Iterative Model for the Calculation of Oxyhemoglobin, Methemoglobin, and Bilirubin in Absorbance Spectra of Cerebrospinal Fluid

The measurement of the blood pigments oxyhemoglobin, methemoglobin, and bilirubin in cerebrospinal fluid (CSF) has been useful in the differential diagnosis of various hemorrhagic and traumatic disorders of the brain (1)(2)(3)(4)(5)(6)(7)(8)(9). Computed tomography often is used for the detection of a hemorrhage, although small vascular bleedings can remain undetectable, in which case the measurement of blood pigments in CSF might be helpful (4)(6)(7). Detection and quantification of the blood pigments from the absorbance spectra are, however, difficult, and attempts to quantify these pigments (1)(10)(11) were not very successful because of factors such as turbidity, pH variability (11), and the large overlap of the absorbance curves, especially those of oxyhemoglobin and methemoglobin (1)(2)(12). The results of the visual interpretation of the spectra are qualitative in nature and depend strongly on the experience of the technicians; results, therefore, are prone to large inter- and intraindividual variation. Our aim was to develop a standardized method for the spectrophotometric examination of CSF and to obtain reliable quantitative results. A solution to this problem was found in the mathematical step-by-step (iterative) unraveling of the main absorbance scan into the scans of the individual blood pigments via the iterative process. On arrival at the laboratory, the CSF samples were centrifuged for 10 min at 1800g and ambient temperature. The supernatants were kept in the dark to avoid degradation of bilirubin or were stored at −20 °C. Before analysis, the pH of each supernatant was adjusted to pH 6.6 (11). Absorbance scans were then made from 350 to 500 nm on a Beckman DU 640 spectrophotometer. The absorbances at 360, 405, 414, and 455 nm were recorded separately (see Fig. 1, A and B ). Application of the iterative method. (A), original absorbance curve of a CSF sample. (B), calculated absorbance curves of the individual pigments as calculated by the iterative method. (C), comparison of the added (columnA) and measured (columnB) concentrations of methemoglobin (▨), oxyhemoglobin (□), and bilirubin (▤) in 15 mixtures of various compositions. Samples 1–10 did not contain Intralipid (▪). The final Intralipid concentrations in samples 11–15 were 27, 27, 7, 20, and 5 mg/L, respectively. The concentrations of the three blood pigments are indicated on the lefty-axis and that of Intralipid on the righty-axis. Concentrations of the blood pigments were then calculated with a mathematical approximation technique (iteration procedure). In these calculations (see below), the absorbances at 405, 414, and 455 nm, which are the λmax of methemoglobin, oxyhemoglobin, and bilirubin, respectively, were each corrected for the absorbance of the other components. For this we used the relative absorbances of these components at the three wavelengths and the turbidity at 360 nm. The relative absorbance of component x is defined as the ratio of the absorbance of a blood pigment x at wavelength y(Ay,x) and the maximum absorbance at wavelength λmax (Amax,x). We determined the relative absorbances at 360, 405, 414, and 455 nm by measuring solutions of pure methemoglobin, oxyhemoglobin, and bilirubin, prepared according to the method of Stroes and van Rijn (11); we used Intralipid (Pharmacia & Upjohn) solutions for turbidity measurements. The absorbance at 360 nm was used as a measure of interference by turbidity. In addition, the measured absorbances were also corrected for a blank CSF sample for which we used fixed absorbance values: 0.011, 0.008, 0.007, and 0.005 absorbance units at 360, 405, 414, and 455 nm, respectively. Repeating the iterative process 20 times appeared to be sufficient in all cases. where A405 is the absorbance at 405 nm, n is the iteration number (1–20), and o405 is the relative absorbance of oxyhemoglobin at 405 nm. The same line of reasoning was used for the other pigments [bilirubin (b), oxyhemoglobin (o), and turbidity (t)]. The recalculated absorbances for the individual pigments were then transformed to concentrations using calibration curves. An example of the original scan of a CSF sample is provided in Fig. 1A . The absorbances contributed by the individual compounds, as calculated by the iterative process, are presented in Fig. 1B . The linearity of the method was investigated by increasing the concentration of one of the blood pigments in a solution containing high concentrations (3 μmol/L) of the other two pigments. In each case, when the calculated concentration was plotted against the expected concentration, we found a straight line with a slope close to 1 and an intercept that was nearly zero. The lower detection limits calculated from these lines were <0.1 μmol/L for all three pigments. The reliability of the iterative calculation method was tested with several mixtures of the four components in blank CSF. The results of the added and calculated concentrations in 15 mixtures are presented in Fig. 1C . A close correlation was observed. In addition, we supplemented 27 blank CSF samples with one or more of the blood pigments. Concentrations were calculated with the computer program, and eight experienced technicians interpreted the scans. The results are presented in Table 1 , with the presence or absence of an added component used as the “gold standard” to calculate positive and negative predictive values, sensitivities, and specificities. Again, the program calculated the concentrations of the pigments correctly, although two results near the cutoff point of 0.1 μmol/L were discrepant. Nevertheless, the results show that the performance of the computer program is substantially better than the visual interpretation by technicians. Comparison of the results from the proposed method with the mean results from eight technicians.1 All 27 CSF samples were supplemented with one or more of the blood pigments. PPV, positive predictive value; NPV, negative predictive value. Comparison of the results from the proposed method with the mean results from eight technicians.1 All 27 CSF samples were supplemented with one or more of the blood pigments. PPV, positive predictive value; NPV, negative predictive value. In another approach, using this computer program as the gold standard, 39 absorbance spectra of pathological and nonpathological CSF samples were visually examined by eight trained technicians. The calculated sensitivities [mean (range)] for methemoglobin, oxyhemoglobin, and bilirubin were 0.66 (0.35–1.00), 0.84 (0.54–1.00), and 0.83 (0.67–0.97), respectively. The specificities [mean (range)] were 0.79 (0.37–0.95), 0.42 (0.31–0.54), and 0.88 (0.67–1.00), respectively. There was a large variation in the results between the technicians, which can be overcome by use of the iterative procedure. In summary, in this study we developed a simple, objective, quantitative, and technician-independent method for the interpretation of CSF spectra. The results show that interpretation of the spectra by individual technicians is subject to large intra- and interindividual variation, which complicates a useful clinical interpretation. Our calculation method bypasses these difficulties. However, even if the method is reliable, the results of CSF spectrophotometry must be interpreted carefully by the clinician because blood pigments can appear in CSF without any relation to a hemorrhage (2)(5)(6)(7)(10). We thank the technicians for their assistance. A protected copy of the Microsoft Excel 97 calculation program and a full-size article can be obtained by e-mail free of charge.