Nvidia AI predicts diabetes and CVD over a decade better than HbA1c results, Nature study suggests
AI could potentially also make continuous glucose monitoring more useful for people without diabetes by predicting glycaemic responses.
An AI model from tech giant Nvidia was better than HbA1c results at predicting the risk of diabetes and cardiovascular disease over an 11-year longitudinal study, a Nature study has concluded.
GluFormer, a model trained on continuous glucose monitoring (CGM) data, could also predict a person’s glucose response to specific foods, thereby improving CGM, the researchers say.
Nvidia says the tool has a similar structure to a large language model but is trained on CGM data rather than text — specifically, 10 million readings from 10,812 adults.
Applied to a longitudinal epidemiological study of 580 Spaniards with prediabetes, GluFormer was used to stratify subjects into quartiles based on their predicted risk of developing diabetes.
