Analytic Bias of Thyroid Function Tests. Analysis of a College of American Pathologists Fresh Frozen Serum Pool by 3900 Clinical Laboratories

Transcription

1 Analytic Bias of Thyroid Function Tests Analysis of a College of American Pathologists Fresh Frozen Serum Pool by 3900 Clinical Laboratories Bernard W. Steele, MD; Edward Wang, PhD; George G. Klee, MD, PhD; Linda M. Thienpont, PhD; Steven J. Soldin, PhD; Lori J. Sokoll, PhD; William E. Winter, MD; Susan A. Fuhrman, MD; Ronald J. Elin, MD, PhD Ideally, participants in a proficiency testing program could compare their performance with that of all laboratories analyzing a particular analyte. However, it has Accepted for publication November 1, From the Department of Pathology, University of Miami School of Medicine, Miami, Fla (Dr Steele); the College of American Pathologists, Northfield, Ill (Dr Wang); the Department of Laboratory Medicine and Pathology, Mayo Clinic, Rochester, Minn (Dr Klee); Laboratory for Analytical Chemistry, Faculty of Pharmaceutical Sciences, Ghent University, Ghent, Belgium (Dr Thienpont); the Department of Laboratory Medicine, Children s National Medical Center, and the Departments of Pediatrics and Pathology, The George Washington University School of Medicine, Washington, DC (Dr Soldin); the Department of Pathology, Johns Hopkins Medical Institutions, Baltimore, Md (Dr Sokoll); the Department of Pathology, University of Florida, Gainesville (Dr Winter); the Grant Medical Center and Riverside Methodist Hospital, Columbus, Ohio (Dr Fuhrman); and the Department of Pathology and Laboratory Medicine, University of Louisville, Louisville, Ky (Dr Elin). Dr Steele declares that he has received research grants from Beckman Coulter and Ortho-Clinical Diagnostics for work unrelated to the preparation of this manuscript. Dr Klee declares that he has received research grants from Beckman Coulter and Biosite for work unrelated to the preparation of this manuscript. All other authors also have no relevant financial interest in the products or companies described in this article. Context. In proficiency testing surveys, there are differences in the values reported by users of various analytic methods. Two contributors to this variation are calibrator bias and matrix effects of proficiency testing materials. Objectives. (1) To quantify the biases of the analytic methods used to measure thyroid-stimulating hormone, thyroxine, triiodothyronine, free thyroxine, and free triiodothyronine levels; (2) to determine if these biases are within allowable limits; and (3) to ascertain if proficiency testing materials correctly identify these biases. Design. A fresh frozen serum specimen was mailed as part of the 2003 College of American Pathologists and Chemistry surveys. The means and SDs for each analytic method were determined for this sample as well as for a proficiency testing sample from both surveys. In the fresh frozen serum sample, target values for thyroxine and triiodothyronine were determined by isotope dilution/liquid chromatography/tandem mass spectrometry. All other target values in the study were the median of the means obtained for the various analytic methods. Main Outcome Measures. Calibration biases were calculated by comparing the mean of each analytic method with the appropriate target values. These biases were evaluated against limits based on intra- and interindividual biological variation. Matrix effects of proficiency testing materials were assessed by comparing the rank of highest to lowest analytic method means (Spearman rank test) for each analyte. Participants. Approximately 3900 clinical laboratories were enrolled in the College of American Pathologists Chemistry and surveys. Results. The number of methods in the that failed to meet the goals for bias was 7 of 17 for thyroid-stimulating hormone and 11 of 13 for free thyroxine. The failure rates were 12 of 16 methods for thyroxine, 8 of 11 for triiodothyronine, and 9 of 11 for free triiodothyronine. The means of the analytic method for the proficiency testing material correlated significantly (P.05) only with the fresh frozen serum means for thyroxine and thyroid-stimulating hormone in the Chemistry and free triiodothyronine in the. Conclusions. A majority of the methods used in thyroid function testing have biases that limit their clinical utility. Traditional proficiency testing materials do not adequately reflect these biases. (Arch Pathol Lab Med. 2005;129: ) long been known that the materials used in proficiency testing demonstrate matrix effects that cause a bias in survey results. 