CATHERINE FILLÉE, JEAN CUMPS, JEAN-MARIE KETELSLEGERS. Cliniques Universitaires St-Luc, Université Catholique de Louvain, Brussels, Belgium SUMMARY

Clin. Lab. 2012;58:725-736 Copyright ORIGINAL ARTICLE Comparison of Three Free T4 (FT4) and Free T3 (FT3) Immunoassays in Healthy Subjects and Patients with Thyroid Diseases and Severe Non-Thyroidal Illnesses CATHERINE FILLÉE, JEAN CUMPS, JEAN-MARIE KETELSLEGERS Cliniques Universitaires St-Luc, Université Catholique de Louvain, Brussels, Belgium SUMMARY Background: Free T4 (FT4) and free T3 (FT3) immunoassays exhibit wide inter-assay variations. We therefore established control values with three different immunoassays and compared their clinical performances in thyroidal diseases and severe/acute non-thyroidal illnesses (NTI). Methods: The UniCel DxI 800, Architect i2000, and Elecsys 2010 assays were used. FT4 and FT3 reference ranges were established in 68 controls, without conditions interfering with thyroid function, with normal TSH (0.35-3.02 mu/l) and negative anti-thyroid peroxidase antibodies. Free hormones were determined in 60 patients with thyroid diseases (TSH: <0.001-31.5 mu/l) and 45 NTI patients (TSH: 0.10-4.72 mu/l). Control values were normalized as Z-scores; patients results were expressed as Z-scores, using the control values of each assay. Pairwise comparisons of the Z-scores were performed by Deming s regression. Classification of patients results based on 95% control values were evaluated by kappa agreement statistics. Results: Control values for FT4 (pmol/l; geometrical means; 95% confidence intervals) were: 11.1 (7.6-16.1), 12.3 (9.1-16.6), 15.6 (11.4-21.4) and for FT3: 4.8 (4.0-5.7), 4.0 (3.0-5.3), 4.3 (3.1-5.09), with DxI 800, Architect, and Elecsys, respectively. Pairs of control Z-scores correlated significantly, but with different strengths (FT4: r = 0.915, 0.740, 0.770; FT3: r = 0.615; 0.589; 0.790, for DxI 800 vs. Elecsys; DxI 800 vs. Architect; Elecsys vs. Architect; p<0.001); slopes and intercepts of paired controls were 1.00 and zero. In thyroid diseases, slopes of FT4 Z-scores among assays differed slightly from 1.00 (1.11, 0.88, 0.87 for DxI versus Elecsys, DxI 800 versus Architect, Elecsys versus Architect, respectively; p<0.05); slopes of FT3 Z-scores were consistent with unity, except for DxI 800 versus Elecsys (0.88; p<0.05). In NTI patients, regression slopes were consistent with unity (p<0.05). The agreement statistics showed moderate to very good inter-assay concordances for thyroid and NTI patients results. Conclusions: FT4 and FT3 assays show moderate to very good agreement, in patients with thyroid diseases or NTI when compared pairwise to their control values. Slight quantitative differences between some pairs of assays are observed in thyroid diseases, after normalization as Z-scores. (Clin. Lab. 2012;58:725-736. DOI: 10.7754/Clin.Lab.2011.110706) KEY WORDS Free thyroxin; free triiodothyronine; immunoassay; thyroid; non-thyroidal illness LIST OF ABBREVIATIONS FT3 - free triiodothyronine FT4 - free thyroxin IRQ - interquartile range LC-MS/MS - liquid chromatography-tandem mass spectrometry NTI - non-thyroidal illness TPO - thyroid peroxidase TSH - thyrotropin TTR - transthyretin Manuscript accepted November 27, 2011 Clin. Lab. 7+8/2012 725

C. FILLÉE et al. INTRODUCTION Thyrotropin (TSH) is the first line assay for the routine evaluation of thyroid function status. However, free thyroxin (FT4) and free triiodothyronine (FT3) are essential biomarkers, together with thyrotropin (TSH) for final diagnosis and management of thyroid diseases [1]. The majority of third generation TSH assays show excellent performances [2]. The measurement of free thyroid hormones remains a complex analytical issue [3]. The various immunoassay systems for FT4 and FT3 show wide inter-assay variations [4-7]. In particular, the normal ranges among manufacturers exhibit important differences, a feature causing much confusion in the comparison of results among different laboratories. This lack of consensus in the reference values of FT4 and FT3 among assays may also shed some doubt on the clinical value of the results generated by different assays. Thus, it could be hypothesized that the FT4 or FT3 results categorized as normal, elevated or low in patients with thyroid diseases or severe/acute non-thyroidal illnesses (NTI) are at variance among assays. Such inter-assay comparisons, based upon the analysis of clinically defined groups of patients is at present poorly documented. To obviate the problems related to the inter-assay variations of free thyroid hormone immunoassays, the use of methods based upon the physiccal separation of the free hormones from