Welcome back to the journal review. This episode covers the January twenty twenty-six issue of the Journal of the American Academy of Dermatology, and we've got four papers to get through — an artificial intelligence meta-analysis, a long-term follow-up on a laser prophylaxis trial, a retrospective look at bleeding after nasolabial interpolation flaps, and a real-world study on reflectance confocal microscopy. Let's get into it.

First up is a meta-analysis looking at ChatGPT's diagnostic accuracy in dermatology, specifically comparing text-based prompts against image-based prompts. This is a systematic review and meta-analysis, not a primary study, so the whole point is synthesizing what's already out there rather than generating new patient data. The background here is familiar to all of us — large language models are being thrown at every diagnostic problem in medicine, and dermatology, with its visual nature, is an obvious testing ground. But individual studies on ChatGPT's accuracy for skin lesions have been all over the map, and the authors reasonably conclude that's probably due to differences in methodology rather than a real signal about the tool itself. So they pooled seventeen studies pulled from PubMed and Scopus, extracting variables like the skin condition tested, lesion classification, Fitzpatrick phototype, which ChatGPT model was used, and where the images came from — public datasets versus private clinical sources. They looked at two flavors of accuracy: whether ChatGPT nailed the top diagnosis on its differential, and whether the correct answer showed up anywhere in the differential at all. Meta-analysis is really the only way to answer "how good is this tool, overall" when the primary literature is this fragmented — pooling gives you power and lets you formally test which variables are driving the variability, which a single study can't do.

The results are genuinely useful for setting expectations. The newer model, ChatGPT-4o, significantly outperformed the older ChatGPT-4, with top-diagnosis accuracy jumping from under forty percent to roughly two-thirds correct, and differential accuracy climbing from about sixty percent to about seventy-six percent — a real, clinically meaningful jump, likely reflecting genuinely improved multimodal reasoning in the newer model. Interestingly, visual prompts actually beat textual prompts overall, around seventy percent versus sixty-five percent — but the regression model told a more nuanced story: when you control for other variables, visual analysis was independently associated with significantly lower odds of correct differential diagnosis compared to text. So the raw numbers and the adjusted analysis pull in different directions here, which the authors don't fully reconcile, and I'd flag that as something to interpret cautiously rather than take at face value. The most clinically important finding, frankly, is the equity signal: accuracy dropped significantly for darker skin types, going from about seventy-eight percent in lighter phototypes down to about sixty-six percent in Fitzpatrick three-and-above. That's not a rounding artifact — that's a meaningful performance gap. Accuracy also improved significantly year over year as models matured, and public datasets performed roughly comparably to private clinical sources. The authors are upfront about the limitations: heterogeneity across these seventeen studies was very high, meaning the pooled numbers should be read as a general trend rather than a precise point estimate, and no publication bias was detected but that doesn't rescue you from apples-to-oranges comparisons in prompt engineering and lesion mix. Practical takeaway: this is interesting, not practice-changing. Nothing here suggests you should be recommending ChatGPT as a standalone diagnostic tool to patients or trainees. If anything, the skin-of-color performance gap is the single most actionable data point — a reason to be explicitly cautious about equity implications before any of these tools get closer to clinical deployment.

Second article: a five-year follow-up on a randomized controlled trial of fractionated laser resurfacing as prophylaxis against actinic neoplasia. This is a continuation of a study most of you may already know — the group has published six-month and three-year data previously, and this brief report extends the observation window out to sixty-six months. The biological premise is that aged, sun-damaged skin loses fibroblast-derived insulin-like growth factor-1 signaling to keratinocytes, and that controlled wounding — in this case, fractionated laser resurfacing — recruits non-senescent fibroblasts that restore that signaling and correct the precancerous ultraviolet-B response. The design is an investigator-initiated, single-site, randomized, single-blinded intrapatient trial: each of forty-eight elderly volunteers, overwhelmingly male, with Fitzpatrick types one and two, had one forearm treated with a single session of fractionated laser resurfacing while the other arm served as an untreated internal control. This within-patient design is a smart methodological choice for a prophylaxis question like this — using each patient as their own control eliminates a huge amount of interpatient variability in sun exposure history, skin type, and field cancerization burden, so you need a much smaller sample to detect a real effect than you would with a between-patient design.

