This was a cross-sectional study. Between October 2015 and January 2016, serum samples were collected from 156 inpatients, including 31 males and 125 females (median age: 40 years, range: 9–91 years), who requested TT4 measurement, at Peking Union Medical College Hospital, Beijing, China. The samples were from cases of hyperthyroidism (N=40), postoperative thyroid cancer (N=18), hypothyroidism (N=15), thyroid nodule (N=14), pregnant women (N=14), and others (N=55). The study was approved by the Ethics Committee of Peking Union Medical College Hospital (ZS-984). All patients involved were made aware of the intended use of their samples and provided written consent. Experiments were carried out in accordance with the Declaration of Helsinki (2013 revision).

Serum was collected in VACUETTE tubes with separator gel and clot activator (Greiner Bio-One, Kremsmunster, Austria). Each sample was divided into seven aliquots, which were stored at −80℃ until analysis within a month. A freshly thawed aliquot was used for each analytical run. ADVIA Centaur XP (Siemens, Munich, Germany) was employed to select samples that had TT4 concentrations evenly distributed between 1.3 and 387 nmol/L, and without hemolysis, icterus, or lipemia.

TT4 was measured using six automated chemiluminescent immunoassays—Architect 4000 (Abbott Diagnostics, Lake Forest, IL, USA), DXI800 (Beckman-Coulter, Brea, CA, USA), E601 (Roche Diagnostics, Basel, Switzerland), ADVIA Centaur XP (Siemens, Munich, Germany), Autolumo A2000 (Autobio, Zhengzhou, China), and CL-1000i (Mindray, Shenzhen, China)—as well as ID-LC-MS/MS. All immunoassays were carried out in the clinical laboratory of Peking Union Medical College Hospital in a blinded manner, with different experienced operators for each. ID-LC-MS/MS was carried out at the National Center for Clinical Laboratory in China, on an API 4000 triple quadrupole mass spectrometer (AB Sciex, Framingham, MA, USA) coupled with an Agilent 1200 LC system (Agilent Technologies, Santa Clara, CA, USA) and equipped with Analyst 1.4.2 software (AB Sciex) [13].

Serum samples or calibrators were sampled by weight, and ¹³C6-T4 internal standards were added volumetrically using automated diluters, followed by equilibration. Then, 1 mL of methanol (0.1% formic acid) was added to precipitate the protein, and the mixture was centrifuged at 3,000×g for 15 minutes. The upper layer was transferred to the cation exchange column. Cation exchange solid-phase extraction was performed using a mixed-mode cation exchange column (Oasis MCX 1 mL, 30 mg; Waters, Milford, MA, USA). The solid-phase extraction cartridge was conditioned by washing with methanol and water. Then, it was washed with 1 mL of a 2% aqueous solution of formic acid and 1 mL of methanol. Next, T4 and the internal standard were eluted from the cartridge with 1 mL of 5% ammonium hydroxide. The eluates were evaporated to dryness under nitrogen, and the residues were reconstituted in 400 µL of mobile phase, 20 µL of which was loaded into the ID-LC-MS/MS system. HPLC separation was carried out with a Zorbax Eclipse XDB C18 column (5 µm, 2.1×150 mm; Waters) and a mobile phase consisting of 0.05% formic acid in water-methanol (30:70) at a flow rate of 0.3 mL/min. MS detection was carried out using the positive electrospray ionization mode with multiple reaction monitoring. Quantitative ion pair transitions were m/z 777.7→731.6 and m/z 783.7→737.6 for T4 and ¹³C6-T4, respectively. Qualitative ion pair transitions at m/z 783.7→639.8 and m/z 783.7→639.8 for T4 and ¹³C6-T4, respectively, were monitored for confirmation. Each sample was injected three times, and average peak area ratios were calculated to quantify T4 concentration in the unknown sample by bracketing calibrations.

Certified reference materials for TT4 (CRM21201 and CRM20202) were provided by Professor Lothar Siekmann of the German Society of Clinical Chemistry and Laboratory Medicine (DGKL). The CRMs were lyophilized human serum samples and were dissolved in 3 mL deionized water. The CRMs were measured in three runs, and with triplicate measurements in each run.

Immunoassays were carried out with the following reagent lots for TT4: 55935UI00 (Abbott), 527911 (Beckman-Coulter), 188364 (Roche), 04003169 (Siemens), 20151121 (Autobio), and 150701 (Mindray). The calibrator lots were: 7K66-01 (Abbott), 2015101503 (Autobio), 33805 (Beckman-Coulter), 20150901 (Mindray), 18150202 (Roche), and 53803A86 (Siemens). Detailed information on the immunoassays is presented in Supplemental Data Table S1. TSH concentration was measured by the Centaur assay (Siemens) using the TSH reagent produced by Siemens.

In addition, three serum pools from Bio-Rad (Hercules, CA, USA; lot 40300, levels 40301, 40302, and 40303) that were used as quality control materials were prepared for assessing immunoassay imprecision prior to comparisons. Following the CLSI EP15-A [15], on five consecutive days, one freshly thawed aliquot of each pool was measured four times by all immunoassays. We ensured that measurement performance met the quality control standard before proceeding with measurements.

