Simplified interpretative format for assessing test interference: studies with hemoglobin-based oxygen carrier solutions

Abstract

Substances such as hemoglobin that interfere with analytical processes are recognized as a frequent source of error in laboratory medicine. Standard guidelines for assessment of test interferences assume that interference effects are not related to the concentration of the analyte being measured. However, previous investigations have demonstrated that interference effects can be markedly different, depending on the concentrations of interferent and analyte within the specimen. An experimental protocol for investigating these different types of interference effects has been developed. This protocol utilizes an orthogonally arranged matrix with progressively increasing concentrations of analyte and interferent. Evaluation of the measured analyte concentrations in specimens within the matrix using multiple regression analysis allows the magnitude, direction, and significance of each type of interference to be determined. Unfortunately, implementation of the interference data derived from the multiple regression analysis for judging the clinical acceptability of test results when an interferent is present is difficult. We describe a two-dimensional graphical format for evaluating the clinical acceptability of test results, based on criteria established under the Clinical Laboratory Improvement Amendments of 1988, in specimens containing hemoglobin-based oxygen carrier solutions.

The clinical utility of laboratory test results depends on the accuracy and precision with which the analyte of interest can be measured. The presence of certain endogenous or exogenous substances in body fluids can adversely affect the determination of many analytes in the clinical chemistry laboratory (1). Guidelines have been established for assessment of the effects of interfering substances on clinical laboratory tests, and a variety of experimental designs have been advocated (2)(3)(4)(5). Unfortunately, these models for evaluating interference assume that interference effects are independent of the concentration of the analyte.

Kroll et al. (6) previously have shown that at least three types of interference can contribute to the overall interference effect. These three types of interference have been classified as analyte-dependent, where the magnitude of interference is dependent on the concentration of the analyte of interest; analyte-independent, where the degree of interference is constant regardless of the analyte concentration; and a combination of the first two, where the effect of an interferent is dependent upon both the concentration of analyte and interferent.

An experimental model for assessing the relative contribution of each of these three types of interference has been described (6). This technique offers a rigorous statistical approach for assessing interference and also provides insight into the mechanism of the interference. Unfortunately, routine implementation of the information derived using this approach for judging the acceptability of laboratory test results when an interferent is present is difficult. It is especially difficult in those cases where the interference is dependent on the concentration of analyte or on both the analyte and interferent concentrations.

The imminent widespread use of hemoglobin-based oxygen carrier solutions in patients requiring blood transfusion will require that clinical laboratories have mechanisms in place to evaluate the acceptability of specimens for analysis which contain this possible interferent. Hemoglobin-based blood substitutes may be given in amounts producing serum hemoglobin concentrations up to 50 g/L (7). Our investigations into the interference effects of diaspirin cross-linked hemoglobin chains (DCLHb; also known as HemAssist^®; Baxter Healthcare Corp., Deerfield, IL) on laboratory testing methods using the method of Kroll et al. (6) have led to the development of a format for assessing interference that can be readily implemented in the clinical laboratory. We report on an easy-to-read two-dimensional graphical display that shows where clinically significant interference occurs as a function of analyte and interferent concentrations. In particular, we evaluated the interference effect of DCLHb on four different chemistry assays.

Materials and Methods

apparatus and analytical methods

We measured total calcium, cholesterol, total bilirubin, and potassium, using a Hitachi 747 (Boehringer Mannheim Corp.) with reagents supplied by the manufacturer. The methods used for measurement of calcium (o-cresolphthalein complexone), cholesterol (cholesterol esterase, cholesterol oxidase, peroxidase/phenol-4-aminophenazone indicator), total bilirubin (2,5-dichlorophenyldiazonium tetrafluoroborate diazonium salt), and potassium (ion-selective electrode, diluted 1:31) were all performed at 37 °C.

reagents

DCLHb (lot no. 96F28AD11) and DCLHb diluent (lot no. PBS-3-96-005A) were obtained from Baxter Healthcare. The DCLHb solution contained 100 g/L hemoglobin. These products were stored at −70 °C until use. Once thawed, DCLHb and DCLHb diluent were used within 24 h. Any product that remained was discarded if not used within this time frame.