the hormonebinding complexes and the subsequent measurement of T4 or T3 by liquid chromatography-tandem mass spectrometry (LC-MS/MS) have been advocated. However, these methods will require further standardization, in particular, in the preanalytical phases and significant effort is still necessary to implement them in automated formats suitable for large scale routine volumes [8-12]. Since routine laboratories will continue for some time to use the available automated immunoassays, we addressed the question of the comparison of three different FT4 and FT3 immunoassays, in a group of patients with known thyroid diseases or NTI, compared to the reference ranges of each assay. To this end, we first determined FT4 and FT3 with the three immunoassays, in a control group. In this group, we also tested if FT4 and FT3 would be influenced by age, in all assays. The results of the control group were expressed as Z-scores. FT4 and FT3 were also determined in a cohort of patients with thyroid diseases or NTI and the data expressed as Z-scores based on the reference control values of each assay. The patients results obtained in each assay were compared pairwise by regression analysis. Concordance among assays, in relation to the reference control values was also computed. MATERIALS AND METHODS Patients The patient characteristics are presented in Table 1. A group of 68 control patients (ambulatory or hospitalized, 35 to 95 years old) were selected on the basis that their medical record mentioned only benign conditions, not prone to alter the thyroid status; their TSH levels were normal (median: 1.24 mu/l; range: 0.35 3.02); all of them had undetectable anti thyroid peroxidase antibodies (atpo) levels. A group of 60 patients with established thyroid diseases (substituted primary hyperthyroidism, secondary hypothyroidism, Basedow diseases, euthyroid goiter) were included. For all of them, the diagnosis was based on clinical grounds, thyroid imaging, as well as several pre-treatment thyroid panels; these patients were either under early or long-term therapy and were either well controlled on the basis of TSH levels or were in the course of therapeutic adjustments; indeed, the TSH values in this subgroup ranged from undetectable to very high values, consistent with the wide range of thyroid dysfunctions encountered at the time of blood sampling. In the subgroup of 45 patients with NTI, the TSH values were normal (n = 38), low (n = 4; 0.09-0.28 µu/ml) or slightly elevated (n = 3; 4.10-4.72 µu/ml). Assays TSH and atpo were determined on the UniCel DxI 800 (Beckman Coulter); FT4 and FT3 were measured on the UniCel DxI 800, the Elecsys 2010 (Roche), and the Architect i2000 (Abbott). All assays were performed on the same day, all three instruments were maintained according to the manufacturers recommendations, and the whole procedure was directly supervised by the same biologist (CF). The procedure was in agreement with the internal rules of the hospital ethical committee. FT4 determination on both DxI 800 and Architect i2000 use a 2-step enzyme immunoassay, in which the FT4 of the sample reacts first with anti-t4 antibody coated microparticles; after washing, the antibodies remaining free on the solid phase are back-titrated by addition of labeled T3 and the signal is inversely related to the FT4 concentrations in the sample. The electrochemiluminescent Elecsys FT4 assay differs from the other two, as it is based upon a first incubation in which the FT4 of the sample reacts with ruthenium labeled anti-t4 antibodies; in a subsequent incubation, microparticle-bound T4 is added to the medium and binds to the remaining free labeled anti-t4 antibodies; the solid phase is then separated from the incubation medium and the amount of labeled anti-t4 antibodies captured by the T4 coated microparticles is inversely related to the FT4. The DxI 800 and Elecsys 2010 FT3 assays are based on the single labeled antibody technique. In contrast, the FT3 Architect i2000 assay makes use of the 2-step assay scheme. The assays imprecisions were in accordance with those stated in the data inserts (data not shown). Statistics Comparisons between FT4 and FT3 results obtained with the three assays from controls and all patients considered together were performed using the non-paramet- 726 Clin. Lab. 7+8/2012

