Welcome back to the journal review. This is your walkthrough of Dermatologic Surgery for December 2025, volume fifty-one, number twelve. We've got four pieces worth your time this month — an artificial intelligence coding comparison, a randomized trial on compression therapy for second-intention wounds, a cross-sectional survey on how patients perceive the cost of Mohs surgery, and a short technique report repurposing an ophthalmologic instrument for vulvar cysts. Let's get into it.

First up is a comparative study — really a benchmarking exercise — looking at artificial intelligence large language models for medical coding in dermatologic surgery, out of UT Southwestern. The clinical problem here is one you already live with: ICD-10 and CPT coding in Mohs and reconstructive practice is genuinely complex, and coding errors translate directly into denied claims and lost revenue. The question the authors asked is whether general-purpose large language models — specifically ChatGPT 4.0 and Doximity GPT, which is a health-care-oriented model built on similar underlying architecture — can accurately generate codes from real clinical documentation.

Methodologically, they pulled two hundred fifty anonymized encounters from their own dermatologic surgery clinic, complete with the original note and the actual billed ICD-10 and CPT codes serving as ground truth. Each encounter's documentation was fed to both models with a standardized prompt, then a clarifying follow-up prompt to give the models their best shot. Responses were graded as correct, partially correct, or incorrect, and then run through the standard machine-learning accuracy metrics — precision, recall, and F1 score, which is just the balanced average of the two. Using an existing coded dataset as ground truth here makes sense methodologically because it lets you benchmark model output against what actually got billed, without needing a prospective workflow — this is a retrospective accuracy check, not a live deployment test, and that's an appropriate first step before anyone considers integrating these tools clinically.

The results were a genuine split decision, and honestly, not encouraging overall. For ICD-10 diagnosis coding, both models performed poorly — ChatGPT got a correct response only about one in fifteen times, Doximity GPT even less often, and while ChatGPT's edge over Doximity GPT was statistically significant, both F1 scores were down in the single digits, which is clinically irrelevant precision. For CPT procedure coding, the story flipped: Doximity GPT was correct roughly six times out of ten, dramatically and significantly outperforming ChatGPT, which was correct less than one time in ten and was outright incorrect nearly two-thirds of the time. That's a meaningful, practice-relevant gap in performance between two consumer-accessible tools. Interestingly, the error patterns differed by model — ChatGPT tended to misclassify procedure type, frequently mistaking Mohs surgery for a standard excision even when the note was explicitly labeled as a Mohs operative report, which the authors candidly say they can't explain. Doximity GPT's main failure mode was repair-site errors, particularly at anatomic junction zones where the lesion site and the repair site documentation diverged, like a hairline lesion repaired onto the forehead.

On the diagnosis-coding side, the authors identified two specific failure mechanisms worth knowing about: incidental past medical history in the note pulling in irrelevant extraneous codes, and a loss of specificity where models defaulted to vague codes like "unspecified malignant neoplasm of the skin" instead of capturing histology and site — which matters because specificity is what drives appropriate reimbursement for higher-complexity repairs.

The authors' bottom line, and this is a fair one, is that neither general-purpose model is ready for autonomous use in dermatologic surgery coding and billing. The health-care-tuned model outperformed its general parent model specifically in the procedural coding domain, suggesting that targeted training on procedural documentation and coding guidelines is a promising direction, but we're not there yet. Limitations are straightforward — single-institution data, only two models tested, and no comparison against specialized coding-specific machine learning systems that other studies have shown reaching accuracy in the seventy to ninety percent range. For your practice, this is interesting but not remotely actionable — you should not be routing Mohs documentation through consumer-facing chatbots for coding, and if anything this reinforces that human coding oversight remains essential for the foreseeable future.

Second article is a randomized controlled trial out of Cooper University Hospital examining whether compression stockings speed up second-intention healing after dermatologic surgery on the distal lower extremity. The clinical rationale is intuitive — venous stasis and age-related vascular decline slow healing on the leg, compression improves venous return and interstitial fluid clearance in chronic venous ulcer patients, and prior small studies of Unna boots after Mohs surgery suggested benefit. But nobody had tested plain compression stockings in a randomized fashion, and critically, prior positive studies weren't randomized and didn't isolate stockings from the zinc oxide component of an Unna boot.

This is a prospective, randomized, evaluator-blinded trial — forty patients, twenty per arm, comparing standard wound care against standard wound care plus knee-length graduated compression stockings worn at least twelve hours a day. Randomization here is the right call because it controls for the wound-size and comorbidity variability that plagued earlier retrospective work, and evaluator blinding for the wound measurements protects against bias in what is otherwise a fairly subjective assessment of percent re-epithelialization. Patients with known venous insufficiency or peripheral neuropathy were deliberately excluded — an important design choice, since that's precisely the population most likely to benefit from compression, and excluding them means this trial is really asking about compression in an otherwise vascularly normal population.

The results were unambiguously negative for benefit. Average time to complete re-epithelialization was about nine and a half weeks with compression versus ten weeks with standard care alone — a difference with no statistical or clinical significance. Stratifying by final defect size — under four square centimeters, four to eight, and over eight — showed no consistent pattern either; in the smallest wounds compression actually trended faster, in the mid-size and largest categories standard care trended faster, and none of it reached significance. Regression analysis confirmed that age, sex, defect size, diabetes, and smoking status also had no significant effect on healing time in this cohort. Safety-wise, compression was well tolerated, with only a single bleeding event tied to incidental trauma rather than the intervention itself, and no infections in either arm.

