Table 2.

Performance comparison on the test dataset

Method	Estimated mortality		Jaccard
	Trajectory-wise	State-wise
Clinician’s policy	0.0956	0.0959	–
Dual-LSTM	0.0887	0.0935	0.3171
AMANet	0.0827	0.0895	0.3250
DPO-LSTM	0.0919	0.0930	0.2069
SRL-Multimorbidity	0.0948	0.0953	0.2610
SL-LSTM ($\epsilon =0$)	0.0956	0.0967	0.3432
RL-LSTM ($\epsilon =1.0$)	0.0752	0.0721	0.0342
SRL-LSTM ($\mathit{\epsilon }=0.4$)	0.0643	0.0878	0.3110
SRL-LSTM($\epsilon =0.4$, w/o diagnosis codes)	0.0741	0.086	0.2233

The bold indicates the prefered AI model to learn dynamic treatment strategies for CHD patients