COMPARISON OF FT4 AND FT3 IMMUNOASSAYS ric Passing-Bablok regression. Relations between age and FT4 and FT3 in the control group were computed by Pearson regression analyses after logarithmic transformation of the free hormone concentrations. Confidence limits of the FT4 and FT3 control values were calculated after logarithmic transformation of the data and were expressed as geometrical means with 95% confidence intervals; the 90% confidence of the upper and lower limits were also computed. The significance of differences between control values for FT4 and FT3 among the three assays was computed using paired t- test (after Bonferoni correction for multiple comparisons). For each assay, Z-scores of the 68 controls group samples were calculated on the basis of the means of the logarithmic transformed data and the standard deviations [(log of each control value - mean of the log of the control group)/sd of log of the control values)]. The Shapiro-Wilk test showed that the distribution pattern of the control Z-scores were consistent with a normal distribution (p>0.05) with the exception of the Architect i2000 FT3 for which the normality test was very close to normality (p = 0.041). The correlations among the control Z-scores were determined, as well as the 95% confidence ellipses for each pair of assays. Deming regression, assuming equal variances, showed a slope of 1.00 and intercept of zero for all pairs of FT4 and FT3 control Z-scores. FT4 and FT3 Z-scores for the groups with thyroid illnesses and NTI were also determined (using the corresponding control means and SD for each assay). Deming regression was also used to determine the slopes and intercepts, with their 95% confidence intervals, for each pair of FT4 and FT3 Z-scores among all three assays, in the patients with thyroid diseases and NTI. The observation of slopes different from unity would suggest differences of immunoassay behaviors for samples from diseased patients. We also computed the between-assay agreements (kappa statistic) for FT4 and FT3 in thyroid diseases and acute NTI after classification of their Z-scores, using the 2.5 SD limits of the control data of each assay. The strength of the agreement was evaluated following the guidelines: <0.20: poor; 0.21-0.40: fair; 0.41-0.60: moderate; 0.61-0.80: good; 0.81-1.00: very good [13]. Statistical analysis was performed with JMP, version 8.0.1, from SAS Institute, Analyse-it for Excel, version 2.05 and MedCalc for Windows, version 10.4.8.0 (MedCalc Software, Mariakerke, Belgium). RESULTS Relations between FT4 values obtained with the three assays including all subjects (n = 173) are depicted in Figure 1 (panels A - C). The Elecsys FT4 values were systematically higher than those generated with DxI 800 or the Architect. In contrast, FT4 levels determined with the DxI 800 compared to the Architect were closer to each other. For FT3 (Figure 1; panels D - F), the larger between-assay differences were observed when Elecsys results were compared to those of the DxI 800. One patient included in the analysis had extremely high FT3 values (15.2, 18.2, and 18.9 pmol/l, with the DxI 800, Architect, and Elecsys assays, respectively). The corresponding FT4 values were also very high (42.60, 35.91, and 53.02 pmol/l), and TSH levels were undetectable (<0.01 µu/ml). These results were from a 49 year old female who at the time of blood sampling presented clinically overt hyperthyroidism, due to acute recurrence of Basedow disease, caused by the fact that the patient had interrupted her anti thyroid drug therapy. FT4 levels in the control group were positively related to age with a high degree of significance, a 20 30% increase being observed over the age span of 20 90 years. This relation was observed with all three assays (r = 0.473, 0.545, and 0.396; DxI 800, Architect, and Elecsys assays, respectively; p<0.001). In contrast, FT3 and age showed no significant correlation. Table 2 represents estimations of normal intervals for FT4 and FT3 assays derived from the control group. Mean FT4 values obtained with the DxI 800 were 10% and 29% lower than those measured with the Architect i2000 or the Elecsys 2010, respectively. The mean FT3 values with the DxI 800 were 20% and 12% higher than those measured with the Architect i2000 or the Elecsys 2010, respectively. These differences between groups were highly significant (p<0.001). For comparison, the estimated normal reference ranges stated by the manufacturers in the data inserts are summarized in Table 3. Figure 2 shows the distribution of the Z-scores for FT4 and FT3 in the control group, as well as their correlations among the three assays; the density ellipse, a surface containing 95% of all observations is also represented. The width of the density ellipse was related to the correlation coefficients between two variables. The coefficients of correlation among the different pairs of assays showed great variability. The strongest correlation of Z-scores was observed for the relation between the FT4 levels obtained with the