Evaluation of hemolysis index thresholds for 18 biochemistry assays: implications for laboratory-developed tests in the era of the IVDR

Introduction

Hemolysis in laboratory blood samples is one of the most prevalent preanalytical errors, posing the risk of clinical misinterpretation and subsequent incorrect patient treatment [1], [2], [3], [4], [5]. It can occur in two different ways, in vivo and in vitro, with the latter being more common due to mechanical or physical influences during blood sampling, handling, transportation or storage [5]. Studies indicate that hemolysis occurs in 3.3 % of all routine samples and accounts for nearly 60 % of all rejected specimens [2], 6]. Conversely, in vivo hemolysis triggered by factors such as antibodies, infections, specific medications, or mechanical causes (e.g. artificial heart valves) constitutes only 3.2 % of hemolytic samples [5], [6], [7].

Hemolysis releases substances from red blood cells, such as hemoglobin, potassium or lactate dehydrogenase (LDH) into the plasma [7], causing undesirable bias in test results. Free hemoglobin (fHb) can interfere with photometric assays due to its distinct absorption spectrum [8], 9]. Depending on the extent of hemolysis, as well as the instrument and method employed, significant deviations in measurements may occur. Consequently, this can yield results that do not adequately represent the patient’s physiological state.

Evaluating hemolysis is crucial for ensuring the quality of laboratory samples [10]. For instance, in accordance with German Federal Medical Council guidelines (Rili-BAEK), rejection criteria must be established, and instances of sample rejection must be documented [11]. Research has shown that approximately one-third of laboratories visually determine hemolysis, while another third utilizes the automatically measured hemolysis index (HI) [10], 12]. Consequently, the HI serves as an important criterion, and an increasing number of laboratory analyzers are equipped to assess it.

According to current IVDR guidelines and the CE (Conformité Européenne) certification process, manufacturers must determine assay interferences and typically provide HI thresholds to prevent impaired results [8]. Laboratories on the other hand are obliged to use these limits otherwise the test becomes a so-called modified analysis procedure and must classify as an in-house production (LDT: laboratory developed test). In this case the requirements of the IVDR apply and the laboratory becomes de facto manufacturer. This comes with certain basic requirements such as safety and performance requirements (cf. Annex I of IVDR), documentation requirements (specifications, validations, risk assessments), limited in-house privilege (LDTs may only be used if no equivalent CE-IVD products are available) and the necessity of a public declaration (healthcare institutions must make a public declaration on their in-house IVDs by 2024) [13].

So reliable thresholds are crucial: overly strict thresholds can lead to unnecessary suppression of results, while overly lenient thresholds can allow the release of error-prone findings, both of which endanger patients.

Despite its importance, there has been limited scrutiny regarding the quality of manufacturers’ HI thresholds [3], 14]. Additionally, original data is frequently inaccessible to the public, rendering assessment of its accuracy is challenging. Therefore, the primary objective of this comprehensive study was to assess the acceptance limits of the HI for critical or notably hemolysis-prone laboratory parameters, with a focus on enhancing patient safety.

Materials and methods

This study received approval from the local Ethics Committee (approval number: 2022–108). Apparently healthy volunteers aged 18 or older (n=77) were primarily recruited from the participating institutions. All samples were pseudonymized using a randomly generated six-digit code.

A total of 40 mL of whole blood was collected from each participant using lithium-heparin (Li-Hep) and serum (S) tubes (manufacturer: Sarstedt, Nümbrecht, Germany). The serum samples were allowed to clot at room temperature for 15 min. Subsequently, all samples were centrifuged at 2,500×g for 15 min utilizing a Sigma 4K15 centrifuge. The resulting clear supernatant for plasma and serum was collected and stored in 10 mL tubes. Approximately 50 μL of erythrocyte suspension from the sediment was transferred to a microcup and then frozen at −80 °C to induce total cell hemolysis.

Meanwhile, the collected plasma and serum was aliquoted into microcups. After 1 h, the lysed erythrocytes were thawed at room temperature and thoroughly mixed. Subsequently, the following dilution levels were prepared (undiluted, 1:140, 1:160, 1:200, 1:300, 1:400, 1:500, 1:1,000, 1:2,000, 1:10,000) and stored at −80 °C until all sample sets were measured.