1,2 Therefore, many proficiency testing programs, including the surveys of the College of American Pathologists (CAP), employ a grading system in which the target values are the means of the individual analytic methods used by the participants. That is, a laboratory using a particular method is evaluated against the mean of all the laboratories (peer-group mean) using that same method. Although the separation of results from different analytic methods corrects for any matrix effects, it potentially allows a laboratory using a method with a large analytic bias to report erroneous patient results. In 1994, in an attempt to reduce method calibration error, the CAP conducted a fresh frozen serum (FFS) study to determine matrix biases for 11 routine chemistry analytes. 3 In that study, calibration bias was defined as the difference in values obtained in clinical samples using a definitive/reference method and a field method. The pres- Reprints: Bernard W. Steele, MD, Jackson Memorial Hospital, Holtz Center Room 2026 (R-5), 1611 NW 12th Ave, Miami, FL ( bsteele@med.miami.edu). 310 Arch Pathol Lab Med Vol 129, March 2005 Analytic Bias of Thyroid Function Tests Steele et al

2 ent FFS study is an extension of the 1994 study and includes endocrine analytes not tested in the original study. Also, the current study was expanded from approximately 550 laboratories to nearly 4000 participants. This report describes the details of the results for tests of thyroid function. In addition to the 2 analytes, thyroxine and triiodothyronine, for which definitive/reference methods are available, the study also includes 3 other thyroid-related tests: thyroid-stimulating hormone (TSH), free thyroxine, and free triiodothyronine. The current study examines whether or not different field methods for thyroid function tests give the same results in a clinical surrogate an aliquot of an FFS pool. Fresh frozen serum precision and bias data were evaluated to determine if they met acceptance criteria for bias and accuracy. Finally, proficiency testing materials (PTM) results were examined to determine if they adequately reflected any biases demonstrated in the FFS challenges. MATERIALS AND METHODS Study Design An aliquot of FFS pool was mailed as 1 of the 5 challenges in the first mailing of the 2003 CAP. Approximately 6 months later, a second aliquot was included in the third mailing of the Chemistry. Both the FFS and the PTMs were shipped in the manner typical of these surveys, that is, as frozen specimens packed in frozen gel packs. The approximately 3900 participants of the and Chemistry surveys tested, in the manner prescribed by the Clinical Laboratory Improvement Amendments of 1988, 4 the fresh frozen serum sample, along with 4 other challenges, as if they were real patient samples. The results obtained for the FFS samples and comparable PTMs were analyzed and reported back to the participants in the manner typical of the respective surveys. The data were then reanalyzed for the purposes of this investigation. First, the histograms of the data were visually inspected, and errors that occurred because participants incorrectly filled out the report form were removed. The data were then subjected to a 2-pass, 3-SD test for outliers in the manner used in the CAP and Chemistry surveys. 5 Those data points that were greater than 3 SDs from their peer-method mean on the first pass were removed. The procedure was repeated. In addition, laboratories using methods not specified or methods with fewer than 10 participants were not included, because these data tended to result in non-gaussian distributions. Further, such data included little information to enable inferences about the shape of the distribution, which might have resulted from a larger population of analyses. Samples Two types of specimens were used for the study. The first was a pool of FFS that was collected and processed as described by Miller et al. 6 In brief, the FFS specimens were prepared by Aalto Scientific (Carlsbad, Calif) using a modification of the NCCLS C37-A Guideline. 