Natural Language Processing Extracted Social And Behavioral Determinants Of Health And Newer Glucose-lowering Drug Initiation Among Real-world Patients With Type 2 Diabetes

Jingchuan Guo,Yonghui Wu,Yi Guo,Aokun Chen,Zehao Yu,Xi Yang,Jared W. Magnani,Inmaculada Hernandez,LaToya O'Neal,Elizabeth Shenkman,Jiang Bian
DOI: https://doi.org/10.1161/circ.145.suppl_1.P108
IF: 37.8
2022-01-01
Circulation
Abstract:Background: Sodium-glucose cotransporter-2 inhibitors ( SGLT2i ) and glucagon-like peptide-1 receptor agonists ( GLP1a ) are two newer glucose-lowering drugs ( GLD ) that can reduce cardiovascular disease ( CVD ) risk in patients with type 2 diabetes ( T2D ). Yet, uptake of these newer drugs remains low, especially among racial minority groups. Prior studies evaluating factors associated with underuse of SGLT2i/GLP1a primarily relied on claims or structured electronic health records ( EHR ) data, which contain little information on patients' social and behavioral determinants of health ( SBDoH ). We hypothesized that SBDoH would affect the use of these cardioprotective drugs among real world patients with T2D. Methods: EHR from University of Florida Health system were used to identify T2D patients who initiated a second line GLD in 2015-2020. We used a deep learning based natural language processing (NLP) pipeline to extract SBDoH from clinical notes: employment, housing, marital status, smoking, alcohol use and drug use. We constructed a multilevel logistic regression to determine the association of race/ethnicity and SBDoH with initiation of SGLT2i/GLP1a versus other second line GLDs, including zip codes as random effect and adjusting for clinical factors. Results: Of 2,964 patients in the cohort, mean age was 59 (14) years and 56% were females. In the adjusted model ( Table ), Black race, male sex, unemployment, housing insecurity, being unmarried, smoking, and drug use were associated with lower likelihood of initiating SGLT2i/GLP1a. In subgroup analyses by race, housing insecurity was only a risk factor for non-initiation of SGLT2i/GLP1a among Black patients, and smoking and drug use were only risk factors for White patients. Conclusion: Social and behavioral risk factors contribute to racial disparities in access to novel GLDs. Advanced NLP methods and EHR data enabled us to investigate the social risk associated with real-world treatment access.
What problem does this paper attempt to address?