Reliability of laboratory tests is quite obviously affected by technical performance within the laboratory. The effect of these technical factors is reflected by test reproducibility and accuracy. Reproducibility (precision or inherent error) is a measure of how closely the laboratory can approach the same answer when the test is performed repeatedly on the same specimen. Theoretically, exactly the same answer should be obtained each time, but in actual practice this does not happen due to equipment and human imperfection. These deviations from the same answer are usually random and thereby form a random or gaussian distribution (Fig. 1-1). Variation from the average (mean) value is expressed in terms of standard deviation (SD). The laboratory frequently converts the standard deviation figure to a percentage of the mean value and calls this the coefficient of variation (CV). The majority of tests in a good laboratory can be shown to have reproducibility—expressed as CV—in the neighborhood of 4% (some may be a little better and some a little worse). This means that two thirds of the values obtained are actually somewhere between 4% above and 4% below the true value. Since ±2 SD (which includes 95% of the values) is customarily used to define acceptable limits (just as in determination of normal ranges), plus or minus twice the CV similarly forms the boundaries of permissible technical error. Returning to the 4% CV example, a deviation up to ±8% would therefore be considered technically acceptable. In some assays, especially if they are very complicated and automated equipment cannot be used, variations greater than ±8% must be permitted. The experience and integrity of the technical personnel, the reagents involved, and the equipment used all affect the final result and influence reproducibility expressed as CV. In general, one can say that the worse the reproducibility (as reflected in higher CVs), the less chance for accuracy (the correct result), although good reproducibility by itself does not guarantee accuracy.

Gaussian (random) value distribution

Fig. 1-1 Gaussian (random) value distribution with a visual display of the area included within increments of standard deviation (SD) above and below the mean: ±1 SD, 68% of total values; ±2 SD, 95% of total values; ±3 SD, 99.7% of total values.

These considerations imply that a small change in a test value may be difficult to evaluate since it could be due to laboratory artifact rather than to disease or therapy. Larger alterations or a continued sequence of change are much more helpful.

Accuracy is defined as the correct answer (the result or value the assay should produce). Besides inherent error, there is the possibility of unexpected error of various kinds, such as human mistake when obtaining the specimen, performing the test, or transcribing the result. Investigators have reported erroneous results in 0.2%–3.5% of reports from one or more areas of the laboratory. The laboratory analyzes so-called control specimens (which have known assay values of the material to be tested) with each group of patient specimens. The assumption is that any technical factor that would produce erroneous patient results would also produce control specimen results different from the expected values. Unfortunately, random inaccuracies may not affect all of the specimens and thus may not alter the control specimens. Examples of such problems are a specimen from the wrong patient, the effect of specimen hemolysis or lipemia, inaccurate pipetting, and insufficient mixing when the assay method uses a whole blood specimen. In addition, clerical errors occasionally occur. In my experience, the majority of clerical difficulties are associated with the patients who have the same last name, patients who have moved from one room to another, decimal point mistakes, transcription of results onto the wrong person’s report sheet, and placement of one person’s report sheet into the chart of someone else. These considerations imply that unexpected laboratory abnormalities greater the time lapse between the original and the new specimen, the more problems will be encountered in differentiating an error in the original specimen from true change that occurred before the next specimen. One of the more frustrating duties of a laboratory director is to receive a question or complaint about a laboratory test result several days or even weeks after the test was performed, when it is usually too late for a proper investigation.