7 Donor blood was collected in an additive-free plastic collection bag and immersed in an ice-water bath. The donor bag was then centrifuged and the platelet-rich plasma transferred to a sterile plastic centrifuge bottle. The plasma was allowed to clot at room temperature for 4 hours. The serum was transferred to another sterile centrifuge bottle and flash frozen. Serum units were shipped to the processing center in frozen CO 2 and stored at 70 C for up to 2 months prior to pooling. Specific details of this procedure and of the quality of the material are published elsewhere. 6 PTMs were prepared by Bio-Rad Laboratories (Hercules, Calif). The individual challenges in both surveys were admixtures of human serum based high and low targeted pools. The high pools were created by the addition of the analytes to be measured. To achieve the low targets in the Chemistry, the low master pool was diluted with de-ionized water, and salts Table 1. List of Reagent Systems Studied Arch Pathol Lab Med Vol 129, March 2005 Analytic Bias of Thyroid Function Tests Steele et al 311 Code A1 A2 A3 B1 B2 B3 B4 C1 C2 D E F1 F2 F3 G1 G2 G3 G4 H I1 I2 I3 I4 J1 J2 K Reagent System* Abbott Architect Abbott Axsym Abbott IMX Bayer ACS:180 Bayer ACS:180 3rd Gen Bayer Advia Centaur Bayer Advia Centaur 3G Beckman Coulter Access/2 Beckman Coulter Synchron System biomerieux CEDIA Dade Behring Dade Dimension Dade Dimension HM DPC Coat-A-Count DPC Immulite DPC Immulite 2000 DPC Immulite G Microgenics DRI Roche Cobas Integra Roche Cobas-FP Roche E170 Roche Elecsys 1010/2010 Tosoh AIA-Pack Tosoh AIA-Pack 3rd Gen Vitros ECi * Abbott Laboratories, Chicago, Ill; Bayer Diagnostics, Tarrytown, NY; Beckman Coulter, Fullerton, Calif; biomérieux, Marcy l Etoile, France; Roche Diagnostics, Indianapolis, Ind; Dade Behring, Deerfield, Ill; Diagnostic Products Corporation, Los Angeles, Calif; Microgenics, Fremont, Calif; TOSOH Bioscience, San Francisco, Calif; and Ortho-Clinical Diagnostics, Inc, Rochester, NY. were added. For the, the low master pool was either diluted with bovine serum albumin, ligand stripped, or both to lower analyte concentration. Samples with analyte values similar to those of the fresh frozen pool were selected for the comparisons in this study. For the challenge specimen, K-03 from the first mailing of the 2003 K/KN was used. From the third mailing of the Chemistry, specimen C-04 was chosen for thyroxine comparison and C-03 for comparing the other thyroid function tests. For both surveys, the designated FFS specimens were K-02 and C-02. Statistical Analyses A mean, SD, and coefficient of variation were calculated for each analytic method listed in Table 1. Target values were established for each analyte by either reference/definitive method or the median of the means of the analytic methods. For thyroxine and triiodothyronine, the target values in the FFS pool were determined by reference laboratories using isotope dilution/liquid chromatography/tandem mass spectrometry The thyroxine value was the average of 2 laboratories 5 and 10 respective measurements. The triiodothyronine value was the average of 1 laboratory s 10 replicates. The target values for all other analytes in the FFS were the medians of the means of the field methods (peer-group means). For thyroxine and triiodothyronine, the differences between the FFS means for each analytic method and the respective reference target values were considered true calibration biases. For the other 3 analytes, the differences between the FFS method means and the median of the peer-group means were considered relative calibration biases. The calibration biases and SDs were compared to analytic goals for bias and accuracy. One set of bias goals was calculated using equation (1), 11,12 using the intraindividual and interindividual biological variation data reported by Sebastian-Gambaro et al. 13 The resulting percentages were multiplied by the target value.

3 Table 2. Summary of Means of the Analytic Methods and Coefficients of Variation (CVs) for the (K-A) * Specimen Fresh frozen serum Proficiency testing material Statistics Number of laboratories Number of methods Target value by isotope dilution/liquid chromatography/tandem mass spectrometry Medians of means of the analytic methods Ranges of means of the analytic methods Ranges of peer-group CVs Ranges of calibration bias Ranges of relative calibration bias No. of peer groups with bias 10% No. of peer groups with bias 20% Medians of means of the analytic methods Ranges of means of the analytic methods Ranges of peer-group CVs Ranges of relative calibration bias No. of peer groups with bias 10% No. of peer groups with bias 20% *T 4 indicates thyroxine; T 3, triiodothyronine; and TSH, thyroid-stimulating hormone. Calibration