Performance of an easy-to-use prediction model for renal patient survival: an external validation study using data from the ERA-EDTA Registry

Background An easy to use prediction model for long-term renal patient survival, based on only four predictors (age, primary renal disease, sex, and therapy at 90 days after the start of renal replacement therapy (RRT)), has been developed in the Netherlands. To assess the usability of this model for use in Europe, we externally validated the model in ten European countries. Methods Data from the ERA-EDTA (European Renal Association - European Dialysis and Transplant Association) Registry were used. Ten countries that reported individual patient data to the registry on patients starting RRT in the period of 1995-2005 were included. Patients under 16 years of age and/or with missing predictor variable data were excluded. The external validation of the prediction model was evaluated for the 10-year (primary endpoint), 5- and 3-year survival predictions by assessing the calibration and discrimination outcomes. Results We used a dataset of 136,304 patients from 10 countries. The calibration in the large and calibration plots for 10 deciles of predicted survival probabilities showed average differences of 1.5%, 3.2% and 3.4% in observed versus predicted 10-, 5-, and 3-year survival, with some small variation on country-level. The C-index, indicating the discriminatory power of the model, was 0.71 in the complete ERA-EDTA Registry cohort and varied according to country level between 0.70 and 0.75. Conclusions A prediction model for long-term renal patient survival developed in a single country, based on only four easily available variables, has a comparably adequate performance in a wide range of other European countries.