Assessing related factors to fasting blood sugar and glycosylated hemoglobin in patients with type 2 diabetes simultaneously by a multivariate longitudinal marginal model

Abstract

AbstractThe multivariate marginal model can be used to simultaneously examine the factors affecting both FBS and HbA1c using longitudinal data. The model fitted to multivariate longitudinal data should prevent redundant parameter estimation in order to have greater efficiency. In this study, a multivariate marginal model is used to simultaneously investigate the factors affecting both FBS and HbA1c with longitudinal data for patients with type 2 diabetes in Northern Iran. The present research is a retrospective cohort study. Overall, 500 medical records with complete information were reviewed. The multivariate marginal model is used to determine the factors associated with FBS and HbA1c using longitudinal data. Data have been analyzed in R-3.4.0 using ‘mmm2’ package. Given that the coefficients for the interactions of rtype with the intercept, time, family history of diabetes, history of hypertension, history of smoking, insulin therapy, systolic/diastolic blood pressure and duration of disease at first visit are significantly different from zero (P < 0.05), the effect of the independent variables on the two response variables is different and different coefficients should be used for each. Therefore, the interactions of these variables with rtype are kept in the final model. The coefficients for the interactions of rtype with sex, age at first visit, history of high cholesterol, and weight are not significantly different from zero (P > 0.05), indicating that their effect on the two response variables is similar and only one coefficient should be used for each. We examined the similarity of coefficients when fitting the longitudinal multivariate model for the relationship between FBS/HbA1c and sex, age, history of high blood cholesterol, and body weight. If an independent variable has similar effects on both responses, only one coefficient should be estimated, which will increase the efficiency of the model and the reliability of the results.