bias for T 4 and T 3 and relative calibration bias for the other analytes. T 4, g/dl (nmol/l) Free T 4, ng/dl (pmol/l) (82.9) (84.1) ( ) 4.01% 8.79% 6.54% 11.32% % 2 (13%) 0 (0%) 6.69 (86.3) ( ) 3.41% 10.1% 15.7% 50.8% 9 (56%) 6 (38%) 1.01 (13.0) ( ) 5.05% 13.4% 21.2% 16.4% 5 (38%) 1 (8%) 1.64 (21.2) ( ) 4.62% 10.4% 39.7% 96.7% 11 (85%) 7 (54%) where CV w 2 2 Bias 0.25 CVw CV b (1) average within-individual biological variation CV average between-individual biological variation b A second set of bias goals used the approach of Klee, 14,15 who defined analytic bias limits using the variations found in cumulative test value distributions. This method uses the batch-tobatch variation observed when distributions of large numbers (eg, 1000) of laboratory test values generated by a stable assay are compared over time. These distributions are used to calculate the proportion of patients having values outside key decision points. The goals for analytic bias are set at the concentration limits corresponding to 1 SD of the variation in these proportions. The goal for inaccuracy, the difference between a single measured result and the expected value, 16 was calculated using equation (2) 17 and the data set of Sebastian-Gambaro et al Inaccuracy CVw 0.25 CVw CV b (2) For each analytic method, a total error value was calculated. Total error was defined as 2 SDs plus the absolute bias, with bias being the difference between the method means and their respective target values. If the total error was greater than the inaccuracy goal, at least 2.5% of the method s results would be expected to be inaccurate. Likewise, if the value of the absolute bias was greater than the accuracy goal, more than 50% of the results would be considered inaccurate. To assess matrix effects, Spearman rank correlations between FFS and PTM peer-group means were performed to test whether FFS and PTM means exhibited similar ascending or descending patterns across all analytic methods. All data analyses were performed using SAS for Windows version 8.2 software (SAS Inc, Cary, NC), and significance was defined as P.05. RESULTS For each analyte, Tables 2 and 3 present a summary of the number of participants and methods, the median value of the means of the analytic methods, and the ranges of the means and coefficients of variation. Listed also are the FFS target values measured by isotope dilution/liquid chromatography/tandem mass spectrometry for thyroxine and triiodothyronine. For the other analytes specimen combinations, the target values are the medians of the analytic methods. In the Chemistry, means of the analytic methods for the FFS thyroxine, free thyroxine, and TSH ranged from 5.95 to 7.31 g/dl ( nmol/l), 0.79 to 1.17 ng/dl ( pmol/l), and 1.24 to 1.73 U/mL ( mu/l), respectively. This equates to calibration biases (true or relative) varying from 7.5% to 13.7%, 21.7% to 15.8%, and 13.4% to 21.0%, respectively. Results were similar for the. Means of the analytic methods for the FFS triiodothyronine and free triiodothyronine ranged from 88.4 to ng/dl ( nmol/l) and 233 to 413 pg/dl ( pmol/l), respectively, with the calibration bias/relative calibration bias being 32.0% to 0.6% and 25.5% to 32.1%, respectively. Except for thyroxine in the Chemistry, the ranges were uniformly wider for the PTMs. Tables 2 and 3 also show the number of methods that had true or relative calibration biases greater than the arbitrary cutoffs of 10% and 20%. In the Chemistry, the percentage of methods with biases greater than 10% and 20%, respectively, were 21% and 0% for thyroxine, 38% and 8% for free thyroxine, and 28% and 11% for TSH. Similar results were obtained in the. For total and free triiodothyronine in the, the percentage of methods with calibration biases greater than 10% or 20% were 58% and 33%, and 55% and 27%, respectively. Results for the PTMs were generally poorer. Figure 1, a through h, illustrates FFS bias and random error for each analytic method. The solid lines represent the allowable biases based on biological parameters, 11,12 and the dashed lines correspond to those based on test distributions (G.G.K., unpublished data, 1991). These limits and the number of analytic methods that fail to meet these bias criteria are tabulated in Table 4. For example, Figure 1, c and d, illustrates that only 4 of 16 and 8 of 19 Chemistry thyroxine methods had 312 Arch Pathol Lab Med Vol 129, March 2005 Analytic Bias of Thyroid Function Tests Steele et al