Shortly before measurement, the samples were thawed at room temperature followed by a centrifugation step (10 min at 18,000×g, Mikro 120, Hettig). Then, the samples were transferred into measuring tubes. In addition to measuring the HI in each dilution level, 18 common clinical analytes were assessed, namely sodium, potassium, chloride, aspartate aminotransferase (ASAT), alanine aminotransferase (ALAT), gamma-glutamyl transferase (GGT), lactate dehydrogenase (LDH), haptoglobin, total bilirubin, direct bilirubin, glutamate dehydrogenase (GLDH), creatine kinase (CK), iron, amylase, phosphate, total protein, enzymatic creatinine, and high-sensitive troponin T. These measurements were conducted by using the cobas^®pro analyzer (Roche Diagnostics, Mannheim, Germany) and ISE, c 503, and e 801 modules, respectively. The HI was determined photometrically on the c 503 module and reported as a dimensionless value between 5 and 1,200 [15]. The results appear to be equivalent to the plasma cell-free Hb concentration in mg/dL [9], 16], 17].

The statistical analysis of interference across increasing hemolysis Indices (H-indices) was conducted per measurand. For each measurand, a linear regression analysis was performed with concentration as the dependent variable and HI as the independent variable. Starting with concentrations at HI=0, thresholds for maximum errors of 5 , 10, and 20 % were derived from the sample-specific regression equations. For each measurand the final cutoff H-indices were subsequently determined as the 5th percentile and the median of all probands. Furthermore, we have calculated the manufacturer’s error tolerance [%]. This was calculated reversely from the linear relationship between the observed HI limits and median errors.

For visualization, boxplots for each dilution step are presented, with lines indicating deviations of 5 and 10 % from the concentration observed in the undiluted samples. Moreover, scatterplots featuring regression lines per proband, as well as a median regression line (determined by the median slope and median intercept of the sample-specific regression lines), are depicted. Additionally, the medians of the calculated cutoff HI values, referencing 5 and 10 % deviation, are plotted.

Results

Lithium-heparin and serum samples were collected from 77 participants. Each sample set comprised 10 different dilution steps for each material, yielding a total number of 1,540 samples. 18 common clinical analytes and their corresponding HI were measured, resulting in nearly 29,000 individual measurements. 57 measurements were missing due to technical errors. The calculated HI thresholds for lithium-heparin plasma and serum are shown in Tables 1 and 2.

To provide a detailed illustration of our findings, we selected the following six common analytes: ASAT, sodium, potassium, LDH, haptoglobin, and high-sensitive troponin T (Figure 1). All plots for lithium-heparin and serum are provided as Supplemental Material.

Figure 1:

Boxplots and scatterplots plots for (A) LDH (B) high-sensitive troponin T (C) haptoglobin (D) sodium (E) potassium and (F) ASAT. Boxplots (left): Comparison of dilution level (x-axis) and deviation for median (grey), 5 % (green) and 10 % (orange) error (y-axis); scatterplots (right): Comparison of H-index; 5 % (green), 10 % (orange), manufacturer cutoffs (blue) (x-axis) and measurand concentration (y-axis).

For aspartate aminotransferase (ASAT) and lactate dehydrogenase (LDH), noticeable deviations in measurement results were observed across the dilution series as hemolysis increased. The manufacturer’s specified HI is 20 for ASAT and 15 for LDH. For instance, by selecting a 10 % error limit and using the median, our calculated HI threshold for ASAT was determined to be 23 (Li-Hep and S), and for LDH, it was 13 (Li-Hep and S). Thus, our calculated thresholds closely align with the manufacturer’s values.

In contrast, for sodium, increasing hemolysis showed minimal impact on the measurement results. The manufacturer’s specified HI is 1,000. However, our extrapolated HI threshold was 3,181 for lithium-heparin and 3,768 for serum. As our HI results did not exceed 300, extrapolated thresholds should be interpreted with caution.

