Viewing interactions

Each row in the table presents two predictor values, which appear in the columns labeled Item1 and Item2, and a response, which appears in the column labeled for the selected event variable, such as PT.

Note: You can also include columns in the table that indicate the type of variable selected: the P_Item1 and P_Item2 columns contain either D, to indicate the drug variable, or the name of a covariate variable.

The scores appearing for each predictor1+predictor2+response combination include LROR1 (LROR for predictor1+response) and LROR2 (LROR for predictor2+response), which are repeated from the main results table. They reappear in the interactions table as a reminder of the individual predictor disproportionalities with the response.

The table displays the following information:

• N_TOT—The total number of times the pair of predictor items appeared together, irrespective of the responses.
• INT_LOF (lack of fit interaction ratio)— A Bayesian “shrunk” version of N/E. For example, if INT_LOF = 2, then the response seems to be appearing with this pair of predictor items 2 times as frequently as the standard or extended logistic regression model formula indicates. That is, the model doesn’t “fit” well when these items are both present.
• INT_REGR—Compares the standard or extended logistic regression expected adverse event frequency when both items are present to the expected frequency when only the item (of the two) having the larger LROR is present. That is, if INT_REGR = 2, then the regression model predicts that the response is 2 times as likely when both items are present than when just the more associated item is present.
• INT_TOT = INT_REGR x INT_LOF—Shows the combined relative effect of having both items present versus having the more associated item present.