An isotope dilution-liquid chromatography-tandem mass spectrometry-based candidate reference measurement procedure for the quantification of cortisol in human serum and plasma

Chemicals and reagents

LC-MS grade solvents used were methanol (Biosolve, Valkenswaard, The Netherlands), ammonium fluoride and ethyl acetate (Merck, Darmstadt, Germany), and water (purified using a Millipore Milli-Q^® IQ 7000 system – Merck). The standard reference materials SRM^® 921 and SRM^® 921a were obtained from NIST, while the qNMR standard tecnazene (TraceCert qNMR Standard traceable to NIST PS1; Lot Nr. BCBW6288), cortisol (European Pharmacopoeia; H1300000), deuterated cortisol-9,11,12,12-d₄ ISTD (100 μg/mL in methanol), and prednisolone were purchased from Merck (Darmstadt, Germany). The certified secondary reference materials utilized in the study included ERM^®-DA192 lyophilized human serum, ERM^®-DA193 lyophilized human serum, and the human serum reference panel ERM^®-DA451/IFCC (European Commission, Joint Research Centre, Geel, Belgium). Steroid-free human serum and native human serum (anonymized samples) were obtained from Roche Diagnostics GmbH (Penzberg, Germany). Native human plasma (K₃EDTA, K₂EDTA and lithium [Li]-Heparin plasma pools) was obtained from Biotrend (Cologne, Germany). In compliance with the Declaration of Helsinki, all patient samples were residual anonymized samples.

General requirements for laboratory equipment

A complete list of laboratory equipment (calibrated and certified by the manufacturer) and relative requirements is provided in Supplementary Material 1.

qNMR for determining the purity of the standard materials

qNMR measurements were performed on a Jeol 500 MHz NMR with a N₂-cooled Cryoprobe (2–3 times ¹H sensitivity when compared to RT-probes). Single-Pulse-¹H{¹³C}NMR was utilized for the quantitation (olefinic proton, 1H; Supplementary Material 2, Table s1 and Figure s2) with an inter-scan delay of 70 s. This olefinic proton at 5.66 ppm is clearly separated from any other resonance and is optimal for the quantitation of cortisol. Other known impurities of cortisol were not found in this European Pharmacopoeia standard (Supplementary Material 2, Figure s1). The other methylene protons (alpha to OH) are unsuitable as quantitation signals owing to their propensity towards abstraction by basic impurities, if present. Further details about NMR acquisition and processing parameters are available in the Supplementary Material 2 (Table s1, Figure s1 and s2).

Preparation of calibrators and quality control (QC) samples

The preparation of the calibrator and QC samples is described in detail in Supplementary Material 1.

To summarize, two independently weighed calibration stock solutions were prepared. For each stock solution, 5 mg of cortisol (in-house qNMR characterized cortisol H1300000) was weighed using an ultra-microbalance (XP6U/M, Mettler Toledo) and dissolved in 10 mL methanol in a volumetric flask, to obtain a concentration of 0.5 mg/mL (1.38 mmol/L). The exact concentration of the stock solutions was calculated based on the purity determined by qNMR (99.6 % ± 0.4 %; uncertainty given as 2 standard deviation (SD) and the exact amount weighed. Each primary stock solution was diluted further with methanol to achieve working solutions with a 50 μg/mL (0.138 mmol/L) concentration. The working solutions were used to prepare nine calibrator spike solutions of different concentrations in 40 % methanol. Each calibrator spike solution was diluted in steroid-free human serum matrix (1 + 49 v + v) and uniformly distributed in a range of 0.800–600 ng/mL (2.21–1,655 nmol/L).

The certified reference materials ERM^®-DA192 (98.8 ng/mL ± 2.0 ng/mL; 273 ± 6 nmol/L) and ERM^®-DA193 (277 ng/mL ± 5 ng/mL; 763 ± 14 nmol/L) were used as control material, as well as two self-spiked controls (QCs) and a native patient sample. The QCs were prepared using a third, independent primary stock solution (0.5 mg/mL, 1.38 mmol/L) and the resulting spike solutions. This stock solution was prepared using the NIST SRM^® 921, and its final concentration was calculated based on the certificate. After further dilution with methanol, the resulting working solution (50 μg/mL, 0.138 mmol/L) was used to prepare the two control spike solutions and final QCs with concentrations of 2.40 and 550 ng/mL (6.62 and 1,517 nmol/L), respectively.

