The Choice of Treatment Effect Scale for the Analysis of Heterogeneity of Treatment Effect (HTE) and Qualitative Interactions
*Ravi Varadhan, Department of Oncology, Johns Hopkins University
Keywords: additive interactions, multiplicative interactions, qualitative interactions, subgroup analysis
A fundamental problem in the analysis of HTE is that the degree of HTE depends on the effect scale. For example, for a binary outcome, the treatment effect may be homogeneous on the multiplicative scale (odds-ratio or risk-ratio) while it is heterogeneous on the additive scale (risk difference). Some have argued that interactions on additive scale are most relevant to assessing public health impact of exposures and more closely related to the notion of mechanistic, while others have proposed that interactions should be evaluated on both multiplicative and additive scales and all the results should be presented. It is not clear which treatment effect scale is more appropriate in PCOR, or whether it is useful to study HTE in terms of all meaningful scales. A quantitative interaction can be nullified by a transformation of the effect scale. A qualitative interaction cannot be removed by a scale transformation. We argue that identifying qualitative interaction is of fundamental importance in the context of PCOR. This talk will discuss the scientific and statistical issues in the choice of treatment effect scale for PCOR studies.