Fig. 4

Approximation of main model by incremental submodels using the top 15 predictors, defined according to the ranking of variables by increase in cross-validated RMSE. The dashed line (posterior median \(R^{2}\)) and the dark and light grey shaded areas (50% and 95% BCI) indicate the full model performance in terms of cross-validated \(R^{2}\). For submodels, the points indicate the posterior median \(R^{2}\), thick and thin bars give 50% and 95% BCIs, respectively. The left panel depicts results when baseline eGFR is used as part of the longitudinal outcome vector, the right panel results when baseline eGFR is used to update predictions for post-baseline eGFR. The variables used in the submodels increase from left to right, starting with Intercept and time, then adding the first predictor according to the ranking (TNFR1 and KIM1, respectively), then adding the next predictor (RAGE and UACR, respectively), and so on. In particular, in the right panel the results show the added predictive performance for the predictors on top of baseline eGFR. The ordering shown is the ordering obtained across all cross-validation folds