(086) Applying Natural Language Processing of Clinical Notes to Determine Social Determinant Risk Factor Status for Patients Receiving Glucagon-Like Peptide-1 Agonist Treatment

Background: Glucagon-Like Peptide-1 (GLP-1) agonists are increasingly being used to treat chronic conditions including type 2 diabetes mellitus and obesity. The social determinants of health (SDoH) status of patients receiving such treatment is poorly understood.

Objectives: Our objective was to determine the SDoH status of patients at the time of GLP-1 agonist prescription or administration using a previously developed N-gram model applied onto unstructured clinical notes.

Methods: In this descriptive analysis, structured and unstructured clinical electronic health record (EHR) data were collected for patient encounters from integrated delivery networks (IDNs) in the OMNY Health real-world data platform (United States) over a 6-year period (2017-2022) during which a GLP-1 agonist (tirzepatide, semaglutide, or liraglutide) was prescribed or administered. Demographic characteristics (including mean, standard deviation, counts, and/or percentages, where appropriate) of these patients, including age, sex, race, and payer status were tabulated. In parallel, previously developed and fine-tuned n-gram models for the detection of SDoH risk status across 5 domains (economic insecurity, housing insecurity, undereducation, social isolation, and transportation insecurity) were applied to the notes of those encounters. Percentages of encounters scoring positively for the SDoH domains were tabulated.

Results: A total of 322,840 encounters across 97,865 patients were included. The mean age during the encounters was 54.8 (standard deviation: 12.5). 209,453 (64.9%) of the encounters were for female patients, and 241,126 (74.7%) of the encounters were for white patients. 103,658 (32.1%) of the encounters were covered by private insurance, while 12,899 (4.0%) of the encounters had Medicaid coverage. 1991 (0.6%) of the encounters were positive for at least one SDoH risk factor [economic insecurity: 1832 (0.6%); social isolation: 198 (0.1%); housing insecurity: 89 ( < 0.1%); transportation insecurity: 63 ( < 0.1%); undereducation: 16 ( < 0.1%)]. At the patient level, 1,732 (1.8%) of included patients were positive for at least one SDoH risk factor at any time during their visit history.

Conclusions: Compared to the overall SDoH risk factor prevalence across all encounters (1.2%), SDoH risk factors were less prevalent in encounters associated with GLP-1 prescription (0.6%). This result may suggest that patients receiving GLP-1 agonists are either less likely to have SDoH risk factors in general or to have them documented in their EHR data. Further research is required to determine definitively the SDoH risk in this population.