simPH: An R Package for Showing Estimates from Cox Proportional Hazard Models Including for Interactive and Nonlinear Effects

Abstract

The R package simPH provides tools for effectively communicating results from Cox Proportional Hazard (PH) models, including models with interactive and nonlinear effects. The Cox Proportional Hazard model is a popular tool for examining event data. However, previously available computational tools have not made it easy to explore and communicate quantities of interest and associated uncertainty estimated from them. This is especially true when the effects are from interactions or nonlinear transformations of continuous variables. These transformation are especially useful with Cox PH models because they can be employed to correctly specifying models that would otherwise violate the nonproportional hazards assumption. simPH makes it easy to simulate and then plot quantities of interest for a variety of effects estimated from Cox PH models including interactive effects, nonlinear effects, as well as ‘vanilla’ linear effects. simPH employs visual weighting in order to effectively communicate estimation uncertainty. There are options to show either the standard central interval of the simulation’s distribution or the shortest probability interval – which can be useful for asymmetrically distributed estimates. This paper uses hypothetical and empirical examples to illustrate simPH’s syntax and capabilities.