This module checks each covariate to determine if it should be included or excluded in model as an adjusting variable. This module is used to identify important covariates that should be adjusted for in multivariate regression models.
Determining which covariates should be put into a model for adjusting is a common question. Different criteria have been used to include or exclude covariates in a model. The Empower(S) “Covariates check" module provides a function of diagnosing the effect of including or excluding a covariate in a model.
For a covariate (C), the module will first model the outcome with the covariate (C) called univariate model, the p-value for the covariate (C) will be taken into account; Then, the module will compare the model that has risk factor(X) and the covariate(C) versus the model that has risk factor(X) only, the change in the regression coefficient for the risk factor(X) will be taken into account; lastly the module will compare the model that has risk factor(X) and all covariates versus the model that has risk factor(X) and all other covariates except the covariate(C), the change of regression coefficient for risk factor(X) will be taken into account. Finally, the module picks up the covariates that meet your specified criteria.
For example, we want to run regression for SBP with BMI, we will adjust for SEX and AGE. We want to determine if we should adjust for SMK2, ALH2, EDU2 and OCCU2. Use SMK2 as an example, we will run the following models:
(1) SBP = SMK2+ SEX + AGE
(2) SBP = BMI + SEX + AGE
(3) SBP = BMI + SEX + AGE + SMK2
(4) SBP = BMI + SEX + AGE + SMK2 + ALH2 + EDU2 + OCCU2
(5) SBP = BMI + SEX + AGE + ALH2 + EDU2 + OCCU2
Model (1) will give us the p value for SMK2 show the significance of the association between SMK2 with SBP. Model 2 is basic model. Model 3 adds SMK2 to basic model. By comparing the regression coefficient of BMI from model 2 to model 3, we calculate the change of the BMI’s effect by adding SMK2 to the basic model. Model 4 is called the full model. Model 5 removes the SMK2 from full model. By comparing the regression coefficient of BMI from model 4 to model 5 we calculate the change of BMI’s effect by removing the SMK2 from the full model.
The module automatically detects whether the outcome variable is binary or continuous and selects logistic or linear regression as appropriate. You also can specify the distribution and link function for each outcome manually, right click the variable and then select “Change functions”. In the popup window, select the distribution and link function.
A screen shot of sample design:
A sample output table: