Missing and clustered data in healthcare research

Abstract

This thesis describes the challenges and methods for dealing with missing and hierarchical or clustered data, which often occur together in epidemiological studies. It provides a comprehensive overview of imputation methods tailored to hierarchical data and introduces a new approach to handle data that is not missing completely at random. Additionally, the thesis highlights the complexity of and introduces methods for generating illustrations of calibration during the external validation of clinical prediction models in the context of datasets with incomplete data. It also discusses the challenges and methods for producing calibration plots based on clustered datasets.