Metabolomics plays a crucial role in understanding metabolic diseases and identifying biomarkers for detection of diseases. However, most existing research in this field has focused on inter-individual differences and disease-state biomarkers, with limited attention given to the identification of intra-individual variability and early metabolic biomarkers. This thesis addresses this challenge by developing a Bayesian statistical modelling approach to detect intra-individual variation in metabolite levels across repeated measurements. A Bayesian generalised linear mixed model approach, termed MetaboVariation, is proposed that detects variations in metabolite levels at the individual level. MetaboVariation assesses intra-individual variability in a metabolite in a univariate manner and flags individuals whose observed metabolite levels fall outside their highest posterior density intervals. Given the inherent dependencies between metabolites, MetaboVariation 2.0 is then proposed which models interdependencies between metabolites when assessing intra-individual variability in metabolite levels, thus providing a more comprehensive assessment of intra-individual variations. Both proposed methodologies are evaluated through simulation studies and application to a real-world metabolomics dataset with results demonstrating the ability of the proposed approaches to detect intra-individual variations. Moreover, the multivariate MetaboVariation 2.0 intuitively outperformed its univariate predecessor in cases with high inter-metabolite dependencies. An open source R package, called \texttt{MetaboVariation}, is available to the scientific community to facilitate broader application in metabolomics research of the approaches developed in this thesis. Given the capability to detect intra-individual variability, this thesis then explores the association of intra-individual metabolic variation with clinical profiles. In the two datasets that were examined, the majority of metabolites were stable and showed minimal variability; however, the the multivariate MetaboVariation 2.0 approach did flag some individuals with intra-individual variations in their metabolite profiles. These flagged individuals were then shown, through fitting a Bayesian logistic regression model, to exhibit slight perturbations in clinical variables. Although the inferred associations were weak, they underscore the importance of studying intra-individual variability to flag individuals that may be at metabolic risk. In conclusion, this thesis makes novel contributions to the field of metabolomics by introducing statistical approaches for detecting intra-individual variations using repeated measurements.