Welcome back to the journal review. This is our March twenty twenty-six run through the Journal of the American Academy of Dermatology, and we've got four pieces to get through — a correspondence exchange bookending the issue, a substantial original study on confocal microscopy for the lentigo maligna spectrum, and a population-based brief report on radiotherapy in head and neck Merkel cell carcinoma. Let's get into it.

First up is a letter to the editor — a comment on the Veldhuizen multicenter cohort study of patient-reported quality of life after Mohs surgery, which some of you will remember from earlier this year. This is a critique letter, not new data, so what we're really doing here is walking through the reviewers' methodological objections. The authors, a group out of Xiangya Hospital in China, raise three points. First, they note that the subgroups showing increased cancer worry — tumors near the eyes and on the upper lip — each represented only about five percent of the total cohort, and they argue that's a recipe for wide, unstable confidence intervals. They push for correction methods like bootstrapping, Bayesian analysis, or Bonferroni adjustment to shore up reliability in those small subgroups. Second, they flag that response rates at six and twelve months hovered around sixty percent, which they call acceptable but potentially bias-introducing, and they suggest that relying on a single electronic contact method probably left engagement on the table — pointing to literature on reminder emails being mistaken for spam and to large cohort studies showing that multi-modal contact, combining email, phone, and mail, improves retention, especially in older populations. Third, and probably the most clinically interesting point, they highlight that the original study found scar appraisal dropping and anxiety rising with increasing defect size — roughly a two-and-a-half to three point shift per centimeter on those scales — yet reconstruction type itself showed no significant satisfaction differences. The letter writers argue that's counterintuitive, since more complex reconstructions typically track with larger defects, and they recommend formal interaction-term modeling between defect size and reconstruction type to actually interrogate that relationship rather than let it sit as a null finding.

Now, rather than leaving that hanging, the issue also carries the original authors' reply, so let's take that now while it's fresh, and then we'll come back to the confocal study. This is Veldhuizen, Dusza, and Lee's formal response, and it's a good example of push-back that's worth hearing in full because it teaches you something about analytic philosophy, not just data. On the multiplicity correction point, they push back directly — their argument is that Bonferroni-type corrections trade type one error for type two error, and that trade-off makes sense in something like genomics with thousands of comparisons, but not in a hypothesis-driven, prespecified clinical outcomes study with a limited number of clinically relevant comparisons. Their point is that broad correction would risk burying real, clinically meaningful signals under unnecessary conservatism, and there's no principled way to decide how many comparisons you'd even be correcting for. On response rates, they defend their single-platform electronic approach as intentional — patients consented to electronic follow-up at enrollment, the design used one standardized platform across four academic centers specifically to preserve consistency in data quality and protocol adherence for nearly a thousand participants, and they note their response rates are in line with other large prospective outcomes cohorts. Their view is that bolting on phone calls or mailed letters would have introduced heterogeneity across sites and potentially biased who participated. And on the defect size and reconstruction interaction — they actually did the analysis the critics asked for. They recoded defect size into quartiles, built the interaction term with reconstruction type, and found no significant interaction, which they candidly frame as underpowered rather than definitively negative, calling for future larger and more heterogeneous cohorts to look at it properly. As a listener, the practical takeaway from this whole exchange is fairly narrow — it doesn't change your practice, but it's a nice worked example of how a well-designed prospective cohort can defend a prespecified, hypothesis-driven analytic plan against post hoc calls for more conservative statistics, and it reinforces that defect size, not reconstruction complexity per se, is what's driving patient-reported scar and anxiety outcomes — something worth keeping in mind when you're counseling patients preoperatively about larger facial defects regardless of which flap or graft you're planning.

Now to the main course. This is a prospective multicenter diagnostic validation study out of two Australian tertiary melanoma centers, looking at cutaneous confocal microscopy across the entire lentigo maligna spectrum — from atypical intraepidermal melanocytic proliferation, or AIMP, which is thought to be a precursor lesion, through lentigo maligna itself, to invasive lentigo maligna melanoma.

The gap here is pretty clean: confocal microscopy is already known to be good at separating lentigo maligna from benign mimickers — a prior meta-analysis put sensitivity and specificity in the low-to-mid nineties — but the existing tool for that, the LM score developed back in 2010, was never validated for the harder task of telling early-stage disease apart from invasive disease within the spectrum. And the sixteen confocal features identified in a 2024 systematic review as relevant across that spectrum had never been quantitatively tested prospectively. So the question driving this paper is really a staging question, which matters enormously for you clinically because early lentigo maligna can sometimes be managed non-surgically, while invasive disease obviously needs surgical treatment — so getting the biopsy site and the diagnostic call right up front has downstream management consequences.

