Poster #: 72B
Session/Time: B
Author: Mohamad Najafi, MS, ODU Computer Science
Mentor: Hongyun (Tracy) Fu, PhD
Research Type: Clinical Research

Abstract

Background

Adolescent syphilis is rising and rarely occurs alone. Syndemic theory holds that infections cluster and interact with behavioral-health, trauma-related, substance-use, and structural conditions within adverse social environments. Because many such factors appear only in narrative clinical notes, analyses based on structured diagnosis codes miss them. Knowledge graphs represent patients, conditions, and social factors as linked, source-anchored entities, providing a computational representation of syndemic theory. We piloted an evidence-linked knowledge graph to characterize syndemic co-occurrence in adolescents with syphilis.

Methods

Retrospective, mixed-methods pilot of 14 adolescents (median age 18 years; 10 male; 10 Black/African American) with confirmed syphilis across 48 encounters at a coastal Eastern Virginia pediatric health system (2017-2024). Structured electronic medical record (EMR) fields (coinfections, insurance status, Area Deprivation Index) and clinical notes were mapped to a six-domain syndemic codebook: infectious coinfection, behavioral health, trauma/vulnerability, substance use, social determinants of health, and healthcare-system fragmentation. Each factor was linked to its source document; narrative factors were extracted using a locally hosted large language model (Llama-3.3-70B). Patient-level graphs were aggregated into a cohort knowledge graph.

Results

Every patient had at least one syndemic domain; the mean was 3.4 of 6 domains, and 71% had three or more. Domain prevalence was: social/structural 86%, healthcare-system fragmentation 86%, infectious coinfection 50% (human immunodeficiency virus [HIV] 21%), behavioral health 50%, substance use 43%, and trauma/vulnerability 29%. Behavioral-health, substance-use, and trauma factors were documented in narrative notes for 7, 6, and 4 patients, respectively, but appeared in zero structured diagnosis codes. The cohort graph showed clustering of neighborhood deprivation, care fragmentation, behavioral health, and substance use.

Conclusion

An evidence-linked knowledge graph feasibly represented co-occurring clinical and social vulnerabilities in adolescent syphilis, operationalizing syndemic theory as an interpretable, auditable structure and revealing burden invisible to structured diagnosis codes. This single-site pilot cannot establish causal syndemic interaction, but it provides a validated foundation for temporal knowledge graphs and a larger adolescent HIV and sexually transmitted infection (STI) cohort.