Citations

Background: Large language models (LLMs) are being used by surgeons for education and reference yet concerns about hallucinations and reliability limit safe adoption. Retrieval-augmented generation (RAG) can offer a potential solution by grounding responses in a high-quality external database (e.g., medical textbooks) to enhance accuracy. However, performance tradeoffs across different RAG configurations—many of which exponentially increase computational cost—remain poorly characterized. Methods: In total, 120 lightweight, open-source RAG configurations were evaluated across 40 plastic surgery-focused question-answering tasks (20 single-hop, 20 multi-hop), spanning multiple subspecialties (4800 total evaluations). Configurations varied by base LLM (Phi-3-mini-128k-instruct vs. BioMistral-7B), embedding model, database size, chunk size, and query hop type. Performance was assessed using semantic similarity (Ragas) to physician-validated reference answers. Performance was analyzed using linear mixed-effects regression with query as a random effect and fixed and interaction effects selected via likelihood testing and AIC. Results: High performance was achievable using lightweight, open-source models. While BioMistral-7B had high mean sematic similarity under specific configurations (mean semantic similarity up to 0.786), Phi-3-mini-128k-instruct demonstrated more consistent performance across query complexity. Larger database sizes significantly improved semantic similarity, with the largest gain at intermediate sizes (e.g., size 5: +0.043, p = 0.001). Embedding choice had a strong effect, with bge-large-en-v1.5 improving performance (p = 0.0016) and Bio_ClinicalBERT markedly reducing it (p < 0.001). Multi-hop queries substantially reduced performance (p < 0.001), though this effect was attenuated for Phi-3-mini-128k-instruct via a strong model × hop-type interaction (p < 0.001). Conclusions: RAG systems for plastic surgery do not require large proprietary models, as performance depends on configuration choices and interaction effects rather than isolated components. With advancements, predictive modeling may enable resource-efficient, safe deployment of clinical RAG systems.

Abstract Introduction: Evidence increasingly shows small nerve fibre pathology in subsets of people with fibromyalgia through skin biopsy, corneal confocal microscopy, and microneurography. However, blood-based biomarkers of ongoing neuronal damage, particularly neurofilament light chain (NfL), an axonal structural protein, are not well established. Objectives: The aim of this study was to evaluate serum NfL concentrations in fibromyalgia compared with other chronic pain conditions and assess their relationship with neuropathic-like pain. Methods: We measured serum NfL using Simoa SR-X NF Light v2 digital-immunoassay in 60 participants with fibromyalgia, 61 with endometriosis (another traditionally considered non-neuropathic pain condition), 24 with small fibre neuropathy (a recognised neuropathic pain condition), and 30 healthy controls. Serum NfL concentrations were compared between the groups controlling for age and body mass index. Neuropathic-like pain was assessed using the painDETECT questionnaire for fibromyalgia and endometriosis and NeuPSIG grading criteria for small fibre neuropathy. Results: Compared with healthy controls (mean concentration = 6.77 pg/mL, Z-score = −1.15), serum NfL concentrations were significantly elevated in fibromyalgia (mean concentration = 8.52 pg/mL P = 0.048; Z-score = −0.30, and P = 0.006) and small fibre neuropathy (mean concentration = 8.98 pg/mL, P = 0.004; Z-score = 0.23, P = 0.0001), with endometriosis representing an intermediate phenotype (mean concentration = 5.57 pg/mL, P = 0.44; Z-score = −0.41, P = 0.024). Using PainDETECT, 44% of fibromyalgia and 26% of endometriosis participants met likely neuropathic pain criteria; using NeupSIG grading, 96% of small fibre neuropathy participants met probable/definite criteria. Conclusion: Our study provides serological evidence of nerve pathology in fibromyalgia at comparable levels with small fibre neuropathy, although this did not correlate with neuropathic-like pain presence or severity. These findings highlight the need for subgroup-specific treatment strategies.