Methodologically, this ran over four years, enrolling consecutive lesions clinically suspicious for lentigo maligna, facial or extrafacial, pigmented or amelanotic. Every lesion got confocal imaging with the VivaScope system before biopsy or excision, and — this is a nice design touch worth flagging — the contralateral normal skin was also imaged in every patient and used as an internal control, specifically to differentiate AIMP from ordinary sun-damaged skin. Having patients serve as their own control is a smart way to cancel out confounding from age, sun exposure, and genetic background, which are exactly the variables that muddy this diagnosis in older, chronically sun-damaged facial skin. Imaging was read blinded to clinical, dermoscopic, and pathology findings by an expert with over five years of confocal experience, and any discordance with the histopathology got a second blinded look from a senior confocalist with over two decades of experience. That blinding structure is the methodological backbone of the whole paper — without it, you'd have no way to know if the confocal reads were being unconsciously anchored by clinical suspicion or biopsy results.

On results: 129 lesions from 111 patients, median age seventy-two, split into 28 AIMP, 52 lentigo maligna, 30 invasive melanoma, and 19 benign mimickers. Cheeks were the dominant site across all stages, and there was no site preference distinguishing early from invasive disease. AIMP skewed younger, with a median age of sixty-five versus seventy-three for the invasive cases, which fits a stepwise progression model.

The feature-level findings are genuinely useful for how you think about reading these scans. Scattered thin dendritic cells sitting within an otherwise preserved honeycomb epidermal pattern were the key discriminator between AIMP and benign mimickers — present in three-quarters of AIMP cases versus only about one in seven benign mimickers, a sixteen-fold difference in odds. Moving from AIMP up to lentigo maligna, the single best discriminator was adnexal spread — seen in three-quarters of lentigo maligna cases but only fourteen percent of AIMP, an eighteen-fold odds difference — alongside architectural disarray and non-edged papillae, which were roughly three-to-five-fold more common in lentigo maligna. And critically, nests were essentially absent in AIMP but showed up in over forty percent of lentigo maligna and nearly two-thirds of invasive cases.

Then, for the step that matters most clinically — separating in-situ lentigo maligna from frankly invasive melanoma — architectural distortion of the dermis was the standout feature: present in two-thirds of invasive cases but only two percent of in-situ lentigo maligna. That's a huge, clinically obvious signal, not a marginal statistical one. Dermal nests, similarly, were seen exclusively in the invasive cases, in nearly half of them, and never in lentigo maligna alone.

Based on those findings, the authors built out what they're calling the LM Spectrum Score — taking the original six-feature LM score and adding nests and dermal architectural distortion. That addition took the diagnostic accuracy, measured by area under the curve, from an unremarkable 0.73 up to 0.95 for distinguishing invasive melanoma from in-situ lentigo maligna — that's a jump from mediocre to excellent discrimination. For separating AIMP from benign mimickers and sun-damaged skin, it moved from 0.75 to 0.85, a smaller but still meaningful improvement. Worth noting too, agreement between the confocal diagnosis and final histopathology was excellent, with a kappa of point nine seven, and using the original LM score alone, every invasive case scored above three, while roughly ninety percent of AIMP cases clustered in the one-to-three range, and all contralateral control skin scored between minus one and one — so there's a workable numeric separation even before you add the two new features.

The authors are appropriately restrained in their conclusions, flagging the obvious limitation up front: this is entirely an Australian population, presumably with the sun-exposure patterns and skin phenotypes that implies, so generalizability to other populations and skin types is unproven. They frame the Spectrum Score explicitly as needing further external validation rather than being ready for prime time.

Where does this land practically? I'd call this genuinely promising but not yet practice-changing in the sense of replacing histopathology — nobody's suggesting you skip biopsy based on a confocal score. But it is immediately useful as a refinement of how you use confocal microscopy that you may already have access to: the presence of dermal architectural distortion or dermal nests on confocal should raise real concern for invasion and can help you target your biopsy or excision to the most concerning subregion of a large, heterogeneous facial lentigo maligna lesion — which is exactly the clinical scenario where sampling error causes people to miss an invasive component. It's also a nice conceptual framework for longitudinal monitoring of biopsy-proven early lentigo maligna managed non-surgically, watching for the emergence of these invasive-pattern features over time. Just don't treat the specific score cutoffs as validated outside this cohort yet.

