Selection bias in case-control studies using relatives as the controls.