Internal standard (ISTD) solution

ISTD solution (600 ng/mL; 1,637 nmol/L) was prepared by transferring 60 µL of the commercially available stock solution (Cortisol-d4 in methanol 100 μg/mL [273 μmol/L]; C-113-1ML; Cerilliant) into a 10 mL volumetric flask and adding 40 % methanol (v/v) up to the calibration mark.

Sample preparation

A sample matrix native serum, plasma (Li-Heparin plasma, K₂EDTA plasma and K₃EDTA plasma) and steroid-free serum serving as surrogate matrix were used. High-concentrated samples (>600 ng/mL; >1,655 nmol/L) were diluted 1+1 (v + v) with steroid-free human serum before sample preparation and mixed for 30 min on an overhead shaker before further use. For sample preparation 100 µL of a native sample, calibrator or control sample were transferred into a 0.5 mL screw cap. 20 µL of the internal standard working solution was added, and the samples were incubated on an overhead shaker for 15 min at room temperature. Milli-Q^® water (330 µL) was added to each sample before shaking them for 5 min on the thermomixer (Eppendorf 5,382 thermomixer C) at 1,400 rpm. Final purification was performed by loading 400 µL of the serum/water mixture onto a 3cc Novum Support Liquid Extraction (SLE) cartridge (Phenomenex) with an initial vacuum of 5 s and allowing it to settle for 20 min. The extraction was performed using 1 mL ethyl acetate (x2), evaporating the extract to dryness in a nitrogen evaporation system (Biotage, Uppsala, Sweden; 50 °C, 1.3 mL/min, 15 min) and reconstituting the residue in 300 µL 40 % methanol. The solution was filtered using centrifuge filter tubes (5 min, 15,000 rcf) and transferred into an HPLC vial with a 450 µL insert.

Liquid chromatography-tandem mass spectrometry (LC-MS/MS)

The chromatographic purification was performed using an Agilent 1290 Infinity II LC system equipped with two binary pumps, (Agilent, Santa Clara, CA, USA). Analyte detection was performed using an AB Sciex Q-Trap 6500 mass spectrometer (AB Sciex, Framingham, MA, USA). Chromatographic separation of cortisol was achieved using a two-dimensional heart-cut liquid chromatography (LC) approach with a combination of two orthogonal stationary phases (in the first dimension a Waters Acquity BEH C18 [1.7 µm, 50 × 2.1 mm] and in the second dimension a Raptor Biphenyl [2.7 µm, 150 × 2.1 mm]), to minimize matrix effects and the co-elution of isobaric interferences. In both dimensions, mobile phases consisted of 0.2 mM ammonium fluoride in Milli-Q-water (A) and methanol (B) and the separation at 50 °C was achieved using individual gradients over 10 min with flow rates of 0.3 mL/min and 0.5 mL/min, respectively. The injection volume was 5 µL.

Cortisol was detected in the multiple reaction monitoring (MRM) mode using an AB Sciex QTrap 6500 mass spectrometer operating in electrospray ionization mode. The quantifier ion transition (cortisol m/z 363.3 → 96.9) and the corresponding internal standard transition (cortisol-d4 m/z 367.1 → 96.9) served as the basis for the cortisol quantitation, while the additional qualifier transition (cortisol m/z 363.3 → 120.9) was monitored to screen for unknown interferences. The LC and MS methods are fully reported in Supplementary Material 1.

System suitability test (SST)

To assure the long-term stability of the method, an SST was established to examine sensitivity and chromatographic resolution before every sequence. Two sample levels (SST1 and SST2) were prepared in 40 % methanol, containing cortisol in concentrations of 0.800 ng/mL (2.21 nmol/L) and 600 ng/mL (1,655 nmol/L), respectively.

For passing the SST, the signal-to-noise ratio (S/N) of the quantifier transition of sample SST1 was required to be ≥10 S/N. It was calculated using the Analyst^® software (AB Sciex, Framingham, MA, USA) as peak height divided by noise. The Analyst software uses the SD (with a mean of zero) of all chromatographic data points between the specified background to calculate noise. Background from 7.20 to 7.70 min was used to calculate the S/N ratio. In addition, a retention time of 8.3 min (±0.5 min) for cortisol was required for both samples (SST1 and SST2).

