Method for analyzing glycated hemoglobin and glycated hemoglobin analysis apparatus

Abstract

To provide method and apparatus for analyzing glycated hemoglobin for easily determining, from the shape of a chromatogram, whether a specimen is from a healthy subject or contains an abnormal hemoglobin variant.SOLUTION: The method and apparatus for analyzing glycated hemoglobin are characterized by determining whether a first component peak indicating the maximum intensity of chromatogram data is monomodal or multimodal; performing height normalization processing for normalizing the height of a second component peak different from the first component peak with respect to a group of monomodal chromatogram data; performing first cluster analysis to classify the data into a first cluster group; and performing second cluster analysis on each group of multimodal chromatogram data to classify the data into a second cluster group.SELECTED DRAWING: Figure 7

Description

[Technical Field]

[0001] The present invention relates to a method and an apparatus for analyzing glycated hemoglobin. [Background technology]

[0002] Liquid chromatography is a technique used to separate and quantify various mixtures. Separation occurs due to differences in the properties of the column (packing material) and sample components, and their interactions with the eluent, and each component is detected using various detectors. In qualitative analysis, the general method is to identify the component by taking advantage of the fact that each component elutes at a specific time under the same separation conditions. In quantitative analysis, the relationship between concentration and detector output (calibration curve) is obtained in advance using multiple standard samples of known concentration, and the concentration is calculated from the detector output obtained with an unknown sample.

[0003] Liquid chromatography can separate a variety of samples, including low molecular weight compounds, synthetic polymers, and proteins, but different "separation modes" are used depending on the purpose. For low molecular weight compounds, "reverse phase chromatography" is often used, while for synthetic polymers, "size exclusion chromatography" is often used, which is based on differences in molecular size. For protein separation, size exclusion chromatography and "ion exchange chromatography," which separates based on differences in charge, are often used.

[0004] In recent years, liquid chromatography has been increasingly used in the field of clinical testing. A typical application is the measurement of HbA1c% (hemoglobin A1c), an index of diabetes. This involves hemolyzing and diluting a blood sample (patient sample), injecting it into a column packed with ion exchange resin, separating the hemoglobin into fractions using multiple eluents with different ionic strengths, detecting the fractions with a visible absorption detector (near 415 nm), and calculating HbA1c% from the amount of the S-A1c peak fraction (ion exchange chromatography). In practice, HbA1c% is calculated from the ratio of the S-A1c peak area to the sum of all peak areas and a calibration curve prepared in advance from standard samples, and is used clinically (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2001-74748 Summary of the Invention [Problem to be solved by the invention]

[0006] However, in rare cases, samples containing "abnormal hemoglobin species" due to genetic abnormalities are encountered. In such cases, the charge differs from that of hemoglobin species in healthy individuals, resulting in different elution positions and peak shapes on the chromatogram. Furthermore, there are many types of abnormal hemoglobin, and each type exhibits different chromatogram patterns. Therefore, if a chromatogram shows a different pattern from that of a healthy individual, accurate HbA1c% calculations may not be possible. To accurately calculate HbA1c%, it is necessary to distinguish between healthy individuals and those with abnormal hemoglobin species based on the chromatogram shape. However, this distinction requires considerable experience and knowledge, and a simpler method for this distinction is desirable.

[0007] An object of the present invention is to provide a glycated hemoglobin analysis method and a glycated hemoglobin analyzer that can easily distinguish whether a sample is from a healthy person or an abnormal hemoglobin species based on the shape of a chromatogram. [Means for solving the problem]

[0008] The glycated hemoglobin analysis method according to the present invention comprises performing offset processing on a group of chromatogram data of glycated hemoglobin obtained by liquid chromatography, determining whether a first component peak showing the maximum intensity of the chromatogram data for each of the offset-processed chromatogram data groups is unimodal or multimodal, performing height normalization processing on unimodal chromatogram data groups determined to be unimodal from the offset-processed chromatogram data groups to normalize the height of a second component peak different from the first component peak, performing first cluster analysis on each of the first chromatogram data groups after height normalization processing of the second component peak to classify them into a first cluster group, and performing second cluster analysis on each of multimodal chromatogram data groups determined to be multimodal from the offset-processed chromatogram data groups to classify them into a second cluster group.

