. 2012 Mar 6;175(7):715–724. doi: 10.1093/aje/kwr374

Table 4.

Weighted Prediction Measures (Empirical Relative Efficiency) of Cardiovascular Disease at 3 Years, TwinGene, 2004–2009

Marker	Average Realizations, the Gold Standard	Unstratified Case-Cohort Design	Unmatched Nested Case-Control Design	Stratified Case-Cohort Design	Matched Nested Case-Control Design
Reclassification^a
BIO1^b	9.1 (100)^c	8.6 (34)	9.6 (39)	8.5 (49)	10.1 (31)
BIO2^b	2.0 (100)	1.7 (55)	2.1 (53)	1.6 (67)	4.0 (34)
HDL-C^b	0.8 (100)	1.2 (47)	1.0 (48)	1.1 (61)	0.4 (49)
BIO1^d	9.1 (100)	8.8 (56)	9.5 (67)	8.9 (71)	9.7 (57)
BIO2^d	2.0 (100)	1.7 (69)	2.0 (80)	1.8 (78)	4.0 (57)
HDL-C^d	0.8 (100)	0.8 (77)	0.7 (79)	0.9 (88)	0.5 (73)
Discrimination^e
BIO1^b	0.779 (100)	0.777 (45)	0.783 (48)	0.780 (56)	0.704 (25)
BIO2^b	0.756 (100)	0.755 (46)	0.760 (50)	0.757 (62)	0.679 (28)
HDL-C^b	0.756 (100)	0.756 (47)	0.761 (51)	0.758 (61)	0.678 (24)
BIO1^d	0.779 (100)	0.778 (72)	0.782 (75)	0.780 (77)	0.701 (44)
BIO2^d	0.756 (100)	0.755 (72)	0.759 (76)	0.756 (81)	0.675 (46)
HDL-C^d	0.756 (100)	0.756 (72)	0.760 (76)	0.757 (80)	0.674 (43)
Calibration^f
BIO1^b	9.7	10.0	5.7	6.5	5.5
BIO2^b	8.3	8.6	5.8	6.0	5.6
HDL-C^b	8.1	7.7	5.0	5.9	6.0
BIO1^d	9.7	7.1	6.9	7.8	6.5
BIO2^d	8.3	6.2	6.3	6.9	6.7
HDL-C^d	8.1	5.8	6.2	6.4	7.5

Abbreviations: BIO1, simulated biomarker 1; BIO2, simulated biomarker 2; HDL-C, high density lipoprotein cholesterol.

The net reclassification improvement percentage.

Sampling ratio: 1:1; subcohort size: n = 229.

Numbers in parentheses represent the “empirical relative efficiency,” defined as the ratio between the empirical variance calculated in 2,000 realizations and the empirical variance calculated for each design, expressed as percentage.

Sampling ratio: 1:3; subcohort size: n = 632.

The average of C-index, which is equivalent to the receiver operating characteristic area under the curve, taking censorship into account.

The Grønnesby and Borgan goodness-of-fit test statistic. Values higher than 9.5 indicate significant lack of calibration.