Medication adherence prediction through temporal modelling in cardiovascular disease management

Table 13 Post-PREDICT adherence prediction

Qtrs		Simple RNN	MLP	RC	RC (aggr)	RC (last)
4	LSTM	0.6711	0.4094	0.648	1.807e−2	< 2.2e−16
	Simple RNN		0.6883	0.9242	4.674e−2	2.84e−15
	MLP			0.7797	0.1005	1.475e−14
	RC				2.368e−2	< 2.2e−16
	RC (aggr)					2.734e−10
8	LSTM	9.606e−3	1.169e−4	2.48e−2	5.847e−13	< 2.2e−16
	Simple RNN		5.614e−2	0.7944	2.469e−9	< 2.2e−16
	MLP			7.817e−2	7.099e−5	2.448e−11
	RC				7.365e−10	< 2.2e−16
	RC (aggr)					1.4444e−3
12	LSTM	7.864e−2	2.818e−8	2.351e−3	< 2.2e−16	< 2.2e−16
	Simple RNN		2.207e−4	0.124	< 2.2e−16	< 2.2e−16
	MLP			8.743e−3	2.936e−11	1.16e−9
	RC				< 2.2e−16	< 2.2e−16
	RC (aggr)					0.8463
16	LSTM	6.124e−2	1.799e−6	5.112e−3	< 2.2e−16	< 2.2e−16
	Simple RNN		6.84e−4	0.1974	< 2.2e−16	< 2.2e−16
	MLP			9.566e−3	< 2.2e−16	2.737e−9
	RC				< 2.2e−16	< 2.2e−16
	RC (aggr)					1.063e−2

Qtrs are the number of quarters in the obervation window. The resulting p-values of pairwise comparison using DeLong’s test. Using significance level of 0.05, values under the Bonferroni adjusted significance level of 3.333e\(-\)3 are in bold

ISSN: 1472-6947