Welcome back to the journal review. This is the January 2026 issue of Dermatologic Surgery, and we've got four pieces worth unpacking today — an AI billing accuracy study, a striking case report on squamous cell carcinoma in a rare keratoderma, a randomized pilot trial of platelet-rich plasma for cancer-therapy-related alopecia, and a practical review on patient adherence in surgical practice. Let's get into it.

First up is a research letter — really a small original study — looking at AI-assisted billing in dermatologic surgery, specifically how well ChatGPT and Gemini assign CPT codes, that's Current Procedural Terminology codes, for common procedures. The background here is straightforward: everybody's talking about large language models creeping into documentation and billing workflows, but nobody had actually stress-tested these tools against the kind of coding complexity we deal with every day — Mohs stages, excisions, and the alphabet soup of intermediate and complex repairs, flaps, and grafts.

Methodologically, the authors built ninety test cases, thirty each of Mohs procedures, excisions, and repairs, spanning a range of anatomic sites and complexity. They queried ChatGPT model 4-o and Gemini 2.0 with a deliberately simple, single-line prompt — just a plain-language procedure description — and compared the output against codes assigned by a board-certified Mohs surgeon as the reference standard. The choice to keep prompting simple wasn't sloppy — the authors explicitly say this mirrors how a clinician would actually use these tools in real time, without elaborate prompt engineering or supplemental billing guides. That's a nice methodological point: it sacrifices maximum achievable accuracy in favor of ecological validity, telling us how these models perform in the hands of a busy end user rather than in the hands of someone who knows how to coax the best possible output.

On results, Gemini clearly outperformed ChatGPT — correct coding on about nine in ten cases overall versus roughly three in four for ChatGPT, and that overall gap was statistically significant. Breaking it down by procedure type, both models did reasonably on Mohs surgery and excisions, with Gemini numerically ahead but not significantly so. The real separation was in complex repairs — flaps, grafts, that kind of thing — where Gemini nailed every single case, one hundred percent, while ChatGPT got only about two-thirds correct, and that difference was clearly statistically significant. The error pattern is instructive too: ChatGPT tended to mismatch procedure site with the wrong billing location, underbill, generate partial codes, or drop modifiers altogether — exactly the kind of errors that translate into claim denials in real practice.

The authors' own conclusion is appropriately conservative: even the better-performing model isn't accurate enough for unsupervised clinical use right now. This is a small, single-institution proof-of-concept study with a modest sample size, a single ground-truth adjudicator, and no exploration of enhanced prompting or retrieval-augmented approaches that might close the gap. So practically speaking — this is interesting, not practice-changing. There's no case here for letting either model auto-generate your billing today. But it's a useful signal that AI-assisted coding, especially for complex repairs, is worth watching as these models and integration strategies mature, and it reinforces that human oversight of coding remains essential, particularly anywhere reimbursement hinges on correctly capturing repair complexity.

Next, a case report that's genuinely novel — the first reported instance of locally advanced cutaneous squamous cell carcinoma arising in a patient with Bothnian-type palmoplantar keratoderma, an entity most of us rarely think about outside of genodermatosis conferences. This is caused by a gain-of-function mutation in AQP5, the gene encoding aquaporin 5, a channel protein expressed in granular layer keratinocytes and sweat gland secretory cells.

The patient is a 68-year-old man with a lifelong history of the characteristic white, waxy, aquagenic-wrinkling palmoplantar keratoderma, with a strong family history — his brother, father, paternal uncles, and paternal cousins all similarly affected, and his brother also treated for plantar squamous cell carcinomas. He'd already had two early-stage plantar squamous cell carcinomas excised with skin grafting before genetic testing identified a variant of uncertain significance in AQP5. Despite that "uncertain" label, his clinical course tells its own story: over the next five years he developed additional plantar tumors, one of which recurred aggressively enough to require pembrolizumab, which failed to control growth, then sixty-six Gray of radiation, and ultimately salvage surgery with free flap reconstruction at an outside institution after he was lost to follow-up mid-treatment. A separate lesion on the contralateral foot was excised with clear margins. He is currently without evidence of disease, but remains on a tight surveillance schedule — MRI of the affected foot every six months — along with niacinamide, topical urea, and a plan to reintroduce low-dose acitretin for chemoprevention, having previously not tolerated a higher dose.