To examine potential carryover, sample SST2 was injected, followed by two solvent blanks. The analyte peak area observed in the first blank was required to be ≤20 % of the analyte peak area of SST1 to pass the SST.

Calibration data processing and structure of analytical series

Spiked calibrators were prepared as described in the section Preparation of calibrators and quality control (QC) samples. Calibrator levels were measured at increasing concentrations at the beginning and the end of every sequence. The calibration functions (using 9 levels) were obtained by quadratic regression (1/x weighting) of the area ratios of the analyte and the internal standard (y) against the analyte concentration (x), resulting in the function y=a₂x² + a₁x + a₀.

To process the raw data file, the Analyst software (v1.6.2 or higher) was used with Intelli Quant as Quantitation Integration Algorithm (ABSciex). Raw data were processed with a smoothing factor of three.

The individual sequence setup varied due to specific measurement requirements. For assignment of reference values, the number of sample preparations (n=x) was dependent on the desired MU, and samples were measured on at least two different days. In case of a method comparison study, or complaint sample measurement, samples were prepared with n=1.

Method validation

Assay validation and determination of MU were performed following the established guidelines of the Clinical and Laboratory Standard Institute: C62A Liquid Chromatography-Mass Spectrometry Methods [25], and the Guide to the Expression of Uncertainty in Measurement (GUM) [26].

Selectivity and specificity

Selectivity was assessed using a native serum pool containing a known concentration of endogenous cortisol and an unknown concentration of cortisone. This pool was fortified with prednisone and prednisolone at concentrations of 100 ng/mL (276 nmol/L) and 200 ng/mL (552 nmol/L), respectively. Samples were then prepared and analyzed by LC-MS/MS in the same manner as the calibrators. Baseline separation of the analytes was evaluated in both the first and second dimensions. Additionally, to account for potential interference of prednisolone with cortisol, the fortified sample was quantified (n=6), and the cortisol concentration was determined.

To examine possible interfering matrix signals for the analyte quantifier and qualifier transition, both a steroid-free serum pool and a native human serum pool were checked at the expected retention time (8.3 [±0.5] min).

In addition, an analyte-free matrix was spiked with the deuterated internal standard to evaluate a possible amount of residual unlabeled analyte within the stable isotope labeled internal standard. As matrices, a native human serum pool and a surrogate matrix (steroid-free human serum) were used.

Matrix effect

To determine possible matrix effects, a qualitative post-column infusion experiment and a quantitative experiment based on the comparison of standard line slopes were performed. In the post-column infusion setting, a neat solution with 100 ng/mL cortisol (276 nmol/L)and cortisol-d4 (273 nmol/L) in 40 % MeOH in Milli-Q Water was infused via a T-piece into the HPLC post-column eluent prior to entering the MS/MS system to generate a stable analyte background signal. The flow rate was set at 10 μL/min. Then, a processed matrix sample was injected, and the change of the background signal was acquired. Different matrices such as neat solution (40 % MeOH in Milli-Q Water), native serum matrix pool (made up from five individual donors), steroid-free serum, as well as three plasma matrices (Li-Heparin, K₃EDTA, K₂EDTA) were analyzed. Any decrease or increase of the SRM analyte signal would indicate a matrix component-mediated effect on the ionization yield of the analytes. In addition to the T-piece experiment, a comparison of standard line slopes was performed comparing the following matrices: neat solution (40 % MeOH in Milli-Q Water), native serum matrix pool (made up from five individual donors), steroid-free serum, and Li-Heparin plasma matrix. Calibrator levels were prepared as described in Section 4.3 of Supplementary Material 1. Neat samples were diluted to the final concentration of processed calibrator levels, whereas matrix calibrator levels were prepared by SLE (Section 5.1 of Supplementary Material 1). Slopes and coefficients of determination were compared. The confidence intervals of slopes need to overlap. In addition, the calibrator samples in a surrogate matrix and native serum were evaluated as controls by applying the neat calibration as standard. Recoveries were reported as the percentage of recovery of the measured concentration relative to the nominal concentration.