Normothermic regional perfusion (NRP) adoption for donation after circulatory death (DCD) liver transplantation in the United States has been growing, but there remains a higher rate of non-utilization of NRP DCD liver grafts as compared with donation after brain death liver grafts. Six transplant centers and 1 organ procurement organization (OPO) with substantial NRP liver procurement experience performed retrospective reviews of all NRP procurement attempts from September 1, 2021, until March 1, 2024, identifying opportunities for utilization. Five of the 6 transplant centers had prospectively collected data on all accepted livers with NRP intent. The transplant center's intention to utilize DCD donors included 329 livers accepted for transplantation before withdrawal, 252 (76.6%) potential donors expired within the hospital timeframe for donation, and 168 (66.7%) livers were transplanted. Fifty-seven opportunities for utilization were identified by transplant centers, plus 11 from the OPO: most were due to exceeding a functional warm ischemic time limit (n=26), other reasons (n=17), and functional assessment (n=16). The lessons learned by centers in this study provide guidance for newer NRP centers and for OPOs in identifying opportunities for utilization of NRP DCD grafts that have historically been declined.

ABSTRACT Candida auris is an emerging pathogen known for causing invasive, multidrug-resistant infections and healthcare-associated outbreaks. Measuring the relatedness of C. auris strains is critical for detecting and preventing healthcare-associated transmission. Whole genome sequencing (WGS) and analysis of single-nucleotide polymorphisms (SNPs) offer high-resolution assessments of genetic relatedness and antifungal susceptibility; however, interpretation requires an understanding of inter- and intra-patient C. auris diversity and genotypic markers of resistance, respectively. This study aimed to define SNP thresholds for assessing C. auris relatedness in healthcare-associated outbreak investigations, predict antifungal resistance via genotypic markers, and evaluate patient risk factors to inform future surveillance strategies. WGS was performed on 68 C. auris isolates (clades I and III) obtained from 31 hospitalized patients across a healthcare system between 2021 and 2024. Using MycoSNP analysis, we observed a maximum intra-patient variation of 14 SNPs. Five probable transmission clusters were identified based on epidemiologic links, with patient isolates differing by a median of 5 SNPs (range: 0−12). Analysis with a commercial pipeline (refMLST, BugSeq) showed similar clustering patterns. All new detections occurring more than 1 month after admission were linked to a cluster, representing a highly specific indicator of healthcare-associated infection. WGS detected known genotypic markers for fluconazole (ERG11 Y132F, F126L, and MRR1 N647T) and micafungin (FKS1 F635C) resistance, with the latter emerging during antimicrobial therapy. In addition, putative FUR1 mutations (G207R and Q16::STOP) associated with flucytosine resistance were identified. These findings emphasize the utility of WGS for identifying healthcare-associated clusters of C. auris and predicting antifungal resistance. IMPORTANCE Candida auris is a difficult-to-treat yeast that causes invasive infections in vulnerable patient populations. Healthcare exposure is a key risk factor for becoming colonized or infected with C. auris , and infection prevention groups focus on curbing the spread of this organism within the healthcare environment. Whole genome sequencing approaches are key for supporting these efforts, as they can help define clusters of C. auris transmission and can also provide insight into antifungal resistance. Our work provides practical guidance for interpreting genomic data in this setting, helping infection prevention teams respond more effectively to outbreaks and expanding the use of genotypic predictions for antifungal resistance.

Abstract Women with a personal history of breast cancer (PHBC) remain at risk for ipsilateral breast tumor recurrence or a second primary tumor, with a cumulative incidence of 14.3% and reported annual risk of approximately 1–2%. Post-treatment changes can obscure imaging findings, making early recognition of recurrence a challenging, but critical skill for breast radiologists. Detection of early breast cancer recurrence, before it becomes symptomatic, is linked to improved survival, highlighting the need for multimodality surveillance imaging. This review summarizes key prognostic factors that influence recurrence risk and should inform imaging interpretation. Incorporating these prognostic factors into practice can aid imaging interpretation by heightening the attention to subtle changes that may reflect early recurrence. Illustrative cases are used to demonstrate the imaging spectrum of locoregional recurrence in post-lumpectomy and post-mastectomy patients. Recurrences often resemble the imaging features of the primary tumor but may also present subtly. The distinction between recurrence and benign post-treatment change is emphasized, with attention to imaging findings that warrant further evaluation. Finally, emerging evidence-based recommendations for risk-adapted surveillance incorporating tumor biology, receptor status, and stage are reviewed to support early detection and improved outcomes