The authors' interpretation is sensible: the benefit of compression in venous ulcer literature is likely mediated through correcting an underlying venous insufficiency, and since that population was excluded here, there was no pathology for compression to correct. They also draw a nice mechanistic distinction from the earlier positive Unna boot studies — an Unna boot isn't just compression, it's compression plus a zinc oxide layer with antibacterial and anti-inflammatory properties, so those two interventions aren't really interchangeable. Limitations are the ones you'd expect from a small single-center trial — self-reported adherence to wear-time, no objective monitoring of the twelve-hour requirement, and a sample too small to detect anything but a large effect size. For your practice, this is a useful negative result: routinely prescribing compression stockings for second-intention lower-extremity wounds in patients without documented venous insufficiency is not supported by this data, though it remains reasonable — and the authors suggest as a future direction — in patients who do have venous disease.

Third, a cross-sectional survey study from Brigham and Women's Hospital looking at how patient perceptions of Mohs surgery cost have shifted between 2014 and 2024. The gap being addressed is squarely economic and behavioral rather than clinical — as cost-sharing burdens shift more heavily onto patients, do patients still value Mohs surgery the same way, and has their willingness to pay kept pace with both inflation and the increasing complexity of the procedures they're actually receiving?

The design is a straightforward, identical survey administered at two time points a decade apart to patients at their first postoperative suture-removal visit — two hundred fifty total respondents, evenly split, with a strong response rate around eighty-three percent. Patients estimated three things: what they thought their Mohs procedure actually cost, what a fair insurance reimbursement would be, and what a fair out-of-pocket price would be if they had no coverage at all. Repeating the identical instrument a decade later is a clean way to isolate a temporal trend while holding the survey instrument itself constant — though it's worth flagging that this is a single academic institution, so the demographic profile of respondents was similar across both cohorts in age, education, and insurance status.

The results are telling. By 2024, the tumors being treated were larger — average diameter of roughly nine and a half millimeters versus about five and a half a decade earlier — and reconstructions were more often flaps or grafts rather than simple closures, meaning patients were, on average, undergoing more extensive surgery. Perceptions of actual cost rose accordingly, with more patients in 2024 estimating the procedure cost more than five thousand dollars, and a larger share also felt a fair insurance reimbursement should exceed five thousand dollars. But out-of-pocket willingness to pay did not move in parallel — if anything, more patients in 2024 thought less than five hundred dollars was a fair out-of-pocket price, and the proportion willing to pay more than five thousand dollars out of pocket stayed essentially flat. The authors make the point explicitly: even adjusting for roughly a third inflation over that decade and generally higher household incomes, patients were not willing to pay more out of pocket in 2024 despite undergoing objectively more extensive procedures.

Discussion-wise, the authors frame this through willingness-to-pay economics — patients consistently value fair insurance reimbursement higher than what they'd personally pay, a gap that persisted and if anything widened. They also note, importantly for anyone tracking reimbursement trends, that inflation-adjusted Medicare reimbursement for the Mohs CPT code has dropped by roughly thirty percent over this same window — so the disconnect here isn't just patient psychology, it's set against a backdrop of falling real-dollar reimbursement to the surgeon performing more complex work. Limitations are the ones inherent to any single-center survey — small sample, a higher socioeconomic cohort than the general population, which the authors argue would probably make the true discrepancy even larger elsewhere, and no ability to model the huge variability in actual health plan cost-sharing structures across the country. Practically, this isn't something that changes your Tuesday clinic, but it's a valuable data point for anyone involved in advocacy around Mohs reimbursement or patient financial counseling — it quantifies, in patients' own words, the widening gap between perceived value and payment tolerance as the field faces reimbursement pressure.

Last is a short technique report, essentially a case-based description, describing use of a chalazion clamp for treating steatocystoma multiplex and epidermoid cysts on the vulva. There's no methods-versus-results structure here — it's a practical pearl. The clinical problem is that removal of these benign cysts, whether steatocystomas or epidermoid cysts, is typically for cosmetic or symptomatic reasons, and there's no established gold-standard technique, particularly in a delicate, non-bony anatomic site like the labia majora where visualization and hemostasis during multiple simultaneous small-lesion extractions can be awkward.

The authors borrow the chalazion clamp — familiar to most of you already for lip or ear lobe lesions — which has a ring-shaped aperture on one arm and a flat plate on the other, cinched down with a thumbscrew to stabilize tissue, achieve local hemostasis, and improve visualization. In their patient, a woman in her late thirties with forty to fifty small steatocystomas and two epidermoid cysts across both labia majora, they marked and anesthetized the field, incised each lesion with an eleven blade, applied the clamp to stabilize and express the cyst contents, then curetted and cauterized residual cyst wall while leaving the overlying epidermis intact, dressing with plain petrolatum afterward. The patient tolerated it well, returned for repeat sessions by request given her satisfaction with the outcome, and had no recurrence at six months.

The teaching point is really about the instrument, not a trial finding — the clamp's value is that it gives you a firm, stable working surface against thin, mobile mucosal skin where you'd otherwise be fighting for traction and hemostasis simultaneously, particularly useful when you're managing high lesion counts in one setting. This is a nice pearl for your toolkit in genital or mucosal cyst cases, but keep it filed as a technique option rather than a validated standard — the authors themselves call for comparative studies before this becomes anything like a preferred approach.

That wraps our December rundown. The through-line this month is a healthy dose of humility about new tools and old assumptions — general-purpose artificial intelligence isn't ready to touch your coding yet, compression stockings don't do much without underlying venous disease to correct, and patients are increasingly aware their surgery is getting more complex without being willing to pay more for it. Thanks for listening, and we'll see you next month.