Table 2: Subtype classification results (macro average of multi-label results).

Approach	Clustering method	Clusters	Acc.	Recall	Precision	F1
Baseline	K-Modes	0, 2, 4, 5	0.573	0.495	0.518	0.491
Layered Axes (all)	K-Modes	0, 1, 2	0.657	0.626	0.637	0.627
Layered Axes (severe)	K-Modes	0b, 1a	0.627	0.547	0.547	0.542
Layered Axes (respiratory)	K-Modes	1a, 1b	0.886	0.829	0.866	0.844
Prognosis Space (all)	DBSCAN	2, 3, 4, 5, 6, 7, 8	0.428	0.316	0.356	0.320
Prognosis Space (severe)	DBSCAN	5, 7, 8	0.651	0.581	0.665	0.579
Prognosis Space (severe, respir.)	DBSCAN	5, 7	0.746	0.715	0.739	0.720