An information theoretic approach for the evaluation of surrogates of protection: A simulation study

Abstract

The Individual Causal Association (ICA) metric of surrogacy based on the information theoretic approach (ITA), has been established as an efficient and reliable metric for the evaluation of a binary outcome as a putative surrogate marker for a true binary endpoint for chronic conditions in literature. In an attempt to apply this methodology in the context of evaluating surrogates of protection, this present simulation based study seeks to explore the behaviour of the ICA metric in evaluating surrogacy in a single-trial setting (STS) under different settings for continuous surrogate and true endpoint and dichotomization. The methodology is illustrated using the fine-tuned R Surrogate package, to generate datasets, analyse and extract results. The ICA was affected by the choice of surrogate, choice of endpoint dichotomization, underlying assumptions and measurement error. Since dichotomizing continuous endpoint and/or surrogate is common place in the context of evaluating surrogates of protection, when applying the ICA on real world data, the choice of cut-point, a sensitivity based analysis, field knowledge based assumptions and the presentation of results with more than one cut-point may be considered.