How low can we (reliably) go? A method comparison of thyroid-stimulating hormone assays with a focus on low concentrations

Introduction

Thyroid disorders are common. When a thyroid disorder is suspected, the standard procedure is measuring the serum or plasma thyroid-stimulating hormone (TSH). Accurate TSH measurements are important for timely diagnosis since symptoms can be subtle and also for monitoring therapy in hypo- or hyperthyroidism and assessing suppressive therapy in patients with thyroid cancer. Unfortunately, TSH assays are poorly standardized to date.

The IFCC Working Group on Standardization of Thyroid Function Tests established in 2010 standardization differences up to 39% between TSH assays and recommended further research into the concentration range close to the limit of quantification (1). The working group deemed the availability of TSH reference material in the mid- to short-term technically unlikely and now focusses on statistical harmonization (2). Currently, the grand majority of TSH assays are not yet harmonized, which means that absolute concentrations reported by laboratories using different assays are not comparable (3, 4). This could impact diagnosis, prognosis and management of disease, as has recently been shown for subclinical hypothyroidism (5).

The lower range of the TSH assay is important because in this pathophysiological range, often clinical decisions are made, and this is also where assay performance decreases (6). Several international guidelines mention absolute TSH concentrations for clinical decision-making such as in subclinical hyperthyroidism, thyroid cancer, pregnancy and assisted reproduction (7, 8, 9, 10, 11). However, to our best knowledge, research is lacking concerning the comparability of results in this low TSH range.

Therefore, we performed a method comparison for commonly used TSH immunoassays, with oversampling in the lower concentration range (<0.4 mIU/L) and established the imprecision in this low concentration range. Based on the standardization differences, we investigated the theoretical impact of the absolute cut-offs mentioned in the guidelines.

Material and methods

Results

The distribution of the samples, as measured using the Cobas assay, is depicted in Supplementary Fig. 1 (see section on supplementary materials given at the end of this article). As the lower concentrations were oversampled, the median TSH concentration used for the method comparison was 0.24 mIU/L (interquartile range: 0.04–3.21 mIU/L; n = 60). For the duplicate measurements, the median was 0.25 mIU/L (interquartile range: 0.06–2.66 mIU/L; n = 20).

Passing–Bablok regression and correlation coefficients between the three methods are depicted in Fig. 1. The slope of the Passing–Bablok regressions indicates the standardization differences, as the y-intercept was negligible. The bias ranged from 7% to 16% in the complete measured range. For the samples with TSH <0.40 mIU/L, the bias differed between 1% and 14%, as is depicted in Fig. 1.

View Full Size

Passing–Bablok regression analysis for TSH for the total sample set (n = 60) (A, B, C) and for TSH <0.4 mIU/L (n = 36) (D, E, F). For all P < 0.0001.

Citation: European Thyroid Journal 12, 5; 10.1530/ETJ-23-0123

• Download Figure
• Download figure as PowerPoint slide

The imprecision of the assays increased as the concentration lowered and is shown in Table 1. Differences were seen between assays and concentrations, with the largest mean variation between duplicate measurements being 5.5% for the Atellica assay in the range [TSH] <0.40 mIU/L.

The Atellica and Cobas assay exhibited the largest bias in the lower concentration range (14%). Based on this bias, 1.5% of all TSH measurements with results <0.4 mIU/L (15 of the 998 measurements) in our laboratory, using the Cobas assay, would cross the absolute cut-off and would be <0.10 mIU/L instead of ≥0.10 mIU/L, when measured with the Atellica assay.

In addition, we compared the initial TSH measurement on the Cobas, which we used to include the samples, to the remeasurements performed on the Cobas, but in one run. The Passing–Bablok regression was TSHCobas_{remeasurement} = 0.935 × TSHCobas_initital + 0.0003; rho = 0.998; P < 0.0001 for all concentrations, and for TSH <0.4 mIU/L: TSHCobas_{remeasurement} = 0.924 × TSHCobas_initial + 0.0013; rho = 0.991; P < 0.0001. This remeasurement would also lead to 1.5% of all TSH measurements <0.4 mIU/L to cross the absolute cut-off. So, the imprecision of one TSH assay causes as many patients crossing the cut-off as using a different TSH assay.

