Scientific Abstract Presentations: Research & Technology Session 7 Saturday, October 17 Sat, Oct 17 5:00-6:00 p.m. CDT

Background: The integrated plastic and reconstructive surgery (PRS) residency match is among the most competitive in medicine, prompting applicants to optimize every component of their application. Although both objective and subjective predictors of match success have been described (1), increasing standardization of applicant profiles makes differentiation challenging. Clarifying which factors evaluators prioritize, and whether applicants' perceptions align with those priorities, may improve advising and promote authentic alignment. Therefore, the objective of this study was to assess the alignment between applicants' and evaluators' perceptions of the relative importance of application components in the integrated PRS match.

Methods: An anonymous cross-sectional Qualtrics survey was distributed to medical school plastic surgery interest groups and ACGME-accredited integrated PRS programs. Respondents self-identified as Applicants (medical students) or Evaluators (residents, faculty, program directors, associate program directors). Participants rated 17 application components on a 7-point Likert scale and completed two optional free-text questions identifying elements they believed receive disproportionate emphasis or insufficient emphasis in the selection process. Group means were compared using Welch's t-tests with Cohen's d effect sizes (0.2 small, 0.5 moderate, 0.8 large). Rank-order agreement was assessed using Spearman's rho. Statistical significance was set at p<0.05. Free-text responses were analyzed using thematic coding.

Results: A total of 50 responses were analyzed, including 21 Applicants and 29 Evaluators. Applicants and Evaluators demonstrated strong rank-order agreement in prioritization of the 17 application components (rho=0.84, p<0.001). Across all respondents, Step 1 pass (6.34±1.39), letter of recommendation quality (5.96±1.08), personality fit (5.85±1.37), and Step 2 score (5.66±1.27) were rated most important. Completion of a dedicated research year was rated least important (2.89±1.23). Between-group comparisons revealed Applicants rated medical school reputation (4.43 vs 3.60; d=0.70; p=0.021) and geographic ties (4.57 vs 3.44; d=0.67; p=0.024) higher than Evaluators. Additional small-to-moderate effect sizes were observed: Applicants emphasized first-author (d=0.58; p=0.06) and total publications (d=0.49; p=0.10), whereas Evaluators emphasized personality fit (d=0.51; p=0.08) and clerkship performance (d=0.33; p=0.27). Thirty-seven respondents (15 Applicants and 21 Evaluators) completed the free-text questions. Research quantity was most frequently identified as disproportionately emphasized in the selection process by both Applicants (n=9, 60%) and Evaluators (n=14, 67%). In contrast, Evaluators most frequently identified clinical performance and interpersonal qualities as insufficiently emphasized (n=15, 71%), with representative themes including authenticity, genuine interest, willingness to help others, curiosity, grit, and longitudinal performance during medical school.

Conclusion: Applicants and evaluators largely agree on the hierarchy of important application components in the integrated PRS match. However, applicants appear to overestimate the importance of institutional prestige and publication volume, whereas evaluators place comparatively greater emphasis on interpersonal qualities and clinical performance. These findings suggest that recalibrating advising toward sustained clinical excellence and authentic engagement may better align applicant effort with evaluator priorities and promote a more balanced and sustainable match process.

References: 1. Wang CY, Mellia JA, Levy L, et al. The Association of a Research Year With Matching Into an Integrated Plastic Surgery Residency. Journal of Surgical Research. 2024;303:22-31. doi:10.1016/j.jss.2024.08.010

Purpose: During preoperative counseling for cleft lip repair, families are typically shown representative postoperative photographs of prior patients. These images are not patient-specific and may not reflect individual anatomy, contributing to uncertainty regarding expected outcomes. Recent advances in multimodal generative artificial intelligence enable photorealistic image synthesis from combined visual and textual inputs. We evaluated whether a contemporary generative AI model could produce anatomically plausible, patient-specific simulations of postoperative outcomes following cleft lip repair.

