|
Abstract:
|
Background: Modeling the progression of the disease using standard approaches to modeling longitudinal trajectories, such as generalized linear mixed model (GLMMs), assume missingness mechanism satisfies the assumption of missing at random (MAR). When this assumption is violated, alternative approaches should be considered for unbiased statistical inferences. Method: We studied a national cohort of surviving patients with dementia to model the change in Neuropsychiatric Inventory Questionnaire (NPI-Q) severity score over a 4-years period. In this dataset patients frequently dropped out due to progression of the disease. The methods we evaluated include jointly modeling patient dropouts and change in NPI-Q severity over time, Inverse Probability weighted (IPW) GLMMs, combining Multiple Imputation (MI) and IPW for a doubly robust estimator, and the naïve approach of the complete case analysis.
Conclusion: We found the joint modeling and inverse probability weighted methods will lead to a similar result, and in the applied field the IPW could be more intuitive and interpretable approach.
|
