Dynamic Risk Prediction

The model could adjust predicted risk upward or downward as new creatinine values became available. Prediction performance was evaluated using 4-fold cross-validation at 0.5-year and 2.0-year landmarks across 1-, 2-, and 3-year prediction windows. Dynamic prediction performance was generally strong, though censoring required caution. Dynamic risk prediction estimated event probability within a future window conditional on survival and biomarker history up to a landmark time. The practical goal was dynamically updated patient-specific risk prediction based on accumulated creatinine history.