Table 2.

Confusion matrices comparing the deep learning model–predicted atherosclerotic cardiovascular disease score vs pooled cohort equation–calculated scores

	UK BioBank		EyePACS 10K		EyePACS 10K + diabetes modifier
	PCE-generated ASCVD score <7.5%	PCE-generated ASCVD score ≥7.5%	PCE-generated ASCVD score <7.5%	PCE-generated ASCVD score ≥7.5%	PCE-generated ASCVD score <7.5%	PCE-generated ASCVD score ≥7.5%
DL model–generated ASCVD score <7.5%	5923 (69%)	331 (4%)	4126 (46%)	2238 (25%)	3015 (34%)	279 (3%)
DL model–generated ASCVD score ≥7.5%	642 (7%)	1710 (20%)	217 (2%)	2388 (27%)	1328 (15%)	4347 (48%)

ASCVD = atherosclerotic cardiovascular disease; DL = deep learning; PCE = pooled cohort equation.

Results are n (%) of people.