The most important single influence on laboratory test interpretation is the concept of a normal range, within which test values are considered normal and outside of which they are considered abnormal. The criteria and assumptions used in differentiating normal from abnormal in a report, therefore, assume great importance. The first step usually employed to establish normal ranges is to assume that all persons who do not demonstrate clinical symptoms or signs of any disease are normal. For some tests, normal is defined as no clinical evidence of one particular disease or group of diseases. A second assumption commonly made is that test results from those persons considered normal will have a random distribution; in other words, no factors that would bias a significant group of these values toward either the low or the high side are present. If the second assumption is correct, a gaussian (random) distribution would result, and a mean value located in the center (median) of the value distribution would be obtained. Next, the average deviation of the different values from the mean (SD) can be calculated. In a truly random or gaussian value distribution, 68% of the values will fall within ±1 SD above and below the mean, 95% within ±2 SD, and 99.7% within ±3 SD. The standard procedure is to select ±2SD from the mean value as the limits of the normal range.

Accepting ±2 SD from the mean value as normal will place 95% of clinically normal persons within the normal range limits. Conversely, it also means that 2.5% of clinically normal persons will have values above and 2.5% will have values below this range. Normal ranges created in this way represent a deliberate compromise. A wider normal range (e.g., ±3 SD) would ensure that almost all normal persons would be included within normal range limits and thus would increase the specificity of abnormal results. However, this would place additional diseased persons with relatively small test abnormality into the expanded normal range and thereby decrease test sensitivity for detection of disease.

Nonparametric calculation of the normal range. The current standard method for determining normal ranges assumes that the data have a gaussian (homogeneous symmetric) value distribution. In fact, many population sample results are not gaussian. In a gaussian value distribution, the mean value (average sample value) and the median value (value in the center of the range) coincide. In nongaussian distributions, the mean value and the median value are not the same, thus indicating skewness (asymmetric distribution). In these cases, statisticians recommend some type of nonparametric statistical method. Nonparametric formulas do not make any assumption regarding data symmetry. Unfortunately, nonparametric methods are much more cumbersome to use and require a larger value sample (e.g., і120 values) One such nonparametric approach is to rank the values obtained in ascending order and then apply the nonparametric percentile estimate formula.

Problems derived from use of normal ranges

1.
A small but definite group of clinically normal persons may have subclinical or undetected disease and may be inadvertently included in the supposedly normal group used to establish normal values. This has two consequences. There will be abnormal persons whose laboratory value will now be falsely considered normal; and the normal limits may be influenced by the values from persons with unsuspected disease, thereby extending the normal limits and accentuating overlap between normal and abnormal persons. For example, we tested serum specimens from 40 clinically normal blood donors to obtain the normal range for a new serum iron kit. The range was found to be 35-171 µg/dl, very close to the values listed in the kit package insert. We then performed a serum ferritin assay on the 10 serum samples with the lowest serum iron values. Five had low ferritin levels suggestive of iron deficiency. After excluding these values, the recalculated serum iron normal range was 60-160, very significantly different from the original range. The kit manufacturer conceded that its results had not been verified by serum ferritin or bone marrow.
2.
Normal ranges are sometimes calculated from a number of values too small to be statistically reliable.
3.
Various factors may affect results in nondiseased persons. The population from which specimens are secured for normal range determination may not be representative of the population to be tested. There may be differences due to age, sex, locality, race, diet, upright versus recumbent posture (Table 1-2), specimen storage time, and so forth. An example is the erythrocyte sedimentation rate (ESR) in which the normal values by the Westergren method for persons under age 60 years, corrected for anemia, are 0-15 mm/hour for men and 0-20 mm/hour for women, whereas in persons over age 60, normal values are 0-25 mm/hour for men and 0-30 mm/hour for women. There may even be significant within-day or between-day variation in some substances in the same person.