Methods: Thirty-nine standardized preoperative frontal facial photographs and corresponding postoperative images were identified from published literature for unilateral cleft lip repairs performed using Mohler, Millard, or Fisher techniques. Preoperative images were input into a multimodal generative AI platform (Gemini 3 Pro Image) along with a structured prompt specifying preservation of facial proportions and expected postoperative features including philtral alignment, Cupid's bow reconstruction, nasal symmetry, and scar appearance. The same prompt was utilized for all pre-operative images, with the postoperative timepoint and surgical technique being specified to match each individual case. AI-generated outputs were compared with the corresponding postoperative photographs. Objective similarity was quantified using the Learned Perceptual Image Patch Similarity (LPIPS) metric, where lower scores indicate greater similarity. Subjective similarity was evaluated through an anonymous survey of twelve medical students using Likert-scale items adapted from the Asher-McDade Aesthetic Index and the Scar Cosmesis Assessment and Rating scale (1 = identical, 5 = not similar). Participants also completed a forced-choice task to distinguish AI-generated from real postoperative images.

Results: AI-generated simulations demonstrated moderate perceptual similarity to true postoperative outcomes. Mean LPIPS scores differed by surgical technique (Mohler 0.341 ± 0.095; Fisher 0.413 ± 0.062; Millard 0.459 ± 0.084; ANOVA p = 0.007), with Mohler repairs demonstrating significantly lower LPIPS scores than Millard repairs (p = 0.005). No association was observed between postoperative timepoint and LPIPS similarity (R² = 0.016, p = 0.44).

In discrimination testing, AI images were correctly identified in 43.1% of cases, indicating frequent visual indistinguishability from true postoperative photographs. Identification rates differed by technique (Mohler 20.8%, Millard 54.2%, Fisher 50.0%; χ² p = 0.038).

Subjective ratings demonstrated highest similarity for Fisher repairs (2.67 ± 0.77), outperforming Millard (p = 0.001) and Mohler (p = 0.008). AI simulation performed best in structural feature replication, particularly nasal form and symmetry, while scar pigmentation and erythema were less accurately reproduced.

Conclusions: Generative AI demonstrates promising capability to produce photorealistic simulations of postoperative cleft lip repair outcomes with moderate perceptual similarity to true results. Structural facial features were reproduced more accurately than scar characteristics, and performance varied across surgical techniques. AI-generated simulations may serve as adjunctive tools for patient counseling and expectation management, although further validation is required prior to clinical implementation.

Background: Burn injuries remain a major global public health concern, accounting for substantial morbidity and mortality across all age groups. Early and effective debridement is critical for preventing infection, promoting wound healing, and optimizing graft outcomes. Hydrosurgical debridement is a relatively novel technique that employs a high-pressure saline jet to selectively remove devitalized tissue while preserving viable structures. Although this technology has been proposed to improve surgical precision and clinical outcomes, the available evidence comparing hydrosurgery with conventional debridement techniques remains heterogeneous and inconclusive.

Methods: We conducted a systematic review in accordance with established methodological standards. PubMed, Embase, and the Cochrane Library were searched through December 2025 for randomized controlled trials and cohort studies comparing hydrosurgical versus conventional debridement in paediatric and adult acute burn patients in multiple centers. The primary outcome was wound healing time. Secondary outcomes were skin graft survival and operative time. Pooled mean differences (MD) with 95% confidence intervals (CI) were calculated using random-effects models. Statistical heterogeneity was assessed using the I² statistic, and prediction intervals were estimated using R software.

Experience and results: Four studies comprising a total of 513 patients met the inclusion criteria. Hydrosurgery (HS) was associated with a faster wound healing time compared with conventional techniques (MD −2.03 days; 95% CI −4.22 to 0.16; p = 0.07; I² = 92.6%) (1,2,3) although this did not reach statistical significance and heterogeneity was considerable. Regarding secondary outcomes, conventional excision (CE) was associated with better results related to skin graft survival than HS (MD 0.03; 95% CI 0.02 to 0.05; p < 0.01; I² = 0%)1. No significant difference was observed in operative time (MD −12.36 minutes, 95% CI −48.26 to 23.54; p = 0.50; I² = 86.7%)1,2 between HS and CE.

Conclusions: Hydrosurgical debridement may offer clinical advantages in burn management, particularly in terms of graft survival, with a potential trend toward reduced healing time. However, substantial heterogeneity and wide prediction intervals limit the certainty of these findings. Further large-scale, high-quality randomized trials, as well as cost-effectiveness analyses, are warranted before widespread implementation in routine clinical practice can be firmly recommended.

