Background: Large language models (LLMs) such as ChatGPT are increasingly used to support health-related queries and decision-making. However, these models can be “jailbroken” through adversarial prompts that bypass safety filters and elicit harmful or medically inappropriate responses. In health care contexts, such vulnerabilities pose serious risks. Understanding how jailbreak susceptibility varies across languages is essential for developing robust safeguards and promoting equitable access to safe health information. This paper may contain examples that may be deemed harmful in terms of violence, self-harm, and drug abuse. Objective: This study aims to systematically compare and contrast the vulnerability of a health LLM for jailbreaking across 3 languages: English, Spanish, and Hindi (transliterated using the Latin alphabet), based on emoji and permutation cipher attacks. Methods: We analyzed 1000 input prompts per language, drawn from the BeaverTails dataset, across 3 harm categories: self-harm, violence, and drug abuse. Each prompt was modified using emoji and permutation cipher techniques, resulting in 6000 input-output pairs. Model responses were evaluated by human coders to determine the success rate of jailbreak attempts across languages and cipher types. Results: Hindi prompts showed the highest vulnerability, with 787 successful jailbreaks using emoji ciphers and 873 using permutation ciphers. Spanish and English followed, with lower success rates across both cipher types. Differences in jailbreak success across languages and cipher strategies were statistically significant. Additionally, attacks targeting violence-related prompts were more successful overall than those targeting drug-related or self-harm content, indicating variation in vulnerability by harm type. Conclusions: The findings of this formative study reveal that LLM safety performance varies substantially across languages and harm categories, raising concerns about equitable protection in multilingual health communication. Disparities in access to harmful content may contribute to downstream health risks. Strengthening multilingual content moderation and developing language-aware safety mechanisms are critical steps toward creating safer and more inclusive health AI systems.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/7cc7e7428d95dc1ba253abcc22be9b19" />
The Role of eHealth Literacy and Patient Adherence in Mediating Health Consciousness and Perceived Severity in Quality of Life Among Young Patients With Ischemic Heart Disease: Cross-Sectional Study
Background: Ischemic heart disease (IHD) is becoming increasingly prevalent, with a rising trend significantly impacting the quality of life (QoL) of young Malaysians. Objective: This study aimed to investigate the direct and indirect relationships between eHealth literacy (eHealth) and patient adherence (PA), as well as their mediating effects on the associations of health consciousness (HC) and perceived severity to chronic disease (PS) with QoL among young patients with IHD. Methods: A cross-sectional study was conducted at 2 hospitals, recruiting eligible patients through consecutive sampling at outpatient cardiology clinics. Data were collected via a validated self-administered questionnaire encompassing sociodemographic and socioeconomic status, medical history, and PS, HC, eHealth, PA, and QoL data. Structural equation modeling analysis was used to evaluate the relationships. The ethics approval was obtained from the Ethical Committee of Universiti Kebangsaan Malaysia Medical Centre (FF-2021-117). Results: A total of 136 young patients with IHD participated in the study. Structural equation modeling analysis revealed that eHealth had the strongest positive effect on QoL (β=0.287, =.002), followed by PA (β=0.245, =.02), and HC (β=0.218, =.02). Two significant mediation models were identified, aligning with the Transactional Model of eHealth Literacy theory. The first model demonstrated parallel mediation, where eHealth (β=0.125, =.008) and PA (β=0.083, =.046) significantly mediated the relationship between HC and QoL. The second model indicated serial mediation through eHealth and PA between HC and QoL (β=0.042, =.049). The parallel mediation model exhibited medium predictive power and was deemed the best-fit model. Conclusions: The parallel model pathway showed significant direct and indirect associations between eHealth, PA, HC, and QoL, with eHealth demonstrating the strongest association. Higher eHealth and PA were associated with QoL among young patients with IHD; interventions to support eHealth warrant further investigation in longitudinal or interventional studies.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/082c226f20ce15b8f73dbb00efde1f4f" />
Virtual Reality–Based Training in Radiologic Technology for Contrast-Enhanced Computed Tomography Brain Imaging: Randomized Controlled Trial
