Correlated Drug Action as a Baseline Model for Combination Therapy in Patient Cohorts and Cell Cultures

Abstract

Identifying and characterizing the effect of combination cancer therapies is of paramount importance in cancer research. The benefit of a combination can either be due to inherent heterogeneity in patient populations or because of molecular synergy between the compounds given in combination, usually studied in cell culture, or both. To shed light and help characterized combinations and their enhanced benefits over single therapies, we introduce Correlated Drug Action (CDA) as a baseline additivity model. We formulate the CDA model as a closed-form expression, which lends itself to be scalable and interpretable, both in the temporal domain (tCDA) to explain survival curves, and in the dose domain (dCDA), to explain dose-response curves. CDA can be used in clinical trials and cell culture experiments. At the level of clinical trials, we demonstrate tCDA's utility in explaining the benefit of clinical combinations, identifying non-additive combinations, and cases where biomarkers may be able to decouple the combination into monotherapies. At the level of cells in culture, dCDA naturally embodies null models such as Bliss additivity and the Highest Single Agent model as special cases, and can be extended to be sham combination compliant. We demonstrate the applicability of dCDA in assessing non-additive combinations and doses. Additionally, we introduce a new synergy metric, Excess over CDA (EOCDA), that incorporates elements of Bliss additivity and dose equivalence concepts in the same measure. CDA is a novel general framework for additivity at the cell line and patient population levels and provides a method to characterize and quantify the action of drug combinations.