Importance: Large language models (LLMs) are increasingly integrated into health care applications; however, their vulnerability to prompt-injection attacks (ie, maliciously crafted inputs that manipulate an LLM’s behavior) capable of altering medical recommendations has not been systematically evaluated. Objective: To evaluate the susceptibility of commercial LLMs to prompt-injection attacks that may induce unsafe clinical advice and to validate man-in-the-middle, client-side injection as a realistic attack vector. Design, Setting, and Participants: This quality improvement study used a controlled simulation design and was conducted between January and October 2025 using standardized patient-LLM dialogues. The main experiment evaluated 3 lightweight models (GPT-4o-mini [LLM 1], Gemini-2.0-flash-lite [LLM 2], and Claude-3-haiku [LLM 3]) across 12 clinical scenarios in 4 categories under controlled conditions. The 12 clinical scenarios were stratified by harm level across 4 categories: supplement recommendations, opioid prescriptions, pregnancy contraindications, and central-nervous-system toxic effects. A proof-of-concept experiment tested 3 flagship models (GPT-5 [LLM 4], Gemini 2.5 Pro [LLM 5], and Claude 4.5 Sonnet [LLM 6]) using client-side injection in a high-risk pregnancy scenario. Exposures: Two prompt-injection strategies: (1) context-aware injection for moderate- and high-risk scenarios and (2) evidence-fabrication injection for extremely high-harm scenarios. Injections were programmatically inserted into user queries within a multiturn dialogue framework. Main Outcomes and Measures: The primary outcome was injection success at the primary decision turn. Secondary outcomes included persistence across dialogue turns and model-specific success rates by harm level. Results: Across 216 evaluations (108 injection vs 108 control), attacks achieved 94.4% (102 of 108 evaluations) success at turn 4 and persisted in 69.4% (75 of 108 evaluations) of follow-ups. LLM 1 and LLM 2 were completely susceptible (36 of 36 dialogues [100%] each), and LLM 3 remained vulnerable in 83.3% of dialogues (30 of 36 dialogues). Extremely high-harm scenarios including US Food and Drug Administration Category X pregnancy drugs (eg, thalidomide) succeeded in 91.7% of dialogues (33 of 36 dialogues). The proof-of-concept experiment demonstrated 100% vulnerability for LLM 4 and LLM 5 (5 of 5 dialogues each) and 80.0% (4 of 5 dialogues) for LLM 6. Conclusions and Relevance: In this quality improvement study using a controlled simulation, commercial LLMs demonstrated substantial vulnerability to prompt-injection attacks that could generate clinically dangerous recommendations; even flagship models with advanced safety mechanisms showed high susceptibility. These findings underscore the need for adversarial robustness testing, system-level safeguards, and regulatory oversight before clinical deployment.

Lee CH, et al. J Korean Med Sci. 2024 Sep;39:e293. https://doi.org/10.3346/jkms.2024.39.e293

Purpose: We performed a study-level meta-analysis to summarize the current evidence on the correlation between pretreatment neutrophil-to-lymphocyte ratios (NLR) and oncological outcomes in each type of management for urothelial carcinoma. Method: All articles published until February 2017 in PubMed, Scopus, and EMBASE database were collected and reviewed. The current evidence on correlations between pretreatment NLR and oncological outcomes in each type of management for urothelial carcinoma, including transurethral resection of bladder tumor (TURBT), radical cystectomy (RCx), chemotherapy (CTx), and nephroureterectomy (NUx), were summarized. Results: Thirty-eight studies containing clinical information on 16,379 patients were analyzed in this study. Pooled hazard ratios (HR) and odds ratios (OR) with 95% confidence intervals were calculated after weighing each study. Heterogeneity among the studies and publication bias were assessed. Pretreatment NLR was significantly associated with muscle invasiveness (OR: 4.27), recurrence free survival (RFS, HR: 2.32), and progression-free survival (PFS, HR: 2.45) in TURBT patients. In the RCx patients, high NLR was negatively associated with both disease status (extravesical extension and lymph-node positivity, OR: 1.14 and 1.43, respectively) and oncological outcomes [overall survival (OS), PFS], and cancer specific survival (CSS, HR: 1.18, 1.12, and 1.35, respectively). Pretreatment NLR was negatively correlated with pathologic downstaging (OR: 0.79) and positively correlated with PFS (HR: 1.30) and OS (HR: 1.44) in CTx patients. For patients who underwent NUx, pretreatment NLR was significantly associated with OS (HR: 1.72), PFS (HR: 1.63), and CSS (HR: 1.68). Conclusions: Pretreatment NLR is a useful biomarker for disease aggressiveness, oncological outcome, and treatment response in the management of patients with urothelial carcinoma. More evidence is needed to clarify these results.

To establish a prospective, comprehensive, multidisciplinary, bio-bank linked genitourinary cancer cohort based on standard real practice. We established the Seoul National University Prospectively Enrolled Registry for Genitourinary Cancer (SUPER-GUC), a prospective cohort clinical database and bio-specimen repository system for prostate cancer (SUPER-PC), renal cell carcinoma (SUPER-RCC), and urothelial cancer (SUPER-UC) at a high-volume, tertiary institution. Each cohort consists of several sub-cohorts based on treatment or disease status. Detailed longitudinal clinical information, and general and disease specific patient-reported outcomes are captured. We use the same evaluation format and questionnaires for all participating departments. Patients' blood, urine, tumor, and normal tissues are collected. The number of registered patients and their basic characteristics are summarized. For the surgical sub-cohort, study participation, bio-specimen, and tissue banking rates are analyzed. Since March 2016, 11 sub-cohorts for all disease statuses have been opened, ranging from low-risk localized to metastatic disease. SUPER-PC, SUPER-RCC, and SUPER-UC enrolled 929, 796, and 1,221 patients, respectively. Study participation, bio-sampling, and fresh frozen tumor banking rates of surgical sub-cohorts were 89.0% to 93.1%, 91.2% to 99.1%, and 56.9% to 79.1%, respectively. SUPER-GUC is a study platform for comparative outcome, quality-of-life, and translational (genetics, biomarkers) research for genitourinary cancer.

PurposeWe aimed to identify, verify, and validate a multiplex urinary biomarker-based prediction model for diagnosis and surveillance of urothelial carcinoma of bladder, using high-throughput proteomics methods. Materials and MethodsLabel-free quantification of data-dependent and data-independent acquisition of 12 and 24 individuals was performed in each of the discovery and verification phases using mass spectrometry, simultaneously using urinary exosome and proteins. Based on five scoring system based on proteomics data and statistical methods, we selected eight proteins. Enzyme-linked immunosorbent assay on urine from 120 patients with bladder mass lesions used for validation. Using multivariable logistic regression, we selected final candidate models for predicting bladder cancer. ResultsComparing the discovery and verification cohorts, 38% (50/132 exosomal differentially expressed proteins [DEPs]) and 44% (109/248 urinary DEPs) are consistent at statistically significance, respectively. The 20 out of 50 exosome proteins and 27 out of 109 urinary proteins were upregulated in cancer patients. From eight selected proteins, we developed two diagnostic models for bladder cancer. The area under the receiver operating characteristic curve (AUROC) of two models were 0.845 and 0.842, which outperformed AUROC of urine cytology. ConclusionThe results showed that the two diagnostic models developed here were more accurate than urine cytology. We successfully developed and validated a multiplex urinary protein-based prediction, which will have wide applications for the rapid diagnosis of urothelial carcinoma of the bladder. External validation for this biomarker panel in large population is required.