Linearity

For the determination of linearity, calibration curves from three individual weighings were prepared. Calibration range was extended by ±20 %, and two additional spike solutions were prepared to obtain spiked serum samples with a final concentration of 0.640 (1.77 nmol/L) ng/mL and 720 ng/mL (1986 nmol/L) cortisol. The peak area ratio of the analyte to the corresponding ISTD was plotted against the respective analyte concentration (ng/mL). Correlation coefficients and residuals for each curve were determined and had to be ≥0.99. The linearity of the method was established by serial recovery of diluted samples (pool 1 up to pool 11) using the preferred regression model for calculation. The ideal sample matrix was set as a patient’s specimen pool with an analyte concentration near the expected upper reportable limit that is diluted with another patient’s sample pool having an analyte concentration at the expected or tested lower concentration limit. If no native materials were available, spiked samples were used. Therefore, a native serum sample with a concentration of approximately 550 ng/mL (1,517 nmol/L) was used as pool 11, whereas the analyte-free surrogate matrix (steroid-free human serum) was used as pool 1. Using these two samples, pool 2 was prepared as a mixture of pool 1 and pool 11 (9 + 1 v + v) and serially diluted till pool 10 (1 + 9 v + v). Measurement results had to show a linear dependency (r≥0.99), and recoveries were reported as the percentage of recovery of the measured concentration relative to the nominal concentration of the sample pool.

Lower limit of measuring interval (LLMI) and limit of detection (LOD)

Precision, trueness and accuracy at the LLMI were determined by measuring spiked serum matrix samples in the expected concentration range of the LOQ. The LOQ matches with the lowest calibrator level (0.800 ng/mL; 2.21 nmol/L). Samples were prepared in five replicates; precision, bias and recovery were determined.

Precision

Precision was evaluated with a 5-day validation experiment, performed as reported by Taibon et al. [27]. Two spiked steroid-free serum samples (2.40 and 550 ng/mL, 6.62 and1517 nmol/L), two certified secondary reference materials (ERM^®-DA192 [98.8 ng/mL ± 2.0 ng/mL; 273 ± 6 nmol/L] and ERM^®-DA193 [277 ng/mL ± 5 ng/mL; 763 ± 14 nmol/L]), and two native patient samples (approximately 45 and 120 ng/mL, 124 and 331 nmol/L) were prepared in triplicate for each part and injected twice (n=12 measurements per day and n=60 measurements over 5 days). Data evaluation was done using Biowarp, an internal statistic program based on the Variance Component Analysis (VCA) Roche Open Source software package in R [28]. Repeatability included between-injection variability and between-preparation variability, whereas intermediate precision included additional between-calibration variability and between-day variability. Repeatability and intermediate precision are expressed as SD and coefficient of variation (CV).

Trueness and accuracy

Accuracy and trueness were assessed using two certified secondary reference materials (ERM^®-DA192 [98.8 ng/mL ± 2.0 ng/mL; 273 ± 6 nmol/L] and ERM^®-DA193 [277 ng/mL ± 5 ng/mL; 763 ± 14 nmol/L]) and two spiked steroid-free serum samples (2.40 and 550 ng/mL, 6.62 and 1,517 nmol/L) and. Validity of dilution was performed using two spiked native serum samples at concentration levels of 800 ng/mL (2,207 nmol/L) and 1,000 ng/mL (2,759 nmol/L). In addition, accuracy and trueness were determined in a Li-Heparin plasma pool at three concentration levels (100, 250 and 550 ng/mL; 276, 690 and 1,517 nmol/L). All samples were prepared in triplicates for each part (n=6 measurements) on one day. Accuracy was reported as the level of agreement between the measured value and the accepted reference value. Trueness, on the other hand, was evaluated by comparing the average value obtained from a series of measurements to the target value, assessing their agreement.

Sample stability

The stability of the processed samples (calibrators and QC levels) on the autosampler was investigated at 4–8 °C for 17 days. Therefore, samples from the accuracy and precision experiment were used and were re-measured against a freshly prepared calibration after 1, 3, 7, 8, 10, and 17 days. Recoveries were calculated by comparing the measured value with the nominal concentration (t=0 h). Stability of spike solutions (50.0, 200 and 500 ng/mL; 138, 552 and 1,379 nmol/L) and matrix-based spiked control material (samples from the precision and accuracy experiment) stored at −80 °C were evaluated for 43 and 17 weeks, respectively. Samples were required to be re-measured and evaluated against freshly prepared samples performing four measurements over the aforementioned time period.

