Generalized additive models have been used to decide that the predictors age and start should be included in the final model.

The model can now be simplified even further. Both the fitted functions suggest a simple parametric transformation.