AI Tool Predicts Patient Risk for 348 Diseases Using Medical Records

Researchers at Dana-Farber Cancer Institute and Massachusetts General Hospital developed an artificial intelligence tool that examines electronic health records to estimate a patient's risk of developing 348 diseases. The algorithm incorporates routine medical data and genetic test results when available. It was trained on 683,000 patient records, many from the UK Biobank.

The tool outperformed online calculators used for cardiovascular disease within 10 years and breast cancer within one year, according to the researchers. It also identified 20 biological patterns linked to higher disease risk. One cluster of 17 conditions was associated with colorectal cancer risk. The algorithm showed particular strength in predicting colorectal cancer within the next year.

How the Tool Works The system is named Aladynoulli.

It combines references to the folk-tale figure Aladdin and the mathematician Jacob Bernoulli. Giovanni Parmigiani, a Dana-Farber researcher and senior author, said medical records contain useful information across time and disease areas that a machine learning model can process.

Sarah Urbut, an MGH cardiologist and lead author, said the tool improves as patients generate more records through additional visits. She noted that notes from one specialty, such as rheumatology, may contain information relevant to another, such as cardiology.

The researchers published their findings Wednesday in the journal Nature. They said the tool is ready for use at the two hospitals but did not provide a timeline for broader availability.