Welcome back to the Journal Review, your monthly deep dive into the Journal of the American Academy of Dermatology. This is the January 2026 issue, and we've got four pieces to get through — a full original cohort study on multiple primary Merkel cell carcinoma, a brief report pilot study on culturally concordant sun-protection education, and a letter-and-response pair debating the Erasmus MC model for cutaneous squamous cell carcinoma metastatic risk. Let's get into it.

First up, this is a retrospective cohort study out of Xiangya Hospital in China, using SEER data, looking at multiple primary Merkel cell carcinoma — MPMCC — and how it differs from single primary disease in both tumor behavior and survival. You already know the background here better than most: Merkel cell carcinoma is rare, aggressive, neuroendocrine-derived, and its incidence has been climbing, with checkpoint inhibitors having meaningfully improved survival over the last several years. But that survival gain creates a new problem — patients are living long enough to develop second primary MCCs, and the existing follow-up guidelines and prognostic literature are essentially all built around single primary disease. This paper set out to characterize what a second primary MCC looks like and whether it changes the prognostic calculus.

Methodologically, they pulled every cutaneous MCC case from the SEER seventeen-registry database between 2004 and 2021, ending up with just over nine thousand five hundred patients. They defined multiple primary disease using the standard SEER multiple-primary rules — essentially two or more primary MCCs diagnosed more than a year apart, or diagnosed within a year but at different anatomic sites. The statistical approach here is worth pausing on, because it's the methodologically clever part of the paper. When you're comparing survival between patients who develop a second cancer and those who don't, there's a well-known trap called immortal time bias — patients who live long enough to develop a second primary are, by definition, patients who survived long enough to be eligible for that diagnosis, which can artificially bias survival comparisons in either direction. The authors addressed this by modeling MPMCC diagnosis as a time-varying covariate in a Cox model rather than as a fixed baseline characteristic. That's the right call, and it's a good general teaching point — any time you see a survival comparison based on a second event that occurs after time zero, you should be asking whether the authors handled the time-varying nature of that exposure correctly. They also ran sensitivity analyses using different latency thresholds and restricted analysis to localized-stage patients at baseline, which shores up robustness against misclassification of recurrence versus true second primary.

Now the results. Multiple primary disease was uncommon — about one in seventy patients, so roughly one and a half percent — but the relative risk signal is striking: MCC survivors had roughly a sixty-fold increased risk of a second primary MCC compared to the general population, and that risk, while it declined over time, was still about forty-fold elevated even a decade out. So there's no point at which these patients should be considered "in the clear." Timing-wise, about a quarter of second primaries showed up within the first year, but nearly four in ten weren't diagnosed until more than three years after the index tumor, and cumulative incidence didn't plateau until around seventy-eight percent by five years — meaning a meaningful fraction, about one in five, of second primaries are showing up even later than that. That has real follow-up-duration implications. Interestingly, the second tumors tended to be smaller than the first — the median dropped from twenty millimeters to fifteen — which the authors note, though they don't overinterpret it; smaller doesn't mean less lethal, it might just reflect earlier detection through surveillance. And that's exactly the point, because despite smaller size at second diagnosis, overall survival was significantly worse in the multiple primary group, with roughly a two-thirds increase in the hazard of death, and cancer-specific survival was hit even harder — well over a two-fold increase in the hazard of MCC-specific death.

The authors are appropriately measured in their discussion — this is retrospective SEER data, there's inherent risk of misclassifying local recurrence versus true second primary since SEER coding relies on registrars applying multiple-primary rules rather than centralized pathologic re-review, and there's no granularity on immunosuppression status, checkpoint inhibitor use, or Merkel cell polyomavirus status, all of which plausibly modify this risk. But the core finding — that a history of MCC confers a massively elevated risk of a second primary, and that developing one carries worse survival despite smaller tumor size — is clinically actionable. For practice, this is a good argument for indefinite, not just five-year, surveillance in your MCC survivors, and it reinforces that a new lesion in this population should be treated with a low threshold for biopsy regardless of how innocuous or small it looks, since size clearly doesn't correlate with the same prognosis you'd expect in a first primary. I'd call the surveillance-intensity message practice-affirming rather than practice-changing in the sense that most of us already watch these patients closely, but this gives you the actual numbers to justify that vigilance to patients and to tumor boards, and it's a nudge to keep surveillance going well past the point where you might otherwise start spacing out visits.

Next, a brief report — a pilot study out of the Medical College of Wisconsin looking at whether racial and ethnic concordance between medical student presenters and low-income Latine high school students affects the impact of a sun protection education workshop. This isn't a full original study with a hypothesis-testing design in the traditional sense; it's a pilot survey study, so I'll walk through it as it's presented rather than force it into the five-beat structure.

The motivating problem: melanoma incidence has risen appreciably among Latine adolescents over the past two decades, prior work has shown Latine males in particular have lower skin cancer awareness and weaker uptake of sun-protective behavior compared to Latine females, and there's a broader literature showing that patient-provider racial and ethnic concordance improves outcomes in adults — but whether that concordance effect extends to health education delivered to low-income Latine youth was unknown.