[0009] Furthermore, in the glycated hemoglobin analysis method according to the present invention, it is preferable to perform offset processing on chromatogram data of unknown glycated hemoglobin obtained by liquid chromatography, determine whether a first component peak showing the maximum intensity of the chromatogram data after the offset processing is unimodal or multimodal, and if the offset processed chromatogram data is determined to be unimodal, perform height normalization processing on the second component peak on the chromatogram data, and compare the height normalized chromatogram data of the second component peak with the first cluster group to estimate the cluster with the highest similarity, and if the offset processed chromatogram data is determined to be multimodal, compare the height normalized chromatogram data of the second component peak with the second cluster group to estimate the cluster with the highest similarity.

[0010] In the glycosylated hemoglobin analysis method according to the present invention, a first cluster group is classified by performing a first cluster analysis on each of unimodal chromatogram data groups in which a first component peak showing the maximum intensity of the glycosylated hemoglobin chromatogram data obtained by liquid chromatography, and a second cluster group is classified by performing a second cluster analysis on each of multimodal chromatogram data groups in which the first component peaks are multimodal, and an offset process is performed on the glycosylated hemoglobin chromatogram data of an unknown specimen obtained by liquid chromatography. and determining whether the first component peak is unimodal or multimodal for the chromatogram data after the offset processing, and if it is determined to be unimodal, performing height normalization processing on the chromatogram data to normalize the height of a second component peak different from the first component peak, and comparing the chromatogram data after the height normalization processing of the second component peak with the first cluster group to estimate the cluster with the highest similarity, and if it is determined to be multimodal, comparing the chromatogram data after the offset processing with the second cluster group to estimate the cluster with the highest similarity.

[0011] The glycated hemoglobin analyzer according to the present invention is characterized by having a memory unit that stores a first cluster group classified by performing a first cluster analysis on each unimodal chromatogram data group in which the first component peak showing the maximum intensity of the glycated hemoglobin chromatogram data obtained by liquid chromatography is unimodal, and a second cluster group classified by performing a second cluster analysis on each multimodal chromatogram data group in which the first component peak is multimodal; an acquisition unit that acquires chromatogram data of the glycated hemoglobin of an unknown specimen obtained by liquid chromatography; a determination unit that determines whether the first component peak showing the maximum intensity of the chromatogram data is unimodal or multimodal; an estimation unit that, if the determination unit determines that the chromatogram data is unimodal, compares the chromatogram data with the first cluster group to estimate the cluster with the highest similarity; and, if the determination unit determines that the chromatogram data is multimodal, compares the chromatogram data with the second cluster group to estimate the cluster with the highest similarity; and an output unit that outputs the estimation result by the estimation unit. [Effects of the Invention]

[0012] According to the glycosylated hemoglobin analysis method and glycosylated hemoglobin analyzer of the present invention, it is possible to easily distinguish whether a sample is from a healthy person or an abnormal hemoglobin species based on the shape of the chromatogram. [Brief explanation of the drawings]