Discussion

In the present study, we performed a method comparison between three frequently used TSH immunoassays focusing on the low concentration range, as this range is important for clinical decision-making in case of (subclinical) hyperthyroidism and thyroid cancer.

The studied TSH immunoassays showed standardization differences with a bias between 7 and 16%. When analysing the standardization differences for the low range, the bias differed between 1 and 14%. Interestingly, the assays which were standardized against the same WHO standard varied more from each other than compared to the assay with a different WHO standard. The imprecision differed between 1.6 and 5.5% and increased in the lower range of the assay.

Several guidelines use absolute cut-off TSH values in the low range for clinical decision-making (7, 8, 9, 10, 11). In the American and European guidelines for thyroid cancer and in the European guideline for subclinical hyperthyroidism an absolute TSH value of 0.10 mIU/L is mentioned as an important value above or below which recommendations are given concerning starting or adjusting medication and performing diagnostic testing (7, 8, 9). In order to establish whether the observed standardization differences translate into clinically relevant differences, we looked at all TSH results reported by our laboratory in the last 6 months. We calculated that 1.5% of all TSH measurements <0.4 mIU/L would theoretically lead to a different clinical outcome, concerning this absolute cut-off of TSH = 0.1 mIU/L. Naturally, not all the measurements around 0.1 mIU/L concerned patients with subclinical hyperthyroidism or thyroid cancer.

The imprecision using the Cobas assay had the same repercussions on the different classification of samples around TSH = 0.1 mIU/L as the greatest bias between different assays. For the Alinity and Atellica assays with a slightly higher imprecision for TSH <0.4 mIU/L, the impact on the clinical management difference would be even higher than the immunoassay bias. Although the standardization differences for the total concentration range are evident, we consider these standardization differences too small to be clinically relevant in the low concentration range. Nonetheless, it is important to be cautious when interpreting concentrations close to a clinical cut-off value, especially if the methods are poorly standardized and the thresholds are not assay specific.

The widespread use of arbitrary dichotomous cut-offs based on continuous data has limitations. Considering imprecision and bias, combined in total laboratory error or measurement uncertainty, are crucial when establishing clinical cut-offs. Total laboratory error or measurement uncertainty encompasses variability throughout the entire testing process, including pre-analytical, analytical and post-analytical factors. Additionally, within-subject biological variation contributes to variability in results and should also be considered when interpreting laboratory results. Both the total laboratory (analytical) variation and within-subject biological variation (estimated at 17.7% for TSH, source: https://biologicalvariation.eu/search?query=TSH) contribute to the critical difference, also called the reference change value, the smallest difference between sequential laboratory results that is associated with a real change in the patient. By accounting for these factors, a better understanding of the limitations associated with arbitrary dichotomous cut-offs is achieved, enabling more informed and accurate interpretations of assay results.

Over time, a trend towards smaller standardization differences for TSH assays is seen in multiple studies (1, 12, 13, 14). Comparing our results to other method comparisons is generally complicated by the limited mention of the assay specifics. For some method comparisons, it is difficult to analyse results side by side because of different statistical methods, such as linear regressions instead of Passing–Bablok regressions or because of anonymization of the used assays (1, 15, 16). The imprecision is known to increase with lower concentration, which we also observed (2, 16). The clinical management difference we observed is smaller than in previous studies assessing the effect on clinical and subclinical hypothyroidism. This may be explained by the proportional difference in low vs high concentrations and improvements in bias between assays over time (5, 15, 17).

Taken together, we performed a method comparison of commonly used TSH assays and focused on the lower range of TSH concentrations. The bias in the lower concentration range showed a maximum of 14% between methods. Based on this proportional bias and the imprecision we estimated the theoretical effect on measured TSH concentrations in the last 6 months in our lab. This effect seems to have limited clinical consequences for the low TSH concentration range. Nevertheless, as guidelines use absolute TSH values to guide clinical decision-making, caution must be applied with TSH values close to these cut-offs.

Supplementary materials

This is linked to the online version of the paper at https://doi.org/10.1530/ETJ-23-0123.

Declaration of interest

Authors state no conflict of interest.

Funding

This research did not receive any specific grant from any funding agency in the public, commercial or not-for-profit sector.

Data availability

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Author contribution statement

All authors have accepted responsibility for the entire content of this manuscript and approved its submission.