Lecture 5 - Association models Flashcards
What is the difference between confounding and effect modification?
- Confounding: a variable related to both the independent as dependent variable. By not taking into account the effect of the confounder variable, the association between the independent and dependent variable can be over- or underestimated.
- Effect modification: the association between the independent and dependent variable is different for certain subgroups.
For researching the effect of confounding, the example is used where it is researched whether cholesterol is associated with age. Here, a regression model is made with age as a linear/scale variable, where it is seen that the mean cholesterol increases by 0.062 mmol/L per year (95% CI [0.046; 0.077], p<0.001). However, it is possible that sex (male/female) is related to both the independent (age) variable and the dependent (cholesterol) variable.
Name two ways the effect of sex as a confounding variable can be researched.
- Stratifying the regression model for males and females. If the ‘Group Statistics’ table then indicates a difference in age and cholesterol for both sexes and this is supported by a significant p-value, you can state that sex is a confounding variable within the association between age and cholesterol.
- Creating a crude and adjusted model. In the crude model you only look at the association between age and cholesterol and in the adjusted model sex is added into the mode. If then: the p-value is <0.05 (of sex) and if the regression coefficient (of age) differ by >10%, sex is called a confounder.
For researching the effect of confounding, the example is used where it is researched whether cholesterol is associated with age. Here, a regression model is made with age as a linear/scale variable, where it is seen that the mean cholesterol increases by 0.062 mmol/L per year (95% CI [0.046; 0.077], p<0.001).
If we establish that sex is a confounder in this assocation, the mean cholesterol can be calculated separately for men and females.
Describe the general linear regression formula and describe the formula with confounding effect of sex.
General formula:
* y = b0 + b1* X
Confounding:
* mean cholesterol (y) = b0 + b1 * age + b2 * sex (sex: female=0, male=1)
Females:
* mean cholesterol (y) = b0 + b1 * age + b2 * 0
Males:
* mean cholesterol (y) = b0 + b1 * age + b2 * 1
For researching effect modification, the example is used where it is researched whether cholesterol is associated with age. Here, a regression model is made with age as a linear/scale variable, where it is seen that the mean cholesterol increases by 0.062 mmol/L per year (95% CI [0.046; 0.077], p<0.001).
How can you investigate sex as an effect modifier?
To investigate possible effect modification of sex on the association between age and cholesterol, you need to make use of an interaction variable. This variable combines sex with the independent variable, i.e. sex*cholesterol. In SPSS, the independent variable is put into block 1, the possible effect modification of sex is put into block 2 and the interaction variable is put into block 3. If for the interaction variable a p-value is found <0.10, you can conclude that sex is an effect modifier and that the two sexes need to be handled separately. This means that you use the values of the third block to fill in the regression formula.