Background: Radiologic technology (RT) education faces challenges in bridging theory and practice due to limited clinical opportunities. While virtual reality (VR) enables safe and repeatable practice, a systematic instructional design framework is needed to develop scalable, procedure-focused modules. Objective: This study evaluates the Radiologic Technology Virtual Reality (RTVR) framework that integrates 360-degree video capture, instructional overlays, interactive assets, and an immersive content authoring platform to deliver a contrast-enhanced computed tomography (CECT) brain scan module. Methods: In this open-label, parallel-group, randomized controlled trial, 36 year-2 and year-3 RT students with no prior clinical training in diagnostic radiology at a university hospital in Thailand were randomly allocated (1:1) to a VR group or a conventional document-based instruction (control) group. The VR group completed the VR module, a grounded instructional design framework using 360-degree videos and a structured prebrief and debrief, for 20 minutes using a head-mounted display. The control group studied standard curriculum materials for the same duration. Blinding of participants was not possible. Outcome assessment was blinded. The primary outcome was declarative knowledge gain, assessed using a 20-item multiple-choice test before and after intervention. Secondary outcomes included technology acceptance, student satisfaction, and physiological responses during VR immersion. Results: All 36 randomized participants (VR: n=18, control: n=18) completed the study and were included in the analysis. Experts validated the module as suitable and highly appropriate. Students reported high technology acceptance and satisfaction. Both VR and conventional methods produced substantial gains in declarative knowledge. No statistically significant difference in knowledge gain was detected between groups (test × group: unstandardized regression coefficient β=.056, 95% CI −1.360 to 1.473, =.94). Year-2 students, who had less prior clinical exposure, showed larger pretest to posttest knowledge gains compared to year-3 students. Physiological monitoring showed a reduction in heart rate across the session, while blood pressure remained stable. No adverse events or VR-related discomfort requiring discontinuation was observed. Conclusions: The RTVR framework, which uses a real 360-degree video of authentic clinical settings, offers a scalable approach to procedural VR content creation without requiring specialist technical skills, distinguishing it from prior VR studies in radiography. These findings support the RTVR framework as a feasible, evidence-informed supplement to RT curricula for knowledge-focused procedural teaching, with learning outcomes comparable to those of conventional instruction in this context. Trial Registration: Thai Clinical Trials Registry TCTR20260309005; https://www.thaiclinicaltrials.org/show/TCTR20260309005 (retrospectively registered)
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/81ae0f89c861f97c5ded943343e25478" />
Helping Older Veterans Use Mental Health Apps: Qualitative Interviews and Development of a New Program
Background: Mobile mental health apps may provide an accessible, scalable, and private avenue for older veterans who may not otherwise seek or receive care to address their mental health concerns. However, older veterans may experience barriers to using these apps that need to be addressed to facilitate effective use. Such support could be effectively implemented within the US Veterans Health Administration to facilitate the use of the United States Department of Veterans Affairs’ established mental health apps and to benefit older veterans with mental health concerns. Objective: This study aimed to (1) assess older veterans’ interest in and barriers to using mental health apps to address problems such as difficulties with social connection, and (2) develop and pilot a coaching program to address barriers that older veterans experience in using mobile devices and apps. Methods: Rapid qualitative analysis of semistructured qualitative interviews with 12 older veterans identified themes regarding interest, barriers, and preferences for support for using mobile apps. These themes informed the development of a coaching program, which was piloted with 13 older veterans to assess acceptability, feasibility, and resultant signals of changes in mobile device proficiency. Results: Most veterans expressed interest in using mental health apps. One of the most common barriers was familiarity and proficiency with mobile devices and app technology. Other common barriers included usability or accessibility of the technology or app, motivation, and memory. Veterans reported interest in receiving coaching support. Though the majority of veterans expressed