Table 2.

Comparison of Individually Tuned (Self) and Portable (Cross-Subject) Model Performance for the Nine Subjects Who Passed Modeling Exclusion Criteri^a Subject

Subject	Model	Prediction horizon
		30 min		60 min
		Root mean square error (mg/dl)	Clarke error grid % (A+B)	Root mean square error (mg/dl)	Clarke error grid % (A+B)
2	Self	17.5	100.0	24.8	99.8
	Cross-subject	21.1 ± 0.8	99.7 ± 0.0	30.4 ± 1.1	99.7 ± 0.0
6	Self	22.2	98.6	35.0	97.3
	Cross-subject	25.9 ± 1.3	95.8 ± 1.0	40.1 ± 1.0	90.9 ± 2.7
8	Self	20.9	99.8	33.6	98.0
	Cross-subject	24.8 ± 1.5	98.7 ± 0.4	37.9 ± 1.0	96.2 ± 0.4
9	Self	26.8	99.6	40.1	98.9
	Cross-subject	26.0 ± 1.3	99.6 ± 0.1	38.3 ± 0.7	99.0 ± 0.2
11	Self	29.6	99.0	41.1	98.4
	Cross-subject	30.8 ± 0.4	99.2 ± 0.1	41.5 ± 0.4	98.1 ± 0.3
12	Self	24.0	97.6	36.0	95.3
	Cross-subject	26.9 ± 1.3	97.8 ± 0.8	40.2 ± 1.0	95.2 ± 0.6
13	Self	20.0	99.3	29.5	98.6
	Cross-subject	24.6 ± 1.3	98.2 ± 0.2	36.1 ± 0.7	96.6 ± 0.1
14	Self	19.2	99.0	25.5	98.8
	Cross-subject	20.6 ± 1.1	98.6 ± 0.2	28.1 ± 0.5	97.6 ± 0.5
15	Self	20.1	98.7	25.7	97.4
	Cross-subject	22.3 ± 0.7	95.9 ± 1.4	29.4 ± 0.7	92.5 ± 2.4

^aFor results based on individually tuned models, analysis is performed on the last 4000 data points. For cross-subject models, where the results for each subject are averaged over the other eight model predictions using their own individualized models, analysis is performed for the entire range of the predicted 6000 data points. Their standard deviation over the other eight predictions is also illustrated.