Elecsys 2010 and the DxI 800 (despite the fact that the absolute values are very different). For FT3, the best correlation of Z-scores was obtained when the Elecsys 2010 was related to the Architect i2000. Deming regressions of the Z-scores between all FT4 and FT3 assays from controls had slopes of 1.00 (95% confidence intervals as on the Figure), with an intercept of zero. In the 60 patients with thyroid diseases, the FT4 Z- scores were very similar in the three assays, as they ranged from -4.6 to 7.7 (interquartile: 0.0-2.2), -4.2 to 7.0 (interquartile: 0.2-2.1) and -6.5 to 7.6 (interquartile: 0.3-2.3), with the DxI 800, the Architect i2000, and the Elecsys 2010, respectively. For FT3, these values were -5.1 to 12.3 (interquartile: -1.2-0.6), -6.6 to 10.8 (interquartile: -1.3-0.6), and -6.1 to 8.9 (interquartile: -0.9-0.7) (Figure 3). Intercepts of the Deming regressions were not significantly different from zero for any of the pairs of FT4 and FT3 assays. The slope of FT4 - DxI 800 versus FT4 - Elecsys 2010 Z-scores was slightly above unity (p<0.05); in contrast, the slopes of Clin. Lab. 7+8/2012 727

C. FILLÉE et al. Table 1. Patients characteristics. Pathology N Age a (years) Gender ratio (M/F) TSH (mu/l) a Controls Patients (outpatients or hospitalized) under stable conditions with pathologies unrelated to thyroid disorders, no severe clinical conditions and undetectable atpo 68 50 (16-94) [35-66] b 29/39 1.24 (0.35-3.02) Basedow disease at various states of treatment by surgery, radioiodone or antithyroid agents 19 54 (31-84) 6/13 1.10 (<0.01-31.5) Thyroidectomy for papillary / follicular cancer (n = 11) or multinodular goiter (n = 2) 13 42 (19-67) 3/10 0.32 (0.02-26.5) Thyroidal illnesses Primary hypothyroidism,t4 substituted 19 46 (2-76) 2/16 1.42 (<0.01-13.1) Secondary hypothyroidism,t4 substituted 3 25; 54; 76 c 2/1 0.02; 0.02; 0.21 Euthyroid goiter 6 48 (35-65) 1/6 0.65 (0.35-2.95) Acute cardiovascular condition / Arrhythmia 20 65 (38-81) 18/3 0.98 (0.10-4.72) Severe / acute nonthyroidal clinical conditions Inflammatory / infectious states 12 64 (38-96) 9/3 1.46 (0.09-4.21) Malignancies under treatment 13 45 (17-72) 5/8 1.29 (0.20-2.70) a median (range) b [interquartile range], values determined with DxI 800 c individual values Table 2. FT4 and FT3 values in 68 control subjects using DxI 800, Architect, and Elecsys. FT4 (pmol/l) FT3 (pmol/l) DxI 800 Architect Elecsys DxI 800 Architect Elecsys Geometrical means 11.1 12.3 15.6 4.75 4.01 4.26 95% Reference limits 7.62-16.1 9.10-16.6 11.4-21.4 3.97-5.74 3.05-5.27 3.07-5.92 90% Confidence interval of lower limits 7.14-8.13 8.63-9.59 10.8-12.1 3.85-4.10 2.91-3.20 2.90-3.25 90% Confidence interval of upper limits 15.1-17.2 15.8-17.5 20.3-22.6 5.56-5.93 5.03-5.53 5.59-6.26 728 Clin. Lab. 7+8/2012

COMPARISON OF FT4 AND FT3 IMMUNOASSAYS Table 3. Expected normal values for FT4 and FT3 in adults as stated in the manufacturer s data insert. Manufacturer N FT4 (pmol/l) N FT3 (pmol/l) Beckman Coulter UniCel DxI 800 Abbott Architect i2000 Roche Elecsys 2010 316 7.9-14.4 a b 200 3.8-6.0 e 411 9.0-19.0 c 436 2.6-5.7 f 801 12.0-22.0 d 5,366 3.1-6.8 g --- --- 870 3.9-6.7 h a 95% confidence interval; USA population (North, Central, West); 18-60 years old; TSH 0.3-3.0 mu/l b FT4 values reported as 7.7 20.6 pmol/l until December 2006, data insert of Access 2 c 99% central interval d 2.5-97.5 % non parametric percentile e 95th percentile (median: 4.9 pmol/l) f central 95% interval g population from a private laboratory, near the German coast; TSH: 1.0-3.0 mu/l; non-parametric 95% confidence limits h samples from apparently healthy blood donors; population from Central Germany; non-parametric 95% confidence limits Table 4. Agreement statistics for the classification of Z-scores obtained with the different FT4 and FT3 assays, in thyroid diseases and acute/critical illness conditions a, b. Thyroid diseases (n = 60) FT4 FT3 DxI 800 vs. Elecsys 0.754 (0.104) 0.844 (0.108) DxI 800 vs. Architect 0.799 (0.096) 0.640 (0.164) Elecsys vs Architect 0.529 (0.138) 0.687 (0.147) Acute/Critical illness (n = 45) FT4 FT3 DxI 800 vs. Elecsys 0.789 (0.204) 0.487 (0.195) DxI 800 vs. Architect 0.844 (0.151) 0.407 (0.164) Elecsys vs. Architect 0.645 (0.229) 0.607 (0.152) a Kappa (standard error) b Classifications are based upon the ±2.5 SD limits of the control group in each assay, as represented in Figures 4 and 5 the Deming regressions relating FT4 DxI 800 versus FT4 - Abbott and FT4 Elecsys 2010 versus FT4 Architect Z-scores were slightly below unity (p<0.05). Comparisons of the FT3 Z-scores showed a slope slightly below unity for the DxI 800 versus the Elecsys 2010 assays (p<0.05), while the slopes were within the limits of unity for DxI 800 versus Architect and Elecsys versus Architect. In 45 patients with acute NTI, the range of pathological FT4 and FT3 Z-scores was narrower than in the thyroid diseases (Figure 4). Moreover, FT4 Z-scores were systematically and significantly (p<0.001) higher than the Clin. Lab. 7+8/2012 729

