A state transition framework for patient-level modeling of engagement and retention in HIV care using longitudinal cohort data

Abstract

The HIV care cascade (or continuum) is a conceptual model describing key benchmarks that people living with HIV (PLWH) must pass through to maximize benefits of antiviral therapy (ART). In most formulations, the optimal pathway consists of (1) HIV diagnosis through testing, (2) linkage to care, (3) engagement and retention in care, (4) initiation of ART through retention, and (5) sustained suppression of viral load. The conceptual model provides a useful framework for defining and evaluating the benchmarks that measure the effectiveness of HIV care, and for developing strategies to improve HIV outcomes for PLWH [1–4]. The HIV care cascade has become a framework for monitoring progress and identifying HIV care needs in the US since the release of the National HIV/AIDS Strategy in 2010. Furthermore, the HIV care cascade is used globally as a monitoring rubric to evaluate the performance of HIV/AIDS health system management; the World Health Organization (WHO) has emphasized the cascade model as the central assessment metric for HIV care programs [5]. The UNAIDS recently announced a new global target based on steps (1), (4), and (5) in the cascade: by the year 2020, 90 percent of PLWH should be diagnosed and know their status, 90 percent of those diagnosed on antiviral therapy, and 90 percent of those on therapy have viral suppression Quantitative analyses such as macro level summaries of proportion meeting specific benchmarks, models that examine predictors of engagement in each stage or progression through cascade stages, can provide important information needed to intervene to minimize the negative outcomes and optimize HIV care and treatment efforts to break the cycle of HIV transmission and morbidity. Despite the global acceptance and utility of the HIV care cascade as a conceptual model, our empirical understanding about patient flow through the continuum is still limited, in part because the statistical methods for analyzing cascade data do not have a unified framework. Broadly speaking, there are three main modes of summarizing data related to the care cascade. Macro-level analyses rely on characterizing targeted aspects of the cascade by presenting aggregated data summaries (e.g., number and/or proportion of patients) in each stage of the cascade at certain time points or across time periods [2, 6, 7]. By looking at numbers or proportions of PLWH at each stage, one can readily identify ‘leaks’ or stages where improvements are needed. Risk-factor and regression analyses use individual-level data to identify or evaluate the effect of patient- or program- level factors associated with reaching specific benchmarks such as linkage, retention, and ART initiation [1, 8–11]. For this type of analysis, data are sometimes aggregated across time period to define outcome, or time to event outcomes are considered. A third mode of analysis uses simulation techniques based on an underlying model of progression through the cascade. The mathematical model is specified in advance, and uses inputs from multiple data sources to inform values or ranges of values for the parameters.