Healthcare AI Daily Briefing
New to OpenRounds? Get the free daily email or see how we work.
Research & Evidence
Multimodal supervised pipelines improve verbal-autopsy cause-of-death assignment in South Africa.
Preprint
In a medRxiv preprint, supervised models combining narrative and structured VA inputs on 4,999 records improved top-1 accuracy over InSilicoVA by 8.5‑15.9 percentage points, using physician-assigned cause-of-death as reference. Narrative and structured unimodal pipelines performed similarly overall but differed by cause, with narrative more accurate for noncommunicable deaths and questionnaire for communicable deaths. [1]
Diagnostics & Pathology
Longitudinal Mamba model predicts IBD ~0.97 AUROC with 20‑month lead time.
Preprint
In a study of adults with symptoms at UCSF (2012‑2024), a longitudinal Mamba model identified IBD six months after presentation among 358,886 patients not used for training, achieving an AUROC of ~0.97 and a median lead time of ~20 months; external validation at ZSFG showed similar discrimination, and prospective evaluation is needed to assess whether earlier identification reduces diagnostic delay. [2]
Clinical Practice
Interview flags two‑year med‑school break as likely misaligning memorization with future duties.
Podcast from AI and Healthcare
The interview describes that delaying medical school for two years would likely leave memorization skills unused, resulting only in missing gradual skill improvement like learning a new sport. [4]
Diagnostics & Pathology
SIEVE framework shows trade‑off between classification accuracy and biological signal recovery.
Podcast from Health and Explainable AI Podcast
The interview describes that in their SIEVE framework, classification performance is best when only genomic architecture is provided, whereas recovery of meaningful biological pathways and functions improves with richer biological annotation. This pattern emerged in pilot studies of cardiovascular cohorts comprising 2,000‑4,000 samples, highlighting a tension between predictive accuracy and interpretability that the team plans to explore further. [3]
Sources
- Multimodal Artificial Intelligence for Cause-of-Death Assignment From Verbal Autopsy Records: A National Benchmark Study in South Africa · medRxiv Original source
- Prediction of incident inflammatory bowel disease from routine electronic health records: development and external validation of a longitudinal machine learning model · medRxiv Original source
- Dr. Francesco Lescai on Computational Genomics, the SIEVE Framework and Explainable AI for Discovery · Health and Explainable AI Podcast Original source
- Will AI Replace Doctors? The Case for Autonomous AI in Medicine with Ezekiel Emanuel, MD, PhD · AI and Healthcare 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. Get a sign-in link by email to finish subscribing. An account is created if needed. Unsubscribe anytime. Already subscribed? Sign in to save and personalize.