Generalized Additive Mixed Model (available for Empower(R) only)

This module applies generalized additive mixed models to test for associations between risk factors and outcomes.  You can specify the smoothing term to fit the non-linear relationship and/or random effect terms (random intercept and/or random time, time square term).

If a random effect variable (e.g. Time) is specified, this module will examine the interactions of risk factors and the random effect variable. It will compare 4 models and look for interaction, time-square term, and random-slope versus random intercept.  These four models and LRT (log likelihood ratio test) test are listed in the example below:

Outcome

Y

Risk factor

X

Model 1:Y = X + T + T2 + X*T + X*T2 + C

ref.

p-LRT(Model 2:Y = X + T + X*T + C)

Pvalue  

p-LRT(Model 3:Y = X + T + T2 + C)

Pvalue

p-LRT(Model 4:Y = X + T + C)

Pvalue

Selected Model

1-4

p-LRT(random slope)

Pvalue

Y: Outcome, X: Exposure, T: Time, T2: Time-square, C: covariates
Model 1:Y = X + T + T2 + X*T + X*T2 + C, Random Intercept, Method: ML
Model 2:Y = X + T + X*T + C, Random Intercept, Method: ML
Model 3:Y = X + T + T2 + C, Random Intercept, Method: ML
Model 4:Y = X + T + C, Random Intercept, Method: ML

p-LRT(model 2 vs. 1): P-value from LRT comparing model 2 vs model 1
p-LRT(model 3 vs. 1): p-value from LRT comparing model 3 vs model 1
p-LRT(model 4 vs. 1): p-value from LRT comparing model 4 vs model 1
Selected Model: choose model with p-LRT > 0.05, in order of model 4->3->2->1
p-LRT(random slope+intercept vs intercept): If model 3 or 4 was selected, LRT comparing model with random slope+intercept vs. random intercept only

This module was used for analyzing repeated measurement/longitudinal data.

Screen shot of sample design:

t9_input.gif    t9_input2.gif

Sample output tables:

t9_mixed.gif    t9_mixed_2.gif    t9_mixed_3.gif