Interaction test

Learn me by follow me exercises

This module tests for the difference of risk factor effect with or without the presence of an effect-modifier, which helps to understand the potential cause-effect pathway.

The risk factor could be continuous variables or categorical.  The effect modifier is limited to categorical variables.

·         If the risk factor is a categorical variable, Empower(S) will first list Mean+SD (if outcome is a continuous variable) or N (%) (if outcome is a binary variable) of the outcome among each subgroup of joint distributions of risk factor and effect-modifier.  It will run two models: A and B.  Model A combines risk factor and effect-modifier and has parameters for each subgroup of the joint distributions of risk factor and effect-modifier; Model B does not join risk factor and effect-modifier, has separate parameters for each subgroup of risk factor and each subgroup of effect-modifier. Then the module will apply log likelihood ratio test to compare model A and B.  If the p-value from log likelihood ratio test is significant, it suggests that there is interaction between the risk factor and effect modifier.  Empower(S) reports the output from model A and the p-value from log likelihood ratio test (p-interaction)

·         If the risk factor is a continuous variable, Empower(S) will first run two models A and B.  Model A has separate risk factor parameters for each subgroup of effect modifier. Model B just has one risk factor parameter for all groups.  The module will then apply log likelihood ratio test to compare model A and B. If the p-value from log likelihood ratio test is significant, it suggests that there is interaction between risk factor and effect-modifier.  Empower(S) will report the output from model A and the p-value from log likelihood ratio test (p-interaction).

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.

What it’s used for

This module could be used to identify effect modifier, and thus helps us to understand the underline potential cause-effect pathway.

Screen shot of sample designs:

t5_input.gif    t5_input2.gif

Sample output tables:

t5_ia.gif    t5_ia_2.gif