Welcome to this July twenty twenty-six review from the Journal of the American Academy of Dermatology. In this episode we're covering four articles relevant to surgical and diagnostic practice in dermatologic oncology: a brief report on barbed suture closure technique, a comparison of natural language processing versus International Classification of Diseases, or ICD, coding for skin cancer subtype identification, a TriNetX-based cohort study on genetic predisposition and melanoma outcomes, and a retrospective cohort study on dermatologist-performed skin cancer screening outcomes. Let's get into it.

Our first article is titled "Barbed Suture-Assisted High-Tension Cutaneous Closure: Brief Report of Four Hundred Ninety-Three Surgical Defects," by Kylie Andonian and John Strasswimmer, from Florida Atlantic University and Dermatology Associates of the Palm Beaches. The clinical question here is whether a barbed suture technique can achieve dermis-to-dermis approximation in high-tension defects where extensive undermining is undesirable, such as in fragile, photodamaged skin or in locations like the upper back, shoulder, or sternal chest, without resorting to skin grafting. This was a retrospective, single-surgeon case series of four hundred ninety-three consecutive surgical defects, without a control group and without institutional review board approval, since that wasn't required for this design. The technique used three-oh or four-oh absorbable monofilament Quill Monoderm, a polyglycolic acid-polycaprolactone suture, on a PS-2 needle, with occasional use of three-oh or four-oh V-Loc PDO, or polydioxanone, when longer-lasting support was needed. The barbed suture was placed as a running deep dermal or subcutaneous plication, secured with an initial locking loop, with additional deep layers added as necessary to achieve tension-free dermal apposition prior to superficial closure. In terms of numbers, mean defect area was eight point one square centimeters with a standard deviation of nine point eight, median four point eight four square centimeters, interquartile range three point one five to nine point one square centimeters, and a range spanning zero point three six up to ninety-three point four five square centimeters. The key finding is that most wounds healed without major complication and no dehiscence was documented across this series. Spitting sutures were noted more often with the longer-absorbing PDO product, which led the authors to favor the shorter-duration Monoderm, although product-specific spitting rates were not systematically collected. The most common adverse event was unexpected postoperative pain, particularly following aggressive deep plication in nerve-sensitive areas such as the postauricular region. The evidence-based takeaway for surgical practice is that barbed suture technique distributes tension progressively along the closure rather than concentrating it at knot points, which may allow avoidance of extensive undermining or grafting in difficult defects, but surgeons should anticipate a learning curve, exercise caution near named sensory nerves, and know that misplaced barbs can be released via a small snip incision if repositioning is needed. Limitations are substantial here: this is a retrospective, single-surgeon series with selective case use, no control group, and no standardized assessment of pain or long-term cosmetic outcome, so these findings should be considered hypothesis-generating rather than definitive.