The results are striking for their durability. Before treatment, actinic keratosis counts were essentially balanced between arms — a ratio around 1.3, confirming randomness. Within three months of treatment, that ratio dropped below zero-point-five and it stayed there, significantly below baseline at every single time point all the way out to sixty-six months — over five years of sustained protection from a single treatment session. On the nonmelanoma skin cancer side, at the sixty-six month mark, the untreated arms had accumulated far more lesions than the treated arms — thirty-four skin cancers on treated arms versus... actually reversed, let me be precise: the treated arms showed fewer, with the untreated arms carrying the greater burden, and among the basal cell carcinomas identified, the great majority were superficial rather than nodular. The authors' conclusion is straightforward — a single fractionated laser session appears to provide durable, multi-year chemoprevention-like protection against actinic keratosis and skin cancer development. The honest limitations here are significant: this is a small, single-site cohort of mostly elderly white men with lighter skin types, so generalizability to women, to skin of color, or to younger high-risk patients like transplant recipients is unknown. It's also a forearm model, not face or scalp, where anatomy and cosmetic tolerance for wounding differ substantially. Practical takeaway — this is interesting and encouraging, but I'd stop short of calling it practice-changing yet. It's a proof-of-concept that a single ablative fractional treatment can meaningfully suppress field actinic damage for over five years, which is a compelling data point for thinking about laser resurfacing as an adjunct in high-risk actinically damaged patients, but the population studied is narrow and this hasn't been tested against our standard field therapies head-to-head.

Third article, and this one's squarely in our procedural wheelhouse: a retrospective, single-center study on bleeding complications after nasolabial interpolation flap repair, with a specific focus on antithrombotic therapy. This is a retrospective cohort study. The background is one we all know intuitively — a large chunk of our Mohs patients are on antiplatelet or anticoagulant therapy, and interpolation flaps, because the pedicle depends on angular artery perforators and involves handling highly vascularized donor tissue, carry a reputation for higher bleeding risk than simpler repairs. Prior data on this has come from small cohorts without real risk-factor analysis, so this group set out to build a larger cohort specifically to enable risk stratification. A retrospective design makes sense here — the authors don't spell out the rationale explicitly, but bleeding complications after this specific flap type are uncommon enough that you'd need either a multi-year retrospective pull or an impractically long prospective study to accumulate enough events for meaningful analysis, and this is exactly what they did, pulling consecutive cases across sixteen years at a single academic center.

Out of three hundred sixty-seven nasolabial interpolation flaps, fourteen patients — under four percent — had a postoperative bleeding complication, and every single one was managed successfully in the office, which is itself a reassuring safety signal. But the distribution by antithrombotic status is the real finding. Patients on no antithrombotic had a bleeding rate around two percent. Antiplatelet alone was around three percent. Anticoagulation alone was around six percent. But patients on combined antiplatelet plus anticoagulation therapy had a bleeding rate over fifty percent — more than half of that small subgroup bled. On multivariable analysis, combined antithrombotic therapy was the only independent predictor of bleeding, with an odds ratio over eighty — a massive effect, though the confidence interval was understandably wide given how few patients were on dual therapy. The authors interpret this as a synergistic rather than simply additive interaction between the two drug classes. Other factors — pacemaker presence, more Mohs stages, larger defect size — showed univariate associations but didn't survive as independent predictors. The authors are transparent that their multivariable model is underpowered given how few bleeding events and how few dual-therapy patients they had, even though this represents the largest published nasolabial interpolation flap cohort to date, and the retrospective design can't fully account for confounding around why certain patients were kept on dual therapy perioperatively in the first place. Practical takeaway, and I'd call this one close to practice-changing at the level of patient counseling: overall bleeding risk with this flap remains low and manageable in-office, so there's no reason to abandon the technique broadly. But patients on combined anticoagulant-plus-antiplatelet therapy represent a genuinely distinct, high-risk subgroup deserving specific preoperative counseling, perhaps closer perioperative coordination with prescribing physicians, and a lower threshold for enhanced hemostasis and closer follow-up.