C. FILLÉE et al. Figure 1. Passing-Bablok regressions of FT4 (left panel) and FT3 (right panel), using data from controls, thyroid diseases, and non thyroidal severe illnesses obtained with the DxI 800, Architect i2000, and Elecsys 2010, compared pairwise. The actual values of one patient with extremely high FT3 values are given in parentheses. 730 Clin. Lab. 7+8/2012

COMPARISON OF FT4 AND FT3 IMMUNOASSAYS Figure 2. Correlations between the Z-scores of FT4 (left panel) and FT3 (right panel) of the control group, determined with the DxI 800, Architect i2000, and Elecsys 2010, compared pairwise. The red lines represent the 95% confidence ellipses of the regression. The distributions of the Z-scores for each assay are represented at the upper and right side of the regression graphs. The dotted lines represent the Deming regressions; the intercepts are zero and the slopes are 1.00; the 95% confidence limits of the slopes are represented on the Figures. Clin. Lab. 7+8/2012 731

C. FILLÉE et al. Figure 3. Z-scores of FT4 (left panel) and FT3 (right panel) of patients with thyroid diseases determined with the DxI 800, Architect i2000, and Elecsys 2010, compared pairwise. The red lines represent the 95% confidence ellipses of the control groups. The dotted lines represent the Deming regressions; the slopes and intercepts with their 95% confidence intervals are represented on the figures. O: Basedow disease; Δ: hypothyroidism; : euthyroid goiter (see Table 1). 732 Clin. Lab. 7+8/2012

COMPARISON OF FT4 AND FT3 IMMUNOASSAYS Figure 4. Z-scores of FT4 (left panel) and FT3 (right panel) of patients with severe non thyroidal illnesses determined with the DxI 800, Architect i2000, and Elecsys 2010, compared pairwise. The red lines represent the 95% confidence ellipses of the control groups. The dotted lines represent the Deming regressions; the slopes and intercepts with their 95% confidence intervals are represented on the figures. О: inflammatory/infection; Δ: malignancies; ; acute cardiovascular condition (see Table 1). Clin. Lab. 7+8/2012 733

C. FILLÉE et al. FT3 Z-scores with all three assays. The intercepts and slopes of the Deming regressions were not significantly different from zero and unity, respectively. Table 4 shows the between assay agreements after classification of FT4 and FT3 Z-scores according to ±2.5 SD of their respective controls. For FT4 Z-scores, agreements in cases of thyroid diseases and acute NTI, the kappa statistic showed good agreement, except for Elecsys 2010 versus Architect i2000, which was moderate. For FT3 Z-scores, agreements were moderate to good in cases of thyroid diseases; the lowest kappa statistics (still moderate) were observed for FT3 Z-scores in acute NTI. DISCUSSION In this report, we have compared the FT4 and FT3 values obtained with three automated systems, DxI 800, Architect i2000, and Elecsys 2010, in a control group as well as in patients with various stages of thyroid diseases and patients with severe/acute non thyroidal illnesses (NTI). Significant inter-assay biases were observed. To allow inter-assay comparison, we normalized the control values on the basis of Z-scores and expressed the FT4 and FT3 from the patient groups as Z- scores, using the corresponding control references. All assay Z-score pairs were compared. There was considerable variation among the correlation coefficients indicating differences in random error among assays. In the control group, slopes and intercepts of the regressions of paired assays did not differ from 1.00 and zero; these results are in agreement with the validity of Z-score normalization in these cases. In the patients with thyroid disease, comparison of the slopes of paired FT4 and FT3 Z-scores revealed modest or no significant departures from unity suggesting some differences in assay behaviors in these cases. In NTI patients, regression analysis did not reveal such between-assay differences. Between-assay classification agreements in all cases were rated from moderate to very good. FT4 values generated with the Elecsys are considerably higher than those reported by DxI 800 or Architect i2000. Results of the DxI 800 and the Architect i2000 were closer to each other than those of the Elecsys 2010. The important divergence of the Elecsys FT4 from DxI and Architect i2000 may possibly result from assay design