982-983 / 1676
Key points
•
Model development in a single (institution) / small dataset yields
•
Large variability in models
•
Unreliable assessments of model performance
•
Validation
•
Improves performance assessment
•
May constitute iterative improvement of models
•
The NTCP is a population characteristic, not an individual number
•
The most useful model
•
Is derived from a large (multi-institutional) dataset
•
Is validated resulting in well-characterized performance
•
May vary with the intended use
•
Don’t only rely on statistical tools: Use other knowledge if available.
December 8, 2015