Citations

Large language models (LLMs) have the potential to transform healthcare by assisting in documentation, diagnosis, patient communication, and medical education. However, their integration into clinical practice remains a challenge. This perspective explores the barriers to implementation by synthesizing recent evidence across five challenge domains: workflow misalignment and diagnostic safety, bias and equity, regulatory and legal governance, technical vulnerabilities such as hallucinations or data poisoning, and the preservation of patient trust and human connection. While the perspective focuses on barriers, LLM capabilities and mitigation strategies are advancing rapidly, raising the likelihood of near-term clinical impact. Drawing on recent empirical studies, we propose a framework for understanding the key technical, ethical, and practical challenges associated with deploying LLMs in clinical environments and provide directions for future research, governance, and responsible deployment.

This review explores how multimodal foundation models (MFMs) are transforming biliary tract cancer (BTC) research. BTCs are aggressive malignancies with poor prognosis, presenting unique challenges due to difficult diagnostic methods, molecular complexity, and rarity. Importantly, intrahepatic cholangiocarcinoma (iCCA), perihilar cholangiocarcinoma (pCCA), and distal bile duct cholangiocarcinoma (dCCA) represent fundamentally distinct clinical entities, with iCCA presenting as mass-forming lesions amenable to biopsy and targeted therapies, while pCCA manifests as infiltrative bile duct lesions with challenging diagnosis and primarily palliative management approaches. MFMs offer potential to advance research by integrating radiological images, histopathology, multi-omics profiles, and clinical data into unified computational frameworks, with applications tailored to these distinct BTC subtypes. Key applications include enhanced biomarker discovery that identifies previously unrecognizable cross-modal patterns, potential for improving currently limited diagnostic accuracy—though validation in BTC-specific cohorts remains essential—accelerated drug repurposing, and advanced patient stratification for personalized treatment. Despite promising results, challenges such as data scarcity, high computational demands, and clinical workflow integration remain to be addressed. Future research should focus on standardized data protocols, architectural innovations, and prospective validation studies. The integration of artificial intelligence (AI)-based methodologies offers new solutions for these historically challenging malignancies. However, current evidence for BTC-specific applications remains largely theoretical, with most studies limited to proof-of-concept designs or related cancer types. Comprehensive clinical validation studies and prospective trials demonstrating patient benefit are essential prerequisites for clinical implementation. The timeline for evidence-based clinical adoption likely extends 7–10 years, contingent on successful completion of validation studies addressing current evidence gaps.

Visceral disseminated varicella-zoster virus infection (VD-VZV) involves the hematogenous spread of VZV from the skin to the internal organs. Though rare, it is potentially life-threatening, predominantly affecting immunocompromised individuals. Diagnosis is often delayed due to nonspecific symptoms mimicking other viral illnesses. While the vesicular rash is a hallmark sign, it is absent in approximately 5% of cases. Visceral involvement may precede cutaneous lesions, complicate early recognition, and increase the risk of severe complications. This scoping review screened 594 articles of which 153 met the inclusion criteria, yielding 156 individual cases. Patients were predominantly male (53.8%), with a mean age of 42.3 years. The overall mortality rate was 25.0%. Multiple organs were involved in 46.1% of cases. The most frequently affected were the lungs (56%), liver (44%), heart (16%), kidneys (11%), pancreas (11%), stomach (10%), and esophagus (6%). Antivirals were administered in 89.1% of cases, while corticosteroids were used in 22.4%, with no significant impact on outcomes. Early diagnosis, achieved in 65.4% of patients, was significantly associated with survival (p = 0.043). Mortality was significantly associated with underlying comorbidities (p = 0.004), especially autoimmune diseases requiring immunosuppression (p = 0.048). Septic shock or multi-organ dysfunction (MODS), hepatitis, acute kidney injury, and acute liver failure were linked to higher mortality in univariate analysis. Multivariate analysis identified comorbidities (p < 0.001), septic shock/MODS (p = 0.008), and acute liver failure (p = 0.039) as independent predictors of mortality. Patients with septic shock/MODS had over twice the risk of death (OR = 2.24; p = 0.008). This review underscores the diagnostic challenges and high mortality of VD-VZV. Early recognition and timely administration of antiviral treatment appear critical for survival. Greater clinical awareness and further research are needed to guide management.

