Abstract:
|
For many illnesses, disease progression is best measured using ordinal scales. Such scales may exhibit a variety of complex properties including non-normality, lack of unit and interval invariance, very slow change in state over time and the presence of absorbing states (e.g., death and disease resolution). These properties often violate the assumptions behind standard methods of handling missing data including commonly used multiple imputation algorithms. In this project, a novel missing data approach is proposed by incorporating Markov chains into the multiple imputation framework. The proposed method was motivated by and applied to a COVID-19 treatment clinical trial. The performance of the proposed method is compared with other standard multiple imputation algorithms using a comprehensive and empirically based simulation study.
|