The better question is just "are you actually just selecting for symptom occurrence by socioeconomic group?"

Like you could modify the question to ask "is the model better at diagnosing people who went to a certain school?" and simplistically the answer would likely seem to be yes.