Confidence intervals of reference limits in small reference sample groups



J.P. Braun, Ecole Nationale Vétérinaire, 23 Chemin des Capelles, 31076 Toulouse Cedex 3, France




Reference intervals are the most common tool used to interpret results of laboratory tests. However, in veterinary clinical pathology, the number of available reference individuals is often small.


The purpose of this study was to investigate the effects of small reference sample groups on the imprecision of the reference limits.


Gaussian and log-Gaussian distributions of 10 ≤ n ≤ 750 values were analyzed. Reference limits and 90% confidence interval of limits (90% CI) were calculated. Imprecision of limits was estimated by the ratio of the width of the 90% CI: width of the reference interval (WCI/WRI).


For Gaussian distributions, the WCI/WRI ratio cannot be expected to be lower than 0.2 when n < 55. In log-Gaussian distributions, the ratio greatly increases for the upper limit with skewness toward high values, whereas it moderately decreases for the lower limit.


Independent of the size of the reference sample group, it is very important to report the CIs of the reference limits, which can be very large for small reference sample groups. When the sample size is very small (n < 20), calculations maybe misleading and it is better to instead report all values.