1. Cao YL, Liu ZC, Chen XL. Efficacy of hydrosurgical excision combined with skin grafting in the treatment of deep partial-thickness and full-thickness burns: A two-year retrospective study. Burns. 2022;49(5):1087-1095. doi:10.1016/j.burns.2022.07.012

2. Gravante G, Delogu D, Esposito G, Montone A. VersaJet Hydrosurgery versus Classic Escharectomy for burn Débridment: A Prospective randomized trial. Journal of Burn Care & Research. 2007;28(5):720-724. doi:10.1097/bcr.0b013e318148c9bd

3. Legemate CM, Kwa K a. A, Goei H, et al. Hydrosurgical and conventional debridement of burns: randomized clinical trial. British Journal of Surgery. 2021;109(4):332-339. doi:10.1093/bjs/znab470

Introduction: Medical adhesive-related skin reactions remain an underrecognized complication of surgical wound closure, with reported rates ranging from 3% to 60%. Silk fibroin, a protein from Bombyx mori, has emerged as an alternative dressing material with favorable immunologic properties. We systematically reviewed the clinical evidence for silk fibroin dressings in postoperative wound closure. Methods: Following PRISMA guidelines, we searched PubMed, Embase, Scopus, and Web of Science for human clinical studies evaluating silk fibroin as a topical dressing or closure device on surgical wounds. Animal-only, in vitro, suture-only, and non-surgical wound studies were excluded. Reviewers independently screened 1,665 records. Google Trends data were analyzed to assess temporal search interest. Data were analyzed with Python (v3.12.3). Results: Four studies met inclusion criteria from 93 full-text reviews. Four studies with extractable data (2 RCTs, 2 retrospective cohorts; N = 591) were analyzed (Table 1). Across all studies, allergic contact dermatitis was 0% with silk fibroin versus 6.4-52% with synthetic controls (all P < .05) (Table 2, Figure 2). Silk dressings were associated with reduced pharmaceutical intervention, faster application, and cost savings up to $466 per case (Figure 3). Google Trends data showed an approximately 10-fold increase in U.S. search interest for "silk fibroin" since 2023 (Figure 1). Conclusion: Preliminary evidence suggest silk fibroin dressings may reduce allergic contact across surgical closure. Evidence for long-term scar quality, barrier function, and infection remains limited (Figure 4). Multicenter randomized trials with standardized outcomes are needed.

Purpose: Burn depth assessment is central to surgical decision-making, both for dedicated burn services and plastic surgery teams. Yet even experienced clinicians achieve only 60–75% accuracy when distinguishing superficial partial-thickness (SPT) from deep partial-thickness (DPT) burns. The SPT–DPT boundary determines whether a wound will epithelialize spontaneously or require excision and grafting. Misclassification drives unnecessary inter-facility transfers, disrupts operative schedules, and delays definitive care, with downstream consequences for scarring and reconstructive burden. Artificial intelligence (AI) has emerged as a potential tool to augment bedside assessment. This systematic review evaluates the performance, limitations, and clinical relevance of AI models targeting the SPT–DPT distinction, with a focus on applications to plastic surgery triage, operative decision-making, and surgical planning. Methods: Following PRISMA guidelines and PROSPERO registration (CRD420251077289), PubMed, Embase, Web of Science, and IEEE Xplore were searched from inception through May 2025. Original human studies using AI for burn depth estimation were included when SPT versus DPT classification was an explicit task boundary. After screening 3,197 titles/abstracts and 215 full texts, 19 studies met inclusion criteria. Accuracy, sensitivity, specificity, F1 score, and mean intersection over union were extracted or reconstructed from confusion matrices, and multivariable regression explored the relationship between augmentation, model type, and reported accuracy. Results: Image-level classification models reported overall accuracies of 73–98%, while pixel-wise segmentation models ranged from 70–95%. Models reporting ≥90% accuracy often relied on small datasets with high developmental training‑to‑testing ratios, while a larger, more heterogeneous dataset yielded lower accuracy, highlighting how dataset structure shapes reported performance. Most studies (15/19) used standard RGB photographs; only two incorporated hyperspectral or multispectral imaging. Higher augmentation ratios were independently associated with increased reported accuracy, raising concern for data leakage and overfitting, whereas classification type (image-wise vs pixel-wise) did not independently predict accuracy. Reporting of unique patient counts, skin tone distribution, and imaging standardization was sparse, and reference standards varied from clinician labels to healing-time–based outcomes. Conclusions: For plastic surgeons, AI-based burn depth tools offer two primary potential use cases. Image-level models may function as telehealth triage adjuncts in community settings to better distinguish burns that truly warrant transfer for operative evaluation, reducing overtriage and preserving burn center and plastic surgery capacity for patients most likely to require excision and grafting. Pixel-wise segmentation models that map depth across a wound could assist with operative planning by delineating zones for tangential excision, estimating graft size, and anticipating donor-site requirements. However, current systems are not yet ready for routine clinical integration: small, non-diverse datasets, heterogeneous reference standards, aggressive augmentation, and limited external validation undermine generalizability. Progress toward clinical adoption will require multicenter, demographically diverse, outcome-anchored image repositories and transparent evaluation frameworks that demonstrate not only accuracy but also meaningful improvements in triage, operative decision-making, and resource utilization in plastic surgery practice.