Equivalence of results between independent laboratories

To assess the agreement of the RMP between two independent laboratories, a method comparison study was performed, including overall 164 anonymized residual patient samples (serum and plasma). The RMP was transferred to the second laboratory the Department of Pediatrics and Adolescent Medicine, University Hospital Erlangen (site 2). In addition, a three-day precision experiment based on the experimental design described above was performed at site 2. Spiked samples were provided by Roche Diagnostics Penzberg (site 1). Both laboratories prepared their own calibrator levels using cortisol characterized by qNMR, as primary reference material. The LC-MS system and laboratory equipment were different in both laboratories. In addition, adaptations to the sample preparation protocol were made in the second laboratory to use the 96-well plate format. All steps before the SLE sample preparation were done in one single 96-well deep well plate from polypropylene. In brief, 100 µL sample and 20 µL internal standard were incubated before adding 330 µL water. The sample was mixed, and 400 µL of the diluted sample was loaded on a 96-well plate (Phenomenex Novum SLE) and incubated. Samples were eluted twice using 1 mL ethyl acetate. The eluate of both parts was collected in one set of glass inserts supplied by Hirschmann (Eberstadt, Germany). After evaporation, the samples were reconstituted, mixed, sealed and measured.

To establish the comparability between the RMP, performed at site 1, and an in-house developed high-throughput LC-MS/MS assay, a second method comparison study including 165 anonymized residual patient samples (serum and plasma) was conducted. This assay utilized a commercially available calibration material and the online solid phase extraction LC-MS/MS method by Gaudl et al. [29], and was performed at site 3, specifically the Institute of Laboratory Medicine, Clinical Chemistry and Molecular Diagnostics at Leipzig University Hospital.

Equivalence to JCTLM listed RMPs

To assess the agreement between the newly established RMP and the JCTLM-listed ID-GC/MS-based RMPs [10], 11], the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) cortisol reference serum panel (ERM^®-DA451/IFCC) was used. The serum panel comprises 34 samples that were measured in two replicates on two different days. The results were then calculated as the arithmetic mean, with a total of four measurements obtained.

Additionally, participation in the RELA – IFCC External Quality Assessment Scheme for Reference Laboratories in Laboratory Medicine has been ongoing since 2022. Provided samples were prepared threefold over two different days.

Estimation of measurement uncertainty (MU)

According to the GUM [26] and based on Taibon et al. [27], MU was estimated accounting for the uncertainty of qNMR target value assignment of reference material, preparation of calibrator materials, and LC-MS/MS method. qNMR measurements were performed in six replicates using an SI traceable internal standard for the quantification of the absolute content (Px, purity of the analyte given as mass fraction) of the analyte. A bottom-up approach was used to estimate the MU related to the serum-based calibration material, including the following uncertainty components: purity of the reference material, weighing, preparation of stock solution and dilutions, preparation of spike solution and preparation of matrix-based calibrator level. Uncertainties of the laboratory equipment used, such as ultra-micro balance, pipettes and volumetric flasks, were provided by the manufacturer. Estimation of total variability of the LC-MS/MS method was performed using a precision experiment (type A uncertainty) and considered the following steps: sample preparation of calibrators, preparation of internal standard, preparation of samples, measurement of calibrators, generation of the calibration curve and measurement and evaluation of sample results. A detailed description of the evaluation of MU is given in Supplementary Material 3.

Traceability to SI units

Traceability to the SI unit of mass (kilogram), the most important parameter for a reference measurement method, has been established by the utilization of tecnazene as the qNMR ISTD which is directly traceable to NIST PS1 (primary qNMR standard). Furthermore, traceability to the SI unit of amount of substance (mole) is also included owing to NIST PS1. Since Planck’s constant is now the main parameter for defining kilogram and Avogadro’s constant for mole, qNMR methodology is best suited to fulfill both the traceability chains and represents a primary RMP per IUPAC [19]. Six individual experiments (Supplementary Material 2, Table S1, Figure S2) involving six individual weightings, yield a final absolute content value of 99.6 % ± 0.4 % (k=2).