experimental design

We followed the protocol of Kroll et al. (6) for assessing interference caused by the presence of DCLHb. For each analyte we initially prepared two pools of serum: one containing the analyte of interest at high concentration and one serum pool containing the analyte at low concentration. The analyte concentrations in the low and high serum pools used to make up the linear series of specimen pools included, at a minimum, the 1% and 99% percentiles of patient results reported in our laboratory. Four pools containing intermediate concentrations of the analyte were prepared by making admixtures of the low and high pools. The six resulting pools had concentrations of the analyte of interest related in linear fashion. Next, each of the six pools was subdivided into seven aliquots. The first aliquot from each pool received DCLHb diluent not containing any hemoglobin, whereas the six remaining aliquots had DCLHb added to achieve hemoglobin concentrations of 2.5, 5.0, 7.5, 10.0, 15.0, and 20.0 g/L. The concentrations of DCLHb that were evaluated spanned the range of DCLHb likely to be encountered in patients receiving the product. Thus, for each analyte we prepared 42 different samples: 6 pools of sera containing the analyte of interest related in linear fashion, each of which was further subdivided into 7 aliquots containing DCLHb diluent or DCLHb. All aliquots were assayed in duplicate for the analyte of interest within 4 h following their creation.

statistical analysis

Data analysis proceeded in two stages. The first stage involved analysis of the test results using the experimental protocol for assessing test interference described previously (6). Data were processed using the PROC GLM program (Ver. 6.09, Statistical Analysis System Institute Inc.) with the model: Result = B0 + (B1 × Analyte) + (B2 × DCLHb) + (B3 × (Analyte × DCLHb)) and the SAS program statement: RESULT = ANALYTE DCLHb ANALYTE ∗ DCLHb. The regression coefficients, B0, B1, B2, and B3, and their standard errors were calculated. The coefficient B0 represents the intercept obtained for the multiple regressions, whereas B1 is the coefficient of the analyte variable. In our application, B0, the intercept, should approximate zero. The B1 coefficient should approximate unity, indicating that the method appropriately measures the analyte of interest. The B2 coefficient represents the effect of the interferent, DCLHb, whereas B3 is the coefficient of the variable representing interaction between analyte and interferent, ANALYTE ∗ DCLHb.

A P value <0.01 was chosen a priori for statistical significance. Following the multiple regression analysis performed in stage one, the B2 and B3 coefficients were evaluated for statistical significance. If only one of the coefficients, B2 or B3, was statistically significant, then the nonsignificant variable was dropped from the regression model and a second regression was performed. For analytes where both the B2 and B3 coefficients were not statistically significant, no further data analysis was performed.

The second stage of data analysis involved an assessment of clinically significant interference caused by the presence of DCLHb, using the model equation obtained in stage one. The limits we used to define the presence of clinically significant interference for the analytes evaluated were those specified in the Clinical Laboratory Improvement Amendments of 1988 (CLIA-88) (8). For each analyte, DCLHb interference was evaluated over a range of analyte concentrations that spanned the reportable range of the assay. First, the lowest analyte concentration to be evaluated was substituted into the model equation. Next, sequential DCLHb values, in increments of 0.1 g/L, were substituted into the model equation. A result, y₀, was obtained from the model equation with the DCLHb concentration set to 0.0 g/L. A second result, y_n, was obtained from the model equation at each 0.1 g/L incremental increase in DCLHb concentration. The difference in analyte concentration between y₀ and y_n was then calculated. If the difference in analyte concentration between y₀ and y_n exceeded the limit for allowable error as specified under CLIA-88, then clinically significant interference was deemed to be present. The analyte concentration and interferent concentration pair was tabulated. The entire process was then repeated by incrementing the analyte concentration by a value equal to the reporting increment for the particular analyte, substituting this new analyte value into the model equation, followed by substitution of incremental DCLHb concentrations into the model. Using this iterative process, we obtained multiple data pairs consisting of sequential analyte concentrations and the corresponding DCLHb concentration producing clinically significant interference.

The data pairs obtained in stage two of the data analysis were plotted with the interferent concentration as the abscissa and the analyte concentration, y_n, as the ordinate. If the resulting curve, defined by the data pairs obtained in stage two, extended into a range of clinically relevant analyte concentrations, then significant interference was inferred to be present.

Results

Using the criteria established under CLIA-88 to evaluate clinically significant interference, we investigated four analytes measured on the Hitachi 747 for interference caused by the presence of DCLHb. The range of analyte concentrations evaluated for interference by DCLHb included the 1st and 99th percentiles of patient test results reported in our laboratory (Table 1).

DCLHb interfered with the measurement of calcium, cholesterol, and total bilirubin, but did not interfere with measurement of potassium. Although statistically significant interference by DCLHb was noted for calcium, the magnitude of interference was not clinically significant. Measurement of cholesterol and total bilirubin showed significant statistical and clinical interference caused by the presence of DCLHb; however, the type of interference was different for each. Table 2 gives the regression coefficients, R, SE, and P values for each the four analytes we evaluated.

