Lexa Jones
Pronouns: she/her
Research Mentor(s): Elham Mahmoudi
Research Mentor School/College/Department: Family Medicine / Medicine
Program: UROP
Session: Session 4 (1:40pm – 2:30pm)
Authors: Lexa Jones, Elham Mahmoudi
Abstract
Although Social Determinants of Health (SDoH) can be critical in determining proactive treatment for individuals with Alzheimer’s disease and related dementia (ADRD), SDoH data is often not stored systematically in hospitals. It is typically kept in unstructured electronic medical records (EMRs). This study develops a rule-based Natural Language Processing algorithm that
identifies social determinants of health from the EMRs of hospital patients with
ADRD. This study used 1,000 medical notes randomly selected from 7,401 emergency department and social work notes created between 2015 and 2019 for 231 patients with ADRD in Michigan Medicine. These notes were used to develop the algorithm to be able to identify 7 domains of SDoH: housing, transportation, food and medication insecurities, social isolation, abuse, neglect and exploitation, and financial difficulties. This rule-based algorithm was also compared with a deep learning and regularized logistic regression approach through their accuracy, sensitivity, specificity, F1 score, and the area under the receiver operating characteristic curve (AUC). The F1 and AUC for the rule-based algorithm were at least 0.94 and 0.95, respectively, for all SDoH categories after using the 700 notes for training. After using 300 notes for validation, the F1 and AUC were at least 0.80 and 0.97, respectively, for all SDoH except housing and medication insecurities. This demonstrates that the rule-based algorithm can accurately extract SDoH information in all domains tested except housing and medication. This algorithm could be used by clinicians to address the social needs of patients with ADRD.
