In a Wald-type interval for a logistic regression parameter $\beta$, the interval is calculated as $\hat{\beta} \pm z_{\alpha/2} \cdot SE(\hat{\beta})$. When is this interval typically problematic?