CHARACTERIZING HIV CARE TRAJECTORIES THROUGH LATENT CLASS ANALYSIS TO ADDRESS RETENTION IN CARE CHALLENGES

Purpose: Retaining people living with HIV in regular medical care is critical for improving HIV-related outcomes. Currently, retention is considered a binary state: a patient is either “in care” or “out of care”. However, engagement in care is an evolving process, with periods of more or less frequent care encounters. The goal of this analysis is to characterize HIV patient care status not as fixed at a single point in time, but as longitudinal care trajectories.

Methods: The study population consisted of all HIV-infected patients ≥18 years-old seen at a public, hospital-based clinic in Minneapolis, MN between 2008-2013. Care patterns were based on clinic visit and laboratory test data from 2008-2014, ensuring ≥1 year of observation time for all patients. Clinic data was merged with surveillance records to account for HIV care at other clinics, out-of-state relocation, and mortality. Patient-specific HIV care trajectories were constructed in six-month intervals starting from first observed clinical encounter until death, relocation, or the end of 2014. Care trajectories were binary-valued, with a value of 1 in intervals where a patient had at least one clinical encounter and 0 otherwise. Latent class analysis was used to identify care trajectory classes, described by the probability of having a clinical encounter in each six-month interval. The number of care classes was chosen to minimize the Bayesian information criterion (BIC).

Results: Dividing the study population into 5 care trajectory classes minimized BIC, resulting in the following care trajectories (Figure): (1) consistent care over time; (2) initial attrition with a return to care; (3) slow attrition; (4) fast attrition; and (5) moderate attrition. Over half the study population was consistently in care (Class 1). Viral suppression differed across classes, with patients in the fast (Class 4) and moderate (Class 5) attrition classes being 2.7 (95%CI: 2.3-3.1) and 2.4 (95%CI: 2.0-2.8) times more likely to have a detectable viral load at last test, respectively, relative to Class 1. Patients in these classes were also the most likely to change provider or relocate out-of-state.

Conclusions: We demonstrated the feasibility of characterizing retention in HIV care in terms of longitudinal care trajectories. Five intuitive HIV care trajectories emerged, including four distinct patterns of suboptimal care. These results could be used to target patient retention efforts.