some preference for more individualized and in-person support, they identified both benefits and drawbacks to all potential coaching modalities (group vs one-on-one, in-person vs remote)—including issues of individualizable and guided assistance, feasibility and accessibility of the support, and group settings as potential avenues for social connection as well as potential susceptibility to challenging social dynamics and interactions. Mobile Device and App Learning (MoDAL)—a 2-session, interactive, remote educational group—was developed and piloted. Most veterans who participated found MoDAL helpful. Participants’ mobile device proficiency showed a statistically significant improvement on average pre- to post-MoDAL, although this effect was small, and the small sample size limits the strength of the conclusions. Conclusions: Older veterans do have some interest in using mobile mental health apps to address mental health–related issues. However, they experience critical barriers, including a lack of familiarity and proficiency with the technology. MoDAL may improve older veterans’ comfort and proficiency with mobile devices and apps, which could address one of the barriers that impacts downstream engagement in mental health apps and other virtual care modalities.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/331b6123baca5e2da0ceee858e8013ba" />
<![CDATA[APA session probes ADHD stimulant prescribing, contrasts population studies with RCTs, and explores education and brain effects.]]>
<![CDATA[Researchers at the ASCP Annual Meeting reported on how impaired brain insulin signaling may predict worsening depression severity and the links between metabolism and mood.]]>
<![CDATA[ASCP 2026 leaders share practical ways to engage patients in acute psychosis—active listening, noncollusive language, and rapport before assessment.]]>
Heart Attack Byproduct Linked to Brain Inflammation and Cognitive Decline
Researchers at the University of Ottawa has found that a heart attack may trigger neurological disorders and cognitive decline, and have identified the reactive molecule methylglyoxal as a potential driver of neuroinflammation after myocardial infarction. The research, published in the journal Advanced Science, adds to a growing body of evidence that injury to the heart can directly alter brain function through the “heart-brain axis.”
“Methylglyoxal has been widely studied for its role in metabolic diseases, including diabetes, but much less is known about its function in other diseases. In a previous study, we discovered that methylglyoxal was produced by dying heart tissue after a heart attack. Based on this evidence, we predicted that methylglyoxal in the blood would target other organs and tissues, including the brain—and this is what we did indeed observe,” said senior author Erik Suuronen, PhD, a professor in the department of surgery and director of the BEaTs research program at the University of Ottawa Heart Institute.
The researchers found that methylglyoxal-derived advanced glycation end products (MG-AGEs) accumulated in the brains of mice within hours after myocardial infarction and remained elevated seven days later. This accumulation resulted in neuroinflammation, activation of microglia and macrophages, increased inflammatory signaling, and disruption of the blood-brain barrier.
The study examined five brain regions in female and male mice after they induced myocardial infarction. Brain tissue was collected at six hours and seven days after the cardiac event. Researchers measured levels of MG-H1, the major MG-derived advanced glycation end product, along with glyoxalase-1, inflammatory cytokines, receptors for AGEs, activated microglia, and macrophages.
The work builds on a multiple earlier research studies linking cardiovascular disease and neurological disorders. Prior clinical studies have found that myocardial infarction and heart failure are associated with increased risks of dementia, depression, anxiety, and cognitive impairment. Patients with depression after myocardial infarction also face a greater risk of future cardiovascular events.
The heart-brain axis is the bidirectional relationship between cardiac and neurological function similar to the gut-brain axis. Changes in heart function can influence the brain, while neurological disease can also affect cardiovascular health. While the biological factors relating to this interaction are still emerging, the new research indicates that methylglyoxal may be one factor contributing to this relationship.
MG is produced as a byproduct of glycolysis and can react with proteins, lipids, and DNA, contributing to cellular dysfunction. Conditions associated with myocardial infarction, such as ischemia, inflammation, oxidative stress, and metabolic shifts toward anaerobic glycolysis, promote MG production while limiting its detoxification.
