Citations

Abstract Background The purpose of this study was to develop a series of benchmarks specifically for large language model (LLM) performance in the identification of drug-drug interactions (DDIs). Methods We evaluated six LLMs using a clinician-designed dataset of 750 DDI scenarios. Tasks included: (1) a pointwise two-drug classification task (2) a pairwise three-drug discrimination task and (3) a listwise 4–6 drug selection task. Zero-shot prompting was applied and performance was assessed using precision, recall, F1 score, accuracy, and self-consistency. Results Model performance varied by task structure and interaction severity. GPT-5-Chat demonstrated the highest recall and F1 score in the pointwise task (80.7% and 80.2%), whereas DrugGPT achieved superior accuracy in the pairwise (93.7%) and listwise task (92.6%). All models had improved performance with more severe compared to less severe DDIs (accuracy 72.0-99.2% vs. 58.3-86.7%). All models tended to have higher self-consistency in the pairwise and listwise experiments compared to the pointwise experiment (68.8-93.6% vs. 31.0-74.2%). No model exhibited uniformly high reliability across all tasks. Conclusions Current LLMs show inconsistent capabilities in identifying DDIs. Performance degrades as reasoning space expands with limited stability. These findings emphasize the need for multi-format evaluation frameworks and reliability-aware assessment when considering LLMs for medication-safety applications.

Localized cutaneous squamous cell carcinoma (cSCC) has a favorable prognosis, unlike advanced disease, especially with clinical perineural invasion (PNI), which poses substantial management challenges due to aggressivity and higher recurrence, metastasis, and mortality risks. PNI, a high-risk staging feature, has worse outcomes, particularly when clinically evident rather than incidental. Clinical PNI (cPNI) is evident by clinical symptoms (such as pain, paresthesia, or motor deficits) or radiologic findings, whereas incidental PNI (iPNI) is identified only histologically without associated symptoms or radiologic evidence. PNI remains a novel area with varying practice patterns across institutions. Improving risk stratification and tailoring multidisciplinary approaches are critical for optimizing outcomes. Our review outlines clinical practice patterns at our institution, providing insights into managing cSCC with PNI, focusing on diagnosis, imaging, staging, and emerging immunotherapies. A structured search was conducted using the terms “perineural invasion,” “cutaneous squamous cell carcinoma,” and “immunotherapy.” cPNI has a poor prognosis and requires nuanced clinical decision-making. Surgery and radiation remain central to management. Adjuvant therapy offers substantial survival benefit in cSCC with PNI, with improved disease-free and overall survival compared with surgery alone, supporting its use in appropriately selected high-risk patients. Traditional systemic therapies, including cisplatin and cetuximab, remain foundational but have shown only moderate response rates and limited durability in advanced or neurotropic cSCC. In contrast, immunotherapy—now preferred for advanced or unresectable cases—has transformed management, with programmed cell death protein-1 (PD-1) inhibitors showing promising results (up to 69% response rate) and disease stabilization. Neoadjuvant immunotherapy may enable tumor downstaging, improve radiation planning, and reduce surgical morbidity. Imaging for squamous cell carcinoma (SCC) with PNI aids staging and surveillance, but symptoms remain key for detecting recurrence. Our multidisciplinary approach emphasizes personalized care. Larger trials are needed to define the optimal role and sequencing of immunotherapy in this high-risk patient population.

