Welcome to this January twenty twenty-six review from the Journal of the American Academy of Dermatology. We've got three pieces on deck this time — two ethics journal club columns and one continuing medical education review on artificial intelligence. Let's get into it.

First up is an ethics journal club piece, structured as an advice column — a reader question answered by "Dr Dermatoethicist" — addressing an ethical framework for managing neglected skin tumors in older patients. You know this patient population well: the large, fungating basal cell or squamous cell carcinoma that's been sitting untreated for months or years before anyone gets a look at it. This isn't an original study, so there's no methods or results section to walk through — it's a structured ethical reasoning exercise, and it's genuinely useful because it names the drivers behind these presentations explicitly. The authors point to distrust of the health care system, cognitive impairment, psychiatric illness, and financial or insurance barriers as the recurring themes, compounded by the fact that these tumors are frequently slow-growing, which paradoxically lowers the perceived urgency for the patient until there's rapid growth, pain, or bleeding — often the trigger that finally brings a family member into the picture. The piece walks through the classic four-principle framework. On autonomy, the key clinical task is a real assessment of decision-making capacity, and importantly, the authors note that even when capacity is impaired, treating over objection is rarely appropriate unless there's genuinely imminent harm, like a limb-threatening infection — so lack of capacity doesn't just default you to aggressive intervention. On nonmaleficence, they cite a sobering number worth remembering: a retrospective cohort of frail patients undergoing skin cancer excision showed a six-month mortality rate of about one in five, and critically, that mortality wasn't attributed to the cancer or the procedure itself but to underlying comorbidity and frailty — a reminder that the excision itself is rarely what kills these patients, but the physiologic stress around it, especially when general anesthesia is involved, absolutely can matter. This is where the case is made for clinic-based, local-anesthesia approaches — Mohs surgery being the obvious example — or alternatives like palliative radiation, systemic therapy, or, in appropriate end-of-life contexts, no cancer-directed treatment at all. They also flag something worth sitting with: even very elderly patients may have a decade or more of remaining life expectancy, so timely intervention still matters and shouldn't be reflexively deprioritized just because of chronological age. On justice, the authors push back against ageist or ableist bias that inappropriately de-escalates otherwise indicated care, and on beneficence, they emphasize that the job isn't just excising the tumor — it's addressing the psychosocial scaffolding around the patient with social work, transitional care, home health, and telemedicine or home-based visits for those with mobility barriers. The practical takeaway here isn't a change in surgical technique — it's a workflow prompt: build capacity assessment, goals-of-care conversation, and social work referral into your standard approach to the neglected-tumor patient, and resist the urge to either over-treat reflexively or under-treat out of therapeutic nihilism based on age alone.

Second article, same ethics journal club format, this time on treating actinic keratosis in patients of advanced age — framed around a specific, very familiar scenario: a nonagenarian with a history of skin cancer, brought in annually by his daughter, getting cryotherapy for AKs that clearly causes him visible discomfort. Again, no methods or data collection here — this is applied ethical reasoning, but it's grounded in some genuinely useful numbers you could use in a patient conversation. On beneficence, the authors reiterate the basic rationale for treating AKs — preventing progression to squamous cell carcinoma, which can locally invade or, less commonly, metastasize. But the nonmaleficence discussion is where it gets interesting: they explicitly state that in elderly patients with significant comorbidity, the potential harm from treatment may outweigh the actually quite low risk of any individual AK progressing to invasive SCC. They also raise an financial-incentive point worth sitting with — cryotherapy is fast, reimbursable, and easy to justify in a busy clinic day, and the authors are blunt that decision-making should be driven by patient benefit, not by what's operationally convenient or well-reimbursed. On justice, they cite the scale of this as a systems issue: AK treatment cost the U.S. system on the order of one point eight billion dollars back in 2013, and Medicare reimbursement for cryotherapy alone runs into the thousands of dollars per thousand beneficiaries — real resource allocation questions, plus the very concrete point that a fragile elderly patient's visits often mean an adult child taking time off and providing transport, which is its own burden worth acknowledging. Their practical suggestion is to consider stretching total body skin exam intervals to every two to three years in patients without a personal or family history of skin cancer and no concerning comorbidities — a concrete, actionable de-intensification recommendation. And on autonomy, the message is simple: ask the patient how bothered they actually are by the AKs and how aggressively they want them treated, because some patients want everything treated for cosmetic or risk-aversion reasons, while others with extensive actinic damage would rather not be touched at all given more pressing health priorities. Their bottom line, which is worth carrying into your own practice, is that in a patient in his 90s, AKs are unlikely to metastasize or contribute to mortality, so the treatment decision should really be anchored in that individual's stated goals rather than a reflexive treat-everything protocol.