Purpose: Ultrasound is an increasingly valuable noninvasive tool in plastic surgery, supporting flap planning, intraoperative decision-making, and postoperative monitoring (1,2). Despite these advantages, there is no structured training in ultrasound for plastic surgery residents. Other specialties, such as emergency medicine, have mandated ultrasound competency requirements, whereas plastic surgery still relies on an informal approach to learning this skill, as no standardized curriculum exists (3).

As the clinical utility of ultrasound continues to expand in plastic surgery, its adoption is increasingly driven not only by surgeon preference but also by regulatory requirements. For example, Florida law mandates the use of ultrasound guidance during gluteal fat grafting to ensure safe subcutaneous injection, reflecting a legislative recognition of ultrasound as a standard of care for certain cosmetic procedures (4). However, despite these mandates, ultrasound training remains inconsistently integrated into plastic surgery residency education (5).

This disconnect between regulatory expectations and formal training highlights the need for structured ultrasound curricula within residency programs. Integrating ultrasound education into plastic surgery training would better align with contemporary clinical practice, equip residents with essential diagnostic and procedural skills, and support safer patient care.

Methods: To bridge this gap, a single-day Plastic Surgery Ultrasound (PSUS) curriculum was developed in consultation with a critical care physician experienced in resident ultrasound education. The curriculum was designed to provide plastic surgery residents (PGY1–6) with foundational skills in ultrasound physics, probe handling, Doppler flow assessment, and recognition of postoperative complications.

The course was structured into a morning and an afternoon session. The morning session focused on Doppler fundamentals, femoral anatomy, and superficial and lymph node anatomy. The afternoon session emphasized the sonographic identification of common postoperative pathologies, including hematomas, seromas, and abscesses. Teaching modalities throughout the day included didactic lectures, live scanning demonstrations, and faculty-supervised hands-on scanning practice.

Curriculum effectiveness was assessed using pre- and post-course self-assessment surveys evaluating resident confidence across multiple ultrasound domains, as well as a pre- and post-training ultrasound knowledge assessment.

Results: Among residents who completed the assessments, mean ultrasound knowledge scores increased from 80.75% pre-training to 86.53% post-training, representing a mean improvement of 5.78 percentage points. Individual score changes were variable. Self-reported confidence improved across all assessed domains following the curriculum. Mean Likert-scale confidence scores increased from approximately 2–3 pre-training to 4–5 post-training in operating the ultrasound machine, identifying anatomy, using Doppler, interpreting sonographic findings, and applying ultrasound to flap care.

Conclusion: Integrating a structured ultrasound curriculum into plastic surgery residency represents an important step toward addressing the current lack of standardized training in this modality. Incorporating ultrasound training early in residency fosters technical competency, reinforces anatomic understanding, and promotes independent, evidence-based assessment of reconstructive patients. By strengthening resident proficiency in these skills, training programs can better prepare surgeons for clinical practice while supporting earlier recognition of complications and potentially improving patient outcomes. This pilot study addresses a critical educational gap and supports the integration of formal ultrasound training into plastic surgery residency curricula.

