Identification of Clinical Phenotypes Among People with HIV Using Electronic Health Record Data

Loss to follow-up (LTFU) from HIV care affects nearly half of people with HIV (PWH). This study aimed to identify clinical phenotypes of PWH using electronic health record (EHR) data and assess their association with LTFU. We analyzed 4,316 visits from 849 adults receiving care at an urban Ryan White Clinic (2017-2020), extracting demographic, social history, laboratory, diagnosis, and clinical note features. Latent Class Analysis identified six distinct patient subgroups, and logistic regression assessed their association with LTFU and high viral load within one year of a clinic visit. Analysis of clinical phenotypes revealed the following characteristics: Class 1: higher rates of substance use and cardiovascular disease and mentions of "heterosexual" in notes; Class 2: younger men, new patients, highest rate of LTFU (OR 1.57, 95% CI 1.20-2.05) and elevated viral load (OR 1.74, 95% CI 1.42-2.13); Class 3: primarily women, with frequent mentions of "pregnancy" in notes; Class 4: more likely to be white or Hispanic, fewer mentions of "substance use disorder" in notes, increased LTFU risk (OR 1.39, 95% CI 1.02-1.89) and decreased risk of unsuppressed HIV viral load (OR 0.65, 95% CI 0.50-0.84); Class 5: more sexually active, with mentions of "good adherence" in notes; Class 6: older men, with fewer mentions of "mental illness" or "STI" in notes. In conclusion, we identified six clinically relevant phenotypes in an urban HIV clinic, two of which were associated with LTFU. Our findings could inform strategies to tailor interventions to improve retention in HIV care for PWH.