Skip to main content

Table 2 Latent class model for women and men

From: Analysis of heterogeneity and epistasis in physiological mixed populations by combined structural equation modelling and latent class analysis

Indicators:

Insulin levels at 0, 30 and 120 min after an oral glucose tolerance test (OGTT).

Covariates in all models:

Glucose and C-peptide levels at 0, 30 and 120 min after OGTT.

 

Stage 1

Stage 2

Final model

Women

         

Number of classes

17

18

19

18

18

18

18

18

19

Additional covariates

   

BMI

WH

TG

HDLC

Chol.

BCagec

df a

551

583

615

746

746

746

746

746

1,188

AIC

85,626

81,531

8,.472

8,.330

81,487

81,473

81,411

81,490

80,962

BIC

86,322

82,870

8,.886

82,769

82,925

82,911

82,849

82,928

82,694

Sample-Size Adjusted BIC

85,947

82,148

8,.124

81,994

82,150

82,136

82,074

82,152

81,759

Entropy

0.876

0.868

0.859

0.872

0.862

0.873

0.872

0.871

0.879

Chi-tests:b

         

   AIC

 

<E-5

<E-5

<E-5

<3*E-4

<E-5

<E-5

<6*E-4

<E-5

   BIC

 

<E-5

-

<E-5

-

-

<0.24

-

<0.33

   BICadj

 

<E-5

<0.027

<E-5

-

<0.75

<E-5

-

<E-5

Men

         

df

551

583

615

746

746

746

746

746

1,188

AIC

83,973

83,840

83,792

83,643

83,792

83,726

83,688

83,722

83,226

BIC

85,237

85,181

85,210

85,084

85,233

85,166

85,129

85,162

84,962

Sample-Size Adjusted BIC

84,556

84,459

84,447

84,309

84,457

84,391

84,354

84,387

84,028

Entropy

0.886

0.883

0.882

0.888

0.881

0.881

0.884

0.884

0.898

Chi-tests:

         

   AIC

 

<E-5

<E-5

<E-5

<E-5

<E-5

<E-5

<E-5

<E-5

   BIC

 

<E-5

-

<E-5

-

<0.65

<E-5

<0.37

<0.39

   BICadj

 

<E-5

<0.51

<E-5

1.00

<E-5

<E-5

<E-5

<E-5

  1. adf, degrees of freedom.
  2. bChi-square tests. The models were compared with the immediately preceding model. The models in Stage 2 were compared to the best model in Stage 1 and so on.
  3. cBCage, BMI, cholesterol and age.
  4. The latent class variable was regressed on all covariates (see Figure 1B). All models with higher numbers of classes have lesser goodness-of-fit statistics and are omitted form the table.