Relative performance of gene- and pathway-level methods as secondary analyses for genome-wide association studies