differences. Indeed, the Elecsys assay is based on a one step labeled anti-t4 antiserum, while the two others make use of the 2-step procedure FT4 extraction on solid phase anti-t4 antibodies followed by back-titration of the remaining free antibodies with labeled T3 analogue. Differences in the composition of the first buffers may also be of importance. Antibody affinities and their interactions with binding proteins (thyroxin binding protein (TBG) and transthyretin (TTR)) may also play a role in the inter-assay differences. FT3 values also exhibited significant differences among assays, the Architect i2000 and the Elecsys showing the least differences. Inter-assay variations observed in this study for free thyroid hormones are in line with many previous reports [7,14-15]. The FT4 limits derived from our control group obtained by DxI 800 (7.6-16.1 pmol/l) are not in line with the manufacturer s data insert (7.9-14.4 pmol/l) nor with the report of d Herbomez et al. (7.8-14.3 pmol/l) [16]. Indeed, we observed an upper control limit ~13% higher. This may be due to an increase of FT4 with high age, together with the fact that our control group consisted of subjects from 20-90 years old, while the cohorts used by the DxI 800 manufacturer and d Herbomez et al. were younger (20-60 years and 18-65 years, respectively). The increase of FT4 with age was clearly evidenced in our control group, not only with the DxI 800 assay but also with the Architect i2000 and the Elecsys. The age-dependence of FT4 observed with all three immunoassays in this study may indirectly and partly be related to changes in the TBG and TTR, especially after 50 years. Our observation of a positive relation between age and FT4 levels determined by immunoassays is not unique and has been documented in other studies [17-20]. Thorough evaluations based on well defined populations are thus required to assess references applicable to specific clinical settings. FT4 values of our control group determined with the Architect i2000 or the Elecsys 2010 were close to the expected values of the data inserts. This was also observed by others with both assays [21,22], with the exception of one report with upper values with the Architect i2000 exceeding by ~40% the expected value [23]. FT3 values of our control group were also similar to the expected values described in the data inserts, as well as those from several other reports [15,21-23]. The recent report of the IFCC working group for standardization of thyroid tests (free thyroxin and free triodothyronine) [7] did highlight the important biases of most immunoassays for these free hormones. The report also shows that these biases may be largely corrected by an international conventional reference measurement procedure. This would not solve all problems, as the report also points to the fact that precision, within-run stability, and between-run consistency should be improved for some assays. Important information from this report is that mathematical recalibration of most of the results of the assays, based upon their relation to a reference LC-MS/MS method eliminates the biases. The authors, however, warn that after such recalibration, large sample-related effects remained. In the present report, we normalized the data of the three assays by expressing the results of control groups in terms of Z-sores. It appeared that all Z-scores describing the control values were, as expected, centered on a null value and that their distribution did not deviate from normality, except for the FT3 Architect i2000 for which the data were lightly skewed (Shapiro-Wilk = 0.041). The correlation coefficients of these Z-scores across assays were highly significant, even though the strength of these relations varied considerably, indicat- 734 Clin. Lab. 7+8/2012

COMPARISON OF FT4 AND FT3 IMMUNOASSAYS ing important variations in inter-assay scatter. For FT4, the strongest coefficient of correlation was observed between the DxI 800 and the Elecsys values. For FT3, there was less disparity, but the best relation was between the Architect i2000 and the Elecsys. These differences in the strength of correlation between Z-scores of the controls were visually reflected by the width of their 95% confidence ellipses. These observations are in line with the conclusions of the IFCC report, concerning the relative variations among assays as well as possible differences in precision. The Deming regression lines, assuming equal variance of both axes for each pair of FT4 and FT3 control Z-scores, were, in all cases, equal to unity with an intercept of zero, providing indirect evidence for the validity of the Z-score transformation. Of the 60 patients