Radically different
Nature Medicine, Published online: 28 April 2026; doi:10.1038/s41591-026-04372-z
Genomic screening is increasing rapidly, but insurance protections globally are outdated. Australia’s new law offers the clearest model yet for safeguarding public trust while supporting population genomics.
Nature Medicine, Published online: 28 April 2026; doi:10.1038/s41591-026-04267-z
A reasonably likely surrogate endpoint for metabolic dysfunction–associated steatohepatitis
Nature Medicine, Published online: 28 April 2026; doi:10.1038/s41591-026-04368-9
As artificial intelligence becomes embedded in clinical workflows, trials must accommodate ongoing monitoring and updates.
Nature Medicine, Published online: 28 April 2026; doi:10.1038/s41591-026-04359-w
Reti-Pioneer, a retinal imaging artificial intelligence framework, identified diverse systemic diseases and demonstrated feasibility in a primary care silent trial, offering a pathway toward scalable clinical evaluation.
The ventral tegmental area (VTA) is a key node within the limbic circuitry. Through dense dopaminergic, glutamatergic, and GABAergic projections, the VTA forms reciprocal loops with prefrontal and limbic cortices that are consistently implicated in major depressive disorder (MDD) and obsessive–compulsive disorder (OCD) (1,2). Decades of animal research have established the VTA as a central hub for motivational drive and reward prediction error signaling (3,4). Despite its presumed critical role in mental disorders, direct electrophysiological recordings from the human VTA have so far remained absent.
Get your daily dose of health and medicine every weekday with STAT’s free newsletter Morning Rounds. Sign up here.
Good morning. The other night I watched a shocking episode of “The Vampire Diaries.” A series of cursed, ghost-like hallucinations attempt to convince a teen vampire to end her own life using some disturbingly coercive, cogent arguments. Ultimately, the character is saved. And while this episode aired more than a decade ago, I was surprised by how many parallels there were to current debates about the risks of AI chatbots and people in mental health crises.
Eli Lilly struck a deal Tuesday to develop new forms of gene editors potentially capable of inserting entire genes into patients.
The collaboration, with artificial intelligence-focused biotech Profluent, is sparse on details, including the number of programs the two companies would work on, the types of diseases they’ll pursue, or how much Lilly was paying upfront. But if every one of its efforts works out, Lilly would pay Profluent $2.25 billion in milestones payments.
The deal is part of a larger push by Lilly into gene editing. The big pharma, flush with record revenues from its obesity and diabetes drugs, has opened a new genetic medicine center in Boston and bought up a series of gene editing or gene therapy companies over the last few years.
Background: Multimodal large language models (MLLMs) capable of integrating visual and textual information represent a promising advancement for clinical applications requiring image interpretation. Wound care assessment, which demands simultaneous analysis of wound photographs and clinical data, provides an ideal domain to evaluate multimodal vs unimodal artificial intelligence capabilities against human expertise. Objective: This study aims to compare the performance of MLLMs, unimodal ChatGPT 5.0, and human clinical experts on a standardized wound care certification examination. Methods: This cross-sectional comparative study evaluated 3 participant groups on a 25-question wound care certification examination spanning 4 clinical domains (Diagnosis, Treatment, Complication Management, and Wound Subtype Knowledge). Participants included 3 MLLMs (Med-PaLM 2, LLaVA-Med, and BioGPT), 1 unimodal large language model (ChatGPT 5.0), and 4 human clinical experts (general surgeon, wound care nurse, and 2 internal medicine physicians). Statistical analyses included one-way ANOVA with Tukey post hoc tests and domain-specific Kruskal-Wallis comparisons. Results: Human experts achieved the highest accuracy (mean 86%, SD 9.1%), followed by MLLMs (mean 78.7%, SD 12.2%), while ChatGPT 5.0 achieved 64% accuracy, failing the 70% certification threshold. Significant overall group differences were observed (=8.42, =.02, η²=0.74). MLLMs significantly outperformed ChatGPT 5.0 (difference=14.7 percentage points, =.03, Cohen =1.38), with the multimodal advantage most pronounced in visually dependent domains: Diagnosis (81% vs 43%, =.008) and Complication Management (72% vs 50%, =.03). No multimodal advantage was observed for text-based Wound Subtype Knowledge (both 67%). Med-PaLM 2 achieved 92% accuracy, matching that of the wound care nurse, while the general surgeon achieved the highest overall performance (96%). Conclusions: MLLMs demonstrate significant performance advantages over unimodal artificial intelligence in wound care assessment, particularly for visually dependent clinical tasks. While human experts with specialized wound care experience maintain overall superiority, the point estimate of the top-performing MLLM (Med-PaLM 2, 92%) fell within the observed range of human scores; however, the underpowered comparison (power=0.52) and wide CIs preclude definitive conclusions regarding noninferiority or equivalence to human experts. These findings support the potential role of MLLMs as clinical decision-support tools, warranting further adequately powered validation studies.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/96b26bb006d808b607e07c1c5b36f0ca" />