The second article is "Comparison of Natural Language Processing versus International Classification of Diseases Coding for Skin Cancer Subtype Identification from Pathology Reports," by Malihehsadat Chavooshi and colleagues from Baylor College of Medicine and the Michael E. DeBakey VA Medical Center. The clinical question is whether a natural language processing pipeline applied to unstructured pathology reports can more accurately identify skin cancer subtype and case counts compared with ICD-code-based extrapolation, given that prior work has shown poor correlation between ICD-based estimates and histologically confirmed skin cancers, with a correlation coefficient of only zero point two two. The study design involved retrospective selection of Mohs micrographic surgery case logs from Baylor College of Medicine spanning two thousand seven through twenty twenty-four. Starting from all Mohs cases in twenty twenty and twenty twenty-one, totaling one thousand six hundred fifty cases, one hundred eighty patients were randomly sampled, yielding one thousand two hundred forty-one pathology reports. After exclusions for insufficient data, the final analytic dataset comprised two thousand seven hundred ninety-three lesions from one thousand two hundred forty-one pathology reports across one hundred fifty-two patients. The model used was Bio-ClinicalBERT, a biomedical-text-pretrained transformer model, fine-tuned to classify reports into squamous cell carcinoma, squamous cell carcinoma in situ, basal cell carcinoma, or an "other" category, evaluated using nested five-fold cross-validation against manual chart review as the gold standard. Key findings: overall accuracy was ninety-seven point eight percent. Diagnosis-specific F-one scores were ninety-eight point six percent for both basal cell carcinoma and the "other" category, ninety-seven point two percent for squamous cell carcinoma, and ninety-six point four percent for squamous cell carcinoma in situ. At the lesion level, natural language processing error compared to manual review was under one percent for all subtypes—zero point six two percent for squamous cell carcinoma, zero percent for squamous cell carcinoma in situ, and zero point seven six percent for basal cell carcinoma—whereas ICD-based estimates showed deviations of seventy-five point nine percent for squamous cell carcinoma and one hundred eight point three percent for basal cell carcinoma. At the patient level, natural language processing achieved an accuracy of ninety-seven point four percent versus eighty-four point three percent for ICD coding, an F-one score of ninety-eight percent versus eighty-eight point nine percent, and a Cohen's kappa of zero point nine three versus zero point five zero. ICD codes showed relatively high recall but poor precision, with kappa values as low as zero point zero eight for squamous cell carcinoma detection despite a recall of zero point seven one, while natural language processing maintained patient-level kappa values above zero point eight zero across all subtypes—zero point eight nine for squamous cell carcinoma, zero point nine five for squamous cell carcinoma in situ, and zero point nine six for basal cell carcinoma. Most residual discrepancies involved borderline squamous cell carcinoma versus squamous cell carcinoma in situ distinctions, or invasive versus in situ melanoma terminology. The takeaway for practice and research infrastructure is that a validated natural language processing pipeline can substantially reduce reliance on manual chart review for institutional skin cancer surveillance and outcomes research, and it clearly outperforms ICD-based case identification in precision while maintaining high recall. The main limitation is that this is a single-institution study, which limits generalizability, and the authors note that future work should incorporate multi-institutional validation.

The third article is a research letter titled "Genetic Predisposition to Malignancy Confers Different Outcomes in Malignant Cutaneous Melanoma: A Retrospective Cohort Matched Study Using TriNetX," by George Nassief, David Kaelber, Amy Nowacki, and Joshua Arbesman from the Cleveland Clinic and Case Western Reserve University. The clinical question addressed is how a co-diagnosis of genetic susceptibility to malignancy affects survival in patients with cutaneous melanoma, building on prior data showing germline predisposition accounts for roughly ten point six to fifteen point eight percent of melanoma diagnoses. This was a retrospective cohort study using the TriNetX federated database, identifying melanoma patients via the ICD-ten code C43, and using the code Z15.0, genetic susceptibility to malignant neoplasm, as a surrogate marker for germline predisposition. Propensity score matching was performed on demographic and comorbidity variables likely to affect five-year survival, using a standardized mean difference cutoff of point one zero. Before matching, the cohorts comprised two hundred eighty-seven thousand four hundred twenty-seven melanoma patients without genetic susceptibility and two thousand four hundred twenty-one with genetic susceptibility; after matching, two thousand three hundred sixty-nine patients remained in each arm, with all standardized mean differences at or below point one zero. The key finding was that patients without genetic susceptibility had significantly worse five-year overall survival compared to those with genetic susceptibility, with a hazard ratio of one point four two and a ninety-five percent confidence interval of one point two zero to one point six seven. The authors note this parallels a prior finding by their group that melanoma patients with a positive germline pathogenic variant had better survival, potentially linked to lower peripheral polymorphonuclear myeloid-derived suppressor cell counts, a biomarker associated with poor prognosis in advanced melanoma. They also cite consistency with a multi-cancer study showing improved survival associated with family history of cancer, and a similar pattern in an ovarian cancer cohort with germline predisposition. The practical takeaway is that identified genetic predisposition to malignancy does not appear to confer worse melanoma-specific outcomes, and may in fact be associated with improved survival, possibly reflecting increased medical surveillance or underlying tumor immunobiology, though this requires further mechanistic study. Limitations are important: staging information is unavailable in TriNetX, preventing adjustment for potential stage differences between cohorts, and among patients classified as lacking genetic predisposition, it's impossible to confirm whether genetic testing was performed at all, so misclassification is likely. Residual confounding from unmeasured factors also cannot be excluded despite propensity matching.