There's no methods or limitations section here in the traditional sense — this is a case report, and its value is entirely in the observation and the mechanistic hypothesis. The authors propose that the AQP5 variant may be pathogenic, driving altered epidermal differentiation and chronic inflammation, and they draw a mechanistic parallel to Olmsted syndrome, where a related channel, TRPV3, physically complexes with AQP5 as an osmosensor — when overactivated, this raises intracellular calcium, drives keratinocyte apoptosis and hyperkeratosis, and AQP5 itself has documented oncogenic activity through RAS/MAPK signaling in other malignancies. Combine that biology with years of chronic mechanical trauma on the soles, and you have a plausible carcinogenic substrate.

The clinical teaching point here is simple and directly actionable: patients with Bothnian palmoplantar keratoderma, and probably other diffuse non-epidermolytic keratodermas with chronic plantar hyperkeratosis, warrant a low threshold for biopsy of any new or changing plaque, ongoing dermatologic surveillance, and consideration of chemoprevention given the apparent aggressiveness of squamous cell carcinoma when it does arise in this substrate — this patient needed systemic immunotherapy and radiation on top of surgery for one lesion. This is one case, so it doesn't establish an incidence or a screening protocol, but it should raise your index of suspicion the next time you see a keratoderma patient with a "just another callus" story.

Now to the platelet-rich plasma trial — this is a proper randomized, controlled pilot study, and it's worth walking through in full because the design choices matter as much as the results. The clinical problem is real: a substantial proportion of breast cancer survivors are left with persistent chemotherapy-induced alopecia or endocrine-therapy-induced alopecia after taxane-based chemo, aromatase inhibitors, or selective estrogen receptor modulators, and once first-line topical therapy fails, options are thin. Platelet-rich plasma has reasonable evidence in androgenetic alopecia, but two things made its use here uncertain — no prior data in this specific population, and cancer itself has traditionally been treated as a relative contraindication to autologous blood-product injections because of theoretical concerns about circulating tumor cells and seeding.

Twenty-seven women were enrolled — fifteen with endocrine-induced alopecia, twelve with persistent chemotherapy-induced alopecia — in a single-center, single-blinded, randomized split-scalp design. One side of the scalp got PRP injections monthly for three sessions, the other side got nothing, with a blinded investigator rating a seven-point global assessment scale at twelve weeks as the primary endpoint, and secondary measures including quality of life via the Hairdex questionnaire, trichoscopic hair density, and — notably — circulating tumor cell assays performed on the PRP itself as a safety check. The split-scalp design is a classic efficient pilot strategy, letting each patient serve as her own control and requiring a smaller sample size, but the authors are upfront that this is a trade-off: PRP is known to diffuse beyond the injection site, which can contaminate the "untreated" side and wash out any real between-side difference. That's worth flagging clearly because it directly shapes how you interpret what follows.

And that's exactly what seems to have happened. Global assessment scores improved significantly from baseline on the treated side, but — and this is the key finding — improved just as much on the untreated control side, with no significant difference between the two. Hair density likewise rose meaningfully on both sides, roughly twenty-one hairs per square centimeter on the treated side and sixteen on the control side, again with no statistically significant difference between them. Quality of life scores didn't move in either group. On safety, the news is reassuring: adverse events were limited to mild-to-moderate injection site pain, and although two of twelve circulating tumor cell assays actually detected malignant cells within the PRP preparation itself, there were no observed tumor seeding events in any patient.

So how do you read this? The authors themselves attribute the lack of between-side separation likely to PRP diffusion contaminating the control side rather than concluding PRP doesn't work — a reasonable interpretation given both sides improved together rather than neither improving at all. But it's also fair to note this is a small pilot, single-center, with a control-side confound baked into the design, so efficacy here is genuinely unproven, not just modestly proven. The safety data, on the other hand, is the more solid takeaway: despite detecting malignant cells in some PRP preparations, no seeding occurred, which is a meaningful, actionable reassurance for a population where PRP has often been avoided reflexively. Practically — this is not practice-changing for efficacy; you shouldn't be promising these patients robust density gains from PRP based on this trial. But it is a legitimate, if preliminary, safety green light that should lower the bar for considering a properly designed parallel-arm trial in this population, and it may reasonably support cautious off-label use in patients who understand the evidence is still thin.