Last article is a brief report, a population-based cohort study using the SEER database — Surveillance, Epidemiology, and End Results — looking at postoperative radiotherapy and overall survival in head and neck Merkel cell carcinoma.

The clinical rationale is straightforward: head and neck Merkel cell carcinoma is rare but aggressive, with a high risk of locoregional failure driven by anatomic constraints and rich lymphatic drainage in that region, and the authors wanted to clarify whether postoperative radiotherapy actually moves the survival needle using a large national dataset spanning 2004 through 2021.

Methodologically, this is registry-based by necessity — Merkel cell carcinoma of the head and neck is rare enough that a prospective randomized trial adequately powered for a survival endpoint is essentially impractical, so a large population database is really the only route to meaningful sample size here, and that's presumably why SEER was chosen. They included patients with nonmetastatic disease who underwent curative-intent resection with known radiotherapy status. The critical design feature is how they handled confounding by indication — patients who got postoperative radiotherapy had systematically higher-risk disease at baseline, more nodal burden, higher grade tumors — so a naive comparison would be biased toward making radiotherapy look worse than it is. They addressed this properly, with a prespecified plan: multivariable Cox regression as the primary analysis, Kaplan-Meier and log-rank as a supportive check, and propensity score matching as a sensitivity analysis, using a fairly tight caliper for the match. They also calculated restricted mean survival time, which is a nice touch — it captures the cumulative area-between-curves survival benefit over a fixed time window, which can be more clinically intuitive than a hazard ratio alone, especially when Kaplan-Meier curves look visually close but are still meaningfully separated over time.

On the numbers: out of 1,776 patients, just over half received postoperative radiotherapy. Radiotherapy was associated with significantly longer overall survival in the unmatched cohort, and that held up after multivariable adjustment, with an adjusted hazard ratio a bit above point eight — a modest but real reduction in death risk. After matching into 655 well-balanced pairs, the benefit was confirmed and remained statistically significant. In absolute terms, which is really what you want for patient conversations: three-year overall survival was about sixty-two percent with radiotherapy versus fifty-eight percent without, a four-point absolute difference, translating to a number needed to treat around twenty-five. At five years, that gap widened to about fifty percent versus forty-two percent, an eight-point absolute difference, with a number needed to treat around thirteen. The restricted mean survival time gain through five years was about three months cumulative — modest, but the authors argue, reasonably, that this is a clinically meaningful survival advantage rather than a statistical artifact, particularly given the improving radiotherapy morbidity profile with modern intensity-modulated and volumetric-modulated techniques.

The subgroup analysis is where this becomes actionable for risk stratification. Benefit was clearest in patients over sixty-five, tumors larger than two centimeters, T2-through-T4 stage, more than four positive nodes, cutaneous primary site, and more recent diagnosis years. Conversely, no survival difference was detected in the lower-risk subgroups — tumors two centimeters or smaller, T1 stage, minimal nodal disease — which the authors explicitly suggest may be candidates for selective de-escalation, meaning skipping radiotherapy in favor of surveillance in truly low-risk resected disease.

Limitations are the ones you'd expect from any SEER-based analysis: residual confounding that even propensity matching can't fully eliminate, and no granularity on radiation dose, field design, technique, patient performance status, or comorbidities — all of which plausibly affect both the decision to treat and the outcome.

Practically, I'd frame this as clinically informative and reasonably practice-reinforcing rather than practice-changing in a dramatic sense — it doesn't overturn existing consensus that postoperative radiotherapy benefits high-risk resected head and neck Merkel cell carcinoma, but it does give you cleaner absolute numbers to use in shared decision-making conversations, and the subgroup data offers a reasonably evidence-based rationale for considering de-escalation in genuinely low-risk, node-negative, small primary disease — while reinforcing that older patients, larger tumors, higher nodal burden, and cutaneous primaries are exactly where you should be pushing hardest for adjuvant radiotherapy.

That wraps our four pieces for this issue — a instructive back-and-forth on analytic philosophy in patient-reported outcomes research, a genuinely useful refinement of confocal diagnostics across the lentigo maligna spectrum that's worth watching for external validation, and solid registry-level reassurance on risk-adapted radiotherapy in head and neck Merkel cell carcinoma. Thanks for listening, and we'll see you next month.