The effect of high prevalence of missing data on estimation of the coefficients of a logistic regression model when using multiple imputation