Abstract Reported prevalence estimates of Lewy body pathology (LBP) vary widely, often without considering brain regional distributions or demographic influences. Large, population-representative autopsy cohorts are needed to provide estimates and clarify the distribution and clinical implications of LBP. Neuropathological, genetic, and clinical data were pooled from nine community- or population-based brain autopsy cohorts in the USA (n=6), Brazil, Austria, and Finland (n=7309 total; 59% women; mean age of death 84.2 years). Cognitive status was available for 6166 participants: dementia (38.5%), mild cognitive impairment (14.8%), and cognitively unimpaired (47.0%). Frequency of LBP was examined by anatomical distribution (neocortical, limbic, brainstem, amygdala, olfactory) and stratified by covariates. Prevalence was calculated in meta-analysis, and mixed-effect logistic regression examined associations with sex, cognitive status, Alzheimer’s disease neuropathological change (ADNC), and limbic-predominant age-related TDP-43 encephalopathy neuropathologic change (LATE-NC) co-pathology. Overall, 27.3% of participants had LBP: neocortical 8.4%, limbic 6.8%, brainstem 4.8%, amygdala 3.7%, and olfactory 3.6%. Neocortical LBP was associated with dementia (present in 15%; OR=4.06 [95%CI 3.24-5.10]) and with increased odds of ADNC (OR=2.30 [95%CI 1.88–2.81]) and LATE-NC (OR=2.03 [95%CI 1.69-2.44]). Sex differences were not observed in the frequencies of neocortical or limbic LBP. However, amygdala-predominant LBP was more frequent in women (OR=1.53 [95%CI 1.06-2.21]) whereas brainstem-predominant LBP was more common in men (OR=0.64 [95%CI 0.49-0.84]). Additionally, amygdala-predominant LBP was associated with increased odds of comorbid ADNC (OR=12.7 [95%CI 5.30–30.6]) while brainstem-predominant LBP was not associated with greater odds of ADNC co-pathology (OR=0.88 [95%CI 0.67–1.16]). Pooling data from large, international community-based autopsy cohorts allows for more robust estimates of region-specific LBP prevalence and their associations with cognitive status. The observed sex- and co-pathology–specific differences in brainstem and amygdala-predominant LBP highlight potential biological heterogeneity and suggest that distinct disease pathways may underlie these patterns.

INTRODUCTION: Advanced therapies infrequently achieve complete fistula closure in patients with Crohn's disease (CD). The primary objective of the STOMP2 clinical trial was to determine whether the implantable autologous cell therapy AVB-114 is effective in inducing remission of persistent Crohn's perianal fistulas versus standard of care. METHODS: Eligible patients had a single perianal fistula tract and had failed previous treatment or had documented medication intolerance to biologic or conventional CD therapy. All enrolled subjects underwent surgical fistula optimization and adipose tissue biopsy collection (to manufacture AVB-114) followed by 1:1 randomization to either standard of care (repeat fistula optimization including seton replacement) or AVB-114 implantation. The primary end point was combined fistula remission at 36 weeks, defined as closure of the external opening, no drainage of fluid despite gentle finger compression, and no collections >2 cm in at least 2 of the 3 dimensions on MRI. RESULTS: Subjects at 14 US sites were equally randomized to SoC (N = 24) or AVB-114 (N = 24). Week 36 combined remission was 8.3% and 45.8% for subjects randomized to SoC and AVB-114, respectively (38% difference, 95% CI 11%–60%; P = 0.0078). At week 36, MRI revealed no subjects with collections >2 cm, radiological improvement (≥50% decrease in the Magnetic Resonance Novel Index for Fistula Imaging in Crohn's Disease (MAGNIFI-CD) score from baseline) in 1 SoC subject and 3 AVB-114 subjects, and radiological healing (complete resolution of T2-weighted hyperintensity in fistula tract) in 2 SoC subjects and 5 AVB-114 subjects. Through week 36, there were 40 and 56 treatment emergent adverse events for SoC and AVB-114, respectively. There were no serious treatment emergent adverse events in the AVB-114 group through week 36. DISCUSSION: A significantly higher proportion of patients treated with AVB-114 achieved combined remission vs SoC, and the treatment was well tolerated. These data support the safety and efficacy of AVB-114 in persistent Crohn's perianal fistulas.

