Step-by-step explanation of PSSP and additional prediction model.

A step-by-step explanation of the Patient Specific Survival Predictor, including management of censored patients, and prediction model including donor variables that are available at the time of the organ offer. Figure A in S1 File: Basic Machine Learning approach: Top-to-bottom: produce a PSSP model from a dataset of historical patients. Left-to-right (across bottom): producing a survival curve for a novel patient, using a description of that patient (that includes only the selected variables), based on the learned PSSP Model. Figure B in S1 File: Example of individual survival curves. Each of the 5 solid lines corresponds to the survival distribution, produced by PSSP, of a single patient using the model including donor variables. The dashed curve is the Kaplan-Meier plot over the entire population (n = 2769 patients). Figure C1 to C6 in S1 File: Step-by-step example to illustrate the concept of distribution-calibration. Figure D in S1 File: Sideways histogram, to visualize D-calibration of the model including donor variables. The “p-value” here (0.999) is the result of the χ2 test, on these values. Table A: Summary of the discrimination (Concordance) and calibration (1-calibration, D-calibration) tests for PSSP and Cox-Kalbfleisch-Prentice models, with and without donor information.