Abstract Pancreatic cancer, projected to become the second-deadliest malignancy in Western societies by 2030, requires urgent innovations in early detection and risk stratification. Intraductal papillary mucinous neoplasms (IPMNs) represent critical precursor lesions, but current clinical guidelines demonstrate suboptimal accuracy in malignancy prediction, leading to either unnecessary surgeries or missed opportunities for early intervention. In this paper, we present Cyst-X, an artificial intelligence (AI) framework that accurately predicts IPMN malignant transformation using multicenter magnetic resonance imaging (MRI) data. Unlike most previous approaches that rely on computed tomography (CT), our method capitalizes on the superior soft tissue contrast of MRI, allowing for more precise identification of subtle imaging biomarkers. We developed and validated deep learning models on 723 T1-weighted and 738 T2-weighted MRI scans from 764 patients among seven international institutions, demonstrating significantly superior performance (AUC=0.82) compared to current clinical (Kyoto) guidelines (AUC=0.75) and expert radiologists. AI-based imaging features correlate strongly with clinically recognized malignancy markers, providing potential biologically relevant insights. This approach holds promise to significantly refine clinical decision-making, reduce unnecessary surgeries, and better identify high-risk IPMN patients for timely intervention. Our approach integrates a novel pancreas segmentation network with robust classification models that identify subtle imaging biomarkers associated with malignancy risk. Importantly, we demonstrate that these models retain high performance in a privacy-preserving federated learning setting, where institutions collaboratively train AI models without exchanging patient data to address key regulatory and ethical barriers. We publicly release the Cyst-X dataset--the first large-scale, multi-center pancreatic cyst MRI collection--to accelerate research in this field. This study addresses a critical clinical need while establishing technical foundations for privacy-preserving AI in radiology that could transform pancreatic cancer management through earlier intervention and reduced unnecessary procedures. The dataset can be accessed at https://osf.io/74vfs/, and the source code for our deep learning segmentation and classification models is available at https://github.com/NUBagciLab/Cyst-X.

Introduction: This study investigated the association between capitate morphology and the success rates of non-vascularized bone graft (NVBG) and vascularized bone graft (VBG) operations for scaphoid nonunion. Methods: We retrospectively analysed 213 patients with established scaphoid non-union who underwent either non-vascularized bone graft ( n = 75) or vascularized bone graft ( n = 138). We analysed risk factors, including demographics, medical history, implant type, previous surgery, bone graft donor site and capitate morphology for their association with treatment outcome. We also assessed the relationship between capitate and scaphoid morphology in the uninjured wrists of 31 patients. Results: The overall success rate was 81% for NVBG and 86% for VBG, with no significant difference. For NVBG, Type II capitate, age 40 or older and tobacco use were significant risk factors for failure. For VBG, only previous surgery was a significant risk factor. In uninjured wrists, a Type II capitate was associated with a significantly thinner scaphoid waist and a lower waist index than a Type I capitate. Conclusion: Our study introduces a new radiographic classification of capitate morphology and establishes its significant association with scaphoid morphology. Beyond traditional factors, capitate morphology can help to assess scaphoid morphology and potentially infer vascularity, aiding the selection of the most appropriate bone graft type for treating scaphoid non-union. Level of evidence: III

Abstract Pyroptosis is a form of programmed cell death characterized by the cleavage of the gasdermin (GSDM) family proteins that form pores in the plasma membrane, cell rupture, and the release of proinflammatory cytokines. In this study, we performed immunohistochemistry for cleaved gasdermin D (GSDMD), gasdermin E (GSDME) N-terminus, and gasdermin C (GSDMC) in 2 different cohorts of diffuse large B-cell lymphoma (DLBCL) and analyzed the impact of GSDM expression on prognosis and immunity. The results showed frequent cleaved GSDMD (N-terminal) expression. Only cytoplasmic GSDMD N-terminal expression correlated with significantly better patient survival in the 2 cohorts. In contrast, GSDME was mainly expressed in the vascular endothelium, and correlated with significantly adverse prognostic effect. Correlating with the multiplex fluorescent immunohistochemistry results, we found that cytoplasmic GSDMD N-terminal expression was associated with increased CD38+ (activated) M1 macrophages in both cohorts, cognate interactions between live DLBCL cells and activated M1 macrophages (and T cells), and lower PD-1/PD-L1 expression in the analyzed cases. In contrast, T-cell pyroptosis, lymphoma cell-resistance to cell death, and phagocytosis by M2 macrophages were observed in tissues with predominantly nuclear GSDMD N-terminal expression. Bulk gene expression profiling and deconvolution analysis revealed associations of cytoplasmic GSDMD N-terminal expression with downregulation of “Don’t eat me”–signaling genes, upregulation of many RNA genes, decreased frequency of “inflammatory” lymphoma microenvironment subtype, increased frequencies of prognostically favorable cell states and ecotypes, and decreased frequency of T-cell exhaustion state. In summary, this study showed distinct cellular and subcellular patterns of 3 GSDM proteins and their associated immune response phenotypes and prognostic effects, with implications for novel therapeutic strategies for B-cell lymphoma.