For potassium, haptoglobin, and high-sensitive troponin T, our calculated HI thresholds differed considerably from the manufacturer’s values. The manufacturer specifies the HI as 20 for potassium, 10 for haptoglobin, and 100 for high-sensitive troponin T. However, for a 10 % error limit using the median, our calculated HI thresholds were determined to be 141 (Li-Hep) and 150 (S) for potassium, 159 (Li-Hep) and 155 (S) for haptoglobin, and 281 (Li-Hep) and 374 (S) for high-sensitive troponin T.

Discussion

Medical diagnosis and treatment decisions highly depend on correct laboratory results. Reliable and traceable manufacturer specifications are important in this context. Furthermore, to comply with current regulations, such as the In Vitro Diagnostic Regulation (IVDR), laboratories are required to fulfil the manufacturer’s instructions [11], 13], 18]. Even in legal matters, such as the context of medical malpractice, laboratories must be able to rely on the manufacturer’s specifications.

Establishing appropriate threshold values is a multifaceted process for manufacturers, encompassing various considerations. Patient safety is of utmost priority. Threshold values must strike a delicate balance, avoiding extremes of stringency or leniency. Overly stringent thresholds may unnecessarily increase the rate of rejected samples, while excessively forgiving thresholds could elevate the risk of erroneous laboratory values, potentially fostering misinterpretation and incorrect patient treatment. Therefore, manufacturers must meticulously weigh these factors to ensure that their threshold values optimize patient safety and the accuracy of laboratory results. In this context different approaches for establishing threshold values can be mentioned like the biological variation concept (RCV), the CLSI recommendations and the technical state-of-the-art which is based on fixed values [19], [20], [21]. Due to a lack of standardization, in this study we focused primarily on fixed values.

To validate the specifications supplied for the cobas^®pro analyzer (Roche Diagnostics, Mannheim, Germany), we established our own HI thresholds for 18 common laboratory parameters through hemolysis dilution series. This comparison serves to assess the alignment of our calculated thresholds with Roche’s specifications, aiding in the evaluation of the system’s performance and accuracy. In this study the manufacturers HI-thresholds correspond to approximately 10 % deviation of measurement results.

In the following descriptions, we primarily focused on the median of the 5 and 10 % error limits in lithium-heparin plasma unless otherwise specified. When using the median, 50 % of our findings exceed the threshold. Thus, as a more conservative approach, we also included the 5th percentile thresholds, which are detailed in Tables 1 and 2. This dual approach allows for a more comprehensive assessment of the HI thresholds across the measured parameters.

Overall, in line with other authors [22] we found that serum inherently exhibits a higher degree of hemolysis compared to lithium-heparin plasma (n=76 undiluted samples, mean HI in serum: 7.7, mean HI in lithium-heparin: 1.7, p<0.01). This is also reflected in the generally higher HI values for serum (e.g. Tables 1 and 2). If manufacturers do not provide material-specific HI values, using lithium-heparin plasma seems appropriate to reduce unnecessary suppression of test results.

Equal HI thresholds, comparable to those provided by the manufacturer, were observed within this study for ALAT, CK, GLDH, amylase, iron, GGT and phosphate.

For analytes known to be particularly susceptible to hemolysis, such as AST, direct bilirubin, and LDH, we observed similar HI values compared to those provided by the manufacturer. However, a notable discrepancy was identified for potassium. While the manufacturer’s specified HI value is 20, our analysis with a 10 % error tolerance yielded an HI value of 141. Even assuming a maximum error tolerance of 5 % for potassium, and applying the 5th percentile instead of the median, our calculated HI value of 57 remains even significantly higher than the manufacturer’s value. This suggests that a HI value of 20 appears too strict in this context.

For haptoglobin we noted a significant deviation from the manufacturer’s specifications. Roche specifies an HI of 10. This low value is justified by the Glick model, which explains how a haptoglobin-hemoglobin complex in vitro can cause a decrease in haptoglobin levels by 10–15 % [23]. However, our calculated HI values, with a maximum error tolerance of 5–10 %, ranged between 62 and 159. Given that haptoglobin is in particular used as a marker for in vivo hemolysis, an HI value of 10 appears likewise too strict, even when considering the Glick model. This suggests that adjustments to the manufacturer’s specifications may be warranted to better reflect clinical realities and ensure accurate laboratory results.

