Interval estimation of the attributable risk for multiple exposure levels in case-control studies.