A novel form of ascertainment bias in case-control studies of cancer screening.