Performance criteria were set on the basis of biological variation, which is often used to evaluate whether an assay is analytically acceptable [16]. With this approach, using the within- and between-subject biological variation (CVw=within-subject biological variation, CVg=between-subject biological variation) for TT4 from Westgard [17] (4.9 and 10.9, respectively), the minimum requirements for TT4 assays were as follows: a mean bias of ≤4.5% [0.375 (CV_w²+CV_g²)^1/2] and imprecision of ≤3.7% (0.75CV_w) [18].

The TT4 results analyzed by different methods were summarized as mean±SD, and CV of the methods was calculated using a one-way ANOVA. In addition, TT4 results from different methods were analyzed using Passing-Bablok regression and Bland-Altman plots. Passing-Bablok regression, which was used to evaluate method agreement, calculates a regression equation (y=ax+b), including 95% confidence intervals (CIs), for the proportional (a, slope) and constant (b, intercept) errors [19]. If the 95% CI for intercept includes the value zero, there is no constant difference between two methods; if the 95% CI for slope includes the value one, there is no proportional difference between two methods [19]. The bias between two methods was evaluated using Bland-Altman plots, which quantify agreement between two quantitative measurements by constructing limits of agreement [20]. Linearity was tested using the cumulative sum (Cusum) linearity test, which is used to evaluate whether residuals are randomly scattered around the regression line without a significant trend. The correlation between TT4 and TSH was evaluated using linear regression. P<0.05 indicated a significant deviation from linearity. Statistical analyses were carried out using MedCalc v. 13.3.3 (MedCalc Software, Ostend, Belgium) and Microsoft Excel 2007 (Microsoft Corporation, Redmond, WA, USA).

We evaluated the imprecision of ID-LC-MS/MS by measuring left-over serum samples in three analytical runs, with triplicate measurements in each run. The samples were from seven patients with T4 concentrations ranging from 20.87 to 245.55 nmol/L. For each concentration, the within-run, between-run, and total CVs were 0.60% (0.35–0.89%), 0.54% (0.27–1.23%), and 0.84% (0.57–1.37%), respectively.

For CRM21201 and CRM20202, the mean TT4 concentration measured by ID-LC-MS/MS was 120.74 nmol/L and 88.70 nmol/L, and the bias against target values (120.90 nmol/L and 88.05 nmol/L) was −0.13% and 0.73%, respectively.

This method was used to participate in the 2011, 2012, 2014, and 2016 International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) reference measurement comparison (Ring trials or Rela) schemes. Our laboratory code was 18. In the four-time comparison, the average difference between the results of our method, and the results from the DGKL reference measurement laboratory (code 1) was <1%.

Three vials of 57.55 nmol/L frozen serum samples were combined, and twelve 0.2 mL aliquots were taken for testing the accuracy of the method. Unlabeled TT4 was added to nine of the 12 aliquots: three each with 29, 58, and 116 nmol/L TT4. No TT4 was added to the other three aliquots. The aliquots were then processed and tested by ID-LC-MS/MS. The amounts of TT4 added and recovered were in very good agreement for all three concentrations, with mean recoveries of 100.7% (100.1–101.0%), 99.6% (99.2–100.5%), and 99.9% (99.5–100.7%), respectively.

Imprecision evaluation showed that all methods had within-run CV and total CV for TT4<10% (Table 1). However, only Abbott and Roche showed acceptable CV (≤3.7%) in all samples, Autobio and Mindray produced relatively high CVs at the lower concentrations, and Beckman-Coulter assay results exceeded the criteria for all three concentrations (Table 1).

To compare the results of immunoassays and ID-LC-MS/MS, we used low, medium, and high TT4 concentrations that covered the analytical measurable range in each immunoassay. Passing-Bablok regression (Fig. 1) revealed that the results obtained by each immunoassay strongly correlated with those obtained by ID-LC-MS/MS, with correlation coefficients (R)>0.945. No proportional error (slope) was observed for Autobio, Beckman-Coulter, and Roche, and the 95% CIs for slope comprised. However, constant error was observed for most immunoassays, except Beckman-Coulter, with a 95% CI for intercept comprising zero. The highest constant bias was noted for Siemens, at −12.8 nmol/L (Fig. 1). Bland-Altman plots showed that the mean bias relative to ID-LC-MS/MS was the lowest for Roche (3.5%) and the highest for Abbott (−10.8%) (Fig. 2).

R was −0.371 (−0.516 to −0.206, P<0.001) between TT4 (ID-LC-MS/MS) and TSH (with 121 samples having a TSH concentration higher than the lower limit of 0.008 µIU/mL). The correlations between log TSH and TT4 measured by immunoassays are presented in Table 2. The correlation between log TSH and TT4 results from ID-LC-MS/MS was not stronger than that between log TSH and TT4 results from immunoassays.