Background: C-Reactive protein (CRP) can provide prognostic information about risk of future coronary events in apparently healthy subjects. This application requires higher sensitivity assays than have traditionally been available in the clinical laboratory.
Methods: Nine high-sensitivity CRP (hs-CRP) methods from Dade Behring, Daiichi, Denka Seiken, Diagnostic Products Corporation, Iatron, Kamiya, Olympus, Roche, and Wako were evaluated for limit of detection, linearity, precision, prozone effect, and comparability with samples from 388 apparently healthy individuals.
Results: All methods had limits of detection that were lower than the manufacturers’ claimed limit of quantification except for the Kamiya, Roche, and Wako methods. All methods were linear at 0.3–10 mg/L. The Diagnostic Products Corporation, Kamiya, Olympus, and Wako methods had imprecision (CVs) >10% at 0.15 mg/L. The Iatron, Olympus, and Wako methods demonstrated prozone effects at hs-CRP concentrations of 12, 206, and 117 mg/L, respectively. hs-CRP concentrations demarcating each quartile in a healthy population were method-dependent. Ninety-two to 95% of subjects were classified into the same quartile of hs-CRP established by the Dade Behring method by the Denka Seiken, Diagnostic Products Corporation, Iatron, and Wako methods. In contrast, 68–77% of subjects were classified into the same quartile by the Daiichi, Kamiya, Olympus, and Roche methods. No subject varied by more than one quartile by any method.
Conclusions: Four of the nine examined hs-CRP methods classified apparently healthy subjects into quartiles of hs-CRP similar to the classifications assigned by the comparison method. Additional standardization efforts are required because an individual patient’s results will be interpreted using population-based cutpoints.
C-Reactive protein (CRP)1 is a pentameric acute phase reactant that is synthesized by the liver. Its production is controlled primarily by interleukin-6. The serum CRP concentration may increase by up to 1000-fold with infection, trauma, surgery, and other acute inflammatory events. Chronic inflammatory disorders, including autoimmune diseases and malignancy, can produce persistent increases of serum CRP concentrations. Traditionally, CRP has been used clinically for diagnosis and monitoring of autoimmune and infectious disorders. Routine automated methods for CRP quantification in the clinical laboratory typically have limits of quantification of 3–8 mg/L.
Chronic inflammation is an important component in the development and progression of atherosclerosis [for recent reviews, see Refs. (1)(2)]. Numerous epidemiologic studies have demonstrated that increased serum CRP concentrations are positively associated with risk of future coronary events. Included in this list are several studies conducted in large populations of apparently healthy men and women who subsequently developed coronary artery disease, cerebrovascular disease, or peripheral arterial disease (3)(4)(5)(6)(7)(8)(9). CRP has also been shown to be predictive of future events in patients with acute coronary syndromes and in patients with stable angina and coronary artery stents (10)(11)(12)(13)(14)(15).
Different methods for quantifying CRP in serum have been used. Studies conducted with apparently healthy individuals require high-sensitivity CRP (hs-CRP) methods. Relative risk of future coronary events can then be determined using hs-CRP cutpoints established in prospective epidemiologic studies. These high-sensitivity methods initially used ELISA methodology, and a single in-house ELISA assay was used for several population studies (3)(4)(16). This methodology is primarily for research and is not ideal for routine use in highly automated clinical laboratories. Traditional CRP methods in the clinical laboratory lack the desired sensitivity and, therefore, are unsuitable for the purpose of predicting future risk of coronary events in apparently healthy individuals. A latex-enhanced immunonephelometric hs-CRP method has recently been evaluated and validated clinically (17)(18). More recently, several automated immunoturbidimetric and immunoluminometric hs-CRP assays have been developed and are commercially available. These assays possess improved sensitivity and precision at low concentrations of CRP. In this study, we build on an earlier report (19) and describe the performance characteristics of nine hs-CRP methods, including method comparability, using samples from 388 adult blood donors.