Selectivity and specificity

To minimize matrix effects and the co-elution of isobaric interferences (both known and unknown), a two-dimensional heart-cut LC approach for the accurate analysis of cortisol in human serum was selected. Using the Waters Acquity BEH C18 column in the first dimension in combination with mobile phases consisting of 0.2 mM ammonium fluoride (A) and methanol (B) allowed the separation of cortisol from cortisone and prednisone. Cortisol and prednisolone were transferred to the second dimension, where a Raptor Biphenyl column with the same mobile phases was used, showing almost a baseline separation of prednisolone from cortisol (Figure 1). The bias observed within the nominal cortisol concentration, and the calculated cortisol concentration was 4.1 %.

Figure 1:

Chromatographic separation of cortisol and prednisolone in a native serum matrix with concentrations of 66.4 ng/mL (183 nmol/L) and 200 ng/mL (555 nmol/L), respectively. Analytes (left) and ISTD (right). ISTD, internal standard.

In addition, no interfering signals for the quantifier transition (m/z 363.3 → 96.9) and the corresponding internal standard quantifier transition (m/z 367.1 → 96.9) were observed at the expected retention time window (8.3 ± 0.5 min) neither in the steroid-free serum nor in the native serum pool (Figure 2).

Figure 2:

Cortisol LC-MS/MS derived analytical readouts. (A) Chromatogram of calibrator level 1 with a concentration of 0.800 ng/mL (2.21 nmol/L) spiked in steroid-free human serum matrix, analyte (left) and ISTD (right); (B) secondary reference material ERM-DA192 in native human serum with a concentration of 98.8 ng/mL (273 nmol/L), analyte (left) and ISTD (right); (C) fortified Li-Heparin patient sample with a concentration of 242 ng/mL (668 nmol/L), analyte (left) and ISTD (right). ISTD, internal standard.

The steroid-free serum sample spiked with internal standard did not show any significant analyte signal. Therefore, cortisol-d4 can be used as the internal standard.

Matrix effect

In the post-column infusion experiment none of the tested matrices (neat solution, serum or plasma) showed ion suppression or enhancement in the region of the retention time of cortisol and its internal standard. For the comparison of standard line slopes, calibrator levels were spiked in neat solution, native serum, steroid-free serum, and Li-Heparin plasma matrix. Slopes were 0.0092 (95 % CI 0.0090–0.0093) for the neat solution, 0.0089 (95 % CI 0.0087–0.0091) for steroid-free serum, 0.0088 (95 % CI 0.0086–0.0090) for the native matrix and 0.0085 (95 % CI 0.0080–0.0090) for the Li-Heparin plasma. The confidence intervals of the slopes overlapped, suggesting that they were not significantly different from each other. These data confirm the absence of a matrix effect. Correlation coefficients (r) were ≥0.999, irrespective of the matrix used for calibration. Furthermore, recoveries for calibrator samples in surrogate matrix, native serum and Li-Heparin plasma ranged from 93 to 104 %, meeting the acceptance criteria.

Linearity

Linearity was demonstrated by analyzing three individually generated matrix-based calibration curves using individual sample weights, expanded for ±20 % of the expected measuring range. The residuals were randomly and equally distributed and correlation coefficients were r≥0.999 for all individual calibration curves. The linearity of the method was confirmed using serially diluted samples, with measurement results demonstrating a linear dependence (r≥0.999). Recoveries ranged between 99 % and 106 %, also meeting the acceptance criteria.

Lower limit of measuring interval (LLMI) and limit of detection (LOD)

The LLMI was determined using a spiked matrix sample with a concentration of 0.820 ng/mL (2.26 nmol/L), corresponding to the lowest calibrator level. The relative deviation (n=6) was 1.7 %, CV was determined at 3.1 %, and recovery ranged from 98 to 106 %. These findings satisfy the defined analytical performance criteria based on biological variability. The LOD was estimated as 0.229 ng/mL (0.632 nmol/L).