The integration of artificial intelligence (AI) in medical imaging has revolutionized diagnostic capabilities, yet the black-box nature of deep learning models poses significant challenges for clinical adoption. Current explainable AI (XAI) approaches, including SHAP, LIME, and Grad-CAM, predominantly focus on post hoc explanations that may inadvertently undermine clinical decision-making by providing misleading confidence in AI outputs. This paper presents a systematic review and meta-analysis of 67 studies (covering 23 radiology, 19 pathology, and 25 ophthalmology applications) evaluating XAI fidelity, stability, and performance trade-offs across medical imaging modalities. Our meta-analysis of 847 initially identified studies reveals that LIME achieves superior fidelity (0.81, 95% CI: 0.78–0.84) compared to SHAP (0.38, 95% CI: 0.35–0.41) and Grad-CAM (0.54, 95% CI: 0.51–0.57) across all modalities. Post hoc explanations demonstrated poor stability under noise perturbation, with SHAP showing 53% degradation in ophthalmology applications (ρ = 0.42 at 10% noise) compared to 11% in radiology (ρ = 0.89). We demonstrate a consistent 5–7% AUC performance penalty for interpretable models but identify modality-specific stability patterns suggesting that tailored XAI approaches are necessary. Based on these empirical findings, we propose a comprehensive three-pillar accountability framework that prioritizes transparency in model development, interpretability in architecture design, and a cautious deployment of post hoc explanations with explicit uncertainty quantification. This approach offers a pathway toward genuinely accountable AI systems that enhance rather than compromise clinical decision-making quality and patient safety.

Abstract Objectives Large language models (LLMs) offer potential in natural language processing tasks in healthcare. Due to the need for high accuracy, understanding their limitations is essential. The purpose of this study was to evaluate the performance of LLMs in classifying radiology reports for the presence of pulmonary embolism (PE) under various conditions, including different prompt designs and data perturbations. Materials and Methods In this retrospective, institutional review board approved study, we evaluated 3 Google’s LLMs including Gemini-1.5-Pro, Gemini-1.5-Flash-001, and Gemini-1.5-Flash-002, in classifying 11 999 pulmonary CT angiography radiology reports for PE. Ground truth labels were determined by concordance between a computer vision-based PE detection (CVPED) algorithm and multiple LLM runs under various configurations. Discrepancies between algorithms’ classifications were aggregated and manually reviewed. We evaluated the effects of prompt design, data perturbations, and repeated analyses across geographic cloud regions. Performance metrics were calculated. Results Of 11 999 reports, 1296 (10.8%) were PE-positive. Accuracy across LLMs ranged between 0.953 and 0.996. The highest recall rate for a prompt modified after a review of the misclassified cases (up to 0.997). Few-shot prompting improved recall (up to 0.99), while chain-of-thought generally degraded performance. Gemini-1.5-Flash-002 demonstrated the highest robustness against data perturbations. Geographic cloud region variability was minimal for Gemini-1.5+-Pro, while the Flash models showed stable performance. Discussion and Conclusion LLMs demonstrated high performance in classifying radiology reports, though results varied with prompt design and data quality. These findings underscore the need for systematic evaluation and validation of LLMs for clinical applications, particularly in high-stakes scenarios.