For calcium, only the B1 and B2 coefficients are statistically different from 0. As expected, the B1 coefficient is near unity, indicating that the method appropriately measures calcium. The B2 coefficient is statistically significant, whereas the B3 coefficient is not. This indicates that DCLHb interferes with the determination of calcium in a manner that is not dependent on calcium concentrations. However, although statistically significant interference from DCLHb is present, none of the DCLHb/calcium combinations caused the measured calcium results to be greater than ±0.25 mmol/L from the result that would be obtained if the specimen did not contain DCLHb, as determined by the model regression equation. Thus, although statistically significant interference was found in the presence of DCLHb, clinically significant interference for this analyte was not present.

For cholesterol, interference was dependent on the concentration of DCLHb only. The B2 coefficient is significantly different from zero and positive, indicating that the presence of DCLHb causes a positive bias in the measurement of cholesterol concentrations. The B3 coefficient was not significantly different from zero, indicating that the interference by DCLHb with measurement of cholesterol was independent of the concentration of cholesterol in the sample. The interference effect of DCLHb is readily apparent in the interference plot for cholesterol as seen in Fig. 1 A. The upward sloping line indicates that DCLHb interferes with measurement of cholesterol. The linear nature of the line indicates that the magnitude of the interference is proportional to the DCLHb concentration and is not dependent on the concentration of cholesterol in the specimen.

Measurement of total bilirubin was affected by the presence of DCLHb. The interference effect observed was proportional to the concentration of DCLHb, as well as the concentration of bilirubin. The presence of DCLHb caused a positive bias in measured bilirubin, as evidenced by the upward slope of the interference plot (Fig. 1B). The nonlinear nature of the plot reflects the interaction between bilirubin and DCLHb at relatively low concentrations of the analyte (i.e., <70 μmol/L). At bilirubin concentrations of ∼70 μmol/L or greater, DCLHb up to 20.0 g/L did not cause clinically significant interference.

Measurement of potassium at all concentrations tested showed no statistically significant interference effect attributable to the presence of DCLHb. Only the B1 coefficient was significantly different from zero, as is to be expected if the assay is appropriately measuring potassium. Because no analyte-independent and analyte-dependent interference was associated with the presence of DCLHb, no interference plot was generated for potassium.

Discussion

Several experimental protocols are in use for the investigation of suspected interference. One method used to evaluate the effect of an interferent is to measure the analyte of interest by use of an alternative method. Unfortunately, if the alternative method is also affected by the interferent being evaluated, the presence and magnitude of interference may be seriously misjudged. The most common approach is to assess interference by adding serially higher concentrations of the suspected interferent to aliquots of the same material and then measuring the substance of interest in each aliquot. Results obtained from measurement of the analyte of interest can be evaluated as a function of interferent concentration by use of regression analysis. If the slope obtained from the regression analysis differs significantly from zero, as determined by use of the Student t-test, then interference is considered to be present. The majority of the experimental models used for the evaluation of interference make the assumption that interferences are not related to the concentration of the analyte being measured (2)(3)(4)(5). For example, a 10 μmol/L positive bias in the measured bilirubin concentration caused by 2.5 g/L of hemoglobin in a specimen with a baseline bilirubin concentration of 10 μmol/L would represent a 100% increase; this finding would be extrapolated to suggest that 2.5 g/L of hemoglobin in a specimen with a baseline bilirubin concentration of 200 μmol/L would have a measured value of 400 μmol/L. This type of analysis has been popularized with the publication of “interferographs” for many instrument-specific applications (9)(10). Interferographs are derived by plotting the percentage of relative error, expressed as (apparent result/actual result) × 100, vs concentration of interferent. Unfortunately, data for these plots are usually derived by evaluating the effect of an interferent at one concentration of analyte only. Thus, extrapolation of the information supplied by these interferographs may be highly inaccurate and misleading when applied to analytes for which interference is dependent on analyte concentration (11).

Using the method of Kroll et al. (6), we evaluated the effect of various concentrations of DCLHb on four different analytes measured on the Hitachi 747. These four analytes were chosen because each demonstrates a unique type of interference effect. Interference studies performed using this experimental protocol are significantly more difficult to conduct, compared with other interference evaluations. The former require the measurement of multiple concentrations of analyte, with each concentration of analyte being subdivided into aliquots containing multiple concentrations of interferent. A minimum of four concentrations of analyte and interferent is required. In addition, the data obtained from the analyte/interferent matrix must then be evaluated with software capable of performing multiple regression analysis. Nevertheless, use of this approach in the investigation of interference effects provides considerable information concerning the interaction between analyte and interferent; information not usually obtained when standard interference protocols are used.

