

The confidence interval for a proportion computes results that only make sense if the variable is binomial. There must be two possible outcomes, and you know the exact number of times each occurred.
If you know that an outcome happens 25% of the time, it makes no sense to compute the confidence interval for 25/100 unless you observed exactly 100 times and saw that outcome in 25 of them.
If you know that a treatment reduces white cell count by 33%, you cannot compute a confidence interval assuming a binomial distribution. Nor if a treatment decreases body weight by 25%.