Bias correction via outcome reassignment for cross-sectional data with binary disease outcome