[0013] [Figure 1] (a) is a schematic diagram of the chromatogram (full scale display), and (b) is an enlarged view of (1). [Figure 2] (a) shows the raw data of chromatogram A, (b) shows chromatogram A after offset processing, (c) shows the raw data of chromatogram B, and (d) shows chromatogram B after offset processing. [Figure 3](a) shows a case where auto-balancing is performed accurately, (b) shows a case where the zero position is slightly shifted to the positive side, (c) shows a case where the zero position is slightly shifted to the negative side, and (d) shows a case where the auto-balancing time is slightly shifted. [Figure 4] (a) illustrates a single-peak A0 peak, (b) illustrates a case where a small peak exists before the A0 peak, (c) illustrates a case where the A0 peak is bimodal, and (d) illustrates a case where the A0 peak is bimodal. [Figure 5] FIG. 1(a) shows chromatograms A and B before height normalization, and FIG. 1(b) shows chromatograms A and B after height normalization. [Figure 6] (a) shows a unimodal chromatogram without height normalization, (b) shows a multimodal chromatogram without height normalization, (c) shows a unimodal chromatogram after height normalization, and (d) shows a multimodal chromatogram after height normalization. [Figure 7] FIG. 1 is a diagram illustrating the procedure of hierarchical cluster analysis. [Figure 8] FIG. 1 is a diagram schematically illustrating an example of a tree diagram obtained as a result of cluster analysis. [Figure 9] FIG. 10 is a diagram for explaining calculation of a central waveform (central chromatogram) from a group of chromatograms belonging to each cluster. [Figure 10] FIG. 10 is a flowchart showing the overall flow of the second stage processing. [Figure 11] FIG. 10 is a flowchart showing the flow of determining the shape of the A0 peak. [Figure 12] FIG. 10 is a flowchart showing the flow of height standardization processing. [Figure 13] (a) is a diagram for explaining a comparison between the central waveform and the chromatogram, and (b) is a diagram showing a schematic diagram of which cluster the chromatogram of the sample resembles. [Figure 14]FIG. 14(a) shows an example of a cluster center waveform (chromatogram), and FIG. 14(b) shows an example of the cluster center waveform (chromatogram) squared. [Figure 15] FIG. 1A shows an example of a chromatogram to be verified, and FIG. 1B shows an example of the chromatogram to be verified squared. [Figure 16] FIG. 10 is a diagram showing an example of a result of multiplying a cluster center waveform (chromatogram) and a chromatogram to be verified. [Figure 17] FIG. 10 is a diagram showing an example of the distance between a cluster center waveform (chromatogram) and a chromatogram to be verified. [Figure 18] 1 is a diagram showing an example of a glycated hemoglobin analyzer for carrying out a glycated hemoglobin analysis method. [Figure 19] FIG. 1 is a diagram showing the flow path of the glycohemoglobin analyzer used in this example. [Figure 20] FIG. 2 is a diagram showing a processing flow in this embodiment. [Figure 21] FIG. 10 is a diagram showing a measurement example (printed report) in the glycohemoglobin analyzer GHbVIII. [Figure 22] (a) shows an example of a chromatogram after offset, (b) shows an example of a chromatogram after height correction, (c) shows an example of a final chromatogram (full scale), and (d) shows an example of a final chromatogram (zoomed). [Figure 23] FIG. 10 is a diagram showing correction values ​​of height normalization processing. [Figure 24] FIG. 10 is a diagram showing a "tree diagram" obtained by performing "hierarchical cluster analysis" on a group of chromatograms that have been subjected to height normalization processing. [Figure 25] FIG. 1 is a diagram showing cluster analysis conditions. [Figure 26] FIG. 10 is a diagram showing the number of chromatograms (number of genera) classified into 26 clusters when height normalization processing is performed. [Figure 27] Figure (1) shows chromatograms classified into each cluster overlaid on top of each other. [Figure 28] Figure (2) shows the chromatograms classified into each cluster overlaid. [Figure 29] Figure (3) shows the chromatograms classified into each cluster overlaid on top of each other. [Figure 30] 1(a) is a diagram showing a group of chromatograms (9 types) classified into cluster 14, and FIG. 1(b) is a diagram showing the central chromatogram of cluster 14. FIG. [Figure 31] FIG. 10 shows a "tree diagram" obtained by "hierarchical cluster analysis" performed on a group of chromatograms that were not subjected to height normalization processing. [Figure 32] This is a table showing the number of chromatograms (number of genera) classified into 26 clusters without height normalization processing. [Figure 33] Figure (1) shows chromatograms classified into each cluster overlaid on top of each other. [Figure 34] Figure (2) shows the chromatograms classified into each cluster overlaid. [Figure 35] Figure (3) shows the chromatograms classified into each cluster overlaid on top of each other. [Figure 36] FIG. 1 shows the characteristics of the peaks of clusters 11 to 26. [Figure 37] 1(a) is a diagram showing a group of chromatograms (11 types) classified into cluster 8, and FIG. 1(b) is a diagram showing the central chromatogram of cluster 8. [Figure 38] FIG. 10 is a diagram showing characteristics when height normalization processing is performed and when it is not performed. [Figure 39] FIG. 1 shows a chromatogram of a sample. [Figure 40] FIG. 2 shows the chromatogram of the sample. [Figure 41] FIG. 3 shows the chromatogram of the sample. [Figure 42] FIG. 1 shows the results of evaluation of the similarity of 12 samples (1). [Figure 43] FIG. 10 shows the results of evaluation of the similarity of 12 samples (2). [Figure 44] FIG. 10 shows the results of evaluation of the similarity of 12 samples (3). [Figure 45] FIG. 10 shows the results of evaluation of the similarity of 12 samples (4). [Figure 46] FIG. 10 is a diagram showing the evaluation results of the similarity of sample 1. [Figure 47] FIG. 10 is a diagram showing the evaluation results of the similarity of sample 2. [Figure 48] FIG. 10 shows the evaluation results of the similarity of sample 3. [Figure 49] FIG. 10 shows the evaluation results of the similarity of sample 4. [Figure 50] FIG. 10 is a diagram showing the evaluation results of the similarity of sample 5. [Figure 51] FIG. 10 is a diagram showing the evaluation results of the similarity of sample 6. [Figure 52] FIG. 10 is a diagram showing the evaluation results of the similarity of sample 7. [Figure 53] FIG. 10 is a diagram showing the evaluation results of the similarity of sample 8. [Figure 54] FIG. 10 is a diagram showing the evaluation results of the similarity of sample 9. [Figure 55] FIG. 10 is a diagram showing the evaluation results of the similarity of the specimen 10. [Figure 56] FIG. 10 is a diagram showing the evaluation results of the similarity of sample 11. [Figure 57] FIG. 10 is a diagram showing the evaluation results of the similarity of sample 12. [Figure 58] This figure shows all the differences between the cluster determined to be the closest (1st) and the second and subsequent clusters (2nd) for each sample. [Figure 59] 1 is a diagram showing an example of a screen displayed on a display unit 103 of the glycosylated hemoglobin analyzer 100. FIG. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, a glycosylated hemoglobin analysis method and a glycosylated hemoglobin analyzer according to embodiments of the present invention will be described with reference to the drawings. However, it should be noted that the technical scope of the present invention is not limited to these embodiments, but extends to the inventions set forth in the claims and their equivalents.