with thyroid diseases, most of the data were confined within the normality ellipses for all three FT4 and FT3 assays and consistent with therapeutic action leading to normalization or near normalization of thyroid function in this group. The slopes of the Deming regressions comparing FT4 Z-scores of assay pairs in these patients showed a slight but significant departure from unity, suggesting that quantitative extrapolation from one of these assays to the other must be taken with some caution. For the FT3 Z-score values in the thyroid cases, the regression slope did not depart from unity except for the relation of the DxI 800 compared to Elecsys 2010. For the group of patients with NTI, the mean FT3 Z- scores were lower than those for FT4. This is in accordance with alterations in peripheral conversion of FT4 to FT3. It is possible that the alternative pathway (conversion of T4 to inactive rt3) is activated, resulting in the production of lower amounts of bioactive T3. This was particularly marked for the patients with acute infections. Intercepts of the regression lines were not different from zero and the slopes of the regression lines were consistent with unity suggesting that differences in the assay behavior in this group of patients was not statistically rejected. We also used the Kappa statistics to evaluate the between-assay agreements of the data expressed as Z- scores. They were ranked as moderate to very good when the Z-scores for the patients with different pathological conditions were classified on the basis of the ±2.5 SD limits of the controls. These results indicate that all three assays are of clinical value, provided that appropriate reference values are used. It remains that for some pairs of assays, the concordance is only moderate, and caution should be exercised in extrapolating the clinical classification from one assay to another. This study has several limitations. First, the control group is relatively small. It represents patients attending a hospital physician, even though they were proven to be free of any conditions which could possibly interfere with the thyroid function, that their TSH values were normal, and atpo was undetectable. However, this is the kind of population, in which decisions have to be made on the presence or absence of significant thyroid dysfunction in a hospital based population, a situation quite different from general practice. Again, as already evidenced by several studies, locally normal values should always be checked by each laboratory, and data insert expected values can only be used for general guidance. The problem of a multicenter approach for derivation of reference intervals for thyroid hormones for laboratories, using identical analyzers has already been discussed by Andrew et al. [24]. All our control subjects had undetectable anti-tpo antibodies, measured with a highly sensitive method, but the anti-tg antibodies could not be measured due to practical constraints in all of them. However, a recent report concludes that anti-tg antibody alone, in the absence of anti-tpo is not significantly associated with thyroid diseases [25]. In conclusion, we confirmed the wide disparity in FT4 and FT3 concentrations derived from three immunoassays. Normalization of the results by transformation as Z-scores allows the elimination of between-assay differences. However, caution must be exercised when precise quantitative inter-assay comparisons are performed, as regression analysis comparing assays show that not all behave similarly, particularly in a group of patients with thyroid diseases. Classification of the results in thyroid diseases and NTI among pairs of assays showed moderate to very good agreement on the basis of their respective control Z-scores. It remains that any clinically or biologically questionable results should be referred to a reference laboratory, using physical separation of free hormones followed by their measurement by LC- MS/MS. In the future, it will be desirable to use international conventional reference measurands, or better, to implement routinely automated and standardized LC- MS/MS methods. Disclaimers: None. Declaration of Interest: None. Financial Support: Department of Clinical Biology and Pathology - Cliniques Universitaires St-Luc, Université Catholique de Louvain, Brussels, Belgium. References: 1. Baloch Z, Carayon P, Conte-Devolx B, Demers LM, Feldt-Rasmussen U, Henry JF, et al. Laboratory medicine practice guidelines. Laboratory support for the diagnosis and monitoring of thyroid disease. Thyroid 2003;13:3-126. 2. Thienpont LM, Van UK, Beastall G, et al. Report of the IFCC Working Group for Standardization of Thyroid Function Tests; part 1: thyroid-stimulating hormone. Clin Chem 2010;56:902-11. Clin. Lab. 7+8/2012 735