The mouse studies showed that the brainstem had the highest accumulation of MG-H1 after myocardial infarction, followed by the cortex. These same regions also showed elevated expression of inflammatory markers and receptors associated with MG-mediated signaling. Male mice exhibited higher MG-AGE accumulation and greater neuroinflammation than females in several brain regions.
The team also documented increases in inflammatory markers including NF-κB, TNF-α, and activated microglia across multiple brain regions. The cortex and brainstem demonstrated the highest inflammatory responses. According to the authors, the findings suggest a “region-specific vulnerability to glycation-induced stress post-MI.”
The findings could have implications for better understanding how heart attacks contribute to long-term neurological disease risk, including dementia and Alzheimer’s disease. The researchers noted that MG-AGEs already have recognized roles in neurodegenerative disease and mental health disorders.
The researchers have already used the information from the study to develop a peptide designed to trap methylglyoxal and prevent cellular damage, as a potential new therapy. The peptide will be tested for its ability to protect the brain after myocardial infarction.
“Given the increased risk of subsequent heart attacks or death in heart attack patients who experience depression or anxiety, being able to alleviate these conditions could reduce subsequent major cardiac events and improve the lives of countless patients, filling an urgent unmet clinical need,” Suuronen says.
The researchers said future work will examine how MG-driven inflammation contributes to neuron death and behavioral changes after myocardial infarction. They also plan further research to better understand the glyoxalase detoxification system, including glyoxalase-1 activity and glutathione levels.
The post Heart Attack Byproduct Linked to Brain Inflammation and Cognitive Decline appeared first on Inside Precision Medicine.
<![CDATA[New trials find collaborative care improves OUD and mental health outcomes.]]>
AI Model Predicts Cancer Treatment Response from Tumor Genotype
Researchers at University of California, San Diego have developed a new artificial intelligence (AI) model that can translate a tumor’s complex genetic profile into predictions about how that cancer may respond to treatment. The foundation model, called MutationProjector, was trained on genomic data from more than 30,000 tumors across 10 solid cancer types, and validated through testing across multiple independent patient cohorts. Led by Trey Ideker, PhD, professor of medicine at UC San Diego School of Medicine and director of the Big Data Institute at the University of Oxford, the researchers say the model offers a new framework for connecting cancer mutations to the biological pathways that drive treatment response.
“Genetic sequencing is already routine in cancer care, but we still struggle to fully interpret the many mutations found in a patient’s tumor,” said Ideker, who also holds a second appointment at UC San Diego Jacobs School of Engineering and is a member of UC San Diego Moores Cancer Center. “Our goal with MutationProjector was to build a general-purpose model that can learn from tens of thousands of tumor genomes and turn those mutation patterns into more precise predictions about treatment response.”
Ideker is co-senior and co-corresponding author of the team’s published paper in Cancer Discovery, titled “A foundation model of cancer genotype enables precise predictions of therapeutic response,” in which the authors stated, “These results establish a unifying framework for connecting tumor genotypes to biological mechanisms and therapeutic outcomes.”
Following a cancer diagnosis, one of the next steps is often genetic testing, which helps doctors classify the tumor and decide which treatments to pursue. “DNA sequencing panels—and in particular those that broadly identify alterations in cancer-associated genes—have been widely adopted in the clinic due to their relatively low cost, rapid turnaround, and established relevance to treatment outcomes,” the authors explained.
