OUP user menu

Comparative Study of Blood Collection Tubes and Thromboplastin Reagents for Correction of INR Discrepancies
A Proposal for Maximum Allowable Magnesium Contamination in Sodium Citrate Anticoagulant Solutions

Anton M.H.P. van den Besselaar PhD, Anton P. van Zanten PhD, Helen M. Brantjes PhD, Marc G.L.M. Elisen PhD, Felix J.M. van der Meer MD, PhD, Dennis C.W. Poland PhD, Augueste Sturk PhD, Anja Leyte PhD, Ad Castel PhD
DOI: http://dx.doi.org/10.1309/AJCPGSB5YPJRREEV 248-254 First published online: 1 August 2012

Abstract

International normalized ratio (INR) discrepancies were noted between clinical laboratories using various prothrombin time (PT) systems. We studied the influence of different commercial blood collection tubes and different PT systems on INR measurements. INRs of fresh patient samples were determined by 3 laboratories, each using different PT systems. In the first part of the study, samples were drawn with Vacutainer tubes and in the second part with Monovette tubes. In the first part of the study, the maximum bias for all patients amounted to 0.46 INR (14%), and in the second part, to 0.14 INR (4.9%). The maximum bias for all patients could be reduced further by local system calibration using frozen pooled plasma specimens. The sodium citrate solutions in the blood collection tubes were contaminated with magnesium ions (approximately 2.7 mmol/L and 0.3 mmol/L in the Vacutainer and Monovette, respectively). INR discrepancies could be explained largely by this influence of blood collection tubes. The maximum allowable magnesium contamination in sodium citrate anticoagulant solutions should be less than 1 mmol/L.

Key Words
  • Blood collection tubes
  • Prothrombin time
  • International normalized ratio
  • Calibration
  • Anticoagulant
  • Contamination

All medical staff and health auxiliaries involved in controlling oral anticoagulant treatment with vitamin K antagonists should be encouraged to use the international normalized ratio (INR) system.1 This is especially important in the case of any treated patient who is likely to be transferred to another laboratory where a different modification of the prothrombin time (PT) test is likely to be used. The Netherlands has developed a nationwide system of regionally centralized anticoagulant control units for outpatients. Each individual Dutch center is called a Thrombosis Service. Control of in-hospital patients is excluded from the duties of a Dutch Thrombosis Service.2 When an outpatient controlled by a Thrombosis Service must undergo a medical procedure in a hospital, the anticoagulant control is deputed to the hospital’s laboratory. We observed an INR discrepancy between a Thrombosis Service and a neighboring hospital laboratory, each using a different reagent-instrument combination. The INR measured by the hospital laboratory was systematically lower than the value provided by the Thrombosis Service. The discrepancy gave rise to postponement of elective cardioversion for a number of patients with atrial fibrillation because the hospital preferred the INR to be higher than 2.5 in the period closest to cardioversion.3 Approximately 3% of planned cardioversions were cancelled because of INR discrepancies. The purpose of the current study was to find the origin of the discrepancies and to suggest potential corrective measures.

Materials and Methods

Blood Collection and Processing

The patients included in this study were recruited by 2 large anticoagulant units, ie, the Thrombosis Service of the Medical Center Zuiderzee at Lelystad (center A) and the Thrombosis Service of Leiden (center B). They were outpatients receiving long-term treatment with vitamin K antagonists, either acenocoumarol or phenprocoumon. The patients were not selected based on indication or target intensity of anticoagulation. All patients gave informed written consent. In the first part of the study, center A used plastic Vacutainer tubes containing 0.109 mol/L sodium citrate for venous blood collection (1.8 mL draw; Ref 363047; Becton Dickinson, Plymouth, England). In the second part, center B used plastic Monovette tubes containing 0.106 mol/L sodium citrate for blood collection (4.3 mL draw; Ref 04.1922.001; Sarstedt, Nümbrecht, Germany). After collection, the blood specimens were transported to the laboratory of each center and centrifuged. The patient’s plasma specimen was taken off the cell layer and divided into 3 equal plastic tubes. The 3 sets of fresh plasma specimens were transported to 3 different laboratories (centers C, D, and E) for PT/INR assessment within 5 hours. The samples were transported by car at room temperature.

PT Assessments

PT was assessed by each laboratory using the laboratory’s routine method. Center C used the reagent STA Hepato Quick (Roche Diagnostics, Mannheim, Germany) on a STA-R Evolution instrument (Diagnostica Stago, Asnières sur Seine, France). Center D used the reagent Innovin (Siemens Healthcare Diagnostics Products, Marburg, Germany) on a Sysmex CA-1500 instrument (Toa Medical Electronics, Kobe, Japan). Center E used the reagent HemosIL Recombi-PlasTin 2G (Instrumentation Laboratory, Bedford, MA) on an ACL-9000 instrument (Instrumentation Laboratory). PTs were reported by centers C, D, and E in both seconds and INRs. All calculations were performed centrally by one of us (A.M.H.P.V).

