|
Abstract:
|
In many longitudinal studies, the responses and covariates may be observed intermittently at different time points, leading to sparse asynchronous longitudinal data. Traditional methods, e.g., last-value-carried-forward could lead to larger bias and higher variation in estimation. Moreover, observation times of responses may carry information about or correlate with response variables. Failing to consider the informative observation times results in biased estimates. In this study, we consider a regression analysis of sparse asynchronous longitudinal data with informative observation times, which has not been fully investigated. A flexible semiparametric transformation conditional model is used to incorporate dependence between observation times and responses and simple kernel-weighted estimating equations are proposed to deal with discrepancies between observation times of responses and covariates. Extensive simulation studies were carried out to demonstrate that the proposed method performs well and yields less bias than the naïve last-value-carried-forward method. An example from a prospective HIV study illustrates that the proposed method works well in practice.
|
