Fig. 3

Calibration of posterior median of marginal predictions from the main model, before and after update by baseline eGFR values and stratified by time of observation. Overall, the calibration of predictions was satisfactory over the whole follow-up period. Updating by baseline eGFR led to better calibration and prediction performance, as demonstrated by a more narrow spread around the diagonal line of perfect prediction, even for later follow-up times. The evaluation is stratified by planned follow-up times, actually observed follow-up times used in the model differ slightly. Cross-validated performances by follow-up are reported in Additional file 1: Table S4