ABSTRACT.
For most quantities measured in clinical medicine, healthy subjects produce a range of values which overlaps the range found in patients with disease, so that there is no sharp distinction between normal and abnormal. Furthermore, the word “normal” is ambiguous and potentially misleading. The “normal range” has therefore been superseded in clinical laboratory work by the 95% reference interval, which includes values from 95% of the healthy population. The probability of a healthy person producing a value outside the 95% reference interval is 0.05, or one in twenty. Determining a reliable reference interval requires a representative sample of at least 120, and preferably 200, subjects from the healthy population, and many more may be needed if the distribution of values is asymmetrical. Subjects should be chosen according to clearly defined criteria. The possibility of correlation with age, height, etc. should be considered, and the reference interval calculated by a method appropriate to the data distribution. The common practice of defining the normal range as (sample mean ± 2 standard deviations) is not applicable to non-Gaussian distributions, which should either be transformed to Gaussian form, or the 95% reference interval should be calculated non-parametrically by determining the 2.5 and 97.5 percentiles. Where a test comprises two or more measurements (latency and amplitude of an evoked potential, for example) the separate reference intervals must be adjusted so that the combined reference region includes 95% of healthy subjects.
KEY WORDS.: