Interaction Trees

Interaction Trees finds a subgroup of rows with a differentiated treatment effect (either positive or negative, depending on the direction specified) by testing interactions of predictor variables with a treatment variable and splitting on the best interaction effect. It currently adopts a prune-as-you-go approach, keeping only the best branch of the tree after each split.

This process is particularly useful for identifying the unique combination of conditions that subjects can exhibit along with the treatment expected to result in the most beneficial or, conversely, the most deleterious, outcome.

What do I need?

One wide format data set is required to run the Interaction Trees process. This data set must contain one column containing the dependent response variable, one column containing the treatment variable, and multiple columns to be used as predictor variables.

The adsl_dii.sas7bdat data set, partially shown below, details results for 902 subjects. Subjects are listed in rows, demographic information, trial details, and findings and results are listed in columns. The ARM column lists the treatment variable. The DEATHFL column lists the dependent variable. The predictor variables are spread across 310 columns.

For detailed information about the files and data sets used or created by JMP Life Sciences software, see Files and Data Sets.

Output/Results

Refer to the Interaction Trees output documentation for detailed descriptions of the output and guides to interpreting your results.