Abstract:
|
We analyze repeated visit data from a Los Angeles area clinic with patient demographics, occurrence of sexually transmitted infections (STIs), drug use, sexual behaviors, possible HIV seroconversion, and proximity to HIV testing centers. We wish to use a spatial model to assess the effects of location on the probability of seroconversion. We develop a latent risk profile for each individual over time using a time varying factor analysis model and use the baseline factors and linear change in factors over time as predictors in a GLMM. We decompose the factors into independent temporally correlated and spatially correlated portions which have multivariate Gaussian distributions with isotropic and stationary covariance matrices. We include an additional spatial random effect into the GLMM to assess the residual spatial variability in the probability of seroconversion. We run this model in a Bayesian framework and evaluate the posterior using Hamiltonian Monte Carlo.
|