Background: Diagnostic informatics (DI) in the context of personalized medicine involves the integration of molecular information to provide "actionable" diagnostic and therapeutic strategies. In many cases, retrospective predictions of clinical outcomes affected by multiple genes are complicated by not having the relevant genes measured within the same study. Multiplicative effect modeling is a statistical method for estimating the net effect of ≥ 2 independent variables. The authors demonstrate a DI approach that uses multiplicative-effect modeling to combine genetic information from ≥ 2 independent studies to predict a net clinical outcome.
Methods: As a hypothetical working model, 2 independent studies were selected each reporting on a unique genetic factor proposed to influence the risk of stent thrombosis (ST) among subjects treated with clopidogrel. A multiplicative effect model was used for developing a hypothesis regarding their combined influence on clinical outcome.
Results: Application of multiplicative risk modeling yielded a revised estimated risk of outcomes based on combined genotype. In this scenario, combined genotype revised the categorical risk level (high versus low) estimated from single gene effects for 41.5% of the subjects. Further, the maximum relative risk based on single gene effects was increased from 4.54 to 7.84 based on combined genotype. The revised relative risk values in conjunction with combined genotype frequency estimates provides the data necessary to frame a trial hypothesis and conduct appropriate power analysis to estimate the number of subjects needed to test that hypothesis.
Conclusions: This DI approach can be used to generate quantitative hypotheses on multiple gene effects derived from independent genotype studies. This approach is useful for estimating parameters needed in designing future studies to evaluate the net effect of ≥ 2 genetic variants on a common clinical endpoint.