Reference ranges for screening preclinical drug safety data.

Reference ranges are used in preclinical drug safety studies to screen experimental data for atypical values. The methods used most often to construct sample reference ranges are essentially large sample methods and may flag too few "atypical" values. It is better to generate finite sample reference ranges by modifying existing methods used for constructing tolerance intervals. We define validity and efficiency for reference ranges and discuss the validity and efficiency of the methods described. A finite sample distribution-free method emerges as the clear winner.