Keywords: LASSO, spline, survival
The Scientific Registry of Transplant Recipients publishes semiannual reports on solid organ transplant program outcomes in the United States. Transplant program evaluations are based on risk-adjusted models to allow programs to treat higher-risk patients and use lower-quality organs without negatively affecting their public reports. The risk-adjustment models consider most patient and donor characteristics including linear splines with approximately 10 knots for continuous predictors. To prevent overfitting of the data, a LASSO-penalized proportional hazards model is built with the penalty term selected through 10-fold cross-validation. To reduce the chance that the choice of the spline basis functions affects the LASSO fit, both left- and right-hand linear splines are included for each knot. The LASSO can then automatically select both the useful knots and simplest parameterization. The entire process is separately completed for 2.5-year cohorts for each combination of adult and pediatric transplant recipients, living- and deceased-donor transplants, and graft and patient survival outcomes. Multiple imputation is used as records cannot be dropped from the evaluation.