Materials and Methods
A total of 388 serum samples were collected from 208 male and 180 female blood donors. The median age of the entire group was 32 years with a range of 17–73 years. Values ranged from the limit of detection to 50 mg/L. A second group of eight serum samples from patients with CRP concentrations ≥50 mg/L were used to investigate the prozone and high-dose hook effects. Serum was separated from the red cells and stored at −70 °C until analysis. All studies with samples from human subjects were approved by the Institutional Review Board of the University of Utah Health Sciences Center.
The BN II nephelometer was from Dade Behring, the IMMULITE 2000 analyzer was from Diagnostics Products Corporation (DPC), the Hitachi 911 and 917 analyzers were from Roche Diagnostics, and the Olympus AU 640 analyzer was from Olympus USA.
Information for each of the nine methods is summarized in Table 1⇓ . In all cases the manufacturers’ reagents were used as directed. Analyses by the DPC method were performed using manual sample dilution because of the limited sample volumes that were available for this study. The Dade Behring BN II N High Sensitivity CRP assay was used as the comparison method based on previous analytical and clinical validations (17)(18)(19). At the time of manuscript submission, only the Dade Behring and Kamiya methods had been approved by the Food and Drug Administration (FDA) for clinical use in the United States, and only the Dade Behring assay had been approved by the FDA for use in assessing the risk of cardiovascular and peripheral vascular disease
The limit of detection of each method was assessed by analyzing a zero calibrator 20 times and calculating a 2 SD limit. Samples for linearity and lower limits of quantification were prepared from two serum pools. The low pool was prepared by combining samples from blood donors with hs-CRP concentrations in the lowest quartile. The high pool was prepared by combining patient samples with hs-CRP concentrations of ∼10 mg/L. The high pool was diluted with the low pool to the following final percentages of the high pool: 100%, 75%, 50%, 30%, 20%, 10%, 5%, 2.5%, and 0%. Samples were assayed in duplicate in one analytical run. Five samples were used for imprecision studies. Level 1 and level 5 were prepared using a solution containing 60 g/L bovine serum albumin and adding purified CRP (Sigma). Levels 2 and 4 were Kamiya calibrators with assigned values of 0.5 and 1.5 mg/L, respectively. Level 3 was serum obtained from a single subject. Each sample was run in duplicate on 5 different days, and the total imprecision was calculated.
EP Evaluator-CLIA software (David G. Rhoads Associates, Kennett Square, PA) was used for Deming regression analysis, calculation of r and Sy|x, and linearity assessment. Identical representative standard deviations were used for the x and y methods. An allowable systematic error limit of 10% was chosen for the linearity assessment, and the limit of quantification was the lowest measured concentration that fulfilled this criterion. hs-CRP concentrations were skewed rightward in samples from blood donors; therefore, percentile values were estimated and hs-CRP concentrations were log-transformed for method comparison plots. Probabilities for the t-test were calculated using Primer of Bio-statistics: The Program, Ver. 4.02 (Stanton A. Glantz, author, McGraw-Hill Companies).
The limits of detection that were determined for all methods are summarized in Table 1⇑ . The limits of detection of the Kamiya (0.32 mg/L), Roche (0.21 mg/L), and Wako (0.06 mg/L) assays did not achieve the manufacturers’ claimed limit of detection, although the Wako method was very close to that claimed. These minor discrepancies could be related to how the limit of detection was defined by the manufacturer or to the particular analyzer used in this study.
To examine the linearity of each method, samples were prepared from two serum pools as described in Materials and Methods. All assays were linear to the lowest concentration tested, which was a pool of the samples from the lowest quartile of hs-CRP concentrations from blood donors, except for the Denka method, which was linear down to 0.23 mg/L, using a 10% systematic error limit (Table 1⇑ ). The range of values observed for the low pool was 0.07–0.30 mg/L, and the range of values observed for the high pool was 8.01–12.1 mg/L. Regression analysis of the data yielded intercepts of 0.00–0.30 mg/L, which were comparable to the measured values for the low pool (data not shown).