Assessment of Anticoagulation Stability

Each patient’s anticoagulation stability was assessed by evaluating the last 3 INR records. Anticoagulation of a patient was regarded as stable if treatment was maintained for at least 6 weeks and if the last 3 INRs were within the appropriate therapeutic range.

Local PT System Calibration

The local PT system was calibrated according to the “direct” INR method with 6 frozen pooled citrate plasma specimens.4,5 One pooled plasma specimen was prepared by mixing 30 normal donor plasma specimens. Five pooled plasma specimens were prepared by mixing approximately 200 individual plasma specimens from patients treated with vitamin K antagonists. The mean INRs for the latter pooled plasma specimens ranged from approximately 1.7 to approximately 3.4. Reagent-specific INR values for the pooled plasmas were assigned and validated as described previously.4,5 The reagent-instrument combinations used for the INR assignment were calibrated according to the International Sensitivity Index (ISI) model with the appropriate international standards for thromboplastin, ie, rTF/95 and RBT/05. The mean normal PT for calculation of the INR for each frozen pooled plasma specimen was determined using at least 20 fresh healthy individual specimens. Each laboratory (C, D, E) analyzed the 6 frozen pooled plasma specimens using their routine methods in duplicate on 3 days. According to the direct INR method, an orthogonal regression line was calculated for log(INR) vs log(PT). INRs of patient plasma specimens could be interpolated directly from local PT results using this line.4,5

Statistical Methods

INRs calculated with different reagent-instrument combinations were compared using the Student paired t test. In addition, INR differences between various methods were analyzed with the method of Bland and Altman.6

Because some of the relationships between INRs assessed with different reagents appeared to be curved, an alternate model for INR calculation was tried. The Tomenson7 model was used to describe the relationship between PTs assessed with one reagent and INRs assessed with another reagent. The Tomenson model can be described by the following equation: INR=ed×Rb=ed×(MNPT)b×(PT)b Equation 1in which MNPT is the mean normal PT, R is the PT ratio (= PT/MNPT), b, the slope of the orthogonal regression line for the patients’ samples alone (ie, without normal plasma specimens), and d, a scale parameter. Equation 1 can be written in logarithmic form as follows: ln (INR)=db×ln (MNPT)+b×ln (PT) Equation 2

Magnesium Analysis

Sodium citrate solutions were analyzed for magnesium content by means of atomic absorption spectroscopy and by using routine laboratory colorimetric methods on a Modular Analytics P800 System (Roche Diagnostics, The Netherlands).

Results

The present study was conducted in 2 parts. In the first part, blood specimens of 1 group of outpatients were collected with Vacutainer tubes and analyzed by 3 laboratories. The laboratories reported measured PTs as well as INRs. Reported INRs were calculated with each center’s routine ISI and mean normal PT Table 1. Pearson correlation coefficients for all patients were 0.962 (Hepato Quick vs Innovin), 0.949 (Hepato Quick vs RecombiPlasTin 2G), and 0.988 (Innovin vs RecombiPlasTin 2G). The INR differences between Hepato Quick and Innovin and RecombiPlasTin 2G were significant with the Student paired t test. The maximum mean bias of all reported INRs amounted to 0.46 (14%). The stability of anticoagulation was assessed retrospectively for all the patients included. For the group of 84 stable patients, the bias of the reported INR between Hepato Quick and RecombiPlasTin 2G was 0.51 (15%). Apparently, the stability of anticoagulant treatment had little effect on the INR bias.

Furthermore, INRs were recalculated after local PT system calibration. Local PT system calibration had hardly any effect on the INRs assessed by center C using Hepato Quick. Local PT system calibration in centers D and E resulted in higher INRs than the reported values. The maximum INR bias was reduced by local PT system calibration.

In the second part of the study, blood specimens of another group of outpatients were collected with a different system, ie, Monovette tubes. The mean reported INRs and the corrected INRs after local calibration are shown in Table 2. Although the INR differences between the reagents were statistically significant in several instances, they appeared to be smaller than in the first part of the study. The maximum bias of the reported INR amounted to 0.14 (4.9%) for all patients and 0.19 (6.9%) for the group of stable patients. After local PT system calibration, the mean INR difference between Hepato Quick and Innovin was only 0.02 (0.7%) for the group of 90 stable patients (not significant). After local PT system calibration, the INR bias increased for the group of not-stable patients (maximum bias, 6.5%).

