Many older patients with heart failure survive hospitalization but lose independence within a year after discharge. Researchers from Japan developed a machine learning model using physical function and clinical data to predict this risk. The model outperformed a standard frailty screening tool and identified a group with low mortality but high functional decline. The findings could help clinicians prioritize rehabilitation and discharge support for patients most likely to lose independence after discharge and improve care.

The glacial Arctic Ocean was not a barren ice desert but a breeding ground for the evolution of cold adaptation in marine fish. This is the main conclusion of a new scientific publication from Stockholm University that examines small fish bones found in deep-sea sediments during an expedition aboard the Swedish Icebreaker Oden.

Public biomedical databases such as NHANES, UK Biobank, FAERS, the Global Burden of Disease Study, and FinnGen have become a focus of the research-integrity debate, following a rapid rise in papers built from broadly accessible datasets. A correspondence published in Medicine Discovery argues that the quality of public database research rests on the study process, while the database name provides little information about scientific quality, and that treating a familiar database name as a marker of low quality risks confusing the infrastructure with the production model that exploits it. Citing a scientometric analysis of 36 health datasets — nine of which showed publication patterns consistent with rapid, formulaic expansion, producing 23,005 indexed publications in 2025 and an estimated excess of 11,577 publications above the prior trend — the authors argue that the same NHANES files support careful epidemiology, surveillance, and methods development as well as formulaic work, and that the recent surge emerged long after the survey architecture and its public documentation were established. They describe how generative AI has changed the economics of secondary data research, distinguishing researcher-directed assistance from more autonomous workflows that extend into question formulation and analytical planning, and warning of a plausible risk of automated hypothesis fishing. The correspondence proposes three actions for editorial policy: a short dataset-specific methods note for heavily reused resources; stronger transparency around analysis code, preregistration, and AI use; and editorial screening focused on study conduct rather than data provenance.