4 T 3, ng/dl (nmol/l) Table 2. Extended Free T 3, pg/dl (pmol/l) TSH, U/mL (mu/l) (2.00) (1.69) ( ) 4.8% 13.4% 32.0% 0.6% 19.8% 18.6% 7 (58%) 4 (33%) (2.14) ( ) 4.5% 11.2% 23.3% 38.0% 7 (58%) 4 (33%) 313 (4.82) ( ) 2.83% 15.31% 25.5% 32.1% 6 (55%) 3 (27%) 461 (7.10) ( ) 3.03% 18.2% 50.0% 90.3% 8 (73%) 4 (36%) 1.46 (1.46) ( ) 4.23% 8.09% 15.1% 18.2% 5 (29%) 0 (0%) 0.60 (0.60) ( ) 4.59% 9.59% 19.1% 35.3% 5 (29%) 3 (18%) calibration biases based on biological variation within the allowable limit. In addition, only 2 of 11 triiodothyronine methods met the criterion (Figure 1, g). The ranges of the means for the TSH, free thyroxine, and free triiodothyronine methods (Table 2) are 3.7, 10.5, and 9.7 times the respective allowable biases based on biological variation (Table 4). Thus, there would always be numerous methods outside the acceptable limits regardless of what method was used to assign target values. Furthermore, if the target values for TSH, free thyroxine, and free triiodothyronine were not in the center of the distribution of the analytic method means, there would be even more failures. The limits for inaccuracy (Table 4) are also plotted in Figure 1, a through h. The intersection of these lines with a total error column indicates the percentage of results reported that would be expected to be inaccurate. For example, if the allowable limit line intersects a column in the 2-SD portion, then at least 2.5% (the area to the right of the 2-SD limit of a Gaussian distribution) of the results reported by participants would be inaccurate. This applies to 9 of the free thyroxine methods in the (Figure 1, e) and 11 in the Chemistry (Figure 1, f) and to only 1 and 2 methods, respectively, for TSH (Figure 1, g and h). If the line intersects the absolute bias portion of the column, then at least 50% of the results (the area to the right of the mean of a Gaussian distribution) would be expected to be inaccurate. This applies to 4 of the free thyroxine methods in the and 2 in the Chemistry (Table 4). Lastly, the Clinical Laboratory Improvement Amendments accuracy limit of 20% of the target value for thyroxine 4 is graphed in Figure 1, c and d. Six of the Chemistry analytic methods and 5 of the analytic methods would have had at least 2.5% of their users fail the challenge if a single target value were used. Because the true values of the PTMs were not known, it was not possible to quantify the exact value of the matrix effect the PTM exerted on each method, that is, the analytic method mean, minus the true value, minus the calibration bias. However, if there were no matrix effects present, then the PTM means should all be higher or lower, by the same amount, than their corresponding FFS mean. The difference between each method s FFS and PTM means would be the difference between the true values of the FFS and PTM. As seen in Figure 2, a through h, which are plots of all the peer-group means for the FFS and PTMs, this relationship was not observed. There were clear matrix differences between FFS and PTM for all analytes in both surveys. For example, in Figure 2, a, the distance between each PTM and FFS mean was not equal, and in Figure 2, c, the PTM peer-group means were both higher and lower, depending on the method, than the FFS means. For TSH, thyroxine, and free thyroxine in the (Figure 2, a, c, and e), the Spearman rank test for the order of the method means of FFS and PTM Specimen Table 3. Fresh frozen serum Proficiency testing material Summary of Means of the Analytic Methods and Coefficients of Variation (CVs) for the Chemistry (C-C) * Statistics Number of laboratories Number of methods Target value by isotope dilution/liquid chromatography/tandem mass spectrometry Medians of means of the analytic methods Ranges of means of the analytic methods Ranges of peer-group CVs Ranges of calibration bias Ranges of relative calibration bias No. of peer groups with bias 10% No. of peer groups with bias 20% Medians of means of the analytic methods Ranges of means of the analytic methods Ranges of peer-group CVs Ranges of relative calibration bias No. of peer groups with bias 10% No. of peer groups with bias 20% *T 4 indicates thyroxine; TSH, thyroid-stimulating hormone. Calibration bias for T 4 and relative calibration bias for the other analytes. T 4, g/dl (nmol/l) Free T 4, ng/dl (pmol/l) TSH, U/mL (mu/l) (82.9) (84.6) ( ) 4.1% 8.6% 7.5% 13.7% 9.4% 11.4% 4 (21%) 0 (0%) 6.88 (88.8) ( ) 4.3% 8.4% 8.38% 5.23% 5 (26%) 0 (0%) 1.01 (13.01) ( ) 4.0% 14.7% 21.7% 15.8% 5 (38%) 1 (8%) 0.88 (11.4) ( ) 5.0% 13.4% 25.3% 45.2% 6 (46%) 3 (23%) 1.43 (1.43) ( ) 2.8% 12.4% 13.4% 21.0% 5 (28%) 2 (11%) 1.35 (1.35) ( ) 3.2% 8.5% 12.6% 23.7% 6 (33%) 1 (6%) Arch Pathol Lab Med Vol 129, March 2005 Analytic Bias of Thyroid Function Tests Steele et al 313

