follow a patient. Average per-visit costs came out around one hundred seventy-five to one hundred eighty dollars, which lines up reasonably well with a prior published estimate, so the model seems to be in the right ballpark. Scaled up to national melanoma incidence, total annual US surveillance spending was estimated at over one hundred twenty-five million dollars. The authors are appropriately honest that this is likely an underestimate — Medicare rates run below private insurance, many patients continue surveillance past the five-year window the model captures, and their biopsy-frequency assumptions were drawn from patients at varying times since diagnosis rather than specifically in the highest-risk early period. They also flag the tension at the heart of this: the risk that matters most, a second primary melanoma, occurs in only about four in a hundred survivors over five years, so the cost of maximal imaging-based surveillance has to be weighed against a fairly modest absolute yield, particularly outside the true high-risk imaging indications. For practice, I'd call this interesting and useful for counseling and health-policy conversations rather than immediately practice-changing at the individual patient level — it doesn't tell you to image less, but it does give you real numbers to have the conversation about intensity of follow-up, especially for stage IIB through IV patients where the maximal-surveillance price tag is substantial, and it reinforces that judicious, risk-stratified use of dermoscopy and clinical follow-up rather than reflexive imaging is where the value is.

Third, a letter to the editor responding to Lauck and colleagues' active-comparator retrospective cohort study on cutaneous malignancy risk after biologic therapy for inflammatory disease. This letter, from a group in Karachi, raises four methodologic concerns rather than presenting new data. First, the original study didn't distinguish prevalent from incident skin cancers — meaning cancers already present but undiagnosed before biologic initiation could get misattributed to the drug, inflating the apparent association, and the letter suggests baseline skin cancer screening before starting biologics would help clarify true incident risk. Second, the study captured race and ethnicity but not Fitzpatrick skin type, which the letter argues is the more mechanistically relevant variable for skin cancer risk, since photosensitivity varies substantially within any given racial group. Third, the original analysis grouped drugs by class — tumor necrosis factor inhibitors, interleukin inhibitors, Janus kinase inhibitors — without breaking out individual agents, even though the underlying database, TriNetX, apparently permits prescription-level granularity; the letter's point is that agents within a class, say infliximab versus adalimumab versus etanercept, differ enough in structure and immunologic effect that lumping them may obscure real differences in skin cancer risk. Fourth, and probably the most important confounding concern, is that the study doesn't account for underlying disease severity or duration — patients severe enough to need biologics also carry more chronic inflammatory burden, which itself has been linked to DNA damage and impaired tumor surveillance, so it's genuinely difficult to separate a drug effect from a disease-severity effect without better activity or duration measures. None of this is new data, just methodologic critique, and the practical message for your own reading of that literature is to treat biologic-class-level skin cancer associations as hypothesis-generating rather than causally settled until studies incorporate incident-only outcomes, phototype, agent-level granularity, and disease-severity adjustment.

Last, a response letter concerning the Leibovit-Reiben et al paper describing a gene-expression panel meant to predict local recurrence, metastasis, and overall survival in intermediate- to high-risk cutaneous squamous cell carcinoma. This letter, from a single author at Henry Ford, is a fairly pointed statistical critique and worth sitting with because gene-expression profiling is very much in active clinical use for cSCC risk stratification. The core issue raised is that the discovery cohort was heavily enriched for poor outcomes — dominated by T2a and T2b tumors with roughly one in three developing metastasis, well above the ten to twenty percent metastatic rate typically seen even in high-risk T2b disease in the broader literature. Because positive and negative predictive values are mathematically dependent on how common the outcome is in the tested population, any predictive value figures reported in this enriched cohort simply won't transfer to a general cSCC population, and shouldn't be benchmarked against staging systems or commercial assays validated in more representative cohorts. The second and, I'd say, more fundamental problem is that the fourteen-gene metastasis score, the three-gene local recurrence score, and the eight-gene overall survival score were all derived and tested on the same dataset, with no independent validation cohort — internal cross-validation checks stability but can't correct for bias introduced during gene selection and threshold-setting, so describing the panel as accurately predicting outcomes is, per the letter, likely reflecting model overfitting to this particular cohort rather than generalizable tumor biology. That concern is sharpened for the overall survival score specifically, since it's built on all-cause mortality rather than squamous-cell-specific death, meaning it could easily be capturing unrelated comorbidity patterns in this particular sample. And there's a biological coherence problem too: since metastasis is the dominant driver of cSCC mortality, you'd expect real overlap between the metastasis signature and the survival signature, but the two panels share only a single gene, ZNF750, and it's regulated in opposite directions between the two models — down in the metastasis panel, up in the worse-survival panel — which is hard to explain if both are tracking the same underlying biology rather than statistical noise specific to this training set. The letter's conclusion, and I think it's the right one to carry forward, is that this remains valuable discovery-stage work generating candidate markers, but that clinical validity claims need to wait for a separate, independently assembled cohort selected on tumor characteristics rather than outcomes, with real-world event prevalence and standardized follow-up, tested in a blinded fashion. This is squarely not practice-changing yet — if anything, it's a caution against over-interpreting internally-validated gene-expression signatures in this space until external validation catches up.

That wraps our four pieces for June. Three critical letters reminding us to look past headline statistics — hazard ratios, predictive values, class-level associations — to the molecular, methodological, and confounding structure underneath them, and one cost model putting real numbers on just how expensive guideline-adherent melanoma surveillance can become. Thanks for listening, and I'll see you next issue.