Retention in care is essential to maximizing antiretroviral therapy coverage and compliance and optimizing patient outcomes. Researchers often rely on cross-sectional snapshots of binary patient status to estimate retention rates and to identify predictors of attrition. This approach provides only a macro-level view, and does not make full use of longitudinal patient information that may describe cyclical processes of engagement, disengagement, and re-entry into care spanning the care continuum. We characterized the longitudinal dynamics and estimated effects of patient-level covariates related to retention in care.
We represent the process of engagement and retention in care using four states: Engaged in care, disengaged from care, lost from care (LFC), and deceased. Then various patient behaviors are described in terms of transition from one state to another such as transition from engaged to engaged (retention), engaged to disengaged, continued disengaged, disengaged to engaged (re-entry into care), disengaged to LFC, and mortality. The state space modeling (SSM) approach for longitudinal data was used to identify barriers of retention and sub-groups at higher risk of cycling out from care. This analysis includes data from the CFAR Network of Integrated Clinical Systems (CNICS), including 31,376 patients who enrolled and followed from 8 different sites in the US between 1996 and 2015.
Following enrollment, we ascertained patient state membership every 200 days to reflect CNICS patient monitoring guidelines. Among engaged patients, probability of retention, disengagement, and death are 86%, 13%, and 1%. Once disengaged, probability of return to care, continued disengagement, LFC, and death are 24%, 58%, 16%, and 2%. The SSM identified some important prognostic factors of retention and other transition dynamics, after controlling for variation due to site and cohort entry year (see Table). In particular, patients with lower CD4 counts, higher viral load, and not on ARV have lower retention rates. Heterosexual males have lower retention rates compared to men who have sex with men. In addition, gender, race/ethnicity, age, and AIDS are associated with disengagement and/or LFC.
Beyond binary retention status, more comprehensive longitudinal patient behaviors uncover dynamic patterns of care engagement. Our findings can be used for policy, clinical, and programmatic purposes to enhance retention in care among those at greatest risk.