Welcome back to this journal review. We're covering the May twenty twenty-six issue of Dermatologic Surgery, the official journal of the American Society for Dermatologic Surgery. Four pieces on the docket today — an original anatomic study on filler spread, a commentary on an artificial intelligence study in Mohs frozen section pathology, a systematic review on delayed full-thickness skin grafting after skin cancer resection, and a technique-and-case-series report on using blepharoplasty skin for Mohs defect closure. Let's get into it.

First up is an original article: "Analysis of Filler Spread: Facial Retinacula Play a Prominent Role," from Schelke and colleagues, a multinational group with heavy overlap in facial ultrasound expertise. The clinical problem here is one you've probably heard patients or injectors describe as filler "migration" — the observation that filler placed at one point can end up somewhere else entirely, which in the aesthetic world gets framed as unpredictable and alarming. The authors' driving question is whether that spread is actually random, or whether it's anatomically constrained and therefore predictable — specifically, whether it's governed by the retinacula superficialis and retinacula profundus, the fibrous septal networks in the superficial and deep fat compartments that you already know well from flap and graft dissection planes.

Methodologically, this is a three-part observational study built entirely around real-time ultrasound guidance, which is really the methodological star of the piece — the authors chose ultrasound specifically because prior work in this space leaned on in vitro rheology testing or cadaveric injection, neither of which can capture how tissue actually deforms and channels filler in a live, mobile, hydrated face. Part one is the primary cohort: four hundred forty injected facial areas across a hundred seven patients, all women, mean age right around fifty-seven, injected by a single experienced injector using two volumizing hyaluronic acid fillers — one cross-linked with the traditional BDDE chemistry, one PEG-cross-linked — plus three biostimulatory agents, poly-L-lactic acid, calcium hydroxylapatite, and a hyaluronic acid–calcium hydroxylapatite hybrid. Part two is a small reference cohort, twenty-two treatment areas from four additional injectors, added specifically to check whether the spread patterns the lead author observed were an artifact of her own technique or genuinely reproducible across different hands. Part three zooms in on the zygomatic arch alone, using a standardized zero-point-two milliliter bolus in a subset of forty areas, precisely because that region gives the cleanest, most reproducible ultrasound window — a horizontal bony plane separating two well-defined fat compartments — making it an ideal site to quantify spread length rather than just describe direction.

The results were strikingly consistent. Regardless of needle-versus-cannula, needle gauge, or injection angle, the direction of filler spread at any given facial site was the same, patient after patient, and matched the architecture of the retinacula in that location — oblique in the midface, temple, and forehead superficial fat, and more parallel to the skin surface along the lateral jawline and in the deep facial fat generally, except the maxillary compartment where the deep retinacula angle obliquely down to the periosteum. Biostimulatory fillers behaved directionally just like the hyaluronic acid fillers — so rheology did not change which way the product traveled. What rheology did affect was distance: the length of bolus spread differed significantly by filler type in the zygomatic arch measurements, meaning thicker or more cohesive products traveled different distances even though they took the same anatomic route.

In their discussion, the authors reframe "migration" as a misnomer — this isn't erratic movement, it's directed flow along a fixed, and honestly quite elegant, connective tissue scaffold, at least in tissue that hasn't been surgically altered. The obvious limitation, and one worth flagging clearly, is generalizability: every primary-cohort patient was female with Fitzpatrick skin types one through three, so we have no data here on how these patterns hold in male facial anatomy, in darker skin types, or — importantly — in patients with prior facelift or other surgery that disrupts the retinacular planes, which the authors explicitly excluded. The reference cohort of twenty-two areas across four injectors is reassuring for reproducibility, but that's still a small corroborating sample, not an independent validation cohort in the statistical sense.

For your practice, this is genuinely useful conceptual grounding even if you're not the one holding the syringe day-to-day — it reinforces that facial fat compartments and their septal boundaries behave the same way whether you're thinking about filler spread, tumor tracking along fascial planes, or flap dissection safety, and it gives you a more precise vocabulary for counseling patients who arrive with filler complications or asking about compartment-based reconstruction. I'd call it more conceptually reinforcing than practice-changing for a Mohs surgeon specifically, but it's a nice piece of applied fascial anatomy.

Next, a communications piece — this is not an original study but a commentary responding to a separate paper by Rios-Duarte and colleagues, which had assessed a deep learning model originally trained on standard formalin-fixed, paraffin-embedded pathology and then tested it on Mohs frozen sections for classifying squamous cell carcinoma. The commentary authors, Qi and Nie, are essentially doing what a good discussant does at a tumor board — walking through what that original study found, unpacking why, and telling you what to make of it.