Aim To determine whether there is a difference in the time to complete a rapid number naming task during the interictal state compared to prodrome, migraine headache, postdrome phases of migraine and non-migraine headache, and characterize non-headache symptoms experienced during these phases. Methods This prospective longitudinal cohort study evaluated 30 adults with migraine with or without aura who completed diary entries and a timed rapid number naming task at least three times daily for up to four months from October 2020 to July 2022 at a tertiary headache center. Diary entries included self-report of current migraine phase (having a migraine headache, non-migraine headache, prodrome, postdrome, or interictal baseline) and detailed symptoms experienced. The outcome of interest for the timed rapid number naming task was the time to complete the task with zero errors. Results Compared to the interictal baseline, the time with zero errors for the rapid number naming task was significantly slower during the prodrome (median (Q1, Q3) 1.0 s slower (0.4, 2.9), p -value 0.010), migraine headache (median (Q1, Q3) 3.8 s slower (2.6, 7.5), p -value <0.001), and postdrome (median (Q1, Q3) 0.9 s slower (0.2, 2.1), p -value 0.011) phases of migraine, and non-migraine headache attacks (median (Q1, Q3) 1.2 s slower (0.1, 3.1), p -value 0.006). For those with migraine with aura (N = 15), the time with zero errors was significantly slower during all phases compared to interictal phase. For those with migraine without aura (N = 15), the time with zero errors was significantly slower only during the migraine headache, but not other phases compared to the interictal phase. The top 5 most reported non-headache symptoms during any phase of migraine and non-migraine headache attacks included fatigue, neck stiffness, light sensitivity, blurred vision/vision changes, and difficulty concentrating. Conclusions Among the overall participant cohort, the time with zero errors to complete a rapid number naming task was significantly slower during all ictal phases compared to interictal phase. This was also the case in the migraine with aura cohort but in the migraine without aura cohort there was only a significant difference during the migraine headache phase versus interictal phase. There is overlap in non-headache symptoms between the prodrome, migraine headache, postdrome, and non-migraine headache phases, which may make distinguishing between these phases difficult. Combining subjective data, such as symptom diaries, with objective measures, such as rapid number naming tasks, may allow for more refined assessment of migraine phases and distinguishing migraine from non-migraine headache attacks.

While most cases of human muscle-invasive bladder cancer (MIBC) are diagnosed as conventional urothelial carcinoma (UC), there is growing recognition of histologic subtypes and divergent differentiation within conventional UC. In this study, 31 dogs with UC (33 slides) were evaluated for histologic subtypes and divergent differentiation, as specified by the recent World Health Organization update on the classification of urinary bladder cancers. Slides were reviewed by a medical uropathologist and 3 board-certified veterinary pathologists and assessed for expression of uroplakin III, vimentin, and E-cadherin. All tumors were classified as UC. Fifteen cases were identified as conventional UC. Among the remainder, 8 displayed glandular differentiation, 4 were classified as sarcomatoid UC, 2 showed squamous differentiation, 1 case was classified as a large nested subtype, and 1 case was classified as a tubular/microcystic subtype. This study found a higher frequency of certain histologic subtypes and divergent differentiation in canine UC, particularly sarcomatoid UC and UC with glandular differentiation, compared to previous reports of both canine UC and human MIBC. The relatively high prevalence of the sarcomatoid UC and UC with glandular differentiation in dogs observed in this study suggests that canine UC may serve as a valuable translational model for evaluating novel therapeutic agents, particularly for these rare and aggressive subtypes in humans.