Representation learning for clinical time series prediction tasks in electronic health records

Table 5 Prediction accuracy of top-k comorbidity

Top-k ^∘	Upper Bound ^⋆	Patient representation learning methods
		RNN-DAE	RNN-AE	SDAs	PCA	k-means	GMM	Hand
k=1	0.962	0.604	0.617	0.212	0.449	0.181	0.607	0.605
k=2	0.878	0.534	0.514	0.195	0.384	0.208	0.503	0.503
k=3	0.769	0.452	0.419	0.177	0.305	0.144	0.416	0.417

^∘“Top-k” represents the average accuracy of all the patients, where accuracy for one patient is the average number of correct results included in its top k predicted comorbiditie(s). The top-k comorbiditie(s) is/are sorted by predicted probabilities, with k=1,2,3.
^⋆“Upper Bound” shows the best results achievable (i.e., all the correct comorbidities assigned to all the patients)

ISSN: 1472-6947