The fourth and final article is "Outcomes of Dermatologist-Performed Skin Cancer Screening in an Academic Setting: A Retrospective Cohort Study," by Olivia Burke and colleagues from the University of Miami Miller School of Medicine. The clinical question is what diagnostic yield, biopsy burden, and temporal efficiency trends look like for dermatologist-led total body skin examinations in an unselected, largely low-risk population, given that the US Preventive Services Task Force does not currently recommend for or against routine skin screening in asymptomatic adults. This was a retrospective cohort study of three thousand one hundred twenty-seven screening encounters from one thousand five hundred seventy-two University of Miami employees and spouses undergoing annual total body skin examinations between January twenty nineteen and June twenty twenty-four, analyzed using generalized estimating equations to account for repeated within-patient observations. The mean participant age was fifty-three point four years, sixty-two point two percent were female, and participants averaged one point nine nine visits over the study period. Key findings: premalignant lesions, defined strictly as actinic keratoses, were detected in eleven percent of screening encounters; in situ lesions, comprising six squamous cell carcinoma in situ and one melanoma in situ, were detected in zero point two two percent; and malignant lesions were detected in two point six percent of encounters, breaking down to sixty-nine basal cell carcinomas and twelve squamous cell carcinomas, with all malignant lesions identified at early stages. Cumulative incidence among initially lesion-free patients was six point five percent for premalignant lesions and three point three percent for malignant lesions over follow-up. Incidence rates were three hundred ninety-six per ten thousand person-years for premalignant lesions, nineteen per ten thousand person-years for in situ lesions, and two hundred two per ten thousand person-years for malignant lesions, with basal cell carcinoma alone at one hundred seventy-four per ten thousand person-years and squamous cell carcinoma at twenty-seven point two per ten thousand person-years. On multivariable analysis, independent predictors of premalignant lesion detection included older age, with an odds ratio of one point zero eight per year and a ninety-five percent confidence interval of one point zero five to one point one zero, non-Hispanic ethnicity with an odds ratio of three point six five and confidence interval one point six two to eight point two one, and history of nonmelanoma skin cancer with an odds ratio of two point eight zero and confidence interval one point five two to five point one four. For malignant lesion detection, history of nonmelanoma skin cancer, odds ratio two point seven two, confidence interval one point three two to five point six zero, and number of prior biopsies, odds ratio one point two three, confidence interval one point zero eight to one point three nine, were significant predictors, while immunosuppression showed a trend toward association with an odds ratio of one point six seven that did not reach significance. Notably, patient-reported "spots of concern" had quite limited predictive value: positive predictive values were only two point seven percent for premalignant lesions, zero point five percent for in situ lesions, and one point two percent for malignant lesions, though sensitivity for in situ lesions specifically was higher at eighty-five point seven percent. Overall, only forty-nine of one thousand one hundred nineteen reported spots of concern turned out to be premalignant or malignant, for a combined positive predictive value of four point four percent. Regarding biopsy burden, across all screenings there were three hundred fifty-two benign biopsies and ninety-two malignant biopsies, translating to a number needed to biopsy of four point eight three to detect one cancer, with a GEE-predicted value of four point eight seven and a ninety-five percent confidence interval of three point nine nine to six point zero one. Importantly, efficiency improved substantially across screening rounds: the number needed to biopsy declined from six point one zero in round one to two point seven nine in round five, while the proportion of visits detecting premalignant lesions more than doubled from eight point five percent to eighteen point nine percent, and malignant detection also rose, albeit more modestly. The evidence-based takeaway is that dermatologist-performed total body skin examinations yield meaningfully higher detection efficiency with repeated annual screening, and that yield is concentrated in older patients and those with a prior history of nonmelanoma skin cancer, supporting a risk-stratified rather than universal screening approach, with patient self-identified lesions serving as a poor stand-alone trigger for biopsy decision-making. Limitations include voluntary participation introducing selection bias, variability in biopsy practices among examining dermatologists, and the likelihood that external procedures performed outside the health system were not captured, potentially undercounting malignant detection or biopsy events.

That concludes this month's review. In summary, we saw a technique-focused case series suggesting barbed sutures may reduce reliance on undermining or grafting in high-tension or fragile-skin closures; a informatics study demonstrating that natural language processing substantially outperforms ICD coding for accurate skin cancer subtype ascertainment from pathology text; a large database study suggesting genetic predisposition to malignancy is associated with improved rather than worsened melanoma survival; and a real-world screening cohort supporting risk-based rather than universal total body skin examination strategies, with efficiency improving over successive annual rounds. Thank you for listening, and we'll see you next month.