Precision

A multi-day validation experiment was performed to estimate the total variability of the RMP. Variability components were estimated using an ANOVA-based variance component analysis. All samples were prepared in triplicate for each part and injected twice resulting in n=60 measurements over 5 days. However, for level 4, only 59 measurements were used for evaluation due to an acquisition error.

The intermediate precision, encompassing variances from between-day calibration, preparation, and injection, was ≤2.6 %, while the repeatability CV ranged from 0.9 to 1.9 % across all concentration levels (Table 1), confirming that the requirements for imprecision were met.

Trueness and accuracy

Relative mean bias ranged from −1.3 to 1.4 %, independent of the matrix and concentration level. Dilution integrity was determined by using two spiked samples. The mean deviation was −1.8 % for the 800 ng/mL (2,207 nmol/L) spiked sample and −0.3 % for the 1,000 ng/mL (2,759 nmol/L), highlighting that the method is suitable for the analysis of samples up to a target concentration of 1,000 ng/mL (2,759 nmol/L) (Table 2). Overall, these findings meet the requirements and show that the method does not exhibit statistically significant bias and is matrix-independent.

Sample stability

Autosampler stability of processed samples (4–8 °C) was shown for 16 days. Recoveries ranged between 97 % and 108 %. The stability of neat spike solutions and spiked control samples were determined for 39 and 16 weeks, respectively; recoveries ranged between 95 % and 108 %.

Equivalence of results between independent laboratories

To demonstrate the transferability of the RMP to a second laboratory, a method comparison study was performed with the Department of Pediatrics and Adolescent Medicine, University Hospital Erlangen. A total of 164 samples were analyzed in both laboratories, of which two samples showed concentrations below the lower limit of measuring interval. Figure depicts the very good agreement observed between the two laboratories, resulting in a regression equation with a slope of 1.01 (95 % CI 1.00 to 1.02) and an intercept of −0.54 (95 % CI −1.74–0.15) and r=0.998 (Figure 3A). Bland-Altman analysis showed very good agreement with a mean bias of −0.1 % which does not differ statistically significant from zero. The 2S interval of the relative difference was 8.3 % (lower limit CI interval from −9.5 to −7.2 %, upper limit CI interval from 7.1 to 9.3 %) (Figure 3B). The three-day precision experiment resulted in CVs that were comparable to those of the first laboratory. Intermediate precision was 0.8–2.4 %, and repeatability was <1.7 % for all levels except for the lowest level (reported at 4.9 % and 3.5 %, respectively). These data indicate that the proposed RMP for cortisol is transferable between laboratories, emphasizing the robustness of the protocol design.

Figure 3:

Results from the cortisol method comparison study performed between two independent laboratories. (A) Passing-Bablok regression plot for the method comparison study of the RMP (n=162) performed between the Roche Diagnostics GmbH Penzberg (site 1) and the Department of Pediatrics and Adolescent Medicine and Adolescents, University Hospital Erlangen (site 2). The analysis resulted in a regression equation with a slope of 1.01 (95 % CI 1.00–1.02) and intercept of −0.54 (95 % CI −1.74–0.15). The correlation coefficient was 0.998. (B) Bland-Altman analysis showed a mean bias of −0.1 % (95 % CI −0.7–0.6) and an associated 2S interval of 8.3 % (lower limit CI interval from −9.5 to −7.2 %, upper limit CI interval from 7.1 to 9.3 %). CI, confidence interval; RMP, reference measurement procedure; SD, standard deviation.

In order to establish comparability with an in-house developed high-throughput LC-MS/MS based assay [29], an additional method comparison study was conducted. This assay used a commercially available kit for calibration. A total of n=165 native patient samples were analysed, with 4 samples falling below the lower limit of the measuring interval and 38 samples exceeding the upper limit of the measuring interval in the in-house developed assay. The Passing-Bablok regression analysis resulted in a regression equation with a slope of 1.09 (95 % CI 1.07 to 1.11) and an intercept of 1.02 (95 % CI −0.42 to 3.17) with r=0.990 (Figure 4A). Furthermore, the Bland-Altman analysis revealed a mean bias of 10 % and a corresponding 2S interval of approximately 12 % (with a lower limit CI interval ranging from −4.3 to −0.4 % and an upper limit CI interval ranging from 20.5 to 24.4 %) (Figure 4B).

