An advantage of our screening procedure is that it provides a ranked list of variables which can then be interpreted. In our application to influenza vaccination, a large number of the top-ranked genes were related to antiviral processes, providing a clear immunological interpretation.

In simulation studies, we demonstrate that our test procedure has desirable properties, including control of the type 1 error rate even at sample sizes as small as 30.

When primary outcomes are long, costly, or infeasible to measure, surrogate markers can act as proxies. Yet their evaluation is difficult, especially with high-dimensional data and small samples. We extend a non-parametric, rank-based procedure by Parast et al. (2024, doi.org/10.1093/biom...)

The method has two stages: (1) screen high-dimensional surrogate candidates with a univariate test; (2) combine and validate them on independent data. We apply it to influenza vaccine gene expression, identifying a subset of genes that predict the vaccine effect on the antibody response.