[0015] Figure 1 shows a schematic diagram of a chromatogram of a blood sample from a healthy individual in a liquid chromatography-based HbA1c% (hemoglobin A1c) measurement. Figure 1(a) shows a full-scale display, and Figure 1(b) shows an enlarged display.

[0016] In healthy individuals, the HbA1c percentage is approximately 5-6%, and the S-A1c peak area percentage is also approximately 5-6%, with 85-90% being hemoglobin A0 and the remainder being peaks of other hemoglobin fractions. A typical chromatogram pattern is one in which the low-abundance "other fractions" elute first, followed by the target S-A1c, and finally the main component A0. As can be seen from the figure, a full-scale chromatogram often obscures the fluctuations of the low-abundance S-A1c peak and the other fraction peaks, so it is often enlarged or printed. Furthermore, although the eluent used for separation does not have significant absorption in the visible light range, it does have a slight background. While this background is theoretically always constant, it fluctuates slightly due to various environmental factors.

[0017] The procedure of the glycosylated hemoglobin analysis method according to the present invention is roughly divided into two stages. In the first stage, a "hierarchical cluster analysis" is performed based on a large number of previously acquired chromatograms, and the chromatograms are sorted into groups with similar shapes (clusters). Furthermore, the central waveform (cluster central chromatogram) of each group (each cluster) that is determined to have similar shapes is calculated. In the second stage, it is estimated which of the central waveforms of each group the chromatogram of the unknown sample is most similar to.

[0018] The shape / similarity of the chromatograms in the first and second stages is determined and compared based on the chromatogram configuration (time, detector output). As mentioned above, the chromatograms obtained by measuring S-A1c have characteristics such as large differences in the intensity ratios of each fraction and offsets in the chromatograms, making it difficult to simply compare the chromatograms. Therefore, it is necessary to perform preprocessing on the chromatograms before performing the first and second stages.

[0019] (offset processing) As part of the preprocessing, offset processing is performed on the chromatogram. Even if the chromatogram is the same, if the offset value is different, the time and detector output will be different and it will be difficult to determine whether they are the same. Also, even if the shape of the chromatogram is slightly different, it will be difficult to determine whether they are the same.

[0020] Figure 2 is a diagram for explaining offset processing. Figure 2(a) shows the raw data of chromatogram A, Figure 2(b) shows chromatogram A after offset processing, Figure 2(c) shows the raw data of chromatogram B, and Figure 2(d) shows chromatogram B after offset processing.

[0021] Offset processing can be performed by storing the output value at the time of sample injection and subtracting it from the output at each time point of the target chromatogram. Some recent HPLC instruments have an "auto-zero function (auto-balance function)" that forcibly zeros the detector output at the start of measurement, so this processing is not necessary for chromatograms obtained from such instruments. However, because the automatic zeroing function varies depending on the model / machine, it is more preferable to perform offset processing here.

