Using principal components for estimating logistic regression with high-dimensional multicollinear data