The core finding they're reacting to: a model trained on standard permanent-section slides performed excellently on the tissue type it was trained on — an area-under-the-curve figure of about zero-point-nine-nine, essentially excellent discrimination — but performance dropped substantially when the same model was applied to frozen sections from actual Mohs cases, down to an area-under-the-curve around zero-point-eight, which is a meaningful, clinically relevant drop, not just a statistical footnote. More concerning for our purposes, sensitivity on frozen sections was poor, only around fifty-five percent, meaning the model missed nearly half of true tumor-positive areas, while specificity was perfect at one hundred percent. The commentary explains the likely mechanism clearly: permanent-section training data, drawn mostly from routine biopsies, is dominated by superficial anatomy, while Mohs frozen sections — especially from later stages — routinely contain deep structures like muscle, nerve, and vessel, plus inflammatory infiltrate, all of which the model apparently confused for tumor. That's a classic out-of-distribution problem in medical imaging artificial intelligence, and the original study apparently backed this up with attention heatmap analysis showing exactly where the model was looking when it got confused. It's also worth noting the original data set was small — only fifteen frozen section slides — and skewed toward tumor-containing specimens, which limits how far any of this should be extrapolated.

The commentary's own contribution is really a set of forward-looking recommendations rather than new data: they suggest domain adaptation, meaning fine-tuning the existing model on a modest number of annotated frozen-section whole-slide images rather than starting from scratch; incorporating clinical metadata such as Mohs stage or tumor location as auxiliary inputs so the model knows to expect deep structures in later stages; and building multi-institutional frozen-section image repositories, since the field so far has concentrated its Mohs artificial intelligence efforts on basal cell carcinoma, leaving squamous cell carcinoma, particularly high-risk variants, underrepresented.

There's no methods or results section to critique here in the traditional sense — it's an editorial-style discussion of someone else's work — so the honest takeaway is this: nothing here is remotely ready for clinical deployment. The perfect specificity is reassuring in that a positive call was reliable, but the mediocre sensitivity means false negatives, and in the Mohs context a false-negative frozen-section read is exactly the error that leads to incomplete excision. File this under interesting and worth watching, not actionable — the message for us is that any commercial frozen-section artificial intelligence tool marketed to Mohs surgeons should be able to show you validation data on actual frozen tissue, not extrapolated performance from permanent sections, before it belongs anywhere near an intraoperative decision.

Third is a systematic review: "Delayed Full-Thickness Skin Grafting After Skin Cancer Resection," from Shah, Shah, Sharma, and Song. The gap they identify is straightforward — split-thickness grafting has a mature evidence base, particularly from the burn literature, but delayed full-thickness grafting after Mohs or wide local excision has never really been consolidated, despite being something most of us do regularly when we want to bank a day or two for granulation or for definitive margin confirmation before committing to a graft.

Methodologically, this followed PRISMA guidelines, searching four major databases, with two independent reviewers screening and a third adjudicating conflicts — standard, appropriate systematic review architecture. The authors pooled data where reporting was consistent enough to allow it, and used Welch's t-test for subgroup comparisons given the very unequal sample sizes across included studies, which is the statistically correct choice when you're comparing groups of wildly different sizes rather than assuming equal variance. It's worth being upfront, and the authors are upfront, about what feeds this review: twenty-four studies, but the level of evidence is not strong — no randomized trials at all, just under half rated Level two-b and the rest Level four, meaning the bulk of the literature is case series, case reports, and retrospective cohorts. That's a real limitation baked into the source material itself, not a flaw introduced by the reviewers.

Across six hundred sixty-five pooled patients, mean age right around sixty-four, roughly even split between men and women, over ninety percent of tumors were basal or squamous cell carcinoma, and the nose was by far the dominant site, at nearly two-thirds of all cases. The headline number: overall graft success was about ninety percent at an average follow-up of thirteen months, and satisfaction data, where reported, was overwhelmingly positive — about three-quarters of patients described themselves as "very satisfied." The clinically important finding, the one that could change how you sequence care, is timing. Grafts placed within the first week had meaningfully lower success, around eighty-two percent, while delaying two to three weeks pushed success up to nearly ninety-nine percent — a real, clinically meaningful gap, not just statistical noise — with delays beyond three weeks actually trending slightly lower again, around ninety-five percent. So the sweet spot the pooled data supports is that two-to-three-week window, not "wait as long as possible."

