First plug in choice

Each answer choice is correct equally often on the GMAT, so you're just as likely to get the right answer first try if you start with C as if you start with D.

On some (rather few, in fact) questions, when you test an answer and discover it's incorrect, you'll know whether the correct answer is smaller or larger than what you've tested. On those questions, if you are testing answers, it's certainly most efficient to begin with either B or D. When you start with C:

1 time in 5, you'll have the answer first try
4 times in 5, you'll need to test one other choice (and after that second test, you'll know by process of elimination which answer is correct)

If you start with B:

2 times in 5, you'll have the answer first try (because if B is too large, A must be right)
3 times in 5, you'll need to test exactly one other choice - answer choice D. If D is right, you're done; if it's too large, C must be right; if it's too small, E must be right.

That's exactly what Jon said above, and yes, C is not the best choice to begin with in questions of this type.

All of that said, I don't think it's all that helpful to know this in practice:

-questions where plugging in answer choices is a good idea aren't all that common, at least at the harder level of the test;

-questions where, when plugging in answer choices, you'll know if the right answer is smaller or larger than what you've tested are much less common still;

-if you can solve the question from start to finish, rather than by doing everything backwards, you're guaranteed to do the work once only. Each time you test an answer choice, you're normally doing all of that work once, so you might be solving the same problem two or more times by backsolving.

It's a worthwhile fallback technique when you aren't able to see how to do a problem directly, and there's no harm in knowing how to backsolve and practicing it - it is a strategy that is potentially useful on lower level questions, and it has the advantage of removing some of the abstraction from some questions. It's not a technique that is especially helpful on more difficult questions, however; I analyzed the more difficult questions in the Official Guide (11th edition) to see how many could be backsolved, and wrote about the results here if you're interested:

https://gmatix.com/blogging/index.php?x=15

https://gmatix.com/blogging/index.php?x=20