Background: Studies have suggested a synergism between lenalidomide (LEN) and ibrutinib (IBR) in multiple myeloma (MM). Both downregulate IRF4, a key target and master transcriptional factor regulating myeloma cell survival. Method: A 3 + 3 phase I trial was conducted to determine the maximum tolerated dose (MTD) of IBR in combination with LEN + dexamethasone (DEX) in patients with relapsed/refractory (RR) MM who had at least one prior line of therapy. Three dose levels (DLs) were planned. The cycle length was 28 days. IBR was administered orally daily in doses of 560 mg on DL1-2 and 840 mg on DL3, LEN was administered orally on days 1–21 in doses of 15 mg on DL1 and 25 mg on DL2-3, and DEX was administered orally on days 1, 8, 15, and 22 in a dose of 40 mg if age < 75 years or in a dose of 20 mg if it was ≥75 years for DL1-3. Patients with a glomerular filtration rate (GFR) <60 but ≥30 mL/min were treated in accordance with the manufacturer’s instructions with LEN 10 mg. Dose-limiting toxicities (DLTs) included the following: grade 4 neutropenia lasting more than 5 days, thrombocytopenia, febrile neutropenia, nausea, vomiting or diarrhea; grade 3 thrombocytopenia with bleeding or platelet transfusion; and grade 3–4 hyperglycemia or a thrombotic/embolic event, and other nonhematologic toxicities. The overall response rate (ORR) was defined as the percentage of patients with a partial response (PR), very good partial response (VGPR), or complete response (CR) according to IMWG criteria on two consecutive evaluations at least 4 weeks apart. The clinical benefit rate (CBR) was defined as the percentage of patients with stable disease (SD) or a better outcome on two consecutive evaluations at weeks apart. Results: Fourteen patients (DL1: six patients; DL2: three patients; DL3: five patients) were registered for the study from March 2019 to May 2023, prior to its closure due to limited accrual. Thirteen patients are included in the summary of toxicities and response as one patient on DL3 halted participation prior to the start of the treatment. Two patients on DL3 were excluded from the determination of MTD: one having discontinued cycle 1 treatment due to COVID-19 infection and the another having mistakenly taken 280 mg/day of IBR instead of the assigned 840 mg/day dose during cycle 1. Only one patient developed a DLT, on DL1 with grade 3 non-viral hepatitis. The median number of cycles administered was 4 (range: 1–56). Severe toxicities reported included grade 4 lymphocytopenia (1), grade 4 thrombocytopenia (1), and grade 5 sepsis in the setting of PD (1). Disease responses included a VGPR on DL1 and CR on DL3. Thus, the ORR was 15.4% (90% CI: 2.8–41.0%). One patient on DL1 maintained SD for 4.6 years before discontinuing the treatment to undergo an alternative therapy. Another five patients maintained SD for ≥ 2 consecutive cycles. Thus, the CBR was 61.5% (90% CI: 35.5–83.4%). Conclusions: The combination of LEN with IBR in RR MM proved feasible, with manageable toxicities and the majority of discontinuations being due to disease progression.

Abstract In recent years, there has been increasing interest in the possible prognostic impact of concomitant medications in patients with cancer treated with immunotherapy combinations. This real‐world analysis aims to evaluate the impact of concomitant medications on survival outcomes in patients with advanced biliary tract cancer (BTCs) treated with cisplatin, gemcitabine and durvalumab (CGD) therapy. The study cohort included patients with a diagnosis of advanced BTCs who were taking concomitant medications for their comorbidities before the start of CGD. The primary objectives were overall survival (OS) and progression‐free survival (PFS). The initial population consisted of 666 patients, who were retrospectively collected from 41 sites in 12 countries. Data on concomitant medications were available for 493 patients. After a median follow‐up of 8.8 months (95% CI: 7.8–9.8), patients who did not take steroids (prednisone >10 mg/day or equivalent) or nonsteroidal anti‐inflammatory drugs (NSAIDs) and opioids, before the start of CGD, had longer OS and PFS in univariate analysis. The multivariate analysis confirmed longer OS for patients who did not take steroids. Patients who did not take steroids had an OS of 14.8 months (95% CI: 13.1–29.1) versus 5.0 months (95% CI: 2.14–11.32) of patients who took prednisone >10 mg/day or equivalent. No differences were reported in terms of overall response rate (ORR), disease control rate (DCR) ( p = 1.0 and p = .16, respectively), and safety profile between the two groups. Our analysis suggests that patients who did not receive steroids before the start of GCD had longer survival and highlighted the relevance of balancing concomitant medications and chemoimmunotherapy.

