Fig. 1.

Data processing and analyses flow diagram of this study. Thirty-eight metabolites were significant following multiple testing in multi-variable cox proportional hazards models. For the development of a prediction model, participants were randomly assigned to the training and testing group for model development. After a 10-fold cross-validation test, 24 metabolites were assigned a nonzero coefficient in the elastic net regression model amongst the 249 included metabolites. Receiver operating characteristic (ROC) curve was created and area under curve (AUC) was calculated for predictive value comparison. Categorical net reclassification improvement (NRI) was calculated to investigate the reclassification ability