Abstract Background The treatment (tx) landscape for chronic lymphocytic leukemia (CLL) has evolved, with National Comprehensive Cancer Network (NCCN) guideline-preferred novel therapies (NTs), such as acalabrutinib, zanubrutinib, and venetoclax-based regimens, now preferred over ibrutinib and chemoimmunotherapy (CIT) in front-line (1L) settings. We previously showed that patients (pts) from different racial/ethnic groups may not have equitable access to NTs. Here, we examined racial/ethnic inequities in 1L NT utilization among US pts with CLL and potential social determinants of health (SDOH) drivers of these inequities. Methods This retrospective cohort study utilized the US-based, electronic health record-derived de-identified Flatiron Health Research Database, linked to neighborhood (US Census track or block group) data from American Community Survey and Agency for Healthcare Research and Quality datasets. Eligible pts included adults with CLL who started 1L tx 01/01/2019-07/31/2024. The primary outcome was receipt of 1L therapy (CIT vs preferred NTs, ibrutinib vs preferred NTs), with NCCN guideline-preferred NTs (v2.2023-v3.2025) as a reference. Associations between race/ethnicity (White, Black, Hispanic) and 1L tx types were assessed using logistic regressions, adjusting for age, sex, year of 1L start, immunoglobulin heavy chain variable region status, and del17p/TP53 status. To better understand which SDOH factors contribute to racial and ethnic inequities in tx utilization, we performed mediation analysis using Multiple Mediation Analysis implemented through nonlinear multiple additive regression tree models. Individual-level (Eastern Cooperative Oncology Group status, practice type, insurance) and 20 area-level SDOH factors measuring social deprivation (defined as limited access to economic, social, neighborhood, physical, or healthcare resources) were assessed as potential mediators. Results Of the 4452 pts included in the study, 3717 were White (83%), 371 were Black (8%), 209 were Hispanic (5%), and 155 were Other race (4%). Black and Hispanic pts were more likely to live in neighborhoods with higher social deprivation, such as residential segregation (predominant race/ethnicity in the area: White, Black, Hispanic, Diverse), with no internet access, no vehicle ownership, and no health insurance coverage. Overall, 37% of pts received a preferred NT, highest among White (38%) pts, followed by Black (35%) and Hispanic (26%) pts. Compared with White pts, Hispanic pts were more likely to receive CIT than preferred NTs (adjusted odds ratio [aOR] 2.12; 95% confidence interval [CI] 1.46-3.09); Black (aOR 1.44; 95% CI 1.05-1.97) and Hispanic pts (aOR 1.83; 95% CI 1.17-2.85) were more likely to receive ibrutinib than preferred NTs. Pts living in areas with the highest social deprivation were less likely than their counterparts in lowest social deprivation neighborhoods to receive preferred NTs. Specifically, residence in predominantly Black or Hispanic neighborhoods (26% or 27% vs 39% predominantly White), areas with lowest socioeconomic status (32% vs 40%), and neighborhoods with the lowest levels of internet access (32% vs 39%), vehicle ownership (32% vs 39%), and health insurance coverage (34% vs 41%) was associated with lowest receipt of preferred NTs. In the mediation analysis, individual- and area-level SDOH factors explained 94.4% (95% CI 2.9-185.9) of the observed Black-White inequity in receipt of ibrutinib versus preferred NTs; residential segregation appeared to be the most important mediator (59.7%; 95% CI −3.5 to 123.0). Similarly, the Hispanic-White inequity in receipt of preferred NTs was driven by the combined effect of area-level SDOH and individual-level variables (88.5%; 95% CI 45.6-131.3), with residential segregation as the main driver (48.3%; 95% CI 10.7-85.9), followed by practice setting (9.8%; 95% CI −3.5 to 23.1). In contrast, measured SDOH factors did not explain the inequity in receipt of CIT versus preferred NTs between Hispanic and White pts (21.5%; 95% CI −8.3 to 51.4), suggesting that other factors may contribute to the observed inequities. Conclusion Black and Hispanic pts with CLL were less likely than White pts to receive preferred NTs. A significant proportion of these disparities was explained by area-level SDOH, particularly residential segregation. These findings underscore the need to address structural barriers to ensure equitable access to emerging, guideline-recommended treatments.

