General Linear Mixed Model (GLMM) Bi-response Applications on Diabetes Mellitus Patients Response

Abstract

General Linear Mixed Model  (GLMM) Bi-respon was  an  alternative  solution for  longitudinal  data  with  bi-responses  which  joining  fixed  effects, random effects and vector of  realization of bi-responses process  into single  statistical model. GLMM can  overcome  the  correlation between  observations  in  longitudinal  data  for  the  response  in  the form of continous data.  In each formation GLMM model beginning with  the  determination  of  a  tentative model  through  exploration  of data.  Exploration  data  covering  several  aspects  of  the  individual profile,  average  structure,  variance  structure,  and  correlation structure. Building GLMM  was done by  selecting  fixed  effects under using  Maximum  Likelihood  (ML)  method,  and  the  selection  of variance  components  (the  number  of  random  effects)  using Restricted  Maximum  Likelihood  (REML)  method.  Based  on  the comparison of AIC value, Diabetes Mellitus Type 2 disease data  was better  to  be  modeled  using  GLMM  with  one  response.  Cross correlations  matrix  elements  were  about  0.3  to  0.6  and  produced unstructured  covariance.  Correlation  coefficient  between  two responses was 0.5526 and  produced unstructured  covariance.