The third piece is a substantial departure from the first two — this is part one of a two-part continuing medical education review on augmented intelligence in dermatology, essentially a primer meant to build AI literacy rather than report new data, so there's no background-gap-methods-results structure to walk through — instead it's a vocabulary and concepts tour, and given how much AI terminology gets thrown around loosely in our literature now, it's worth having the definitions pinned down precisely. The authors start by distinguishing artificial intelligence, which describes autonomous systems, from augmented intelligence, which they define as the collaborative model where technology enhances human performance rather than replacing it — a distinction that matters for how we think about AI's role in our own diagnostic workflow, whether that's melanoma triage or margin assessment. They then work through the machine learning hierarchy: machine learning as the broader discipline of building algorithms that improve with experience, computer vision as the specialized subset handling image interpretation, and neural networks as the underlying architecture, loosely modeled on biological neurons, where weighted inputs pass through layers and produce an output once a threshold is crossed. They explain convolutional neural networks in some detail, since these underpin most of the image classification tools relevant to us — early layers pick up simple features like color, and deeper layers extract increasingly abstract shapes and patterns, with a network qualifying as "deep" once it has three or more hidden layers. There's a nicely practical explanation of the standard train-validation-test split — commonly an eighty-ten-ten division — where the training set builds the model, the validation set fine-tunes parameters and guards against overfitting, and the test set, entirely unseen during training, gives the real-world performance check, ideally on out-of-network data that differs meaningfully from the training source. That's a useful lens for critically reading any AI validation paper that crosses your desk going forward — always ask whether the test set was genuinely independent. They also cover backpropagation, learning rate tuning, activation functions, and transfer learning, the last of which is particularly relevant to dermatology since it's the mechanism by which a model pretrained on a massive general image data set gets fine-tuned on a smaller, dermatology-specific data set — this is how many of our lesion classifiers actually get built without needing millions of labeled skin images from scratch. The review then distinguishes supervised learning, where models train on labeled data like lesion images tagged with diagnoses, from unsupervised learning, which finds structure in unlabeled data — they give examples like clustering patients by risk of immunotherapy toxicity or clustering rash phenotypes by clinical features — and from self-supervised learning, a hybrid approach that generates its own labels from inherent data structure, which is particularly valuable when labeled dermatology data is scarce or expensive to produce, with one example being models trained to predict poor-outcome squamous cell carcinoma from histologic images without extensive manual annotation. From there the article moves into generative AI and natural language processing territory — diffusion models, encoders and decoders, transformers with their self-attention mechanism that lets a model retain context across long text, large language models like ChatGPT, Gemini, and Claude, multimodal and visual language models that integrate images with text, reinforcement learning with human feedback, and retrieval-augmented generation, where a foundation model draws on external sources like dermatology textbooks to improve domain-specific accuracy. They note dermatology-specific applications already in the literature — generative models predicting reconstruction options from post-Mohs defect images, assisting with prior authorization language, and generating patient education materials. Since this is a foundational CME primer rather than a study, there's no results or limitations section to report — the actual practical takeaway is about literacy, not practice change. The authors' explicit goal is to equip you to critically evaluate the next AI-in-dermatology paper you read — to ask whether the test set was truly held out, whether the training data reflects your patient population, and whether a tool is being marketed as autonomous when it's really meant to be an augmentation of your judgment. Nothing here is practice-changing today, but it's the vocabulary you'll need for part two, which presumably gets into specific validated tools and their performance.

That wraps this episode. Two ethics columns reinforcing the same core message — that age alone should never be a proxy for treatment intensity in either direction, and a foundational primer to get your AI vocabulary sharp before the follow-up piece on specific applications. Thanks for listening, and I'll see you next time.