Abstract INTRODUCTION Immune checkpoint inhibitors (ICIs) have emerged as a cornerstone of cancer therapy. However, despite their efficacy, ICIs can cause a unique spectrum of autoimmune toxicities known as immune-related adverse events (irAEs). In contrast to irAEs involving other organ systems, hematologic irAEs such as ICI-associated immune thrombocytopenia (ICI-ITP) are less common. As a result, existing data on ICI-ITP are largely confined to isolated case reports, single-center case series, and pharmacovigilance databases lacking granular clinical data. To address this key knowledge gap, we sought to comprehensively characterize the incidence, risk factors, clinical features, treatments, and outcomes associated with ICI-ITP. METHODS We performed a multicenter cohort study of adult patients who initiated ICI therapy at 29 hospitals across 7 major cancer centers in 11 geographically diverse states in the US between 2016-2023. ICI-ITP was defined as a decline in platelet count to <100 x109/L in the absence of an alternative more plausible etiology of thrombocytopenia. Patients with hematologic malignancies were excluded, as were those who received cytotoxic chemotherapy within 30 days prior to thrombocytopenia onset. For each patient identified as having ICI-ITP, we collected data on demographics, comorbidities, medications, laboratory values, treatments, and outcomes. All data were extracted by manual chart review. To identify risk factors for ICI-ITP and to examine its association with death, for each case of ICI-ITP we also collected data on two control patients who initiated ICI therapy at the same site and in the same year but without development of ICI-ITP. We used multivariable logistic regression to identify independent risk factors for ICI-ITP and its recovery (the latter defined according to criteria established by the ITP International Working Group). We used a multivariable Cox model to examine the association between ICI-ITP and its severity with all-cause mortality. RESULTS We identified a total of 214 patients who developed ICI-ITP among 86,648 examined (0.25%). The incidence was highly consistent across sites (range, 0.15–0.30%). Risk factors for ICI-ITP included simultaneous receipt of two different classes of ICIs, additional irAEs, and a lower baseline platelet count. ICI-ITP developed at a median of 8 weeks (IQR, 4–18) following ICI initiation. The median nadir platelet count was 41 x109/L (IQR, 17–64). Six out of 214 patients (2.8%) developed a major bleed, and 16 patients (7.5%) developed a clinically relevant non-major bleed. Cell-bound anti-platelet antibody testing was positive in 6 of 8 patients (75.0%) for whom it was assessed, and median circulating thrombopoietin levels (201 pg/mL [IQR, 178–295]) were similar to those seen in primary ITP. Treatments used included glucocorticoids (49.5%), immune globulin (18.2%), and thrombopoietin receptor agonists (13.6%). Overall, 161 patients (75.2%) recovered from ICI-ITP at a median of 2 weeks (IQR, 1–5) following ICI-ITP onset. Predictors of recovery included additional irAEs and a higher nadir platelet count. Treatment response was observed in 57.0%, 56.4%, and 58.3% of patients within 60 days following initiation of glucocorticoids, immune globulin, and thrombopoietin receptor agonists, respectively. Of 76 patients rechallenged with an ICI, 23 (30.3%) developed recurrent ICI-ITP at a median of 9 weeks (IQR, 2–15); 17 of these 23 patients (73.9%) recovered. Finally, ICI-ITP and its severity were monotonically associated with a higher risk of death (adjusted hazard ratio 2.96 [95% CI, 2.14–4.08] for those with severe ICI-ITP compared to those without ICI-ITP). CONCLUSION In this multicenter study involving patients treated with ICIs at 29 hospitals across the US, we identified 214 patients with ICI-ITP. For the first time, we characterize in detail the incidence, risk factors, clinical features, treatments, and outcomes of ICI-ITP, a relatively novel disease entity. We found that more than half of patients responded to conventional ITP-directed therapies, and that most patients rechallenged did not develop recurrent ICI-ITP. We also found that ICI-ITP was independently associated with an increased risk of death, including a nearly 3-fold increased risk in those with severe ICI-ITP. These findings highlight ICI-ITP as a rare but clinically significant complication of ICI therapy that hematologists must be equipped to recognize and manage.

