GLOBAL TESTS FOR COMBINATION DRUG STUDIES IN FACTORIAL TRIALS

To test the hypothesis that there are some studied dose combinations more effective in treating a disease than their respective component doses of two drugs, Hung, Chi and Lipicky proposed two α‐level tests for normally distributed data. This paper extends the utilities of these tests to the outcome variable that has variance as a function of its mean, such as with a binomially distributed outcome, and to incomplete factorial design settings where some cells are not studied. I explore the impacts of excluding cells from study on the power performances of these tests.