To quantify the imprecision of the nine methods in the hs-CRP concentration range of the reference interval, samples containing five concentrations of hs-CRP were assayed in duplicate on 5 different days (Table 2⇓ ). The DPC, Kamiya, Olympus, and Wako methods all had CVs >10% at a CRP concentration of 0.15 mg/L. The imprecision of each method improved at a CRP concentration of 0.5 mg/L, and all assays had CVs <10% at this concentration except for the Olympus method. This method achieved a CV <10% at a CRP concentration of 0.6 mg/L.
Samples with very high CRP concentrations were analyzed to test each method for susceptibility to falsely low results with very high CRP concentrations caused by either a prozone effect for the single antibody nephelometric and turbidimetric methods or a high-dose hook effect for the double antibody luminometric method. The DPC method showed no evidence of a hook effect at the highest CRP concentration tested, which was 480 mg/L. The Dade Behring, Daiichi, Denka Seiken, Kamiya, and Roche methods showed no evidence of a prozone effect at CRP concentrations up to 480 mg/L. The Iatron, Olympus, and Wako methods showed evidence of prozone effects at CRP concentrations of <50, 206, and 117 mg/L, respectively. Additional samples were analyzed, and the Iatron method demonstrated a prozone effect at CRP concentrations as low as 12 mg/L.
The hs-CRP concentrations of 388 serum samples collected from apparently healthy adult blood donors were measured by all nine methods simultaneously. Inspection of the data revealed a highly skewed population. Therefore, values for the 25th, 50th, 75th, 80th, 90th, 95th, and 97.5th percentile were determined for each method (Table 3⇓ ).
The agreement of the nine methods with samples from apparently healthy individuals donors was assessed graphically (Fig. 1⇓ ). The Dade Behring method had previously been compared with an in-house ELISA method that was used in several hs-CRP epidemiologic studies and has been validated clinically (6)(17)(18); it currently is being used in several prospective epidemiological studies and clinical trials, including the Nurses’ Health Study, Women’s Health Study, Health Professionals’ Study, Women’s Health Initiative, Air Force/Texas Coronary Atherosclerosis Prevention Study, and Cholesterol and Recurrent Events. Furthermore, at the time this study was initiated, it was the only method approved by the FDA for cardiovascular and peripheral vascular risk assessment. Therefore, it was chosen as the comparison method for evaluating the other eight methods. Visual inspection of the log-log plots indicated good concordance between the Denka Seiken, DPC, Iatron, Kamiya, Roche, and Wako methods and the comparison method (Fig. 1⇓ ). The Daiichi and Olympus methods showed more scatter for results that were less than the median value. Deming regression analysis was performed on all data before log transformation (Table 4⇓ ).
To better compare concordance among methods, the distribution of hs-CRP results from the studied population measured by each of the eight new methods was examined within the quartile cutpoints established by the Dade Behring assay. The results of this analysis (Fig. 2⇓ ) indicate there are two groups of methods. The first group, which agrees the best with the Dade Behring method, included the Denka Seiken, DPC, Iatron, and Wako methods. The agreement between these methods and the comparison method in quartile assignments was 92–95%. The second group, which gave slightly higher hs-CRP results, consisted of the Daiichi, Olympus, Kamiya, and Roche methods. The agreement between these methods and the comparison method in quartile assignments was 68–77%. However, in no case did any result disagree with the quartile assignment of the Dade Behring method by more than one quartile. Statistical analysis of the agreement between the Dade Behring method and the other eight methods for each quartile was performed (20). The results (Table 5⇓ ) indicate that the mean differences for the lowest three quartile were <0.1 mg/L for the Denka Seiken, DPC, Iatron, and Wako methods and >0.1 mg/L for the Daiichi, Kamiya, Olympus, and Roche methods. The Daiichi, Kamiya, Olympus, and Roche methods showed statistically significant differences from the comparative method (P <0.001) for the three lowest quartiles. The Denka Seiken method showed a statistically significant difference from the comparative method for the second quartile (P = 0.022). The Iatron method showed a statistically significant difference from the comparative method for the first quartile (P = 0.048).
