Given an undecidable collection of first-order sentences, is there necessarily a complete undecidable theory containing it? 
A direct attempt to prove it seems to require some control over the completions which need not exist, but I don't see a counterexample.