Forty individual-donor sera (PS), three serum pools (Pool1–3), and one sample from a pregnant woman (PN) from the third trimester used in this study were collected using protocols described in the supplemental materials. ^10,11 Use of blood by CDC is consistent with the IRB approval and donor consent. No personal identifiers were provided to CDC. The blood samples used in this project were from commercial sources. This activity was reviewed by CDC, deemed research not involving human subjects, and was conducted consistent with applicable federal law and CDC policy.

Concentrations for the PS were 11.3–32.1 pmol/L (0.881–2.49 ng/dL) for FT4 and 0.337–21.5 mIU/L for TSH. Pool1–3 concentrations were 15.0, 16.5, and 16.9 pmol/L FT4 (1.17, 1.28, and 1.31 ng/dL) and 1.09, 2.26, and 1.65 mIU/L (TSH). The FT4 concentration of PN was 11.4 pmol/L (0.885 ng/dL), but due to volume limitations TSH was not measured.

Procedures for material shipment, storage, and analysis by participants are described in the supplementary materials. The FT4 reference values were assigned to all samples using the CDC FT4 reference measurement procedure (RMP) described previously, and details can be found in the supplementary materials.¹²

A total of twenty-one FT4 assays, four LDTs based on equilibrium dialysis (ED) liquid chromatography tandem mass spectrometry and seventeen IAs were included in the study. Among the FT4 IAs, eight (comprising fifteen different platforms) were operated by the independent assay manufacturer and two by clinical laboratories. Eight IA manufacturers (comprising sixteen different platforms) and one clinical laboratory measured TSH. Several independent manufacturers were represented by more than one platform. Further information about the assays used in the study is summarized in Supplemental Table S2.

The study design was based on the Clinical and Laboratory Standards Institute (CLSI) document EP-09-A2.¹³ Agreement among LDT and IA manufacturers was assessed using the 40 PS. The concentration of one PS was outside the reportable range for most assays and therefore excluded from TSH analysis. Percent bias of sample replicate means to either the FT4 reference value or the all-lab mean (ALM) for TSH, median bias, 95% confidence interval of the median bias, and the percent coefficient of variation (CV) for replicate measurements were calculated for each participant. Outliers detected among replicate FT4 or TSH data were removed using the guidelines described in CLSI document EP9-A2.¹³ Suspected transcription errors (with biases to the all lab mean of 585–609%) for replicate TSH measurements of the same study sample by participant “O” were removed prior to analysis. Analysis of Pool1-3 or PN sera assessing the impact of serum pooling and pregnancy on FT4 measurement was performed independently of the analysis of the forty PS.

Participating assay manufacturers (LDT and IA) and clinical laboratories were compared to the CDC FT4 RMP (or the TSH ALM) using Passing-Bablok regression, chosen to account for the measurement error in the participant and CDC RMP data without making assumptions about the distribution of residuals. The mean biases of replicate measurements of the forty PS samples among participants were compared to the criteria for acceptable performance, the greater of 0.3 ng/dL or ± 15% for FT4 and 0.2 mIU/L or ± 20 % for TSH, set by the Centers for Medicare & Medicaid Services (CMS) to assess method performance pre- and post-recalibration.¹⁴ Methods for estimating the impact of pregnancy, sample pooling, and T4 supplementation on measurement accuracy are described in the supplemental materials. Data were analyzed using Microsoft Excel^® with the Analyse-it^® (version 5.90) add-in and the R statistical environment in R Studio^® (version 4.1.2).

Linear regression-based assay recalibration was performed in-silico using participant PS results and the reference value (or TSH ALM) for that sample. Pooled and PN samples were not included in linear regression models. In-silico recalibration was conducted by subtracting the intercept from the original submitted data and dividing by the slope. Classification of serum samples was performed post-recalibration by comparing the recalibrated results to a unified RI, which was calculated according to the principles of transference discussed in CLSI document EP28-A3c.¹⁵ Prospective transferred reference intervals (RI) were chosen from the RIs of participants (Figure 1) with linear regression slopes closest to 1 and intercepts closest to 0.