Fourth and final article: a prospective real-life study on reflectance confocal microscopy, or R-C-M, and its ability to reduce unnecessary biopsies in both adults and children. This is an original prospective single-center study, not a randomized trial — the authors are explicit that this was an uncontrolled, real-world clinical implementation study. The clinical gap they're addressing is one we all feel: the number needed to biopsy to catch one skin cancer remains high, and biopsy carries real costs — for kids in particular, sedation or general anesthesia is sometimes required, making noninvasive alternatives especially valuable in that population. Reflectance confocal microscopy sits between dermoscopy and histopathology, giving quasi-histologic resolution noninvasively, but most of its supporting evidence to date comes from specialized skin cancer units rather than everyday general dermatology practice. So the design choice here — a prospective, consecutive-case study pulling from both general dermatologists and skin cancer specialists across a large academic department — was specifically meant to test whether the technology's benefits generalize beyond expert referral centers, which is a sensible and clinically important question that a controlled trial in a single specialized unit couldn't answer.

They enrolled twelve hundred eighty-five equivocal lesions from just under nine hundred patients over roughly three years, the vast majority adults but with a meaningful pediatric subgroup. Overall reflectance confocal microscopy to histopathology concordance for specific diagnosis was about seventy percent, and concordance was significantly better in older patients, head and neck locations, nonpigmented lesions, and nonmelanocytic lesions — basal cell carcinoma in particular performed very well, with sensitivity around ninety-nine percent and specificity around ninety-two percent. Melanoma detection was also excellent from a sensitivity standpoint — essentially catching all melanomas in the biopsied cohort — though specificity and positive predictive value were lower, meaning plenty of benign pigmented lesions still triggered a biopsy recommendation, which is exactly the trade-off you'd want for a rule-out tool applied to melanocytic lesions. The headline number is the biopsy-sparing rate: nearly fifty-nine percent of all referred lesions were spared a biopsy after reflectance confocal microscopy evaluation, and in the pediatric subgroup that number climbed to an impressive eighty-eight percent. Biopsies were saved more often in younger patients, pigmented and melanocytic lesions, and when referrals came from general dermatologists rather than skin cancer specialists — which makes sense, since general dermatologists are presumably referring a lower pretest-probability population to begin with. Only three lesions initially called benign on reflectance confocal microscopy turned out later to be malignant, which is a reassuringly small miss rate given the scale of the cohort. The authors' own limitations are worth repeating plainly: this was single-center, not every lesion had histopathologic confirmation since some were simply followed clinically, and the pediatric group, while showing the most dramatic benefit, was still a relatively small absolute number of patients. Practical takeaway — this one leans toward practice-changing, at least for anyone with access to the technology or considering acquiring it. The biopsy-sparing numbers, especially in children where avoiding sedation is a real clinical win, and the strong performance specifically for basal cell carcinoma, make a solid case for incorporating reflectance confocal microscopy into equivocal-lesion workflows, with the caveat that pigmented and melanocytic lesions still carry a real miss rate on the malignant side and shouldn't be over-trusted to rule out melanoma outright.

That wraps this month's four papers — an AI diagnostic tool that still needs equity fixes before it's clinic-ready, encouraging five-year durability data on laser prophylaxis in a narrow population, a clear bleeding-risk signal for dual antithrombotic patients undergoing interpolation flaps, and a real-world validation of confocal microscopy as a genuine biopsy-sparing tool, particularly in kids. Thanks for listening, and we'll catch you next month.