Online Program

A Simulation Study of Statistical Power to Detect Statistical Interaction in Cox Regression Model in the Case of Non-Proportional Hazards in One of the Covariates Involved in the Interaction Effect (304821)

Keywords: Cox regression, statistical interaction, non-proportonality in hazards

Misspecified regression models due not included existing interaction term result with biased regression coefficient estimates. Common method for assessing statistical interaction is testing the product term of two or more variables included in the regression model. The Cox proportional hazards regression is a semi-parametric statistical method to analyze time-to-event data with a key assumption that the hazard ratio is constant over time for each of the covariates included in the model. We conducted a simulation study to investigate the diagnostic of interactions in the case of non-proportionality in hazards in one of the covariates involved in the interaction effect. We performed and evaluated the power of Therneu-Grambsuch non-proportionality test.

The results of the analysis suggested that the identification of a statistical interaction with the same covariate required more statistical power when the magnitude of the interaction term coefficient was higher in compare to the magnitude of coefficient of the term creating non-proportionality. Statistical testing for interaction effect may need to be performed after testing and correcting for proportionality in hazards assumption.