First-order correction of statistical significance for screening two-way epistatic interactions