Table 5.

Diagnostic performance of GPT for predicting all-cause mortality from ICU admission to day 30 with a zero-shot approach

Day	GPT-3.5		GPT-4		Gradient boosting
	Accuracy	AUROC	Accuracy	AUROC	Accuracy	AUROC
0	0.26	0.49	0.55	0.75	0.92	0.88
1	0.36	0.54	0.59	0.73	0.90	0.88
2	0.40	0.53	0.63	0.73	0.88	0.86
3	0.48	0.63	0.62	0.71	0.87	0.85
4	0.38	0.49	0.62	0.71	0.85	0.85
5	0.41	0.57	0.62	0.72	0.85	0.83
6	0.41	0.53	0.61	0.70	0.85	0.84
7	0.41	0.54	0.61	0.69	0.84	0.83
14	0.40	0.47	0.61	0.66	0.79	0.79
21	0.54	0.59	0.63	0.66	0.77	0.78
28	0.52	0.58	0.63	0.66	0.75	0.78
30	0.48	0.54	0.63	0.66	0.73	0.77

ICU: intensive care unit, AUROC: area under the receiver operating characteristic curve.