General Linear Models in a Missing Outcome Environment of Clinical Trials Incorporating with Splines for Time-Invariant Continuous Adjustment

Abstract

Missing data is a common occurrence in longitudinal studies of health care research. Although many studies have shown the potential usefulness of current missing analyses, e.g., (1) Complete Case (CC) analysis; (2) imputation methods such as Last Observation Carried Forward (LOCF), multiple imputations, Expectation-Maximization algorithm approach; and (3) methods using all available data such as linear mixed model and generalized estimation equations approach, the CC analysis or LOCF imputation method have been popular due to their simplicity of execution regardless of some critical drawbacks. The proposed approach employs the generalized least squares method using all available data without deletion or imputations for missing outcomes, producing the best linear unbiased estimate. A simulation study was conducted to compare the proposed approach to commonly used missing analyses under each missing data mechanism and showed the validity of the proposed approach, especially with the first order autoregressive correlation structure. B-spline is applied to the proposed model to manage non-linear relationships between outcome and continuous covariate. Application to a cell therapy clinical trial is presented.