Key points are not available for this paper at this time.
Introduction: An increasing emphasis has been placed on addressing SDOH needs. However, SDOH encompass a wide range of intercorrelated social constructs, and these needs often have deeper roots and correlations with one another. This study aims to leverage advanced machine learning algorithms to cluster the US population into groups with similar SDOH characteristics, providing valuable insights into the intricate landscape of SDOH needs. Methods: Survey data on individual-level SDOH and linked electronic health records from 28,778 participants of the All of Us research program is used. An extreme gradient boosting algorithm was performed to choose and rank the forty most influential variables out of 266, and then we run the PAM-lite to determine the optimal number of variables and clusters. Results: Clustering the pre-retirees (45-64) and the elderly (65+) revealed six distinct groups (Table). Insurance, race/ethnicity, and neighborhood safety played a dominating role in clustering pre-retirees into distinct SDOH groups. However, these factors gave way to partnerships and religion among the elderly, signifying a shift in the determinants of SDOH clusters. Clusters with higher SDOH status had substantial lower rates of T2D, Obesity (chronic), and Pneumonia (infectious). Conclusion: Findings can inform the development of targeted SDOH screening and interventions. Disclosure J. Lee: None. Q. Xue: None. P. Li: None. M. Weber: None. Y. Xi: None. G. Garcia: None. M.K. Ali: Advisory Panel; Eli Lilly and Company. H. Shao: Consultant; Eli Lilly and Company.
Lee et al. (Fri,) studied this question.