It sounds as if you're applying some sort of linear estimation method to a variable that is constrained to non-negative values. Doing so can, as your query suggests, result in nonsensical values for either a predicted value (e.g., the point estimate) or interval estimates of a value.
If I understand your situation, to interpret the specific result you have, convert the CI to (0, 31).
To avoid this in the future, select a procedure that recognizes the restricted range (e.g., for rates, or counts).
As David Morse says, however you calculated the confidence interval of this proportion is not appropriate (it is unclear what you are trying to estimate the interval for, you might want to say what you want in your question). Using the package Hmisc, here are the estimates from a few methods for this proportion.
David Morse Considering The Total lack of information as to what the experiment was, the variables discussed, etc , and the number of possible censored and/or other possible data ranges and types of regression available for each situation and being a diabetic myself, aren't you making quite a few assumptions to get that (0,31) CI?. I hope my medical team wouldn't suggest anything but redo the entire study. Couldn't someone get killed here? David Booth
Not trying to kill anyone here! However, it's true that the original post--like many on RGate--was light on context and detail about variables, research question, etc. So, my initial inference was that the target was an estimation of cases; now it seems that the target was eGFR (scaled). If eGFR cannot be negative, then the lower bound for the CI can't be less than zero.
However, all that aside, the estimation method is clearly inappropriate (or, inappropriate assumptions were made) if bound estimates include impossible values. That's the reason for suggesting that an alternate approach would likely be preferred.
Antonious Bolas I am still not getting your model. What is a mean base difference at baseline? The difference of what? Please be more precise in describing your model. I think most of us are not physicians or biologists, so it is hard (at least for me) to derive it without context. Has the difference been calculated before to make a DV or is it part of a model (like ANCOVA or multilevel model)?
And if the DV is not only the eGFR but the difference of scores (whatever has been subtracted here), negative values are possible and would indicate that you cannot be sure about the direction of effect.