References 1. Mortada H, Al Mazrou F, Alghareeb A, AlEnezi M, Alalawi S, Neel OF. Overview of the role of ultrasound imaging applications in plastic and reconstructive surgery: is ultrasound imaging the stethoscope of a plastic surgeon? A narrative review of the literature. Eur J Plast Surg. 2023;46(1):15-24. doi:10.1007/s00238-022-01981-y 2. Oni G, Chow W, Ramakrishnan V, Griffiths M. Plastic Surgeon-Led Ultrasound. Plast Reconstr Surg. 2018;141(2):300e-309e. doi:10.1097/PRS.0000000000004071 3. Jang TB, Coates WC, Liu YT. The Competency-Based Mandate for Emergency Bedside Sonography Training and a Tale of Two Residency Programs. J Ultrasound Med. 2012;31(4):515-521. doi:10.7863/jum.2012.31.4.515 4. Chapter 458 - 2024 Florida Statutes - The Florida Senate. https://www.flsenate.gov/Laws/Statutes/2024/Chapter458 5. Miller JP, Carney MJ, Lim S, Lindsey JT. Ultrasound and Plastic Surgery: Clinical Applications of the Newest Technology. Ann Plast Surg. 2018;80(6S):S356. doi:10.1097/SAP.0000000000001422

Background: Recent studies have explored how artificial intelligence (AI) can enhance standards of care and lower costs by supporting administrative functions in healthcare. As a developing field, lymphatic surgery lacks standardized billing and coding across physicians and institutions, and only one recently introduced specific CPT code exists for lymphovenous bypass (LVB). This study aims to describe the capabilities of large language models (LLMs) in assigning Current Procedural Terminology (CPT) codes to LVB operative notes.

Methods: We evaluated three LLMs (ChatGPT 5.2, Google Gemini 3.0, and OpenEvidence), including both fast/instant and thinking versions, for assigning CPT codes to 10 de-identified LVB operative notes from a single surgeon. Responses were scored as correct (1) if the recommended code(s) matched the 2026 CPT reference code, partially correct (0.5) if the response aligned with survey-identified commonly used LVB CPT codes, and incorrect (0) otherwise. Overall differences in scores were evaluated with the Friedman rank-sum test, followed by pairwise Wilcoxon signed-rank tests (two-sided; α = 0.05).

Results: ChatGPT 5.2 (Thinking) achieved the highest total score (10.0), followed by Gemini 3.0 (Thinking) (7.0). The remaining models had lower and comparable scores: ChatGPT 5.2 (Instant) 4.5, Gemini 3.0 (Fast) 4.0, and OpenEvidence 4.5. ChatGPT 5.2 (Thinking) significantly outperformed ChatGPT 5.2 (Instant), Gemini 3.0 (Fast), and OpenEvidence (all p < 0.05). No other pairwise comparisons reached statistical significance, including Gemini 3.0 (Thinking) versus Gemini 3.0 (Fast).

Conclusion: The lymphatic specialty coding practices are currently evolving with the recent introduction of an LVB-specific CPT code, LLMs demonstrated variable accuracy in extracting appropriate CPT codes from operative notes. Thinking-mode models performed best, supporting continued investigation of LLMs as decision-support tools to promote coding consistency and administrative efficiency in lymphatic surgery. Given the promising future of HIPAA-compliant AI integration within electronic medical record systems (e.g., EPIC), this study provides preliminary evidence evaluating the ability of language models to interpret operative documentation and accurately associate surgical procedures with appropriate CPT coding.

