Deriving a mapping algorithm for converting SF-36 scores to EQ-5D utility score in a Korean population

Table 3 Two-part modeling (logistic + ordinary least-square regression): predicting EQ-5D index scores from SF-36

	Model 7^a		Model 8^b		Model 9^c		Model 10^d
Derivation set
MAE	0.089		0.089		0.081		0.085
AE > 0.05 (%)	50.3		49.7		47.6		48.6
AE > 0.1 (%)	31.9		31.9		29.3		32.2
RMSE	0.298		0.299		0.284		0.292
Predicted EQ-5D index
Mean (SD)	0.829	(0.233)	0.829	(0.233)	0.829	(0.239)	0.828	(0.234)
Min/max	0.193	1.078	0.198	1.079	0.076	1.000	0.088	1.000
Internal validation set
MAE	0.090		0.089		0.083		0.090
AE > 0.05 (%)	50.8		52.3		49.7		50.8
AE > 0.1 (%)	34.3		47.7		30.3		35.2
RMSE	0.301		0.303		0.289		0.300
Predicted EQ-5D index
Mean (SD)	0.823	(0.231)	0.829	(0.233)	0.825	(0.237)	0.826	(0.229)
Min/max	0.222	1.082	0.198	1.079	0.102	1.000	0.057	1.000
External validation set
MAE	0.091		0.091		0.086		0.081
AE > 0.05 (%)	57.7		59.4		59.4		53.7
AE > 0.1 (%)	38.2		39.0		38.2		34.2
RMSE	0.302		0.302		0.293		0.285
Predicted EQ-5D index
Mean (SD)	0.885	(0.113)	0.881	(0.154)	0.908	(0.125)	0.890	(0.129)
Min/max	0.377	1.034	0.537	1.000	0.421	1.000	0.353	1.000

MAE, mean absolute error; AE, absolute error; RMSE, root mean squared error.
^aIndependent variables: PF, RP, BP, GH, VT, SF, RE, MH.
^bIndependent variables: PF, BP, SF, RE, MH.
^cIndependent variables: PF, BP, GH, SF, RE, MH, PF squared, SF squared, RE squared.
^dIndependent variables: PCS, MCS, PCS × PCS, MCS × MCS, PCS × MCS.