We calculated the number of INR discrepancies between centers, based on a critical INR value of 2. Because the INRs assessed with Hepato Quick were systematically higher than those with the other 2 thromboplastin reagents, we counted the number of patients with INRs (Hepato Quick) more than or equal to 2 and INRs (other reagent) less than 2, both before and after local PT system calibration Table 3. The number of discrepancies was reduced after local PT system calibration in both parts of the study.

To analyze the INR differences further, Bland-Altman plots were constructed. There was a significant negative trend in the INR difference between Hepato Quick and Innovin, both before and after local PT system calibration Figure 1. A similar trend was observed for the INR difference between Hepato Quick and RecombiPlasTin 2G Figure 2. These observations suggested that the ISI model for INR calculation is not completely true for these PT systems. Therefore we tried an alternative calibration model proposed by Tomenson.7 In Tomenson’s model a linear relationship is calculated between the logarithms of the patients’ PT or INR with the reference system and the logarithms of the PT with the system to be calibrated. We chose Hepato Quick as the reference system and Innovin or RecombiPlasTin 2G as the system to be calibrated. The data were fitted to Tomenson’s model using orthogonal regression analysis. In doing so, the following equations were obtained: INR=exp(1.66+0.79×ln PTINNOVIN, Center D) Equation 3and INR=exp(1.81+0.86×ln PTRECOMBIPLASTIN 2G, Center E) Equation 4

View this table:
Table 1
View this table:
Table 2
View this table:
Table 3

Bland-Altman plots were constructed for the INRs with Hepato Quick and the INRs calculated with the aforementioned equations according to Tomenson’s model Figure 3 and Figure 4. No significant trend of the INR difference with the mean INR was observed. The range of INR differences obtained with Tomenson’s model was smaller than the range of differences obtained with the ISI model.

Magnesium concentrations were measured in the sodium citrate solutions from 2 tubes of each commercial blood collection system using atomic absorption spectroscopy and a routine colorimetric method. Both methods gave similar results Table 4. The magnesium concentrations in the Monovette tubes were approximately 10-fold lower than the concentrations in the Vacutainer tubes.

Figure 1

Bland-Altman plot of reported international normalized ratio (INR) differences between centers C (Hepato Quick) and D (Innovin) for all patient samples collected with Monovette tubes (second part of the study). The linear regression line for the INR differences showed a negative trend. The correlation was significant (P < .001). R2 = 0.387. For comparison, the horizontal line for zero difference is shown as well.

Figure 2

Bland-Altman plot of reported international normalized ratio (INR) differences between centers C (Hepato Quick) and E (RecombiPlasTin 2G) for all patient samples collected with Monovette tubes (second part of the study). The linear regression line for the INR differences showed a negative trend. The correlation was significant (P < .001). R2 = 0.094. For comparison, the horizontal line for zero difference is shown as well.

Figure 3

Bland-Altman plot of international normalized ratio (INR) differences between centers C (Hepato Quick) and D (Innovin) for all patient samples collected with Monovette tubes (second part of the study). INRs for Innovin were calculated with Tomenson’s model (equation 3). The linear regression line for the INR differences did not show a significant trend. R2 = 2.271E-4. For comparison, the horizontal line for zero difference is shown as well.

Discussion

The present study was performed with 2 different groups of outpatients treated with vitamin K antagonists. Each group was controlled by a different thrombosis center using different blood collection and PT systems (ie, reagent-analyzer combination). The 2 thrombosis centers were similar with regard to their therapeutic target ranges. The purpose of our study was to assess the magnitude of INR differences generated by using different blood collection tubes and different PT systems. Although we did not compare the Vacutainer and Monovette blood collection tubes directly by drawing blood specimens from the same patients into the 2 tubes, we could compare the differences observed with 3 different PT reagents. If the Vacutainer and Monovette tubes would have been equivalent, the results obtained with the 3 reagents would not have been different. In contrast, the INR differences between the reagents for the Vacutainer samples were larger than the corresponding differences for the Monovette samples (compare Tables 1 and 2). The INR differences between the Vacutainer and Monovette samples are likely to be explained by the different levels of contamination with magnesium ions in the sodium citrate solutions (Table 4). The Vacutainer solutions contained approximately 10 times higher magnesium concentrations than the Monovette solutions. The PT and INR is shortened by exogenous magnesium ions.8 Previous studies have shown that human recombinant thromboplastins are more responsive to magnesium ions than rabbit thromboplastins.9

Figure 4

Bland-Altman plot of international normalized ratio (INR) differences between centers C (Hepato Quick) and E (RecombiPlasTin 2G) for all patient samples collected with Monovette tubes (second part of the study). INRs for RecombiPlasTin 2G were calculated with Tomenson’s model (equation 4). The linear regression line for the INR differences did not show a significant trend. Squared coefficient of linear correlation R2 = 3.341E-4. For comparison, the horizontal line for zero difference is shown as well.