The analytic performance requirements for CRP assays have changed as new clinical applications have been developed. hs-CRP assays are necessary for atherosclerotic risk prediction in apparently healthy adults. Of the nine methods we evaluated, all had limits of detection <0.2 mg/L except for the Kamiya and Roche assays, which had limits of detection of 0.32 and 0.21 mg/L, respectively. However, the limits of quantification based on an allowable systematic error limit of 10% for linearity for both of these assays were slightly lower at 0.25 and 0.19 mg/L, respectively (Table 1⇑ ), and both assays had CVs ≤15% at these CRP concentrations (Table 2⇑ ). Criteria for both accuracy and precision need to be clearly defined. hs-CRP results will be interpreted in quartiles or quintiles for risk assessment (21). Therefore, hs-CRP assays will need to be standardized for concentrations of 0.2–10 mg/L so that results obtained in large population studies can be applied to individual patients. All nine methods we evaluated were linear over this concentration range.
A definition of functional assay sensitivity similar to that used for measurement of thyroid-stimulating hormone is required to ensure that assays have the requisite precision at low CRP concentrations. We previously proposed that for risk stratification for cardiovascular, cerebrovascular, and peripheral vascular disease, the hs-CRP assay imprecision should be <10% at a concentration of 0.2 mg/L (19). The methods we evaluated all had CVs <10% at an hs-CRP concentration of 0.15 mg/L except for the DPC, Kamiya, Olympus, and Wako methods. This lack of precision at low CRP concentrations was most noticeable for the Olympus method when results from blood donors were compared (Fig. 2F⇑ ). When reviewing the precision data in Table 2⇑ , it is noteworthy that different protein-based matrices yielded different relative recoveries. Levels 2 and 4 were calibrators from the Kamiya method with assigned values of 0.5 and 1.5 mg/L, respectively. Mean values determined by each of the nine methods were within −18% and 14% of the mean of all methods. For level 3, which was serum from a single subject, mean values for each method ranged from −31% to 28% of the mean calculated for all methods. These results suggest that even liquid-stabilized protein-based calibrators do not simulate the matrix of human serum samples. These matrix effects may contribute to the lack of standardization between methods.
Several previous studies that examined serum hs-CRP concentrations in apparently healthy populations using highly sensitive ELISA, nephelometric, and turbidimetric methods found median values ranging from 0.58 to 1.13 mg/L (3)(5)(17)(22)(23)(24)(25). Median values determined by the nine methods examined here were 0.78–1.14 mg/L, consistent with these earlier studies. Four of these studies found 75th percentile values of 1.44–2.10 mg/L, whereas we found values of 1.89–2.49 mg/L, consistent with the earlier studies (3)(23)(25). Two previous studies found the 90th percentile hs-CRP concentration to be 3 mg/L, and a third found a range for four methods of 4.1–5.3 mg/L, whereas we found values of 4.41–5.43 mg/L for the nine methods we investigated, consistent with earlier studies (19)(22)(23). Differences between the Dade Behring comparison method and results from the other eight methods for the 25th percentile ranged from −20% to 38% of the comparison method, results for the 50th percentile ranged from −13% to 27% of the comparison method, and results for the 75th percentile ranged from −6% to 25% of the comparison method. These results indicate that further standardization efforts are required if hs-CRP results from different methods are to be used interchangeably. These standardization activities will be similar to those conducted for cholesterol, whose concentration is also interpreted for the individual patient based on population-based cutpoints.