[0022] Figure 3 illustrates the autozero function. Figure 3(a) shows a case where autobalancing is performed accurately, Figure 3(b) shows a case where the zero position is slightly shifted to the positive side, Figure 3(c) shows a case where the zero position is slightly shifted to the negative side, and Figure 3(d) shows a case where the autobalance time is slightly shifted. The function of automatically zeroing the output is useful, but as shown in Figures 3(b) to 3(d), there are cases where the zero point is not exactly zero but is slightly positive or negative, or where the timing of autozero execution is off. Furthermore, because a large number of chromatograms are used in the first stage described above, and the chromatograms may have been acquired at different facilities, performing offset processing is more reliable and preferable.

[0023] (Peak shape determination process) The second preprocessing step involves determining the peak shape of hemoglobin A0. While the A0 peak shape of healthy samples is unimodal, it is known that the A0 peak shape of some abnormal hemoglobin species can become multimodal. By determining the A0 peak shape in advance, the subsequent "hierarchical cluster analysis" can function more effectively. The effects of this process will be described later. The peak shape is determined by whether A0 has one peak top or two or more. The A0 peak corresponds to the first component peak, which shows the maximum intensity in the chromatogram data.

[0024] Figure 4 shows four examples of different A0 peak shapes. Figure 4(a) (CASE_1) shows a unimodal A0 peak, Figure 4(b) (CASE_2) shows a case where a small peak is present before the A0 peak, Figure 4(c) (CASE_3) shows a case where the A0 peak is bimodal, and Figure 4(d) (CASE_4) shows a case where the A0 peak is bimodal. Figures 4(a) to 4(d) show, from the left, a chromatogram, an enlarged portion of the chromatogram, a first-order derivative curve, and a second-order derivative curve.

[0025] The time range in which the A0 peak elutes is determined in advance, and a differential curve (time, output) within that range is obtained. In the first derivative, the points where the value is zero represent the peak apex, the valley of a multimodal peak, or the baseline where there is no output fluctuation. Identifying only the peak apex requires combining other conditions, which can make it difficult to determine the peak shape (see the first derivative curve in Figure 4). In the second derivative, the peak apex is the minimum point with a negative value. Therefore, a single minimum point can be determined as a unimodal peak, and two minimum points as a bimodal peak (see the second derivative in Figure 4). It is more preferable to set a certain threshold and determine the multimodality based on the number of minimum points in the second derivative. In this case, Figures 4(a) and 4(b) can be determined as unimodal peaks, and Figures 4(c) and 4(d) as bimodal peaks. Furthermore, because it can be difficult to distinguish the minimum points due to noise, smoothing is more effective. In this step, since the purpose is to find the number of minimum points, there is no problem even if strong smoothing is performed.

[0026] (Height standardization processing) If the above process determines that the A0 peak shape is multimodal, the chromatogram is not processed further and is sent to the next step. If the above process determines that the A0 peak shape is unimodal, the third preprocessing step is to normalize the peak height (output).

[0027] FIG. 5 is a diagram (1) for explaining height normalization processing. FIG. 5(a) shows chromatograms A and B before height normalization processing, and FIG. 5(b) shows chromatograms A and B after height normalization processing. The third preprocessing is implemented to make the variations in hemoglobin fractions other than the A0 peak clearer, as shown in FIG. 5(a). First, the S-A1c peak is identified from the qualitative results of the chromatogram, and the peak height (output value) (s) thereof is obtained. For all chromatograms, a correction coefficient is calculated such that the height of the S-A1c peak always becomes a constant value (h), and the correction coefficient is multiplied by the entire chromatogram. If the result exceeds a certain value (M), it is replaced with the value (M). The S-A1c peak corresponds to a second component peak different from the first component peak (A0 peak).

[0028] The correction coefficient f is f = h / s. The corrected output value is calculated by the following formula (1). IF output value × f ≧ M, corrected output value = M IF output value × f < M, corrected output value = output value × f (1)

[0029] By performing the above-mentioned normalization processing of the peak height, the variations in fractions other than the large A0 peak become easier to understand. Regarding the A0 peak, the upper part of the peak becomes a saturated shape (see FIG. 5(b)).[[ID=**15**]]

[0030] FIG. 6 is a diagram (2) for explaining height normalization processing. FIG. 6(a) shows a chromatogram with a unimodal A0 before height normalization processing, FIG. 6(b) shows a chromatogram with a multimodal A0 before height normalization processing, FIG. 6(c) shows a chromatogram with a unimodal A0 after height normalization processing, and FIG. 6(d) shows a chromatogram with a multimodal A0 after height normalization processing. If height normalization processing is performed on a chromatogram in which the A0 peak shows multimodality, as shown in FIG. 6(d), even if A0 is a multimodal peak, the upper part becomes saturated in the same manner as a unimodal peak, making it difficult to discriminate. Therefore, when the A0 peak is multimodal, it can be more accurately discriminated without performing height normalization processing.