C. FILLÉE et al. 3. Ekins R. The free hormone hypothesis and measurement of free hormones. Clin Chem 1992;38:1289-93. 4. Nelson JC, Wilcox RB. Analytical performance of free and total thyroxine assays. Clin Chem 1996;42:146-54. 5. Christofides ND, Wilkinson E, Stoddart M, Ray DC, Beckett GJ. Assessment of serum thyroxine binding capacity-dependent biases in free thyroxine assays. Clin Chem 1999;45:520-5. 6. Stockigt JR. Free thyroid hormone measurement. A critical appraisal. Endocrinol Metab Clin North Am 2001;30:265-89. 7. Thienpont LM, Van UK, Beastall G, et al. Report of the IFCC Working Group for Standardization of Thyroid Function Tests; part 2: free thyroxine and free triiodothyronine. Clin Chem 2010; 56:912-20. 8. Jonklaas J, Kahric-Janicic N, Soldin OP, Soldin SJ. Correlations of free thyroid hormones measured by tandem mass spectrometry and immunoassay with thyroid-stimulating hormone across 4 patient populations. Clin Chem 2009;55:1380-8. 9. Holm SS, Hansen SH, Faber J, Staun-Olsen P. Reference methods for the measurement of free thyroid hormones in blood: evaluation of potential reference methods for free thyroxine. Clin Biochem 2004;37:85-93. 10. Van Uytfanghe K, Stockl D, Ross HA, Thienpont LM. Use of frozen sera for FT4 standardization: investigation by equilibrium dialysis combined with isotope dilution-mass spectrometry and immunoassay. Clin Chem 2006;52:1817-21. 11. Thienpont LM, Beastall G, Christofides ND, et al. Proposal of a candidate international conventional reference measurement procedure for free thyroxine in serum. Clin Chem Lab Med 2007;45: 934-6. 12. Yue B, Rockwood AL, Sandrock T, La'ulu SL, Kushnir MM, Meikle AW. Free thyroid hormones in serum by direct equilibrium dialysis and online solid-phase extraction-liquid chromatography/tandem mass spectrometry. Clin Chem 2008;54:642-51. 13. Altman DG. Practical Statisitcs for Medical Research. Chapman & Hall/CRC, 1991. 14. Steele BW, Wang E, Klee GG, et al. Analytic bias of thyroid function tests: analysis of a College of American Pathologists fresh frozen serum pool by 3900 clinical laboratories. Arch Pathol Lab Med 2005;129:310-7. 15. Sapin R, D'Herbomez M. Free thyroxine measured by equilibrium dialysis and nine immunoassays in sera with various serum thyroxine-binding capacities. Clin Chem 2003;49:1531-5. 16. D'Herbomez M, Jarrige V, Darte C. Reference intervals for serum thyrotropin (TSH) and free thyroxine (FT4) in adults using the Access Immunoassay System. Clin Chem Lab Med 2005;43:102-5. 17. Hoogendoorn EH, Hermus AR, de VF, et al. Thyroid function and prevalence of anti-thyroperoxidase antibodies in a population with borderline sufficient iodine intake: influences of age and sex. Clin Chem 2006;52:104-11. 18. Ross HA, den HM, Hermus AR, Sweep FC. Composite reference interval for thyroid-stimulating hormone and free thyroxine, comparison with common cutoff values, and reconsideration of subclinical thyroid disease. Clin Chem 2009;55:2019-25. 19. Davey R. Thyroxine, thyrotropin, and age in a euthyroid hospital patient population. Clin Chem 1997;43:2143-8. 20. Hershman JM, Pekary AE, Berg L, Solomon DH, Sawin CT. Serum thyrotropin and thyroid hormone levels in elderly and middle-aged euthyroid persons. J Am Geriatr Soc 1993;41:823-8. 21. Gonzalez-Sagrado M, Martin-Gil FJ. Population-specific reference values for thyroid hormones on the Abbott ARCHITECT i2000 analyzer. Clin Chem Lab Med 2004;42:540-2. 22. Kratzsch J, Fiedler GM, Leichtle A, et al. New reference intervals for thyrotropin and thyroid hormones based on National Academy of Clinical Biochemistry criteria and regular ultrasonography of the thyroid. Clin Chem 2005;51:1480-6. 23. Hubl W, Schmieder J, Gladrow E, Demant T. Reference intervals for thyroid hormones on the Architect analyser. Clin Chem Lab Med 2002;40:165-6. 24. Andrew CE, Hanning I, McBain AM, Moody D, Price A. A model for a multicentre approach to the derivation of reference intervals for thyroid hormones and testosterone for laboratories using identical analysers. Clin Chem Lab Med 2000;38:1013-9. 25. Hollowell JG, Staehling NW, Flanders WD, Hannon WH, Gunter EW, Spencer CA, Braverman LE, et al. Serum TSH, T(4), and thyroid antibodies in the United States population (1988 to 1994): National Health and Nutrition Examination Survey (NHANES III). J Clin Endocrinol Metab 2002;87:489-99. Correspondence: Catherine Fillée Cliniques Universitaires St-Luc Université Catholique de Louvain Avenue Hippocrate, 10 1200 Brussels, Belgium Tel.: +32 764 6716 Fax: +32 764 6905 Email: catherine.fillee@uclouvain.be 736 Clin. Lab. 7+8/2012