Abstract Preiser's disease (PD), or idiopathic avascular necrosis of the scaphoid, is a debilitating condition leading to wrist pain, dysfunction, and pan-carpal arthritis. While impaired scaphoid vascularity is central to its pathogenesis, identifying individuals at risk is challenging due to the morphological distortion of the scaphoid as the disease progresses. This study aims to investigate the association between a novel capitate morphology classification, scaphoid morphology, and their impact on PD prevalence and characteristics. We also re-evaluate negative ulnar variance (NUV) as a potential risk factor for PD. This retrospective case–control study included 34 PD patients and 340 age-matched controls. Capitate morphology was classified into Type I/II on anteroposterior (AP) radiographs based on capitotrapezoid–scaphocapitate (CT–SC) joint line angle and ratio. Scaphoid morphology was assessed via the waist index on contralateral healthy wrists. Radiographic measurements of ulnar variance and PD staging were also performed. Statistical analyses compared morphological prevalence and risk factors between groups. Type II capitate was significantly more prevalent in the PD group (35.3%) than in controls (17.1%, p = 0.018). Similarly, NUV was significantly higher in PD patients (29%) compared with controls (6.8%, p < 0.001). We found a strong correlation between Type II capitate and slender (Type II) scaphoid morphology in healthy wrists. No significant differences in age, gender, or affected side were observed. Interobserver and intraobserver agreements for capitate classification were substantial (kappa ≥0.882). Our study introduces a reliable radiographic method for classifying capitate morphology, which strongly correlates with intrinsic scaphoid shape and potential vascular vulnerability. Both Type II capitate morphology and NUV are significant predisposing factors for PD. These findings offer valuable insights for identifying at-risk individuals and improving understanding of PD's multifactorial etiology. Diagnostic Level III.

Abstract Background: Pyroptosis is a form of programmed cell death characterized by cleavage of gasdermin (GSDM) family proteins that form pores in the plasma membrane, cell rupture, and release of pro-inflammatory cytokines. Its pro-inflammatory potential has garnered significant research attention in recent years but has not been studied in lymphoma. Patients and Methods: We performed immunohistochemistry for cleaved GSDMD (n-terminal) in two large cohorts of diffuse large B-cell lymphoma (DLBCL) with multiplex fluorescent IHC (mfIHC) data, and analyzed for prognostic and immune impact. Results: Cleaved GSDMD was frequently expressed in DLBCL, and only the cytoplasmic (versus nuclear) form of expression correlated with significantly better patient survival in two DLBCL cohorts. To understand the expression at the single-cell level, we examined mfIHC results, which revealed that GSDMD+cells were tumor or immune cells, and varied in Ki-67, BCL2, and MYC expression. Cases with cytoplasmic cleaved GSDMD expression had higher mean and median levels of CD38+ (activated) M1 macrophages in two cohorts and lower PD-1/PD-L1 expression, as well as phenotypes of cognate interactions between live activated M1 macrophages, DLBCL cells, and T cells in spatially resolved immune landscape by mfIHC. In contrast, phagocytosis of pyroptotic lymphoma cells by M2 macrophages and cell death of T cells were observed in cases with predominantly nuclear form of cleaved GSDMD expression. Bulk gene expression profiling analysis identified a large proportion of RNA genes upregulated in DLBCLs with cytoplasmic cleaved GSDMD expression, and deconvolution analysis found decreased frequency of the “Inflammatory” lymphoma microenvironment (LME) subtype, increased frequency of cell states and ecotypes with favorable prognostic associations, and decreased frequency of T cell exhaustion states. Summary: Analysis of pyroptosis-inducing cleaved GSDMD (n-terminal) expression in a large number of DLBCL patients revealed dinstinctive cytoplasmic and nuclear expression patterns in live and pyroptotic cells associated with different prognostic and immune effects. Cytoplasmic cleaved GSDMD expression is a marker of cognate interactions between live cells (e.g., activated M1 macrophages) associated with adaptive responses and favorable prognostic effects, whereas nuclear cleaved GSDMD expression is associated with immunosuppressive mechanisms, including T cell death and phagocytosis by M2 macrophages.