The setup was simple and pragmatic: they used a validated hour-long workshop called Sun Protection Outreach Taught by Students, delivered by pairs of medical students, with each session led by either a presenter identifying as underrepresented in medicine — Black or Latine — or a White presenter, with at least one male presenter in every session. They surveyed close to a hundred high schoolers, overwhelmingly Latine and majority female, immediately before and after the workshop.

The headline result is that the workshop itself worked — knowledge, attitudes, and intended behavior around sun protection all improved significantly after the session, which is reassuring but not the interesting part. The interesting part is the presenter-concordance effect: at baseline and after workshops led by White presenters, Latine males showed lower confidence connecting tanning to photoaging, lower knowledge, and weaker intent to adopt sun-protective behavior compared to their female peers — recreating that known gender gap. But in sessions led by underrepresented-in-medicine presenters, that gender gap simply wasn't there. In other words, the concordant presenter appeared to specifically close the gap for the group — Latine males — who are otherwise hardest to reach.

The authors are appropriately humble about this — it's a single-site pilot, they don't have long-term behavioral follow-up, and the subgroup sizes here are small once you slice by gender and presenter type, so I'd treat this as hypothesis-generating rather than definitive. But as a piece of practical program design, it's a useful signal: this isn't practice-changing for how you run your Mohs clinic, but if you're involved in any community skin cancer education or outreach programming — and plenty of us are asked to do exactly that — this is a reasonable evidence-based nudge to prioritize diverse, concordant presenter teams, especially when your target audience includes young Latine males who are otherwise the least engaged group.

Finally, let's cover the letter-and-response pair on the Erasmus MC model for cutaneous squamous cell carcinoma metastatic risk stratification. This is a Notes and Comments exchange, not original data, so I'll walk through the actual points of disagreement rather than imposing a results-and-limitations structure that doesn't exist here.

Quick context for why this exchange matters: the original paper being critiqued combined clinical guideline-defined high-risk groups with the Erasmus MC — or EMC — model to refine absolute three-year metastatic risk prediction in cutaneous squamous cell carcinoma, based on two nationwide nested case-control datasets from the Netherlands and the UK.

The letter from Zheng and Wei raises four concerns. First, generalizability — the Dutch and UK healthcare systems both have structured cancer registries and ready access to specialized dermatopathology, so performance in less-resourced or more ethnically diverse settings remains unproven. Second, and this is the meatiest methodological point, the extremely low metastatic event rate — under two percent — raises concerns about calibration and overfitting; the letter argues that at such low absolute risk, even small misclassification could disproportionately distort risk estimates, and that this needs formal calibration curve assessment rather than just discrimination statistics like the C-statistic. Third, they flag that the model assumes guideline variables like perineural invasion and tumor depth are recorded consistently, when in real-world pathology reporting that ascertainment is known to be heterogeneous. Fourth, they note the absence of decision-analytic evaluation — net benefit, decision curve analysis — meaning it's unclear whether reclassifying patients into new risk strata would actually change outcomes or resource use.

The response from the original authors, Steijlen and colleagues, addresses each point directly, largely by pointing back to the model's original development paper. On generalizability, they agree external validation is needed but note that validating in under-resourced settings requires formally quantifying how similar the discovery and validation populations actually are — you can't just apply a model to a different population and assume equivalence. On the low event rate and calibration concern, they push back somewhat firmly: the model was built specifically using a nested case-control design suited for rare outcomes, calibration was already assessed in an independent English validation cohort using observed-to-expected ratios and calibration slope, and the events-per-variable ratio in development exceeded twenty-four, comfortably above the standard threshold of ten needed for a stable multivariable Cox model. On data heterogeneity in variables like perineural invasion, they concede this is a real limitation of pathology reporting generally, but point out the model still validated successfully in the independent English dataset using real-world pathology reports, which is some reassurance of robustness despite that noise — and they agree molecular biomarkers would be a better long-term solution. And on clinical utility, they clarify that decision curve analysis using expert-informed thresholds was already performed in the original model development publication, and that this newer paper's specific aim was demonstrating incremental value on top of existing guidelines, not re-litigating overall clinical utility.

There's no new data here, so there's nothing practice-changing to extract directly — but as a piece of critical appraisal instruction, this exchange is a nice compact refresher on exactly the questions you should be asking of any risk-prediction model before adopting it clinically: how was it validated, was calibration assessed with the right tools for a rare-event setting, are the input variables reliably ascertainable in your own practice setting, and has anyone shown that using it changes decisions or outcomes rather than just improving a discrimination statistic. For now, the EMC model remains promising but externally unproven outside Dutch and UK cohorts, and I wouldn't call it ready to override your existing guideline-based risk stratification in the office.

That wraps up this January 2026 review — a strong reminder from the Merkel cell paper to keep your surveillance indefinite rather than time-limited, a nice pilot signal on the value of concordant presenters in community education, and a good methodological sparring match on how rigorously we should be interrogating risk-prediction models before we let them change how we counsel patients. Thanks for listening, and I'll see you next month.