To the Editor.—It has been suggested that a ratio relating measured enzyme activity to the upper limit of normal (ULN) for each laboratory, a conversion frequently made in clinical literature, may be a dimensionless common or relative unit (RU) that is the same for all laboratories because the computation renders it independent of the laboratory and the method. Clinicians and laboratorians have recommended that all laboratories report enzyme results in RUs, thereby relieving clinicians of the need to know the reference range for each laboratory.1–4 It is the expectation that the error of RUs is about the same as that of activity units or, at least, that the error of results converted to RUs is satisfactory for clinical purposes.The measurement of catalytic enzyme activity most clearly presents the problem of interlaboratory comparisons of method-dependent numerical results. The College of American Pathologists (CAP) addressed this issue in an enzyme survey conducted from 1977 to 1997. During this time, Drs George Grannis, Charles Massion, and John Lott investigated ways to merge method-dependent enzyme activity results and to compare enzyme activity measurements and their ULNs across laboratories using different enzyme methods.During the last 15 years of the CAP enzyme survey, the primary database consisted of RUs computed by dividing each laboratory’s measured activity by the ULN for adult males reported by that laboratory. After merging the data in RUs, the average of the results in RUs across all laboratories was taken as the consensus value for each specimen. The all-laboratory mean in RUs became the point of reference from which to measure interlaboratory variability. For example, if the grand mean in RUs for a sample was 1.00, then, on the average, the enzyme activity measured by a laboratory equaled the ULN reported by that laboratory.The enzyme survey was repeated 4 times each year, so that linear regression of enzyme activity on the consensus RU values gave a characteristic activity slope for each laboratory and method. The slope was the average change in measured enzyme activity per consensus RU. Because each laboratory’s true ULN was by definition that laboratory’s characteristic activity slope, each laboratory’s “true” ULN in RUs was computed by dividing the laboratory’s ULN by its activity slope. Then the average ULN in RUs and its variance were computed for each method and for all laboratories in the survey. For example, if the all-laboratory ULN in RUs was 1.00, then on the average each laboratory reported a ULN equal to its activity slope.The ULN in RUs is a global statistic that incorporates both the error of the activity measurement and the error in the ULNs. Within the range of the linear relationship established by the multilevel enzyme survey samples, the variance of laboratory results in RUs can be computed at multiples of the ULN for an average laboratory, for a method, and for all laboratories.The data in the Table are summarized from the 1985 CAP enzyme survey results, the last year that the survey included alanine aminotransferase (ALT) and aspartate aminotransferase (AST).5 About 1400 laboratories participated. The results have been expressed in the Table as if the RU system was in effect: (1) all laboratories report activities in RUs; (2) the ULN is 1.00 RU; and (3) the RU is the same unit across all laboratories. As examples, the 95% confidence limits for results in RUs for 4 enzymes assayed by 2 methods are given at 1.5 times normal (1.50 RU) for 2 clinical applications. The first application is diagnostic classification of a patient given a result of 1.50 RU (diagnostic testing in the Table). The second is detecting a significant analytic change on follow-up of a patient when the first result was 1.50 RU (serial monitoring in the Table).The within-laboratory error is the baseline for the within-method and all-laboratory RU comparisons, because on the relative scale it is the same as the analytic error in the originally reported activity units. The within-laboratory 95% confidence limits range from 1.41 to 1.59 RU up to 1.10 to 1.90 RU over the 8 cases. This means that a 1.50 RU result is not likely to be reported on a subject judged to be normal by the laboratory (<1.00 RU) or on a patient whose actual value is higher than 2.00 RU. Furthermore, a result of 1.00 or 2.00 RU on follow-up of a patient in the same laboratory with an original value of 1.50 is an analytically significant change.If a result of 1.50 RU for a patient can be selected from any laboratory using 1 of the given methods, its 95% confidence limits are wider. The within-method limits range up to 1.06 to 1.94 RU and still do not include 1.00 and 2.00 RU in the 4 single-point diagnostic testing examples in the Table. However, in the cases of ALT and AST, because of the wider serial-monitoring confidence limits, a follow-up result of 1.00 and 2.00 RU from the within-method laboratory population does not indicate an analytically significant difference.If a result of 1.50 RU can be chosen from any laboratory, the confidence limits are even wider. The all-laboratory 95% confidence limits now include 1.00 and 2.00 RU in both clinical settings for all 4 enzymes. This means that, when comparing a result of 1.50 RU from 1 laboratory to a follow-up result from another laboratory using a different method, the clinician cannot be confident of a change unless the second value is 0 to 0.37 or higher than 2.63 to 3.02 RU. Similarly, given a result