According to information provided by each of the assay manufacturers, all of the methods we investigated were standardized against the IFCC Certified Reference Material (CRM) 470 standard rather than against the older WHO International Reference Standard for CRP Immunoassay 85/506. The methods evaluated in this study could be divided into two groups based on their ability to classify subjects into quartiles of hs-CRP. For one group of methods, which included the Denka Seiken, DPC, Iatron, and Wako assays, 92–95% of subjects were classified into the same quartile of hs-CRP concentrations. Furthermore, these assays showed minimal or negative bias relative to the comparison method. In the other group, which included the Daiichi, Kamiya, Olympus, and Roche assays, 68–77% of subjects were classified into the same quartile of hs-CRP. In addition, substantial positive bias relative to the comparison method was noted. hs-CRP concentrations that correspond to the 75th percentile for this latter group are comparable to 80th percentile concentrations for the former group. It is important to indicate, however, that none of the examined subjects was misclassified by more than one quartile. Therefore, although more standardization is needed to harmonize the results from the various methods, the performance of the examined assays in this report is promising.
A previous comparison between the BN II and Hemagen ELISA methods showed a slope of 0.75 and an intercept of −0.25 mg/L (17). The BN II assay was standardized against CRM 470, whereas the Hemagen ELISA was standardized the WHO CRP reference material (24). Differences in standardization materials or the use of suboptimal value transfer protocols likely explain the difference between these two methods. We have made a similar observation with a different group of hs-CRP assays, all which claim to be standardized against the same CRM 470 material. These findings further support the earlier notion of standardizing hs-CRP assays at concentrations comparable to those seen in healthy subjects.
Another issue that merits discussion is the possibility of underestimating the true CRP concentration because of a prozone effect. The Iatron, Olympus, and Wako methods are particularly susceptible to this problem. A review of 2222 results for hs-CRP performed in a reference laboratory revealed that 55 (2.5%) had values >50 mg/L and 21 (0.9%) had values >100 mg/L. Ideally, one might have one CRP method that could routinely provide both high sensitivity and traditional measurements. This option might minimize confusion in ordering. Physicians could order a single test and obtain either an hs-CRP result for atherosclerotic risk prediction or a higher CRP result as an indicator of more severe inflammation. Of the methods examined in this report, the Dade Behring, Daiichi, and DPC assays currently best fulfill this criterion.
In conclusion, the nine examined methods exhibited some differences in their ability to classify apparently healthy subjects into quartiles of hs-CRP concentrations. Additional standardization efforts are required to further ensure that results obtained by automated hs-CRP methods on an individual patient can be interpreted using cutpoints established by prospective epidemiologic studies. Once standardization has been achieved, hs-CRP assays can provide useful data for coronary risk stratification in apparently healthy individuals.
Support for this study was provided by the ARUP Institute for Clinical & Experimental Pathology (Salt Lake City, UT), Daiichi (Tokyo, Japan), Denka Seiken (Tokyo, Japan), Diagnostic Products Corporation (Los Angeles, CA), Iatron (Tokyo, Japan), Olympus America (Melville, NY), Roche Diagnostics (Indianapolis, IN), and Wako Chemicals USA (Richmond, VA).
1 IN, immunonephelometric; IT, immunoturbidimetric; IL, immunoluminometric.
2 The limits of detection and quantification were determined as described in Materials and Methods.
1 CRP was quantified in samples from 388 blood donors. Of these, 180 were female and 208 were male.
1 The Dade Behring assay was the comparison method.
1 Samples were assigned to a quartile based on their BN II result. The mean and SD of the difference between each method and the BN II method were calculated for each quartile.
2 A t-test was performed to determine whether the means of each quartile were statistically significantly different from those obtained by the BN II method.
↵1 Nonstandard abbreviations: CRP, C-reactive protein; hs-CRP, high-sensitivity CRP; DPC, Diagnostic Products Corporation; FDA, Food and Drug Administration; and CRM, Certified Reference Material.
- © 2001 The American Association for Clinical Chemistry