[0031] (First stage processing) Below we explain the first stage, "hierarchical cluster analysis," which is performed based on a large number of previously acquired chromatograms. Cluster analysis is a data analysis method (machine learning technique) that collects items with similar properties from a collection of items with different properties and creates clusters. It is also one of the methods known as "multivariate analysis," which statistically analyzes the relationships between two or more items (variables). A cluster is a group of items with similar characteristics.

[0032] FIG. 7 is a diagram illustrating the procedure for hierarchical cluster analysis. First, offset processing 702 is performed on a large number of accumulated chromatograms 701 without determining the shape (unimodal / multimodal) of the A0 peak. Next, height normalization processing 703 is performed to form a large number of chromatograms 704 after height normalization processing and a large number of chromatograms 705 without height normalization processing. Next, cluster analysis 706 is performed on the large number of chromatograms 704 after height normalization processing, and similarly, cluster analysis 707 is performed on the large number of chromatograms without height normalization processing. In other words, cluster analysis is performed independently on two groups of chromatograms: a group of chromatograms with peak height normalization and a group of chromatograms without peak height normalization processing, and two results are obtained.

[0033] Cluster analysis can be broadly divided into two types: hierarchical cluster analysis and non-hierarchical cluster analysis. In this study, we use hierarchical cluster analysis. This method forms clusters from the most similar combinations, and the process is represented hierarchically, ultimately displaying them as a dendrogram. Generating a dendrogram requires measuring the distance between clusters. Methods for this measurement include Ward's method (calculating the sum of squares of deviations between the centroid and each individual), the centroid method (using combinations of the centroids of each cluster), the group average method (using the average of all pairwise distances between individuals in each cluster), the shortest distance method (using combinations of individuals with the shortest distances in each cluster), and the median method (using the median between the centroids of two clusters and the centroid of another cluster). The shortest distance method (distance type: cosine distance) is preferred. The resulting dendrogram allows visual confirmation of how the clusters formed during the classification process are combined, and the number of clusters can be determined later.

[0034] Figure 8 is a diagram showing an example of a tree diagram obtained as a result of cluster analysis. In Figure 8, the horizontal axis represents components (chromatograms) and the vertical axis represents distance (cosine distance). Here, the number of chromatograms analyzed is shown as 17 (C1 to C17). In this case, dividing the cosine distance by A (Distance_A line) results in two crafters, dividing the cosine distance by B (Distance_B line) results in three crafters, dividing the cosine distance by C (Distance_C line) results in four crafters, dividing the cosine distance by D (Distance_D line) results in five crafters, and dividing the distance by E (Distance_E line) results in nine crafters. In this case, based on the cosine distance C, C1 to C3 are in one cluster, C4 to C9 are in one cluster, C10 to C15 are in one cluster, and C16 to C17 are in one cluster, and chromatograms with similar shapes belong to each cluster.

[0035] Next, a cluster center waveform group 708 for height normalization processing and a cluster center waveform group 709 for height normalization processing are obtained. When classified by cosine distance C, the average value of the chromatograms of C1 to C3 is the cluster 1 center waveform (chromatogram), the average value of the chromatograms of C4 to C9 is the cluster 2 center waveform (chromatogram), the average value of the chromatograms of C10 to C15 is the cluster 3 center waveform (chromatogram), and the average value of the chromatograms of C16 to C17 is the cluster 4 center waveform (chromatogram).

[0036] 9 is a diagram illustrating the calculation of the central waveform (central chromatogram) from the chromatogram group belonging to each cluster. The finally obtained cluster central waveform (chromatogram) will have different characteristics (chromatogram pattern) for each cluster. For example, clusters can be classified into clusters showing a standard chromatogram pattern obtained from samples from healthy individuals, clusters showing that a specific hemoglobin fraction is frequently abnormal, clusters showing that a specific hemoglobin fraction is rarely abnormal, clusters showing patterns in which the gene sequence differs slightly and the chromatogram shape differs slightly, and clusters showing patterns in which the gene sequence differs significantly and the chromatogram shape differs greatly.