Background: Medical student perceptions of plastic surgery, whether influenced by education, exposure, or stereotypes, determine specialty recruitment and career selection. Formal education about plastic surgery has been suggested, yet the delineation of student perspectives remains poorly characterized. This study investigates how medical students perceive plastic surgery to identify educational targets for intervention. Methods: A PubMed-based narrative literature review was conducted using search terms related to plastic surgery, reconstructive surgery, and medical student perceptions. Twenty articles were identified and reviewed for relevance; 9 studies containing extractable data on medical student knowledge and perceptions of plastic surgery were analyzed across United States (U.S.) and international training contexts. Results: Across studies encompassing over 4,000 medical students at institutions in the U.S., Israel, Saudi Arabia, Kuwait, and India, students showed a consistent pattern in attributing surgical procedures to the managing specialty. Aesthetic and cosmetic procedures were reliably attributed to plastic surgery across all populations, with correct identification exceeding 90% in U.S. and Israeli cohorts. Medical student recognition declined sharply for reconstructive subspecialties: in the largest U.S. study (n=2,434, 44 medical schools), correct identification was 66.0% for general reconstruction and breast, 51.0% for craniofacial, and only 33.4% for hand and lower extremity procedures. Plastic surgery was selected for hand and peripheral nerve conditions by fewer than 30% of students across multiple studies, with carpal tunnel syndrome (6-28%), brachial plexus injury (4-15%), and scaphoid fracture (10%) among the least recognized. Burns were more consistently attributed to plastic surgery (70-100%) though recognition varied by institution. Clinical exposure to plastic surgery improved student performance across reconstructive domains including hand/peripheral nerve, craniofacial, burn/reconstruction, breast/cosmetic, and craniofacial (p<0.001 to p=0.007). Even so, approximately half of students remained unable to identify plastic surgery as the managing specialty for complex reconstructive scenarios following clinical exposure and the presence of a home plastic training program did not independently improve misattribution (53% vs. 53.5%, p=0.936). Media was the most frequently cited influence on student perceptions, cited more often than direct clinical interaction or personal experience. Interest in plastic surgery as a career was not related to knowledge of the specialty's scope. Conclusion: Medical students consistently underestimate the reconstructive scope of plastic surgery. Medical education leaves significant knowledge gaps regarding the full scope of plastic surgery and the breadth of reconstruction. This gap persists despite clinical exposure and is not resolved by institutional program presence alone. Medical schools should incorporate structured didactic education into undergraduate curricula addressing the full scope of plastic surgery, particularly hand, peripheral nerve, and craniofacial subspecialties, to improve future referral accuracy and support workforce development.

Purpose Training language models for clinical tasks requires a lot of data, but obtaining sufficient data can be challenging due to limited availability or restricted access. For example, sorting cleft and craniofacial procedures by type based on the operative note requires adequate examples of notes for each procedure type. However, certain patient subgroups or surgical techniques may be underrepresented. In such cases of data scarcity, augmenting datasets with synthetically generated notes can increase sample size for training the language model. Synthetic data can also safeguard patient privacy by providing detailed, representative text without exposing protected health information. We evaluate the ability of generative artificial intelligence models, such as GPT-4o, to create text that closely mirrors the structure and language contained in real-world operative notes. We additionally assess the utility of synthetically generated operative notes for enhancement of the performance of a classification model on a prediction task.

Methods and Results Operative notes from 656 patients undergoing cleft lip/palate repair were collected from a multidisciplinary research network. Some operative procedures were underrepresented (e.g., only 95 patients undergoing primary cleft nasolabial repair); thus, to augment the dataset, an additional 95 synthetic operative notes were generated using multishot prompting with GPT-4. The linguistic characteristics of the synthetic vs. real notes were compared using three metrics. BERT-Score indicated that synthetic notes retained much of the meaning found in real notes (precision 0.78, recall 0.81, and F1 score 0.79). Jensen-Shannon Divergence (0.14) indicated that the syntactic style of synthetic notes was comparable to that of real notes. BLEU score (0.073) reflected differences in lexical overlap between the two note types. To assess the utility of synthetic notes for augmenting training data for predictive models, a 70%-30% train-test split was employed. Word embeddings were generated with ClinicalBERT, and logistic regression models were trained to differentiate primary cleft nasolabial repair from other cleft-related procedures. Model performance was appraised under two conditions: (a) training on real notes only; (b) training on real notes plus synthetic notes. Additionally, we evaluated the impact of data scarcity by further restricting the training set to include only 10% of real notes. Performance was evaluated based on accuracy and Area Under the Curve (AUC). Using the full dataset of real notes for training, the model achieved an accuracy of 94.7% and an AUC of 0.91, which remained unchanged after addition of synthetic notes. In the data-scarcity scenario, the model achieved an accuracy of 88.6% and AUC of 0.73, which improved to 90.1% and 0.89, respectively, after addition of synthetic notes. This supports our expectation that synthetic data provides valuable augmentation for data-limited situations.

