Week 8: Confounding and Effect Modification

Question 1

What is effect modification?

Answer 1

Effect modification occurs when the effect of an exposure on an outcome varies across levels of another variable

Question 2

How can you identify effect modification?

Answer 2

By examining whether the relationship between exposure and outcome changes across strata of a third variable or through interaction terms in regression

Question 3

What is the role of stratification in assessing confounding and effect modification?

Answer 3

Stratification allows you to separate the data into groups (strata) to evaluate the relationship between variables while accounting for potential confounders or effect modifiers

Question 4

What statistical tools are used to assess confounding?

Answer 4

Mantel-Haenszel stratification, comparison of crude and adjusted estimates, and regression models

Question 5

How do regression models help in addressing confounding?

Answer 5

Regression models adjust for multiple variables simultaneously, isolating the effect of the exposure on the outcome

Question 6

What is an interaction term in regression analysis?

Answer 6

An interaction term represents the combined effect of two variables on an outcome, used to test for effect modification

Question 7

What are the steps for preparing data for confounding and effect modification analysis?

Answer 7

1. Check and recode variables as needed
2. Categorise variables if required
3. Verify the distribution and associations among variables

Question 8

Why is coding consistency important in statistical analysis?

Answer 8

Consistent coding ensures accurate interpretation and analysis, especially when categorising variables or dealing with missing data

Question 9

How can graphical tools aid in identifying effect modification?

Answer 9

Graphs, such as stratified plots, visually demonstrate differences in relationships across subgroups, highlighting potential effect modification

Question 10

What is the difference between confounding and effect modification?

Answer 10

• Confounding distorts the true relationship between exposure and outcome
• Effect modification indicates that the relationship itself changes based on a third variable

Question 11

Why is testing for homogeneity important in stratified analyses?

Answer 11

Homogeneity tests assess whether the effect size is consistent across strata, helping to identify confounding or effect modification

Question 12

What are the potential pitfalls of adjusting for variables on the causal pathway?

Answer 12

Adjusting for variables on the causal pathway can underestimate or eliminate the true effect of the exposure on the outcome

Question 13

how can variables be recoded into binary form (0/1) in Stata?

Answer 13

<gen>
<recode> (1=0) (2=1)
<label define newvar01 0 "no" 1 "yes">
<label>
</label></recode></gen>

Question 14

What command estimates ORs?

Answer 14

<tabodds var1 var2, or>

Question 15

How do you stratify odds ratios by a third variable?

Answer 15

<cc outcome exposure, by(strata)>

Question 16

How do you generate a histogram by group?

Answer 16

<histogram varname, by(group)>

Question 17

What command runs a logistic regression model?

Answer 17

<logistic>
</logistic>

Question 18

How can you include categorical predictors in regression models?

Answer 18

use the i. prefix for categorical variables:

<logistic>
</logistic>

Question 19

What command generates a linear regression model?

Answer 19

<regress>
</regress>

Question 20

How can interaction terms be included in regression?

Answer 20

<regress outcome predictor1##predictor2>

Question 21

How do you store regression models in memory for comparison?

Answer 21

<est>
</est>

Question 22

What command performs a likelihood ratio test between two models?

Answer 22

<lrtest>
</lrtest>

Question 23

How do you calculate summary statistics for a variable by group?

Answer 23

<bysort group1: summarize varname>

Question 24

What commands plot linear relationships?

Answer 24

<twoway (lfit outcome predictor>
<twoway (lfit outcome predictor if group1) (lfit outcome predictor if group2>

Question 25

What command labels variables and values?

Answer 25

<label variable varname “label”>
<label define labelname value “label”>
<label></label>

Question 26

How do you formally test for effect modification?

Answer 26

1. Run a simple regression model with outcome and explanatory variables and store it
2. Run a more complex model with ‘##’ to indicate effect modification for a variable i.e <regress cfex1 c.dhager##i.sex01> and store it
3. Run a likelihood ratio test to see if the complex model with the effect modifier gives a better fit for the data than the simple model (H0: the complex model does not describe the data better than the simple model i.e., the association between exec function and age is not modified by sex)
4. If no evidence against H0, we use the simple model to describe the association