Time-resolved trajectory of glucose lowering medications and cardiovascular outcomes in type 2 diabetes: a recurrent neural network analysis

Table 2 Model discrimination performance

	RNN model (2D input: GLMs and time)
Outcome	True sequence	Inverted sequence	Random sequence
4P-MACE	0.911 (0.904–0.919)	0.892 (0.883–0.900)*	0.905 (0.897–0.912)*
Heart failure	0.807 (0.790–0.824)	0.808 (0.790–0.826)	0.807 (0.789–0.824)
Myocardial infarction	0.811 (0.795–0.826)	0.799 (0.783–0.815)*	0.804 (0.789–0.819)*
Stroke	0.835 (0.814–0.855)	0.828 (0.808–0.848)	0.831 (0.810–0.852)
All-cause mortality	0.752 (0.734–0.770)	0.794 (0.777–0.811)*	0.777 (0.760–0.795)*

The table shows the AUROC of the proposed model on 4P-MACE and its four components on the test set (N = 10,000) when fed by the actual sequence of GLMs (second column), and an inverted and a randomised versions thereof (third and fourth columns). *p < 0.05 versus the true sequence

ISSN: 1475-2840