Conclusions Synthetic notes may enable robust development of clinically useful NLP tools in rare procedure settings, as they exhibit consistent meaning preservation and style similarity compared to notes extracted from patient data. However, they demonstrate semantic divergence, indicating that while essential meaning is captured, the surface-level language and syntax deviate from real operative notes. Despite these differences, synthetic notes can augment training datasets in data-scarce conditions and improve model performance.

PURPOSE: Hidradenitis suppurativa (HS) tunnels are epithelialized tracts that drive pain, chronic drainage, and scarring and often require wide excision and reconstruction (1,2). Prior work suggests tunnels are immunologically active and that HS lesions feature IL-23/Th17 inflammation and stromal fibrotic programs, including Hippo-associated signaling (3,4,5). We built a consensus single-cell tunnel atlas to define recurrent fibro-inflammatory microenvironmental endotypes and resolve the cellular programs that may explain medical recalcitrance and surgical necessity.

METHODS: We constructed a unified single-cell atlas of 46 lesional hidradenitis suppurativa samples drawn from 8 independent cohorts, including whole-skin and enriched epidermal, dermal, immune, keratinocyte, and follicular/tunnel fractions. After standardized quality control, normalization, and highly variable gene selection, Harmony integration generated a shared latent space. We applied unsupervised clustering, cell-type annotation, and gene-set scoring to assign sample-level dominance of inflammatory and fibrotic programs.

RESULTS: Integrated embeddings showed appropriate intermixing by study within major lineages, with enriched fractions localizing to expected myeloid, lymphoid, and epithelial neighborhoods, supporting preserved biological structure across platforms. Within tunneled lesions, we identified recurrent endotypes across cohorts: a pyogenic myeloid endotype enriched for IL1B/CXCL8/NLRP3; a Th17/JAK endotype with IL17/IL23/STAT activity; and a fibro-ECM endotype enriched for TGFβ/PDGF/collagen and Hippo-associated stromal signatures. Pathway scores demonstrated marked inter-patient heterogeneity, with samples dominated by distinct programs despite similar clinical labels.

CONCLUSIONS: As the largest integrated single-cell atlas of HS tunnels to date, these findings support a shift from empiric therapeutic escalation toward endotype-informed, mechanism-aligned management. A fibrosis-dominant, low-inflammatory tunnel state explains biologic non-response and supports earlier wide excision and reconstruction as definitive therapy in selected patients. Atlas-informed stratification could guide operative timing, reconstruction planning, and rational perioperative pairing with targeted agents in prospective, endotype-guided studies.

REFERENCES: 1. Alikhan A, Sayed C, Alavi A, et al. North American clinical management guidelines for hidradenitis suppurativa: A publication from the United States and Canadian Hidradenitis Suppurativa Foundations: Part I: Diagnosis, evaluation, and the use of complementary and procedural management. J Am Acad Dermatol. 2019;81(1):76-90. doi:10.1016/j.jaad.2019.02.067 2. Mehdizadeh A, Hazen PG, Bechara FG, et al. Recurrence of hidradenitis suppurativa after surgical management: A systematic review and meta-analysis. J Am Acad Dermatol. 2015;73(5 suppl 1):S70-7. doi:10.1016/j.jaad.2015.07.044 3. Navrazhina K, Frew JW, Gilleaudeau P, Sullivan-Whalen M, Garcet S, Krueger JG. Epithelialized tunnels are a source of inflammation in hidradenitis suppurativa. J Allergy Clin Immunol. 2021;147(6):2213-24. doi:10.1016/j.jaci.2020.12.651 4. Schlapbach C, Hänni T, Yawalkar N, Hunger RE. Expression of the IL-23/Th17 pathway in lesions of hidradenitis suppurativa. J Am Acad Dermatol. 2011;65(4):790-8. doi:10.1016/j.jaad.2010.07.010 5. van Straalen KR, Ma F, Tsou PS, et al. Single-cell sequencing reveals Hippo signaling as a driver of fibrosis in hidradenitis suppurativa. J Clin Invest. 2024;134(3):e169225. doi:10.1172/JCI169225