DENSE electronic health record analysis has revealed several distinct obesity clusters among patients beginning antiobesity medication, suggesting deep phenotyping could guide more personalised treatment. The findings show the potential of health records for identifying clinically meaningful subgroups before therapy is started.
Obesity now affects nearly 40 per cent of adults and up to 20 per cent of children in the United States, creating major challenges for health systems. Clinicians currently face difficulties in predicting which patients will respond best to available antiobesity medications, driving interest in precision medicine and deep phenotyping approaches. Identifying subgroups with unique characteristics could improve therapeutic strategies and outcomes.
Researchers examined 53,688 time periods preceding antiobesity medication in 32,969 patients with obesity or overweight, using data from electronic health records. Ninety-two laboratory and vital signs measures were analysed alongside 79 diagnostic categories from ICD-derived codes in the year prior to starting treatment. The team applied a longitudinal deep learning model to the data, followed by principal component analysis and Gaussian mixture modelling to uncover clusters. Their work revealed at least nine distinct clusters, five of which had clear and explainable clinical features. Several clusters overlapped with traditional obesity phenotypes, while others suggested novel subgroups. Importantly, clustering was reproducible across multiple training folds, confirming clinical significance. Despite promising results, some challenges remained, especially in visualising complex disease patterns, imputing missing data and ensuring consistency across input features.
For clinical practice, this study supports the use of detailed health record modelling to segment obesity into subtypes before prescribing medication. Further research with larger, independent populations is needed to validate these clusters and refine their clinical utility, potentially transforming how clinicians approach obesity pharmacotherapy.
Reference
Ruan X et al. Deep phenotyping of obesity: electronic health record–based temporal modeling study. J Med Internet Res 2025;27:e70140.