For creatinine and total bilirubin, our determined HI values are significantly lower than the manufacturer’s specified value of 800. With an error tolerance of 5–10 %, we found HI values for creatinine between 299 (5 %) and 598 (10 %), and for total bilirubin between 166 (5 %) and 333 (10 %). As shown in Table 1, an HI value of 800 would correspond to a deviation of nearly 20 % in this context. Therefore, the HI values specified by the manufacturer should be adjusted to lower levels.

In contrast, we found higher HI values than the manufacturer claims for four more measurands. For chloride (1,000 (Roche) vs. 1957 (5 %)/3,914 (10 %)) and sodium (1,000 (Roche) vs. 1,590 (5 %)/3,181 (10 %)) these higher values are unlikely to play a significant role in laboratory routine. An HI over 1,000 corresponds to a total cell hemolysis which is why the sample is typically rejected by the laboratory before any measurement.

For total protein, we identified a 10 % error limit that was higher by a factor of almost two compared to the manufacturer’s specification. The manufacturer’s specified HI value is 500, whereas our calculated HI value for a 5 % error tolerance is 578, and for a 10 % error tolerance, it is 1,157. Based on these findings, a higher HI limit might be acceptable.

For high-sensitive troponin T, Roche specifies a threshold of 100 mg/dl Hemoglobin (equals HI=100) However, our determined HI-thresholds were consistently higher. For lithium-heparin plasma, with an error tolerance of 5–10 %, we found HI values ranging between 140 and 281, and for serum, between 187 and 374. Therefore, a HI of 100 appears overly stringent.

Troponin T is a time critical measurand and too broad limits for sample rejection may impact patient safety. In addition, it should be noted that a time-dependent effect of proteases on troponin T has been described in further studies, so the age of the sample may also need to be considered [24]. Furthermore, the impact or severity of the medical decision and the biological variability influence the applied acceptance limits. It should also be considered that acceptance limits can be derived from reference change values or permissible uncertainty [19], 20].

The example of troponin illustrates that acceptance limits in general should be determined based on the medical context of a measurand rather than adopting a “one size fits all” approach. Thus, in accordance with the current applicable ESC guidelines [25] the discussion should focus on how a tolerated error limit of 5–10 % affects the diagnosis of an acute myocardial infarction. Therefore, further data should be collected to define well balanced HI-thresholds.

As mentioned before, reliable and traceable manufacturer specifications are important for good medical practice. If laboratories deviate from manufacturer specifications, they are responsible for independently validating their threshold values according to applicable guidelines [11], 13], 18].

Strengths and limitations

With blood from predominantly healthy participants, we were able to determine hemolysis thresholds by preparing and measuring simple hemolysis dilution series. Over 50 % of our threshold values corresponded with those provided by the manufacturer. This demonstrates that our established method is applicable within the context of a routine laboratory setting.

Some limitations need to be considered. In a different context with a higher proportion of pathological values, the calculated HI thresholds might slightly deviate. In addition, it should be noted that a reliable interpretation of the results is only feasible for a maximum HI of 300. Extrapolated values beyond this point should be interpreted with caution. Furthermore, we only compared measurements on the cobas^®pro analyzer (Roche Diagnostics, Mannheim, Germany). Thus, our findings may not directly be transferable to other laboratory systems and underscore the need for platform-specific validation to ensure accuracy and reliability across different analytical instruments.

Conclusions

We established HI thresholds for 18 selected biochemistry assays. For 10 of the assays the calculated thresholds were consistent with those provided by the manufacturer. The remaining eight showed considerable deviations of which the following six may have potential clinical relevance. For creatinine and total bilirubin, the manufacturer’s specifications appear too forgiving. Conversely, for the hemolysis-sensitive measurands potassium and haptoglobin as well as total protein and troponin T we found manufacturer HI thresholds which appear to be too strict.

Based on our findings, we conclude that the specifications provided by the manufacturer do not represent the most suitable HI thresholds in all cases and should be revised by the manufacturer. This study could serve as a blueprint for laboratories to adjust their thresholds as LDTs in line with current guidelines. Adjusting thresholds can help mitigate the risk of blocking results unnecessarily while preventing the release of considerably flawed results.