Despite the number of algorithms developed to inform clinical care, there has been no evidence that scores are routinely used in practice. This talk will present the main steps to build a score for clinical practice. Then, it will focus on a use case to predict heart failure after myocardial infarction from proteomic data. The analysis of the dataset presents two main challenges : 1) high dimension with much more variables than individuals 2) repeated measurements (4 timepoints). Some preliminary work using only the baseline timepoint will illustrate the high dimension problem. Then, work in progress using joint models will be presented, based on the regularized latent class model for joint analysis of high-dimensional longitudinal biomarkers and a time to-event outcome paper (Sun et al., Biometrics, 2019).