Location also mattered a great deal. Central face grafts — overwhelmingly nasal — succeeded at about eighty-nine percent, significantly lower than lateral face, upper face, or upper extremity sites, which all clustered near one hundred percent. Lower extremity grafts were the clear outlier on the bad end, succeeding only about eighty-four percent, which tracks with what we already know clinically about lower leg perfusion, venous disease, and weight-bearing mechanical stress undermining graft take. Donor site, by contrast, didn't matter much — lower abdomen and groin donor skin trended slightly higher than periauricular or clavicular donor skin, around ninety-six versus ninety-two percent, but that difference wasn't statistically significant, so no strong donor-site preference is supported here. Interestingly, tumor histology itself — nonmelanoma versus melanoma — made no difference to graft success, which suggests it's wound bed biology and anatomic location driving outcomes, not the underlying diagnosis.

The authors' own discussion is appropriately cautious: this is the first review to consolidate this literature, but it's built on predominantly low-level evidence with heterogeneous reporting, no true comparator arm of immediate versus delayed grafting head-to-head within a single design, and inconsistent comorbidity reporting that leaves confounders like smoking, diabetes, and peripheral vascular disease impossible to fully account for.

Practically, I'd call the timing finding genuinely practice-relevant, if not fully practice-changing, since most experienced Mohs surgeons already lean toward delaying grafts on marginal beds — this gives you pooled numbers to actually cite to a patient or a resident: waiting two to three weeks versus grafting within the first week is the difference between roughly ninety-nine percent and eighty-two percent success. The location data is a useful reminder to counsel patients on the nose and especially the lower leg that success rates, while still generally good, are meaningfully lower and revision may be more likely. The donor-site equivalence is reassuring but not a strong enough signal to override your own preference based on color and texture match.

Last is a technique-and-case-series report: "Mohs Surgery Defect Closure Using Blepharoplasty." This one doesn't follow the original-study scaffolding — there's no hypothesis-testing design here, just a retrospective description of twenty-eight consecutive patients whose periocular, nasal, cheek, or digit Mohs defects were reconstructed using full-thickness skin grafts harvested from the upper eyelid via standard blepharoplasty technique, done as a bilateral procedure even when only one eyelid's tissue was needed, specifically to preserve symmetry.

The rationale the authors lay out is essentially a donor-site argument: upper eyelid skin is the thinnest skin on the body, on the order of a quarter of a millimeter, hairless, pliable, and a close color and texture match to facial skin, and it comes with a built-in high elastin content that resists the wound contracture you'd worry about with a thicker graft. Compared with postauricular or supraclavicular donor grafts, it avoids creating a second visible donor scar elsewhere and instead piggybacks on a procedure — blepharoplasty — that's already cosmetically desirable to many of these patients, several of whom also underwent adjunctive lateral canthoplasty, defatting, or canthopexy to optimize the aesthetic result. Patient selection involved routine ophthalmologic clearance — visual fields, tear production, intraocular pressure, ptosis assessment, extraocular movement — and excluded anyone with insufficient eyelid tissue or who couldn't safely pause anticoagulation.

Across the case series, most defects were basal cell carcinoma, averaging about one and a half square centimeters, ranging up to just over four square centimeters, mostly clustered in the periocular, nasal, and cheek regions. Outcomes were excellent: one hundred percent graft survival, no lagophthalmos, no significant ocular complications, and color-texture match rated excellent in the large majority of cases, around eighty-six percent, with only minor mismatch in the remainder — and even those were manageable, with two patients needing a touch-up pulsed dye laser session to blend graft height into the surrounding skin.

Since this is a technique report rather than a comparative study, there's no formal limitations section to relay, but the authors are honest about the practical trade-offs: this approach demands real surgical facility with periocular anatomy, it risks eyelid-specific complications like asymmetry or partial graft loss from inadequate vascularization, and it's not appropriate for larger defects — they cite roughly four by one and a half centimeters as an upper threshold before you'd need bilateral harvest or should really be reaching for a local flap instead. Older patients may also not clear ophthalmologic screening, which narrows the applicable population.

The practical takeaway here is fairly clear-cut for anyone doing reconstructive work in the periorbital, paranasal, or upper cheek zones with access to a comfortable oculoplastic partnership: for the right small-to-moderate defect in a patient who either wants cosmetic blepharoplasty anyway or simply has redundant upper eyelid skin, this is a legitimate, well-matched, low-morbidity donor option worth having in your reconstructive toolkit alongside postauricular grafting and local flaps — genuinely useful, though it's a niche solution dependent on defect size, location, and access to eyelid surgical expertise, not a wholesale replacement for your existing reconstructive ladder.

That wraps our four articles for May. To summarize the through-line: solid new anatomic grounding for filler behavior, a sobering reminder that frozen-section artificial intelligence is not yet ready for prime time, pooled evidence supporting a two-to-three-week delay when you're banking full-thickness grafts, and one more well-matched donor option for periorbital reconstruction. Thanks for listening, and we'll see you next month.