How does it work?

Model every variable independently. Based upon some cutoff, keep that variable and fix it in the model and then rerun all of them (p-1). Keep the best version of that and repeat until no variables reach your cutoff (they give p ≤ 0.15).