Abstract Background Cisplatin, gemcitabine, and durvalumab combination is a standard first-line treatment for advanced biliary tract cancer. This study aimed to assess the impact of genetic alterations on outcomes in patients with advanced biliary tract cancer treated with cisplatin, gemcitabine, and durvalumab in real-world clinical practice. Methods Patients with unresectable, locally advanced, or metastatic biliary tract cancer treated with cisplatin and gemcitabine plus durvalumab across 39 centers in 11 countries in Europe, the United States, and Asia were included in this analysis. Results The cohort included 513 patients with advanced biliary tract cancer. The 5 most frequently altered genes were TP53 (22.1%), KRAS (13.7%), CDKN2A/B (13.6%), ARID1A (12.2%), and IDH1 (9.2%). In multivariate analysis, SMAD4 mutations were associated with improved progression-free survival (PFS) (hazard ratio [HR] = 0.49, P = .018) and overall survival (HR = 0.11, P = .023), while TP53 mutations were linked to worse PFS (HR = 1.62, P = .0047) and TERT mutations to worse overall survival (HR = 8.92, P = .0012). No other genomic alterations were statistically associated with outcomes. Subgroup analysis showed that TP53 mutations negatively affected PFS and overall survival in intrahepatic cholangiocarcinoma, while KRAS mutations were associated with poorer PFS in extrahepatic cholangiocarcinoma. No gene alterations were linked to outcomes in gallbladder cancer. Conclusions This large-scale analysis, with comprehensive molecular profiling, supports the positive prognostic impact of SMAD4 mutations for PFS and overall survival and highlights the negative prognostic roles of TP53 (PFS) and TERT (overall survival) mutations, providing valuable insights for personalized treatment strategies in biliary tract cancer.

ABSTRACT Background Over the years, prognostic indexes have been developed to help clinicians stratify patients with biliary tract cancers (BTC) into risk groups. This study aims to identify a new prognostic index for patients with BTC treated with cisplatin, gemcitabine and durvalumab (CGD) in the first‐line setting. Patients and Methods The study population consisted of patients with BTC from 11 Eastern and Western Countries. Using multivariate analysis for overall survival (OS), we identified 5 baseline statistically significant variables: stage, carcinoembryonic antigen (CEA) levels, albumin levels, gamma glutamyl transferase (GGT) levels, neutrophil‐to‐lymphocyte ratio (NLR). Metastatic disease is a prognostic factor with a superior weight considering the HR of 3.62, while all the others can be considered prognostic factors with equivalent weight as the HRs are quite similar (HRs between 1.55 and 1.92). Based on these reasons, we developed a prognostic model called the MAGIC‐D index by assigning a score of 2 for metastatic disease, and a score of 1 for CEA increased levels, albumin decreased levels, GGT increased levels, NLR ≥ 3. Patients were stratified into three risk groups as follows: low‐risk group (Score from 0 to 2), intermediate‐risk group (Score from 3 to 4) and high‐risk group (Score from 5 to 6). At the first data cutoff (April 2024), these data were available for 319 patients that composed the training cohort used for the analysis. At the second data cutoff (May 2025), 79 patients were further enrolled and composed the validation cohort. Results Median progression‐free survival was 11.7 months in low‐risk group (20.7%), 8.7 months in intermediate‐risk group (46.4%) and 5.4 months in high‐risk group (32.9%) [low‐risk hazard ratio (HR): 0.27, intermediate‐risk HR: 0.55, high‐risk HR: 1, p < 0.0001]. Median OS was 18.4 months in low‐risk group,15.9 months in intermediate‐risk group and 7.8 months in high‐risk group (low‐risk HR: 0.17, intermediate‐risk HR: 0.43, high‐risk HR: 1, p < 0.0001). There was no difference in overall response rate (low‐risk: 31.8%, intermediate‐risk: 36.5% and high‐risk: 25.7%; p = 0.0718), while disease control rate was significantly different across the three risk groups (low‐risk: 83.3%, intermediate‐risk: 70.9% and high‐risk: 60.9%; p < 0.0001) as well as the rate of patients receiving a second‐line therapy (low‐risk: 54.5%, intermediate‐risk: 48.6% and high‐risk: 25.7%; p = 0.0061). Finally, the prognostic role in terms of OS and PFS of the MAGIC‐D index was confirmed in a validation cohort of 79 patients. Conclusion The MAGIC‐D index is an easy‐to‐use tool able to stratify patients with BTC with different prognoses undergoing first‐line therapy with CGD.