Figure 4:

Results from the cortisol method comparison study performed between the RMP and a well-established routine LC-MS/MS based assay. (A) Passing-Bablok regression plot for the method comparison study of the RMP (n=123) performed between the Roche Diagnostics GmbH Penzberg (site 1) and the Institute of Laboratory Medicine, University Hospital Leipzig (site 3). The analysis resulted in a regression equation with a slope of 1.09 (95 % CI 1.07–1.12) and intercept of 1.02 (95 % CI −0.42 to 3.17). The correlation coefficient was 0.990. (B) Bland-Altman analysis showed a mean bias of 10 % (95 % CI 8.9–11.2) and an corresponding 2 SD of approximately 12 % (with a lower limit CI interval ranging from −4.3 to −0.4 % and an upper limit CI interval ranging from 20.5 to 24.4 %). CI, confidence interval; RMP, reference measurement procedure; SD, standard deviation.

Equivalence to JCTLM listed RMPs

Equivalence between the candidate RMP and the JCTLM-listed RMPs, NRMeth 57 and NRMeth 8, was assessed by measuring the certified IFCC reference panel. Passing-Bablok analysis showed a very good agreement resulting in a regression equation with a slope of 1.00 (95 % CI 0.99 to 1.01) an intercept of 2.33 (95 % CI −1.02–4.70) and r=1.000 (Figure 5A). The Bland-Altman analysis showed a mean bias of 1.0 % (95 % CI 0.6–1.5) and a corresponding 2S interval of approximately 4.5 % (with a lower limit CI interval ranging from −2.3 to −0.7 % and an upper limit CI interval ranging from 2.8 to 4.4 %) (Figure 5B).

Figure 5:

Results from equivalence study between the candidate RMP and the IFCC Cortisol Reference Serum Panel ERM-DA451. The results provided in the ERM Panel’s certificate are given in nmol/L. (A) Passing-Bablok regression between the candidate RMP and the certified concentration for the IFCC reference panel ERM-DA451 (n=34). The analysis resulted in a regression equation with a slope of 1.00 (95 % CI 0.99–1.02) and an intercept of 2.33 (95 % CI −1.02–4.70). The correlation coefficient was 1.000. (B) Bland-Altman analysis showed a mean bias of 1.0 % (95 % CI 0.6–1.5) and corresponding 2SD of approximately 4.5 % (with a lower limit CI interval ranging from −2.3 to −0.7 % and an upper limit CI interval ranging from 2.8 to 4.4 %). Conversion factor ng/mL to nmol/L: 2.76; CI, confidence interval; RMP, reference measurement procedure; SD, standard deviation.

The relative difference between the candidate RMP results (average of n=4) and the certified concentration for the IFCC reference panel ERM^®-DA451 ranged between −1.3 and 3.8 %. Total measurement uncertainty was calculated as combined uncertainty of the calibrator preparation (unc _cal ) and uncertainty (SD) of the mean of measurement results (unc _mean ), resulting in uncertainties between 0.7 and 2.3 % (k=1).

Results from the RELA Scheme for 2022 and 2023 align with those from listed laboratories, including the JCTLM-listed services, one of which is the Reference Institute for Bioanalytics (see Table 3). Data from the 2024 participation have not yet been published.

Estimation of MU

The measurement uncertainties (k=1) for cortisol in single measurements were ≤2.8 % regardless of the concentration level and sample type as shown in Table 4. These uncertainties were estimated by combining the uncertainty of calibrator preparation (unc _cal ) and the precision experiment (unc _prec ). To ensure the error is not underestimated the calibrator level with the higher uncertainty was chosen for each individual sample concentration level. The derived expanded uncertainty was then multiplied by a coverage factor of k=2, which corresponds to an approximate confidence level of 95 %, assuming a normal distribution. The resulting expanded uncertainty and was ≤5.7 %.

For target value assignment, the uncertainty was reduced due to a higher number of measurements (n=6, 3 measurements on 2 days). As a result, the uncertainty (k=1) ranged from 0.7–1.9 %, leading to an expanded measurement uncertainty (k=2) of 1.4–3.8 %, meeting the defined analytical performance specification for the maximum allowable expanded measurement uncertainty (Table 5).