Based on the data as received from study participants, the all-sample median percent bias and 95% confidence interval (95% CI) to the CDC RMP observed for the forty PS among all participants was −17.1% (−18.1 – −15.7%). LDTs had better accuracy, with median bias (95% CI) of −4.5% (−6.1–−3.2), compared to −20.3% (−21.5–−19.4) for IAs (Figure 2 and Table 1). Pre-recalibration, three of the twenty-one FT4 assays met the CMS criteria for all PS, with five to forty of the forty PS meeting the bias requirement among FT4 assays (Table 3). The mean percent of samples meeting the CMS requirements (± 95% CI) for individual FT4 measurements was higher (99.0 ± 1.8%) among FT4 LDTs than IAs (58.9 ± 12.9%), which improved to 99.5 ± 1.6% for FT4 LDTs and 97.0 ± 1.9% for FT4 IAs post-recalibration. The bias distributions were narrow with an average interquartile range (IQR) across all participants of 9.0% (Figure 2) for FT4. The FT4 bias becomes more negative with increasing FT4 concentrations for IA while LDT bias remains constant (Figure S2). Most assays were well correlated with the CDC RMP with correlation coefficients >0.8 (Figure S1). The Passing-Bablok regression parameters among LDTs indicate well-calibrated assays with slope and intercept 95% CIs encompassing 1 and 0, respectively. Only one of the seventeen Passing-Bablok regression parameters among the IAs met the same criteria for slope and intercept (Table 1 and Figure S3). Mean precision (range of mean CV) was 3.8% (1.7–8.0) for all FT4 assays pre-recalibration and 4.3% (1.9–10.3) post-recalibration (Table 1). Comparing the FT4 replicate mean biases among the four PS from donors taking levothyroxine to the FT4 mean biases of the remaining 36 PS by two sample t-test suggest a significant difference in biases between these two groups for IAs (p-value < 0.05) but not for LDTs (p-value > 0.05) at the 5% significance level. No significant difference between the two groups was observed among TSH participants (p-values > 0.05).

The seventeen participating TSH IAs were in better agreement compared to FT4 assays with a median percent bias (Figure 3 and Table 2) to the ALM of −1.2% (95% CI: −1.8–−0.4%). The sample replicate mean bias distribution among all TSH assays showed a narrow 5.4% mean IQR. Of the seventeen TSH participants, ten met the CMS measurement criteria for all PS, with thirty-one to thirty-nine of the thirty-nine PS meeting the bias requirement among TSH assays (Table 3). Results of Passing-Bablok regression analysis indicate five of the seventeen TSH assays are well-harmonized to the ALM, with slope and intercept 95% CIs including 1 and 0, respectively (Table 2 and Figure S6). Mean precision (range of mean CV) was 2.6% (1.0–5.6) for all TSH assays pre-recalibration and 2.2% (0.9–5.4) after recalibration (Table 2).

Agreement between manufacturer and end-user results varied. For IA manufacturer “P”, Passing-Bablok regression analysis of FT4 data indicated no significant difference between manufacturer “P” and end-user (“G”) with 95% CIs of slope and intercept including 1 and 0, respectively (Figure 4). This contrasts with the comparison of manufacturer “P” to the clinical lab (“G”) for TSH measurement (Figure 4) and the comparison of FT4 manufacturer “C” to clinical lab “U” (Figure 4), where the 95% CI of the slope parameter is > 1. Use of the same lots of calibrators and reagents were confirmed for FT4 measurement between assays “P” and “G”, however, different calibrator and reagent lots were used in TSH assays “P” and “G.” Reagent and calibrator lot numbers could not be determined for FT4 assays “U” and “C”.

After recalibration of the FT4 assays, classification was repeated using the recalibrated PS sera results and the RI transferred from participant “H” (9.75–25.1 pmol/L or 0.758–1.95 ng/dL). Classification of TSH results post-recalibration was performed in a similar way, by first performing transference testing using the ALM for comparison, and then re-classifying assay results using the transferred RI from TSH assay “M” (0.606–4.32 mIU/L).

Agreement among FT4 assays on sample classification was poor before recalibration; twenty-one of the forty PS were classified uniformly by all assays (Table 4). FT4 classification agreement improved upon recalibration to thirty-three of the forty PS uniformly classified by all assays. TSH classification agreement was consistent pre-recalibration (thirty-three of thirty-nine PS) and post-recalibration (thirty-two of thirty-nine PS) (Table 5). Post-recalibration FT4 participant median percent biases (95% CI) improved to −0.2% (−1.1 – 0.5) overall, −0.2% (−1. 2 – 0.6) for IA, and −0.3% (−2.5 – 1.4) for LDT (Figure 2). Median percent bias (95% CI) among TSH assays improved to 0.5% (0.0– 0.8) post-recalibration (Table 2).

Pool1–3 sera were found to be comparable to individual donor units among all IAs; median percent biases (−21.0% for FT4 and 0.5% for TSH assays) of these samples were similar to the −20.3% median percent bias for FT4 and −1.2% median percent bias for TSH IAs measuring the 40 PS (Table 6).

IA participants generally reported higher values (−1.9% median percent bias to the RMP, range of −27.8–19.8) for the PN compared to the −20.3% median bias (95% CI of −21.5 – −19.4) of PS materials (Table 6). This increase occurred despite assay-specific RI being lower in general during pregnancy than the adult RI. ¹⁶