Table 3. Confusion matrix of classification rates for individuals’ assignment to each cohort by matching individual topological profiles to cohort-level topological profiles.

Without parentheses: results using full GP Progression Model -driven topological profiles. Within parentheses: results using topological profiles estimated from only late-stage data. Higher numbers are better in diagonal entries (correct assignment), and lower numbers are better in off-diagonal entries (incorrect assignment).

	AD	PPMS	HA
AD	84% (57%)	4% (9%)	12% (34%)
PPMS	2% (16%)	68% (45%)	29% (38%)
HA	7% (14%)	2% (12%)	91% (74%)