Improved Mortality Prediction in Dialysis Patients Using Specific Clinical and Laboratory Data

Area of Science:

• Nephrology
• Medical Statistics
• Public Health

Background:

• Risk prediction models are vital for patients on renal replacement therapy (RRT).
• Registry data offers convenience but may have limited predictive power.
• Enhancing existing models with clinical and laboratory variables can improve survival predictions.

Purpose of the Study:

• To improve a basic survival prediction model for RRT patients.
• To assess the added value of clinical and laboratory variables to a registry-based model.
• To evaluate model performance using calibration and discrimination metrics.

Main Methods:

• Utilized data from 1,835 Dutch RRT patients.
• Categorized predictors by data availability: clinical, laboratory, and GFR/Kt/V.
• Employed multivariate Cox regression and backward selection for model refinement.
• Assessed model calibration and discrimination (C-index, IDI, NRI) on a separate cohort.

Main Results:

• The baseline registry model showed good calibration (C-index=0.724).
• Adding easily available clinical parameters improved discrimination (C-index=0.784) and reclassification, particularly for short-term survival.
• Inclusion of laboratory values or GFR/Kt/V did not substantially enhance model performance.

Conclusions:

• A simple registry-based model can be effectively enhanced by incorporating readily available clinical parameters.
• This improved model offers better survival prediction for RRT patients.
• The findings emphasize the utility of accessible clinical data in refining prognostic models.