Welcome back to the journal review. This is JAMA Dermatology, October 2026, and we've got two pieces worth sitting with today — an editor's note that uses a single study to open up a much bigger question about artificial intelligence and overdiagnosis, and a striking case report on a rare cutaneous manifestation of end-stage melanoma. Let's get into it.

Our first piece is an editor's note — a commentary written by the journal's associate editor, Veronica Rotemberg, reacting to and contextualizing a separate original investigation in the same issue by Ronicke and colleagues. So what we're covering here is really the editorial lens on that study, not the full primary methods paper itself, and I'll flag where the detail comes from the editorial's own summary rather than a complete methods section.

The underlying study is a prospective, single-center investigation asking a genuinely novel question: can AI-assisted line-field confocal optical coherence tomography — think of it as a high-resolution, non-invasive optical imaging modality already validated for basal cell carcinoma margins and subtype — be turned into a screening tool to find basal cell carcinomas that have no visible clinical correlate at all? Not clinically atypical macules worked up with imaging, but completely normal-appearing facial skin in high-risk patients, interrogated by AI as a sort of spatial navigation tool to flag regions of interest, which a clinician then reviews for BCC-specific optical features before deciding whether to biopsy.

Why this design makes sense, at least as the editorial frames it: this is fundamentally a feasibility and detection question, not a treatment-outcome question, so a single-center prospective screening cohort is a reasonable first step — you're asking "can we find anything at all," not yet "does finding it change outcomes." The human-AI collaboration structure is deliberate too — the AI isn't making an autonomous call, it's cueing a clinician who still applies pattern recognition, which is the realistic way most of these tools will actually enter practice.

The numbers: among 150 high-risk patients screened, subclinical lesions were flagged in 14 of them — call it about one in eleven, or roughly 9%. That yielded 17 total candidate lesions. Of those, 18 lesions overall were deemed suspicious enough by the human-AI combination to warrant biopsy, and 15 came back histologically as basal cell carcinoma. That's a positive predictive value of about 83%, with a confidence interval roughly spanning the high 50s to the mid-90s — so a wide interval, which matters given how few lesions this is based on.

Here's where the editorial does real teaching work, and it's worth slowing down on. A positive predictive value isn't a fixed property of a test — it's entangled with pretest probability, meaning the prevalence of disease in the population you're applying the test to. This was a high-risk population by design, so the baseline odds of any flagged lesion actually being BCC were already elevated before the AI did anything. Pull this same tool into a general screening population with a much lower baseline prevalence, and that 83% PPV would almost certainly fall, even if the test itself performed identically. Separately, there's a more fundamental problem here: because only AI-and-clinician-flagged lesions got biopsied, there's no denominator of imaged-but-unflagged skin to tell us what was missed. That means you simply cannot calculate sensitivity or specificity for the AI algorithm from this design — this is verification bias, and it means PPV is essentially the only metric available to us here, and even that one is a function of a selected, enriched sample rather than something generalizable.

The deeper point the editorial raises, and the real reason this is worth your attention, is conceptual rather than statistical: these subclinical, invisible BCCs are a previously uncharacterized category of disease. We don't know their natural history. Some may progress, some may sit indolently, some may even regress — and remember, even for clinically visible BCC, watchful waiting is already an accepted strategy in appropriate patients. The editorial notes the predominance of superficial subtype among these subclinical lesions, which hints that nonsurgical management could be reasonable for at least some of them — but there is currently no framework at all for deciding which subclinical lesions, if any, warrant treatment. So detection capability is outrunning clinical decision-making, and that's the overdiagnosis risk baked into the title of this piece.

Practical takeaway: this is interesting, not practice-changing. There is no indication here to start screening normal-appearing facial skin with LC-OCT in your high-risk patients, and even if you had access to this exact technology, a positive subclinical finding wouldn't currently come with a validated management pathway. What this really calls for, per the editorial's own close, is natural history studies of these subclinical lesions and defined treatment thresholds before this moves from a research curiosity into anything resembling a screening tool you'd deploy.

Our second piece is a case report — an Images in Dermatology submission describing diffuse melanosis cutis and melanuria in a patient with advanced BRAF V600E–mutant melanoma. No methods or statistics here; this is a single patient's clinical course plus a synthesis of what's known about a rare phenomenon, so we'll walk it the way the authors did.

The patient was a man in his fifties with a decade-long melanoma history — two primary BRAF V600E–mutant cutaneous melanomas, both deep, Breslow thickness around 5 millimeters, Clark level four and five. He went through the expected trajectory for a case like this: sentinel node mapping that didn't correlate intraoperatively, nodal and cutaneous recurrences, several lines of systemic therapy including a BRAF-MEK-checkpoint combination trial, and eventually further nodal and then visceral progression to liver, spleen, and adrenal glands. Lactate dehydrogenase and S100 were markedly elevated, consistent with a high tumor burden. At that point, he developed a progressive slate-gray discoloration of the skin, most pronounced on the face, along with darkening urine. A biopsy of the discolored skin showed epidermal pigmentation with a normal melanocyte count and dermal melanophages — not melanoma infiltrating the skin itself, but pigment deposition in the dermis — which is the histologic signature that, in this clinical context, defines diffuse melanosis cutis. Despite further targeted and immune therapy, he progressed, developed darkening of his sclerae as well, and died ten months after the onset of the skin discoloration.

The teaching points here are the real payload of the article. Diffuse melanosis cutis is exceedingly rare but highly specific for advanced, usually end-stage, melanoma. A systematic review of roughly seventy cases cited in the discussion found a mean interval of about eleven months from original melanoma diagnosis to onset of melanosis — though obviously that's hugely variable depending on disease course, as this patient's own decade-long timeline shows. Once it appears, prognosis is grim: mean survival from onset is only four to six months, clearly worse than the ten-year overall survival of 40 to 50% quoted for stage four melanoma in general. The rare outlier survivors beyond two years tended to be patients on targeted therapy, which is a useful detail if you're counseling a patient and family on what this finding means. Clinically, the clues are slate-gray discoloration favoring sun-exposed sites, a cephalocaudal pattern of spread, concurrent melanuria — which the authors note occurs in roughly half of reported cases, reflecting renal excretion of melanin precursors, predominantly pheomelanin — and that histologic combination of normal melanocyte density with dermal melanophages. The differential is worth keeping in your back pocket: drug-induced hyperpigmentation from agents like minocycline, amiodarone, antimalarials, clofazimine, or busulfan-based chemotherapy, plus ochronosis, argyria, Addison disease, Wilson disease, and hemochromatosis. Pathogenesis remains incompletely understood, but the leading hypothesis is that lysis of melanoma cells releases melanin precursors systemically, which are then synthesized into melanin in the dermis and taken up by native histiocytes — with an alternative theory implicating tumor-derived growth factors stimulating melanocyte or dermal micrometastatic melanin production.

For practice, this isn't a management-changing paper in the sense of giving you a new intervention — there isn't one here. The value is purely diagnostic pattern recognition: if you see a melanoma patient developing new slate-gray discoloration with dark urine, this entity should come to mind immediately, both because it will spare the patient an unnecessary workup chasing other pigmentary causes, and because, unfortunately, it should prompt a frank conversation about prognosis, since this finding tracks with a very short survival window.

That's it for this episode. One editorial pushing back on how we interpret AI-driven early detection of subclinical BCC, and one sobering case reminding us what diffuse melanosis cutis signals in advanced melanoma. Thanks for listening, and I'll see you next time.