Last, a review article on practical interventions to optimize patient adherence in dermatologic surgery and Mohs micrographic surgery. This isn't an original study — no new patient data — so think of it as a curated, evidence-graded toolkit rather than a hypothesis-testing paper. The author frames the problem with some sobering numbers pulled from the existing literature: more than one in ten skin cancers referred for Mohs surgery go untreated because of missed appointments or comprehension gaps, nearly half of Mohs patients don't really understand concepts like healing by secondary intention, roughly two-thirds don't follow postoperative antibiotic regimens, one in five never fill their opioid prescription, and about one in five skip follow-up entirely. These aren't trivial numbers — they translate directly into flap and graft complications and suboptimal cosmetic outcomes.

Methodologically, this was a literature search across MEDLINE and Scopus, with every recommendation graded for level of evidence using the Oxford Centre for Evidence-Based Medicine system, evaluated by a single author — worth noting as an inherent limitation of a solo-reviewer narrative synthesis rather than a systematic review with independent dual extraction. The result is thirty-six discrete interventions organized across the preoperative, intraoperative, and postoperative periods, built around behavioral principles like the "IKEA effect" — patients value a plan more when they help build it — nudge theory, anchoring, salience, and what the author calls "white coat compliance."

A few of the higher-grade, more actionable ones are worth calling out specifically. In the preoperative space: showing patients their actual defect in a mirror before reconstruction to anchor realistic expectations, explicitly telling patients to expect to spend the entire day in the Mohs suite regardless of stage number, and disclosing trainee involvement up front — all backed by decent-level evidence and tied to measurably higher satisfaction scores. For smoking, rather than insisting on full cessation, which fails in the vast majority of patients preoperatively, offering nicotine replacement therapy for as little as two weeks before and one week after surgery is a pragmatic harm-reduction compromise.

Postoperatively, the list runs from the obvious — pairing written instructions with video content, since nearly half of patients forget at least one verbal instruction — to the more clever behavioral tricks: using nonabsorbable sutures specifically because they force a return visit, which the author terms "white coat compliance," or asking patients to text a photo of the wound for a quick virtual check. Then there's a long list of burden-reduction strategies with practical, low-cost application in most practices — complimentary wound care supplies, favoring absorbable sutures or adhesive strips over ones needing removal, waterproof dressings to allow normal showering, tranexamic acid to cut postoperative bleeding events, and choosing full-thickness over split-thickness grafts when feasible to minimize donor-site care burden. On the pharmacologic side, small substitutions carry decent evidence — long-acting bupivacaine with epinephrine at closure, higher-dose ibuprofen instead of over-the-counter strength, and cefadroxil over cephalexin for prophylaxis given its longer dosing interval and presumably better adherence.

Because this is a review rather than a primary study, there's no results-versus-limitations tension to walk through in the usual sense — the honest caveat is simply that the evidence grades span a wide range, with plenty of level three and four, grade C or D recommendations mixed in among the stronger level one, grade A items, and a good portion of the underlying data is extrapolated from general dermatology or general surgery adherence literature rather than Mohs-specific trials. Practically, though, this is one of the more immediately usable pieces in the issue — most of these interventions cost nothing, require no new equipment, and can be adopted into your workflow starting tomorrow, even though few have been formally validated in dedicated Mohs-specific adherence trials.

That wraps our four articles for this January issue. The throughline across all of them is really a theme of technology and process meeting the limits of current evidence — AI coding tools showing promise but not yet reliability, PRP showing safety before showing proven efficacy, a rare genodermatosis reminding us that inflammation and mechanical trauma can be genuinely oncogenic, and a reminder that some of the most impactful things we do for our patients' outcomes have nothing to do with the scalpel at all. Thanks for listening, and we'll see you next month.