of 1.50 the clinician can be confident only that the value obtained from another laboratory on the same sample will be no lower than 0.40 to 0.69 RU and no higher than 2.31 to 2.60 RU.Although some variation in the amount of analytic error exists between laboratories, the inflated error of results expressed in RUs is attributable primarily to the effect of error in the estimate of the ULN. Lott et al6 found that the nonparametric central 95% of ULNs expressed in RUs for AST ranged from 0.25 to 1.75 RU and for ALT from 0.50 to 1.50 RU. In the case of ALT, this means that 95% of laboratories with the identical enzyme activity slope of 40 U/L used ULNs ranging from 20 to 60 U/L. In the case of AST, the 95% range at 40 U/L was 10 to 70 U/L.Because there is no agreement that the choice of type of reference population or reference limit should be the same for all laboratories, the variance of the ULN is attributable in part to variation in medical utility needs. For example, ALT testing to exclude ethanol or other drug effects lowers the reference limit, whereas ALT testing for good evidence of viral disease raises it. Some laboratories may choose a midpoint to optimize both sensitivity and specificity of the test. Other medical settings may lead to weighting of sensitivity or specificity, thereby skewing the reference limit to favor the weighted false result. The reference population chosen for some enzymes, for example, amylase, may include subjects who are not well but lack the disease of clinical interest, for example, acute pancreatitis. Two reference limits may be specified, one indicating absence of disease and the other absence of the disease of primary interest.Some of the variation in the ULN appears to be simply error in the estimate of a reference limit or inappropriate use of a vendor’s reference limit. For example, the 95% confidence limit of the nonparametric 97.5th percentile of a typical, positivity-skewed reference population of 310 nondiseased subjects evaluated recently in my laboratory was 16.1%. Lott et al7 reported that the error in the ULN recommended by 12 vendors for AST averaged 15.7% (range, 0%–54.2%), and for ALT averaged 21.7% (range, 0%–48.9%).There are several caveats for these data. First, the all-laboratory means in RUs used as the true values for linear regression of activity on RUs were standardized by reference limits for adult males. Because these values are not applicable in all cases to females and young or old subjects, accurate age- and sex-specific ULNs in RUs are necessary. The interlaboratory variation of age- and sex-specific RUs is likely to be larger than those given in the Table in some cases.Second, because data regarding the average biases of individual laboratories and methods or their distributions are not available, I have not included in the variances the bias terms for the individual laboratories and the individual methods to compute the total error of the average laboratory and method. These biases randomize within method and within the survey and are included in the all-laboratory variation.Last, the 2 methods presented as examples had better within-method standardization than the majority of those evaluated in 1985.Careful use of ULN ratios by 1 laboratory can be clinically and analytically appropriate. Division of activity units by an accurate, reproducible ULN decreases the mean and SD proportionately, and the coefficient of variation is unchanged. The clinician can have the same confidence in the results in RUs as before. It is possible that the RU for the laboratory will not change if the method is changed. However, this stability is not automatic. It requires use of the same type of reference population, the same reference limit of the reference population, and the same error in the estimate of the reference limit.Efforts to standardize enzyme measurements by reference materials, calibrators, and methods continue.89 Relative units lessen the method dependency of catalytic activity units, but their use introduces a parameter of a reference population as a new variable. If RUs have a place in the standardization process, then the process will include at some point a laboratory-dependent step. This may be an avenue toward improvement, but certainly further investigation would be necessary. For example, estimating a high percentile of a reference population is error prone. A more stable statistic, the population geometric mean, was chosen for the international normalized ratio approach to prothrombin time standardization. Use of a specific reference population is necessary. The best reference population for international normalized ratio standardization is subjects in ideal health. Medical utility must be addressed. For example, the international normalized ratio is useful only in 1 clinical setting, that of long-term warfarin therapy.The appearance (or assumption) of equivalency is the main danger of widespread use of RUs for enzymes. If the unwary clinician does not interpret the results with the appropriate confidence limits, the risk of errors in diagnosis is increased.The use of RUs may be practical for 1 laboratory or a closely coordinated group of laboratories using the same method. However, their universal implementation at this time will impair rather than improve the quality of patient care.
No takes yet. Share an insight, caveat, or question.
John W. Ross (1999) studied this question.
Synapse has enriched 5 related papers. Consider them for comparative context: