Table 1 Model comparison without and with profile allocation probabilities

From: Subclassification of obesity for precision prediction of cardiometabolic diseases

Outcome

Follow-up time (years)

Sex

Cohort

LRT P value

5% FDR

Added variance explained (%)

C-statistic baseline model

C-statistic baseline + profiles

C-statistic difference

Difference P value

5% FDR

MACE

5

Female

UKB

3.63 × 10−4

True

2.31

0.733

0.735

0.002

0.075

False

MACE

5

Female

RS

0.997

False

0.16

0.768

0.769

0.001

0.399

False

MACE

5

Female

GHS

0.133

False

5.00

0.816

0.820

0.004

0.376

False

MACE

5

Male

UKB

2.70 × 10−16

True

5.44

0.704

0.709

0.005

3.33 × 10−5

True

MACE

5

Male

RS

0.530

False

1.78

0.721

0.721

0.000

0.953

False

MACE

5

Male

GHS

0.998

False

0.06

0.771

0.771

0.000

0.581

False

MACE

10

Female

UKB

5.23 × 10−5

True

1.36

0.726

0.728

0.002

0.035

True

MACE

10

Female

RS

0.886

False

0.43

0.763

0.763

0.000

9.62 × 10−1

False

MACE

10

Male

UKB

4.22 × 10−20

True

4.23

0.685

0.690

0.005

1.19 × 10−6

True

MACE

10

Male

RS

0.271

False

1.88

0.711

0.712

0.001

0.840

False

DM

5

Female

UKB

0.893

False

0.11

0.872

0.872

0.000

0.558

False

DM

5

Female

RS

3.80 × 10−6

True

10.79

0.811

0.822

0.011

0.043

False

DM

5

Female

GHS

1.41 × 10−3

True

8.97

0.806

0.814

0.008

0.126

False

DM

5

Male

UKB

0.872

False

0.07

0.841

0.841

0.000

0.439

False

DM

5

Male

RS

4.03 × 10−4

True

8.38

0.816

0.821

0.005

0.318

False

DM

5

Male

GHS

0.606

False

1.24

0.802

0.804

0.002

0.388

False

DM

10

Female

UKB

0.046

False

0.49

0.854

0.855

0.001

0.162

False

DM

10

Female

RS

3.92 × 10−11

True

12.83

0.796

0.808

0.012

0.005

True

DM

10

Male

UKB

0.048

False

0.38

0.821

0.822

0.001

0.001

True

DM

10

Male

RS

3.47 × 10−8

True

11.05

0.804

0.808

0.004

0.244

False

  1. LRTs, fraction of additional variance explained and C-statistics comparing nested models with versus without discordant profile allocation probabilities. LRT P values are derived from the chi-square distribution with degrees of freedom equal to the number of additional parameters in the full compared with the reduced model. The variance of the difference in C-statistic considers the covariance of the C-statistics from the two models. The resulting P value is derived from a two-sided normal test. DM, diabetes mellitus; LRT, likelihood ratio test.
Back to article page