Table 4.

Random forest algorithm performance for predicting clinical outcomes at 2-year follow-up

Outcome	Features	Precision	Recall	F1	Accuracy*	AUC
EDSS (delta)	C	0.60	0.58	0.59	0.5360	0.60
	C/I	0.62	0.60	0.61	0.566	0.56
	C/I/G	0.64	0.61	0.62	0.564	0.54
EDSS (CDA)	C	0.86	0.86	0.86	0.6777	0.77
	C/I	0.81	0.78	0.79	0.5775	0.75
	C/I/G	0.83	0.79	0.81	0.5776	0.76
9HPT (CDA)	C	0.83	0.89	0.86	0.4965	0.65
	C/I	0.83	0.88	0.86	0.4863	0.63
	C/I/G	0.83	0.88	0.86	0.4864	0.64
T25WT (CDA)	C	0.67	0.69	0.68	0.564	0.54
	C/I	0.69	0.62	0.64	0.6148	0.48
	C/I/G	0.69	0.62	0.64	0.6148	0.48
SDMT (CDA)	C	0.95	0.96	0.95	0.686	0.66
	C/I	0.91	0.95	0.93	0.4965	0.65
	C/I/G	0.91	0.95	0.93	0.4965	0.65
SL25 (CDA)	C	0.62	0.62	0.62	0.627	0.67
	C/I	0.63	0.62	0.62	0.634	0.64
	C/I/G	0.66	0.64	0.62	0.638	0.68

Features tested are shown by the type of variables (clinical (C), imaging (I) and genetics (G) and the number of features used. Results are shown as precision (positive predictive value), recall (sensitivity), F1 (harmonic mean of precision and recall), the *balanced accuracy and the area under the receiver operator curve (AUC)

CDA, confirmed disability accumulation; EDSS, Expanded disability status scale; 9HPT, nine-hole peg test; T25WT, timed 25-feet walking test; SDMT, symbol digit modality test; SL25, Sloan low-contrast visual acuity 2.5%; NEDA, no evidence of disease activity; DMD, disease-modifying drug