![Trey Ideker is a professor of medicine at UC San Diego School of Medicine and director of the Big Data Institute at the University of Oxford. [Erik Jepsen / UC San Diego]](https://www.genengnews.com/wp-content/uploads/2026/05/Low-Res_TreyIdekerLab-ErikJepsen-800px-300x218.jpg)
However, while genetic testing is relatively low cost, fast, and has a strong track record in cases where validated genetic biomarkers are available, those cases remain limited, because this type of treatment stratification is currently based on only a small number of known biomarkers. Today, only about 8% of cases are successfully matched to an FDA-approved therapy on the basis of genetics and usually on the basis of a single gene, the team continued. “While this situation may reflect the incomplete scope of genes covered by current sequencing panels, it clearly also reflects a fundamental lack of knowledge about how gene mutations should be interpreted.”
They suggest that an “average” tumor has approximately 11 distinct genetic alterations identified by clinical sequencing, representing a potentially rich source of molecular information, if this information could be used to help select treatment. One of the challenges to matching cancer mutations with treatment outcomes is that most mutations are rare, the investigators pointed out. Another is that individual biomarkers do not function in isolation, but act together to influence drug response.
Unlike existing approaches that rely on a small number of biomarkers, MutationProjector analyzes the broader combination of genetic alterations present in a tumor. The model then uses this information to generate a compact representation of the tumor’s biological state, helping researchers interpret which molecular pathways may be disrupted and, by extension, which treatments may be most effective. “Foundation models, which are pre-trained on large datasets and then applied to solve diverse new challenges with relatively few samples, are especially well positioned to advance precision oncology,” Ideker and team noted.
The investigators trained their foundational model, MutationProjector, using genetic profile data from more than 30,000 tumors samples across different cancer types. They then showed that across several independent cohorts of cancer patients, including those with bladder cancer, lung cancer and melanoma, MutationProjector matched or exceeded existing methods for predicting response to common immunotherapy and chemotherapy treatments. The model also identified both known and unexpected biomarkers associated with treatment outcomes, which could help improve current approaches to genetic testing and patient stratification.
“When applied to predict immunotherapy or chemotherapy resistance across multiple cancer types and cohorts, MutationProjector achieves or exceeds state-of-the-art performance in all contexts,” the team wrote. “It identifies unexpected biomarkers, including KMT2D mutation in immunotherapy sensitivity and joint alteration of SMARCA4 and STK11 in immunotherapy resistance.”
JungHo Kong, PhD, first author of the study and a postdoctoral researcher in the department of medicine at UC San Diego School of Medicine, said, “Many cancer mutations are individually rare, which makes them difficult to study one at a time. By pretraining on a large collection of tumors and integrating molecular network knowledge, MutationProjector can detect patterns that would be easy to miss with conventional biomarker approaches. That gives us a way to move from long lists of mutations toward a more functional understanding of the tumor.”
![First study author JungHo Kong, shown here, is a postdoctoral researcher at UC San Diego School of Medicine. [UC San Diego Health Sciences]](https://www.genengnews.com/wp-content/uploads/2026/05/Low-Res_IMG_8731-1024x768-1-300x225.jpg)
The researchers emphasize that the model was designed not only to make predictions, but also to provide insight into why those predictions are made, which could help when refining biomarkers and treatment strategies. This interpretability is especially important in precision oncology, where clinicians need to understand how tumor genotypes relate to treatment decisions.
The team also hopes to expand the model to additional cancer types and data sources, including international cancer genome datasets and other forms of clinical information, such as imaging, transcriptomics, and electronic health records. “While 30,000+ genomes representing 10 solid tumor types were considered in our study, numerous additional tumor samples are available for expansion of MutationProjector to tumor types such as pancreatic cancer, prostate cancer or sarcomas,” the authors said. “Other future studies should explore the extent to which the MutationProject concept can be applied to further clinical tasks of interest, including application to liquid biopsies of circulating tumor DNAs for early cancer detection.”
Ideker added, “Our results suggest that tumor genome foundation models may help extend the clinical value of sequencing beyond a handful of well-known genes. This could support a more comprehensive and biologically grounded approach to precision oncology.”
The post AI Model Predicts Cancer Treatment Response from Tumor Genotype appeared first on GEN – Genetic Engineering and Biotechnology News.