Abstract Introduction: The CARTITUDE-4 study (NCT04181827), which enrolled patients with lenalidomide-refractory multiple myeloma (MM) after 1–3 prior lines of therapy (pLOT), demonstrated a significant benefit of ciltacabtagene autoleucel (cilta-cel) over established triplet regimens. However, the benefit of cilta-cel for patients with standard-risk cytogenetics remains less defined. Here, we report outcomes in patients with standard-risk cytogenetics from the intent-to-treat and as-treated populations in CARTITUDE-4. Methods: InCARTITUDE-4, patients randomized to the cilta-cel arm underwent apheresis and bridging treatment with either pomalidomide, bortezomib, and dexamethasone (PVd) or daratumumab, pomalidomide, and dexamethasone (DPd), followed by lymphodepletion therapy with cyclophosphamide and fludarabine, and then a single cilta-cel infusion. Progression-free survival (PFS) was assessed using a validated computerized algorithm. The intent-to-treat population included 208 patients; 32 patients progressed or died on bridging therapy, resulting in an as-treated population of 176 patients. Patients with high-risk cytogenetics, defined as del(17p), t(14;16), t(4;14), gain/amp(1q) (n=105), or with unknown cytogenetics (n=12), were excluded from the as-treated analysis. The 12-month minimal residual disease (MRD)-negative complete response (CR) was defined per the International Myeloma Working Group criteria as the proportion of participants with CR or better prior to and at 12 months (±3 months), achieving MRD-negative status at 12 months (+3 months) after cilta-cel infusion, as determined by next-generation sequencing (10-5), prior to progressive disease or subsequent anti-myeloma therapy. For the as-treated population, PFS rates were measured from the time of cilta-cel infusion. Results: In CARTITUDE-4, in the intent-to-treat population, at a median follow-up of 33.6 months, patients with standard-risk cytogenetics had a 30-month PFS rate (95% CI) of 71.0% (58.8–80.2) in the cilta-cel arm (N=69) vs 43.2% (31.3–54.5) in the standard-of-care (SOC) arm (N=70). Patients with standard-risk cytogenetics in the as-treated population (n=59) had a 30-month PFS rate of 80.5% (95% CI, 67.2–88.8). In CARTITUDE-1 (NCT03548207), which evaluated cilta-cel in patients with heavily pretreated relapsed/refractory MM (RRMM; ≥3 pLOT), the 30-month PFS rate among patients with standard-risk cytogenetics (negative for del(17p), t(14;16), or t(4;14); n=68) was 59.9% (95% CI, 47.2–70.5). In the CARTITUDE-4 as-treated population with standard-risk cytogenetics, 8 PFS events occurred within 1 year and 4 PFS events beyond 1 year of cilta-cel infusion. Twenty-six patients achieved MRD-negative CR at 12 months after cilta-cel; 100.0% of these patients were progression free at 30 months. Fourteen patients were not evaluable for MRD due to: calibration failure (n=12), no sample availability for testing (n=1), or indeterminate results post baseline (n=1). Conclusions: The PFS rate at 2.5 years for patients with standard-risk RRMM was higher in CARTITUDE-4 compared with CARTITUDE-1, supporting the use of cilta-cel as early as second line in the treatment course. In CARTITUDE-4 (as-treated population), 80.0% of patients with standard-risk cytogenetics were progression free and off treatment at 2.5 years. In patients with standard-risk disease who achieved MRD-negative CR at 1 year, this rate increased to 100.0%. The low rate of progression events in cilta-cel-treated patients with standard-risk cytogenetics shows the profound benefit of a single cilta-cel infusion in this population.