Table 2.

The overall and SPMS-specific cross-validated (CV) error rates (ER) achieved by the respective models. The overall CV error is a balanced error rate (BER), which is adjusted for class sizes so that 0.50 corresponds to random chance. Both the metabolomics and CRP models did not perform well for distinguishing between all groups (BER values close to 0.5). However, both models were moderately able to distinguish the SPMS patients from remaining groups. As for RRMS vs. SPMS, the CRP model obtained a BER value of 0.22 and the metabolomics model a BER value of 0.20. When reducing the variables to the overlapping top ten variables, the BER value was improved for the CRP model whereas it increased for the metabolomics model (0.20 and 0.30 respectively). Combining the top CRP and metabolic variables (CRPM) resulted in a BER value of 0.23, comparable to the value achieved by the full models.

CV error, mean (±SD)	CRP model n=46	Metabolomics model n=606	Reduced CRP model n=1-10	Reduced metabolomics model n=1-10	CRPM model n=2-20
Global
Overall (BER)	0.48(±0.111)	0.42(±0.119)	0.38(±0.135)	0.55(±0.120)	0.45(±0.145)
SPMS (ER)	0.28(±0.271)	0.26(±0.248)	0.23(±0.249)	0.39(±0.302)	0.34(±0.230)
RRMS vs. SPMS
Overall (BER)	0.22(±0.103)	0.20(±0.142)	0.20(±0.138)	0.30(±0.167)	0.23(±0.149)

BER: balancer error rate; CV: cross-validation; CRP: clinical, radiological and protein; CRPM: clinical, radiological, protein and metabolite; ER: error rate; RRMS: relapsing-remitting multiple sclerosis; SD: standard deviation; SPMS: secondary progressive multiple sclerosis.