Incorporating medical code descriptions for diagnosis prediction in healthcare

Table 4 The code-level accuracy @k of diagnosis prediction task

Dataset	@k	MLP	MLP +	RNN	RNN +	RNN_a	RNN_a+	Dipole	Dipole +	RETAIN	RETAIN +	GRAM	GRAM +
MIMIC-III	5	0.3104	0.3181	0.2952	0.3193	0.2910	0.3162	0.2941	0.3155	0.3056	0.3198 ^∗	0.3072	0.3183
	10	0.5040	0.5138	0.4796	0.5111	0.4693	0.5085	0.4767	0.5086	0.4980	0.5160 ^∗	0.5003	0.5138
	15	0.6286	0.6352	0.6019	0.6335	0.5889	0.6290	0.5971	0.6325	0.6258	0.6360 ^∗	0.6267	0.6348
	20	0.7114	0.7239 ^∗	0.6894	0.7198	0.6822	0.7144	0.6845	0.7168	0.7129	0.7202	0.7130	0.7196
	25	0.7754	0.7852 ^∗	0.7545	0.7804	0.7491	0.7785	0.7501	0.7795	0.7735	0.7806	0.7728	0.7794
	30	0.8214	0.8294 ^∗	0.8040	0.8279	0.7987	0.8269	0.7990	0.8280	0.8198	0.8286	0.8220	0.8283
Heart failure	5	0.4580	0.5132	0.5599	0.5960	0.5699	0.5882	0.5687	0.5868	0.4085	0.5808	0.6152	0.6227 ^∗
	10	0.6266	0.6412	0.6835	0.7169	0.6920	0.7109	0.6953	0.7105	0.5460	0.7042	0.7393	0.7455 ^∗
	15	0.7124	0.7254	0.7603	0.7876	0.7645	0.7845	0.7702	0.7841	0.6512	0.7765	0.8088	0.8130 ^∗
	20	0.7717	0.7827	0.8132	0.8355	0.8153	0.8334	0.8209	0.8307	0.7162	0.8261	0.8544	0.8580 ^∗
	25	0.8206	0.8283	0.8516	0.8698	0.8532	0.8673	0.8580	0.8655	0.7684	0.8622	0.8872	0.8902 ^∗
	30	0.8572	0.8635	0.8812	0.8958	0.8825	0.8943	0.8860	0.8923	0.8100	0.8899	0.9113	0.9134 ^∗

ISSN: 1472-6947