Four Percent: Who Actually Answers a Patient's Skin Question

Roughly seven in ten people living with a chronic skin disease never see a clinician about it. That single estimate, which a new viewpoint in JMIR Dermatology borrows from the access literature, frames the whole problem. Where there is no appointment, no coverage, or no dermatologist within reach, the gap does not stay empty. It fills with whatever is closest to hand, and for most people what is closest is a phone.

On that phone the supply of dermatological advice is large and almost entirely unaccredited. The paper cites a figure that four percent of dermatology influencers on Instagram are board-certified dermatologists; on TikTok, more than a third of skin-disease videos come from people with no professional training. For a patient with no other route, that feed is not background noise. It is the first medical opinion they receive — and four percent is a fair description of its evidentiary quality.

What kind of paper this is

The piece is the work of sixteen authors, among them contributors at the World Health Organization, Massachusetts General Hospital, the Medical University of Vienna and UCLA, and it grew out of a professional summit rather than a dataset. A viewpoint is an argued opinion, not a study: it ran no cohort, measured no endpoint, and generated no data of its own. Its competing-interests statement reads "none declared." The fair test for a paper like this is not whether it proves anything — it cannot — but whether it reads the existing evidence honestly. On that test it largely holds up, provided the reader keeps the same discipline the authors mostly do.

Two numbers that need labels

Most of the paper's optimism rests on two borrowed findings, and both should be read for what they are. The first is a 2024 review counting more than 900 dermatology apps, 41 of them using AI. The authors invoke this not as a working market but as a warning: many of these tools were never tested systematically across skin tones, and training data that over-represent lighter skin make diagnostic accuracy systematically worse on darker skin. For software meant to catch skin cancer early, that is not cosmetic. It is later diagnoses and worse outcomes, concentrated in populations already underserved. Their suggestion to move from the older Fitzpatrick phototypes to the Monk Skin Tone scale is a sensible response to a documented bias, not a fix for it.

The second is a WHO pilot in Kenya. In 2024 an AI-assisted mobile app was deployed across five counties to help frontline workers recognise thirteen neglected tropical diseases and twenty-four common skin conditions; forty Ministry of Health workers captured 605 patient images, which dermatologists then reviewed. Preliminary findings, the paper reports, show better than eighty percent agreement between the app's calls and that specialist review, with the tool framed for triage rather than diagnosis. The authors are explicit that peer-reviewed accuracy data are still pending — and the careful reader should add two caveats they leave implicit. Agreement with a reviewer is not the same as being correct against a confirmed diagnosis, and an agreement rate says nothing on its own about how a tool performs at the prevalence of the condition that actually matters. This is a signal that something might work in a setting with no specialist on the ground. It is not yet evidence that it does.

A third proposal is honest about its own status. The authors sketch a chatbot built on retrieval-augmented generation, fine-tuned on dermatology content and drawing only on curated sources — Cochrane reviews, the guidelines of the American and European dermatology academies, patient-association material, WHO and CDC repositories — with evidence ranked internally by the GRADE framework. They state plainly that it remains conceptual, with no trials behind it. As an illustration of a principle it is useful: an AI system that works as a quality-controlled route to good information rather than a substitute for clinical judgement. As something that exists and works, it does not yet.

The argument worth keeping

The paper's most durable contribution is a stance, which it labels "radical dermatology": the claim that clinicians and their professional bodies should lead the digital transformation of the field rather than watch it happen to them. The label is chosen for effect; the content is unremarkable and sound. In practice it means specialists publishing quality-controlled material in public channels as part of the job, societies treating digital communication as core work, institutions funding the validation of AI tools, and regulators setting and enforcing a quality floor for anything sold as a dermatology app. The contrast it draws comes down to being in the room first or arriving last.

For European systems the structural picture is familiar: expertise pooled in a few centres, long waits, geography deciding who reaches a specialist. A viewpoint solves none of it. But it has the order of operations right. The unaccredited accounts are already defining what good dermatological information looks like for most people who go looking, and four percent measures how far that has gone. Crucially, the case for clinicians helping to set the standard does not depend on any tool in this paper turning out to work. It depends only on the vacuum being real — and that much is not in question.

Source: du Crest D, Madhumita M, Enbiale W, et al. AI and Digital Tools in Dermatology: Addressing Access and Misinformation. JMIR Dermatology 2026;9:e79044 (PMID 41719483). A viewpoint article with no primary data of its own; the figures it cites — including the WHO Kenya pilot — are drawn from secondary and in part non-peer-reviewed sources, and should be weighed accordingly.