5 Figure 1. Total error across laboratories in the first mailing of the (K-A) and the third mailing of the Chemistry (C-C). Each column represents the total error, that is, absolute bias plus 2 peer-group SDs. indicates allowable bias based on biological variation;, allowable bias based on test populations;, allowable inaccuracy; and, 20% of target value fixed Clinical Laboratory Improvement Amendments criterion for thyroxine only. 314 Arch Pathol Lab Med Vol 129, March 2005 Analytic Bias of Thyroid Function Tests Steele et al

6 Figure 2. The peer-group means of the fresh frozen serum (FFS) and the comparable (closest in value) proficiency testing material (PTM) in the (K-A) and Chemistry (C-C) surveys are plotted against analytic method. Also presented in the figures are 95% confidence intervals (CI) of each analytic method mean and the Spearman rank correlation coefficient (r) between the analytic method means of FFS and PTM. Arch Pathol Lab Med Vol 129, March 2005 Analytic Bias of Thyroid Function Tests Steele et al 315

7 Table 4. Allowable bias criteria based on biological variation Allowable bias criteria based on population test distributions The Number of Methods Not Meeting Criteria for Bias and Accuracy* T 4, g/dl (nmol/l) 3.33% 0.21 (2.71) 5 (64.5) 0.33 (4.26) Allowable inaccuracy 10.3% 0.66 (8.51) Chemistry Free T 4, ng/dl (pmol/l) 3.59% (0.464) Analyte 1.8 (23.22) (1.30) 12.5% (1.62) Chemistry T 3, ng/dl (nmol/l) 4.72% 6.14 (0.095) Free T 3, pg/dl (pmol/l) 5.96% 18.6 (0.286) TSH, U/mL (mu/l) 8.89% (0.129) 5 (5) (0.325) 13.8% 18.0 (0.277) 15.2% 47.4 (0.730) 32.2% (0.469) Chemistry No. of methods No. of methods not meeting allowable bias criteria based on biological variation No. of methods not meeting allowable bias criteria based on population test distributions No. of methods having at least 2.5% of responses not meeting the allowable inaccuracy criteria No. of methods having at least 50% of responses not meeting the allowable inaccuracy criteria *T 4 indicates thyroxine; T 3, triiodothyronine; and TSH, thyroid-stimulating hormone. For T 4 and T 3, the isotope dilution/liquid chromatography/ tandem mass spectrometry values were used, and for other analytes, the median of the means of the analytic methods. revealed poor correlation, ranging from 0.41 to 0.1. Overall, the Chemistry illustrates fewer PTM effects, suggesting that different PTMs have different matrix effects. COMMENT This study shows that there are absolute or relative calibration biases in popular methods used to measure 5 major thyroid function analytes. We found that 47 of 68 analytic methods in the and 32 of 50 methods in the Chemistry had unacceptable analytic bias, using acceptability criteria based on biological variation within a population. Although the technique of using criteria based on biological variation has been criticized by Doumas 18 as being too arbitrary, it allowed us to objectively apply the same published formulae and the same biological variability data source for all of the analytes. The second purpose of this study was to examine the ability of PTMs to detect biases among different analytic methods. As Figure 2 shows, the magnitude of the matrix effect varies not only with the analytic method, but with the PTM itself. The issue of lot-to-lot variation was not examined, and there is no assurance that making different batches of a PTM would produce products with the same matrix effect. Thus, it appears at this time that PTMs cannot be used to detect analytic biases, because of confounding and variable matrix effects. There is a clinical need for assuring that different analytic methods give similar results. To accomplish this harmonization, all field methods must use calibration materials that make their results similar to those of other methods. It follows that manufacturers would need to agree on reference/definitive methods, and several manufacturers would likely have to recalibrate their kits. The benefits of this harmonization are 2-fold. First, harmonization of results would allow patients to transport their laboratory data between health organizations that use different immunoassay methods. Second, harmonization of results would allow development of clinical protocols that are independent of analytic platforms. An example is found in the National Cholesterol Education Program, which includes clinical guidelines 19 that are not method dependent. 