The utilization and implementation of interference data obtained using the protocol of Kroll et al. (6) in the routine clinical laboratory setting is difficult. These authors utilized three-dimensional response surface plots for illustrating the interaction of interferent and analyte at various concentrations of each. Although these three-dimensional plots allow one to appreciate the magnitude of interference that is influenced by concentrations of analyte and interferent, it is still extremely difficult to judge whether any bias caused by an interferent at a particular analyte concentration is clinically significant and might lead to inappropriate changes in patient management if reported to the physician. More recent attempts to graphically display interference data from multiple regressions analysis have led to the development of “contour” plots (12). These plots show the changes that occur in measured analyte concentration at a single concentration of analyte as a function of interferent concentration. A unique contour plot must be constructed for each particular concentration of analyte to demonstrate the effect of the interferent.

Expanding on these earlier studies, we developed a two-dimensional interference plot that shows where clinically significant interference occurs as a function of both analyte and interferent concentrations. Interpretation of our interference plot simply requires knowledge of the DCLHb concentration, measured independently, and the apparent analyte concentration measured in the specimen. Plotting the DCLHb concentration and apparent analyte concentration obtained from the specimen on the interference plot enables one to determine if the measured analyte result can be reported and be within the limits specified under the CLIA-88 rules for clinically significant bias. Thus, as seen in Fig. 1 , if the plotted DCLHb/analyte results for a particular specimen are to the left of the interference plot line, then the measured analyte concentration can be reported with the knowledge that any measurement bias caused by the presence of DCLHb is less than the criteria for allowable error specified in CLIA-88. However, if the plotted results fall to the right of the interference plot line, then one can be assured that the combination of DCLHb and analyte concentrations in the specimen have yielded a test result bias that is clinically significant.

The use of the limits for allowable error established under CLIA-88 for judging the presence of clinically significant test interference represented a compromise between the somewhat arbitrary use of the 10% limits applied by some (9)(10) and the use of more rigorous approaches for defining acceptable laboratory performance advocated by others, such as use of total variance limits (including analytical and biological variability), loss of diagnostic efficiency, and use of medical-usefulness criteria (13)(14)(15)(16). Because the requirements for test precision and accuracy may differ for different clinical situations as well as whether a test is being used for diagnostic or monitoring decisions, no single set of criteria for delineating clinically significant interference will be acceptable in all circumstances (17)(18). Thus, one advantage in presenting interference data using the format we propose, in addition to its ease of interpretation, is that different interference plots can be constructed for the same analyte, using various limits for clinically significant interference based on the clinical situation of the patient. Fig. 2 shows the interference plots for cholesterol using guidelines for maximum allowable error advocated by the National Cholesterol Education Program (3%) (19), CLIA-88 (10%) (8), and the German Federal Medical Association (18%) (20). As can readily be seen in the interference plots, the more stringent the criteria for allowable bias, the more likely that the presence of DCLHb will preclude reporting of the measured value.

In summary, we present an easy-to-interpret visual format for judging the acceptability of laboratory test results when the quantity of an interfering substance is known. The visual format we present can also be automated once the multiple regression coefficients have been determined for a particular instrument and reagent combination and the maximum allowable bias attributable to presence of the interferent has been decided. One potential drawback to using our method for judging test acceptability is that it requires quantitative measurement of interferent concentrations in the specimen being analyzed. However, some of the clinical chemistry analyzers manufactured today have the ability to perform accurate quantitative measurements of interferents such as hemoglobin, bilirubin, and lipids by use of spectrophotometric techniques. The accuracy of the Hitachi 747 analyzer for measurement of hemoglobin concentrations has been verified by previous investigators (21). The imminent widespread use of hemoglobin-based blood substitutes makes it imperative that instrument manufacturers develop systems that can measure plasma or serum hemoglobin in a manner that is quantitative, precise, and linear over a useful range. The use of such systems coupled with use of detailed information regarding the effect of an interferent at various concentrations of analyte will enable more reliable reporting of laboratory test results and enable clinicians to be more confident about the accuracy of unexpected laboratory data (22).

We thank Baxter Healthcare Corp. (Deerfield, IL) for supplying the DCLHb (HemAssist) and DCLHb diluent used in this study.

References