Apollo: AI model trained on 33 years of hospital records
A preprint describes Apollo, an AI model trained on records from 7.2 million patients at US hospitals. The authors report good results on 322 clinical tasks, but the work has not yet been peer reviewed.
Source: Apollo: AI model trained on 33 years of hospital records (arxiv.org). Summary written with AI from the source and reviewed by rules; read the original.
What the research shows
The preprint describes Apollo, an artificial intelligence model trained on records from 7.2 million patients at hospitals in the United States, collected over 33 years. The model combines text, images and structured data, such as information organized in fields. According to the authors, Apollo performed well on 322 clinical tasks, including disease prediction.
Work of this kind has not yet been reviewed by other experts before publication, so the results should be read as preliminary. A preprint is a version shared before formal review by a scientific journal, and its conclusions may change after review. A good performance on computer tasks also does not prove that the model improves patient care in practice. That would require testing in real health services, with follow-up of effects on patients.
What it means
The study suggests that complete patient histories could be used by AI systems to estimate risks and support clinical tasks. It is not yet known whether these results hold in other hospitals, populations and countries, or whether they help patients safely.
How to read it with care
A result like this does not replace a medical consultation. If you have questions about your health, talk with your doctor or another qualified professional before making any decision.