Decrease in test values after change from upright to supine position

Table 1-2 Decrease in test values after change from upright to supine position

4.
Normal values obtained by one analytical method may be inappropriately used with another method. For example, there are several well-accepted techniques for assay of serum albumin. The assay values differ somewhat because the techniques do not measure the same thing. Dye-binding methods measure dye-binding capacity of the albumin molecule, biuret procedures react with nitrogen atoms, immunologic methods depend on antibodies against antigenic components, and electrophoresis is influenced primarily by the electric charge of certain chemical groups in the molecule. In fact, different versions of the same method may not yield identical results, and even the same version of the same method, when performed on different equipment, may display variance.
5.
As pointed out previously, normal values supplied by the manufacturers of test kits rather frequently do not correspond to the results obtained on a local population by a local laboratory, sometimes without any demonstrable reason. The same problem is encountered with normal values obtained from the medical literature. In some assays, such as fasting serum glucose using so-called true glucose methods, there is relatively little difference in normal ranges established by laboratories using the same method. In other assays there may be a significant difference. For example, one reference book suggests a normal range for serum sodium by flame photometry of 136-142 mEq/L, whereas another suggests 135-155 mEq/L. A related problem is the fact that normal ranges given in the literature may be derived from a laboratory or group of laboratories using one equipment and reagent system, whereas results may be considerably different when other equipment and reagents are used. The only way to compensate for this would be for each laboratory to establish its own normal ranges. Since this is time-consuming, expensive, and a considerable amount of trouble, it is most often not done; and even laboratories that do establish their own normal ranges are not able to do so for every test.
6.
Population values may not be randomly distributed and may be skewed toward one end or the other of the range. This would affect the calculation of standard deviation and distort the normal range width. In such instances, some other way of establishing normal limits, such as a nonparametric method, would be better, but this is rarely done in most laboratories.

One can draw certain conclusions about problems derived from the use of the traditional concept and construction of normal ranges:

1.
Some normal persons may have abnormal laboratory test values. This may be due to ordinary technical variables. An example is a person with a true value just below the upper limit of normal that is lifted just outside of the range by laboratory method imprecision. Another difficulty is the 2.5% of normal persons arbitrarily placed both above and below normal limits by using ±2 SD as the limit criterion. It can be mathematically demonstrated that the greater the number of tests employed, the greater the chance that at least one will yield a falsely abnormal result. In fact, if a physician uses one of the popular 12-test biochemical profiles, there is a 46% chance that at least one test result will be falsely abnormal. Once the result falls outside normal limits, without other information there is nothing to differentiate a truly abnormal from a falsely abnormal value, no matter how small the distance from the upper normal limit. Of course, the farther the values are from the normal limits, the greater the likelihood of a true abnormality. Also, if two or more tests that are diagnosis related in some way are simultaneously abnormal, it reinforces the probability that true abnormality exists. Examples could be elevation of aspartate aminotransferase (SGOT) and alkaline phosphatase levels in an adult nonpregnant woman, a combination that suggests liver disease; or elevation of both blood urea nitrogen (BUN) and creatinine levels, which occurring together strongly suggest a considerable degree of renal function impairment.
2.
Persons with disease may have normal test values. Depending on the width of the normal range, considerable pathologic change in the assay value of any individual person may occur without exceeding normal limits of the population. For example, if the person’s test value is normally in the lower half of the population limits, his or her test value might double or undergo even more change without exceeding population limits.(Fig. 1-2). Comparison with previous baseline values would be the only way to demonstrate that substantial change had occurred.

Population reference

Fig. 1-2 How patient abnormality may be hidden within population reference (“normal”) range. Patients A and B had the same degree of test increase, but the new value for patient B remains within the reference range because the baseline value was sufficiently low.

Because of the various considerations outlined previously, there is a definite trend toward avoiding the term “normal range.” The most frequently used replacement term is reference range (or reference limits). Therefore, the term “reference range” will be used throughout this book instead of “normal range.”