A Simple Tool to Predict Outcomes After Kidney Transplant

Performance of models for graft loss based on information available (A) pretransplant and at (B) 7 days and (C) 1 year posttransplant. The y-axes reflect the percentage of model performance obtained including all candidate predictors resulting from the backward-selection process. As the number of variables in each model is iteratively reduced (moving to the right on the x-axes), the lines depict loss of discriminatory ability and percentage of variation explained by the model. Final full models were defined when the percentage of variation explained was not <99% of the full candidate set; abbreviated models, when the percentage of variation explained was not <80% of the full candidate set. Of note, fewer variables are needed to attain the same model performance posttransplant compared with pretransplant.

Conversion of risk level to predicted probability for the full models for graft loss within 5 years based on information available (A) pretransplant and (B) 7 days and (C) 1 year posttransplant. The distribution of risk in the population is shown in the histogram (gray lines, right axis). Predicted probabilities (black lines, left axis) associated with risk levels more extreme than the 1st and 99th percentiles are shown with dashed lines. A patient's risk score can be calculated as follows: (A) risk score = (x + 1.25)/4.05; (B) (x + 1.10)/4.46; (C) (x + 1.28)/5.94, where x is equal to the patient's linear predictor calculated from , respectively.

Calibration statistics (slopes of the prognostic index) for the full and abbreviated models for graft loss within 5 years based on information available pretransplant and 7 days and 1 year posttransplant. Slopes of 1.0 indicate good model calibration. Shown are calibration slopes and 95% confidence intervals.