Table 4. Binary classification accuracy comparing each algorithm for predicting “high” risk of mortality in the test cohort (n = 125,657).

Algorithm	Optimal Threshold	Correctly Classified Death	Correctly Classified Alive	Sensitivity	Specificity
Deep Learning	> 2%	2,343/3,608	92,978/122,049	64.9%	76.2%
Random Forest	> 5%	2,300/3,608	94,603/122,049	63.7%	77.5%
Adjusted Cox Model	> 6%	2,197/3,608	92,832/122,049	60.9%	76.1%
Age/Gender Cox Model	> 8.4%	1,728/3,608	93,661/122,049	43.7%	76.7%