Table 8.

Performance of the predictive models based on leave-one-site-out cross-validation

Measure	Out-of-samplea mean	In-sampleb mean	Sample sizec	Response	Model
C-Index	0.579	0.652	307,784	CDP	predictive
C-Index	0.646	0.675	505,724	Relapse	predictive
MSE	0.125	0.119	3119	CDP	predictive
MSE	0.748	0.711	3119	Relapse	predictive
NLL	1254.330	1192.544	3119	CDP	predictive
NLL	2581.261	2523.731	3119	Relapse	predictive

CDP confirmed disability progression, C-Index Harrell’s concordance statistic, MSE mean squared error, NLL negative log-likelihood.

^a Based on predictions for patients from a clinical site not used during training

^b Based on predictions for all patients from all clinical sites used for training

^c Refers either to the number of observations (MSE, NLL) or the number of matched pairs (C-Index)