It's easy to fool yourself: Case studies on identifying bias and confounding in bio-medical datasets