20,21 Until there is harmonization of the common field methods for thyroid function, the full potential of the laboratory to improve medical care for thyroid disease will not be realized. References 1. Gilbert RK. Analysis of results of the 1969 comprehensive chemistry survey of the College of American Pathologists. Am J Clin Pathol. 1970;54: Gilbert RK. The size and the source of analytic error in clinical chemistry. Am J Clin Pathol. 1974;61: Ross JW, Miller WG, Myers GL, Praestgaard J. The accuracy of laboratory measurements in clinical chemistry: a study of 11 routine chemistry analytes in the College of American Pathologists Chemistry with fresh frozen serum, definitive methods, and reference methods. Arch Pathol Lab Med. 1998;122: Clinical Laboratory Improvement Amendments of 1988, Final Rule, 57 Federal Register (1992). 5. Barnett V, Lewis T. Outliers in Statistical Data. New York, NY: Wiley; Miller WG, Myers GL, Ashwood ER, et al. Creatinine measurement: state of the art in accuracy and interlaboratory harmonization. Arch Pathol Lab Med. 2005;129: Preparation and Validation of Commutable Frozen Human Serum Pools as Secondary Reference Materials for Cholesterol Measurement Procedures: Approved Guideline. Wayne, Pa: NCCLS; NCCLS document C37-A. 8. De Brabandere VI, Hou P, Stockl D, Thienpont LM, De Leenheer AP. Isotope dilution-liquid chromatography/electrospray ionization-tandem mass spectrometry for the determination of serum thyroxine as a potential reference method. Rapid Commun Mass Spectrom. 1998;12: Soukhova N, Soldin OP, Soldin SJ. Isotope dilution tandem mass spectrometry for T3/T4. Clin Chim Acta. 2004;343: Soldin OP, Tractenberg RE, Soldin SJ. Differences between measurements of T4 and T3 in pregnancy and in non-pregnant women using isotope dilution 316 Arch Pathol Lab Med Vol 129, March 2005 Analytic Bias of Thyroid Function Tests Steele et al

8 tandem mass spectrometry and immunoassays: are there clinical implications? Clin Chim Acta. 2004;347: Fraser CG, Petersen PH. Analytical performance characteristics should be judged against objective quality specifications [editorial]. Clin Chem. 1999;45: Fraser CG. Quality specifications in laboratory medicine: current consensus views. Accred Qual Assur. 1999;4: Sebastian-Gambaro MA, Liron-Hernandez FJ, Fuentes-Arderiu X. Intra- and inter-individual biological variability data bank. Eur J Clin Chem Clin Biochem. 1997;35: Klee G. A conceptual model for establishing tolerance limits for analytic bias and imprecision based on variations in population test distributions. Clin Chim Acta. 1997;260: Klee GG. Tolerance limits for short-term analytical bias and analytical imprecision derived from clinical assay specificity. Clin Chem. 1993;39: Bureau International des Poids et Mesures, International Electrotechnical Commission, International Federation of Clinical Chemistry and Laboratory Medicine, International Organization for Standardization, International Union of Pure and Applied Chemistry, International Union of Pure and Applied Physics, International Organization of Legal Metrology. International Vocabulary of Basic and General Terms in Metrology. 2nd ed. Geneva, Switzerland: International Organization for Standardization; Ricos C, Juvany R, Simon M, et al. Commutability and traceability: their repercussions on analytical bias and inaccuracy. Clin Chim Acta. 1999;280: Doumas BT. The evolution and limitations of accuracy and precision standards. Clin Chim Acta. 1997;260: National Cholesterol Education Program. Summary of the second report of the NCEP Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel II). JAMA. 1993;269: National Cholesterol Education Program Laboratory Standardization Panel. Current status of blood cholesterol measurements in clinical laboratories in the United States. Clin Chem. 1988;34: National Cholesterol Education Program Laboratory Standardization Panel. Recommendations for Improving Cholesterol Measurement: A Report From the Laboratory Standardization Panel of the National Cholesterol Education Program. Bethesda, Md: National Institutes of Health; NIH Publication No Arch Pathol Lab Med Vol 129, March 2005 Analytic Bias of Thyroid Function Tests Steele et al 317