Cancer polygenic scores were weaker in African-ancestry than European-ancestry veterans

Among 560,287 US veterans, all 13 cancer-specific polygenic risk scores were associated with their own cancer in European-ancestry participants (OR 1.05 to 1.70 per standard deviation), but only 5 of 13 reached significance in African-ancestry participants, with smaller effects (OR 1.01 to 1.48).

Thirteen cancer-specific polygenic risk scores were tested against their matched cancers and 340 cancer and cardiometabolic phenotype codes in 560,287 Million Veteran Program participants, stratified by genetic ancestry. All 13 were associated with their respective cancers among European-ancestry participants (OR 1.05 to 1.70 per standard deviation), whereas among African-ancestry participants effects were smaller (OR 1.01 to 1.48) and only 5 of 13 reached significance. Associations with other conditions were rare, with one between cancers (the renal score with skin cancer, OR 1.04) and seven with cardiometabolic conditions.

Why it is interesting: Quantifies the gap in how well cancer polygenic scores transfer across ancestries, in a cohort large enough to test all 13 scores in African-ancestry participants.

Source: JCO Precision Oncology, 17 September 2026
DOI: 10.1200/po-25-00957
Type: Peer-reviewed article
Design: Cross-sectional association study of 13 cancer polygenic risk scores against 340 phecodes in 560,287 Million Veteran Program participants (114,255 African ancestry, 446,032 European ancestry)
Verdict: Refines prior evidence
Driver: Biotechnology against inherited risk and ageing
Driver: Cancer

Self-guided lipid app cut LDL-cholesterol by 3.9% more than usual care over six months

In a pragmatic online trial of 278 adults with LDL-cholesterol above their risk-adapted target, a fully self-guided lifestyle app added to usual care reduced LDL by 3.9% more than usual care alone at six months, with 41.7% versus 23.8% achieving a fall of at least 10%.

Emergent technology New medicines Cardiovascular disease What changes behaviour at scale

Item new peer-reviewed 18 Sep 2026

Machine learning predicted six-month mortality after Taiwanese care-home admission

An ensemble machine learning model built only from routine intake assessments in 636 Taiwanese long-term care facilities predicted death within 180 days of first admission with an AUROC of 0.90 in a held-out 2024 cohort, where 28.7% of new residents died.

Emergent technology AI in medicine Capacity, funding and care Frailty and multimorbidity at the oldest ages Taiwan

Item new preprint 16 Sep 2026

LLM-extracted autopsy brain features classify neurodegenerative disease with 75% accuracy

A fine-tuned large language model scored 39 features from narrative gross descriptions in 5,613 Mayo Clinic brain autopsies, and classification reached 0.75 accuracy, from 0.93 sensitivity for progressive supranuclear palsy to 0.03 for mixed Alzheimer–Lewy body disease.

Emergent technology AI in medicine United States

Item new preprint 16 Sep 2026

A local multi-agent AI system deidentified multimodal hospital data at 99.8% sensitivity

A locally deployable multi-agent system combining multimodal large language models, task-specific networks and rule-based steps removed 99.82% of injected identifiers across text, imaging, handwriting and audio while preserving 99.61% of clinically important content.

Emergent technology AI in medicine What linked data reveals Germany

Cite this item