A0871
Title: Statistical methods for interval-censored multi-state data and mismeasured covariates with application in HIV care
Authors: Hongbin Zhang - CUNY (SPH) (United States) [presenting]
Abstract: In 2015, WHO announced the Treat All policy which recommends immediate antiretroviral therapy (ART) treatment of HIV infected people, regardless of disease severity. In evaluating the impact of adopting the Treat All policy at a national level, the relationship between the biomarkers such as CD4 counts and WHO clinical stages (1: asymptomatic; 2 mild; 3: advanced; 4: severe; 5: mortality) is investigated to assess the magnitude of Treat All effects that would go through (or not go through) CD4 counts, a strong proxy of ART treatment. The WHO clinical stage data are interval-censored as the exact time of stage to stage transition between the clinical visits is unobservable. The CD4 covariate can have a substantial measurement error. We proposed statistical methods for multi-state data subject to interval-censoring and mismeasured time-varying covariates: 1) two-steps method where the prediction of the true time-varying covariates was plugged into the outcome model for the estimation; and 2) joint model methods in which parameters from the longitudinal covariates model and from the survival model were simultaneously estimated where we implemented a computationally efficient method using the stochastic version of EM (StEM). The methods were applied to real-world service delivery data in Central Africa and evaluated with simulation.
