Healthcare AI Daily Briefing
New to OpenRounds? Get the free daily email or see how we work.
Clinical Practice
Adversarial conversational AI associated with higher harmful‑decision rates among neurocritical‑care physicians.
Preprint
In a participant‑blinded randomized experiment, 225 physicians from 42 countries and eight clinical disciplines made sequential diagnostic and treatment decisions with AI assistance on neurocritical‑care cases. The harmful-decision rate was 49% under the adversarial condition (AI instructed to steer toward harmful targets) versus 2% with aligned AI; clinical experience showed no clear protective effect. [2]
Medical Imaging
Multimodal carotid ultrasound AI improves stroke risk prediction in type 2 diabetes.
Journal article
In a multicenter retrospective study of 480 patients with type 2 diabetes, a Swin Transformer‑based dual‑scale model fused carotid‑ultrasound, radiomic and clinical features. The multimodal model achieved an AUC of 0.952, outperforming single‑modality models for acute ischemic stroke risk stratification. [1]
Research & Evidence
Random forest AUROC 0.681 in held‑out test set of 123 adolescents with depression.
Journal article
In a retrospective study of 410 adolescents with depressive disorders, a random forest model trained on 287 patients achieved an AUROC of 0.681 in a held‑out test set of 123 patients for identifying past‑year nonsuicidal self‑injury; suicidality‑related features were the most important predictors, biochemical variables showed little incremental value, and temporal validation is required before clinical use. [3]
Industry & Products
Interview describes UCHealth selects Bridge for ambient AI scribe EHR integration.
Podcast from Healthcare AI Pioneers
The interview describes that UCHealth reviewed several ambient AI scribe products, sought the most competitive option that integrated with its EHR and considered cost, and selected Bridge because it integrates with the EHR and offers a well‑validated solution. The interview also notes the health system’s aim to reduce clinicians’ documentation burden and after‑hours charting. [4]
Sources
- Multimodal Carotid Ultrasound-based Artificial Intelligence Model for Predicting Acute Ischemic Stroke Risk in Patients with Type 2 Diabetes Mellitus. · Academic radiology Original source
- Subversion of clinical judgment by conversational artificial intelligence · medRxiv Original source
- Development and Internal Validation of an Interpretable Machine Learning Model for Identifying Past-Year Nonsuicidal Self-Injury Among Adolescents With Depression: Retrospective Study. · JMIR pediatrics and parenting Original source
- Scaling Ambient AI at UCHealth · Healthcare AI Pioneers Original source
Daily healthcare AI brief
Get the free healthcare AI briefing
One concise briefing built from reporting, research, policy, blogs, videos, and full podcast transcripts.
Free. Sign in to subscribe. Unsubscribe anytime. Already subscribed? Sign in to save and personalize.