View this table:
Table 4

We chose Hepato Quick as a reference reagent in the present study because its PTs were hardly affected by exogenous magnesium ions.10 Hepato Quick consists of rabbit tissue factor combined with adsorbed bovine plasma. Furthermore, the plasma sample is prediluted in the reaction mixture so that the concentration of magnesium ions is reduced as well. In contrast, PTs and INRs assessed with the human thromboplastin reagents Innovin and RecombiPlasTin were shortened by magnesium.10,11 The current study showed that PTs assessed with RecombiPlasTin 2G were also shortened by magnesium in a similar way as Innovin clotting times. The properties of Innovin and RecombiPlasTin 2G are similar because both are recombinant human tissue factor preparations. The larger similarity is also reflected by the high Pearson correlation coefficient for their INRs (0.988).

Comparison of Tables 1 and 2 shows that with Hepato Quick, the mean INR of the patients evaluated by center A is higher than that of the patients evaluated by center B. The difference may be explained by the fact that center A was using Vacutainer tubes for routine blood collection in combination with Innovin for routine laboratory control, whereas center B was using Monovette tubes in combination with Hepato Quick for laboratory control. The difference in the mean INR probably reflects a real but so far undetected difference in treatment intensity between the 2 centers.

The relationship between INR (Hepato Quick) and INR (Innovin) based on the ISI calibration model appeared to be slightly curved. The curvature of the relationship is reflected by a significant negative slope of the difference plot shown in Figure 1. In other words, the INR bias depends on the level of the INR. Similar curvature was observed in the relationship between INR (Hepato Quick) and INR (RecombiPlasTin 2G), as reflected in the significant slope of the corresponding difference plot (Figure 2). Apparently the ISI model is not completely true for one or more of the aforementioned thromboplastin reagents. It is not known why there is a curvature in the relationship between Hepato Quick and Innovin (or RecombiPlasTin 2G), but it may be related to the different composition of these reagents. Another case of a curved relationship was reported by Norwegian investigators for Normotest (rabbit brain thromboplastin) calibrated with OBT/79 (bovine brain thromboplastin).12 In the current study, we used INRs assessed with Hepato Quick as the reference values, but this was a provisional choice. By using Tomenson’s model it is possible to correct for the curvature so that better agreement is obtained between the INRs. The better agreement is demonstrated with the Bland-Altman plots (Figures 3 and 4) showing no significant trends in the INR differences.

The frozen pooled plasma specimens for local PT system calibration had reagent-specific values. Furthermore, these values had been assigned with the ISI model. Local PT system calibration using the original assigned INR values could not correct for the curvature in the relationships between Hepato Quick and Innovin (or RecombiPlasTin 2G). It was then possible to recalculate the INR values for the frozen pooled plasma specimens using equations 3 and 4. The corrected INR values for the frozen pooled plasmas might be used in the future by other laboratories using Innovin or RecombiPlasTin 2G for local calibration using Tomenson’s model, so that INRs would be in better agreement with Hepato Quick.

It should be noted that local calibration cannot correct for INR differences caused by magnesium contamination in blood collection tubes. Because it is probably impossible to manufacture commercial tubes that are completely deficient in magnesium, an upper limit may be advised. Previous studies suggested an upper limit of 1 mmol/L magnesium.8 The Section Coagulation of the Dutch Foundation for Quality Assessment in Clinical Laboratories advised against the use of blood collection tubes containing 1 mmol/L or more magnesium ions in the citrate solution.

In conclusion, we have shown that INR discrepancies between laboratories may be caused in part by blood collection tubes with different levels of magnesium contamination. Tubes with excessive contamination should be replaced by tubes containing less than 1 mmol/L magnesium. In addition, minor INR differences may be caused by inaccurate calibration of the reagent-instrument combination. Local PT system calibration with a set of deep-frozen plasma specimens gave rise to partial correction of the INR bias. The ISI calibration model may not be accurate for particular reagent-instrument combinations. In those cases, an alternative calibration system using the Tomenson model may be appropriate.

Because of registration and introduction on the market of new oral anticoagulants such as dabigatran, rivaroxaban, and apixaban the use of vitamin K antagonists is expected to decrease in the coming years. However, the vitamin K antagonists are expected to be used for at least the next 5 to 10 years, and as long as they are being used, INR assessments should be optimal.

Acknowledgments

Technical assistance was provided by E. Witteveen, J.H. Didden, Y. van der Heide, M. Spanjersberg, E. Spruijt, T. Booij, P. Appelman, A. Tol, P. van der Pol. We thank P. van ‘t Sant, PhD, for performing the atomic absorption spectroscopy.

Footnotes

  • This study was supported by the Section Coagulation of the Dutch Foundation for Quality Assessment in Clinical Laboratories.

References

  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
View Abstract