An approximate method for generalized linear and nonlinear mixed effects models with a mechanistic nonlinear covariate measurement error model (Q2312022)

A generalized linear mixed effect model is considered, with the time-varying covariate measurement error for a longitudinal outcome. Let the response value for individual \(i\) at time \(t_{ij}\) be \(y_{ij},~i=1,\ldots,n;~ j=1,\ldots,m_i.\) It is assumed that conditioned on the random effects, \(y_{ij},~ j=1,\ldots,m_i\) are independent and follow the distribution in the given exponential family. An inference procedure is implemented that uses first-order Taylor approximation to linearize both the covariate model and the response model. The corresponding approximate estimators of fixed effects are consistent and asymptotically normal, and moreover, the estimators are asymptotically equivalent to the exact maximum likelihood estimators; the asymptotic covariance matrix is consistently estimated. The performance of the estimates in evaluated in finite sample scenario through simulation. The new method is applied to real data in an HIV/AIDS study.