[0037] (Second stage processing) In the second stage of processing, the cluster analysis results obtained in the first stage with "height normalization processing" (cluster center waveform group 708) and without "height normalization processing" (cluster center waveform group 709) are used to determine which cluster the chromatogram of the unknown sample belongs to.

[0038] Fig. 10 is a flow chart showing the overall flow of the second-stage processing, Fig. 11 is a flow chart showing the flow of determining the shape of the A0 peak, and Fig. 12 is a flow chart showing the flow of the height normalization processing. As will be described later, the processing flow shown in Fig. 10 is executed by the control unit 101 according to a program stored in advance in the storage unit 102 included in the glycated hemoglobin analyzer 100.

[0039] In the second stage of processing, the chromatogram data (time, output) of the patient sample is first subjected to offset processing (S1001) using the procedure described above. As described above, if the chromatogram is acquired with a reliable offset (output at time zero is zero), the processing in S1001 is not necessary.

[0040] Next, it is determined whether or not a peak of the first specific component (A0 peak) is present in the chromatogram data of the patient sample after the offset processing (S1002), and if there is no peak of the first specific component, an error is displayed (S1003).

[0041] Next, the shape of the peak (A0 peak) of the first specific component in the unknown chromatogram is determined using the procedure described above (S1004), and it is determined whether the shape of the A0 peak is unimodal or multimodal (S1005). The determination of the peak shape of the first specific component will be described later with reference to FIG.

[0042] If the shape of the peak of the first specific component (A0 peak) is determined to be unimodal, then it is determined whether or not there is a peak of the second specific component (S-A1c peak) (S1006), and if there is no peak of the second specific component, an error is displayed (S1007).

[0043] Next, the height normalization process is performed using the procedure described above (S1007). The height normalization process will be described later with reference to Fig. 12. Note that if the shape of the peak (A0 peak) of the first specific component is determined to be multimodal, the height normalization process is not performed.

[0044] If the cluster is determined to be unimodal, the cluster center waveforms (chromatograms) for height normalization obtained in the first step are compared with the chromatogram data after height normalization to identify the cluster with the highest similarity (S1008). On the other hand, if the cluster is determined to be multimodal, the cluster center waveforms (chromatograms) for no height normalization obtained in the first step are compared with the chromatogram data without height normalization to identify the cluster with the highest similarity (S1009). The evaluation of the similarity with the cluster center waveforms (chromatograms) will be described later with reference to FIG. 13.

[0045] FIG. 11 is a flow diagram showing the flow of determining the shape of the peak of the first specific component. First, a first derivative curve (see FIG. 4) is calculated from the chromatogram data of the patient sample after offset processing within a certain time range including the first specific component peak (S1101).

[0046] Next, a second derivative curve (see FIG. 4) is calculated from the first derivative curve in the same time range (S1102).

[0047] Next, the negative minimum points below a certain threshold are calculated in the second derivative curve (S1103), and the number of minimum points is counted (S1104). If there is one minimum point, the chromatogram data of the patient sample is determined to have a unimodal peak (S1105). If there are two or more minimum points, the chromatogram data of the patient sample is determined to have a multimodal peak (S1106).

[0048] 12 is a flow chart showing the flow of the height normalization process, in which the correction coefficient f is calculated first (S1201). The method of calculating the correction coefficient f is as described above.

[0049] Next, the chromatogram data of the patient sample is multiplied by a correction coefficient f to calculate the result, and it is determined whether the result is less than a certain value (M) (S1203). If the result is less than the certain value (M), the corrected output is set to the chromatogram data of the patient sample multiplied by the correction coefficient f (S1204). If the result is equal to or greater than the certain value (M), the corrected output is set to the certain value (M) (S1205).

[0050] Fig. 13 shows the procedure for evaluating the similarity with a group of cluster center waveforms (chromatograms). Fig. 13(a) is a diagram for explaining the comparison between the center waveform and a chromatogram, and Fig. 13(b) is a diagram showing the cosine distance.

[0051] As shown in Figure 13(a), the cosine distance between each cluster center waveform (chromatogram) and the chromatogram of the unknown sample processed as described above is calculated, and the cluster closest to 1.0 is determined to be the first candidate with the highest similarity. The cluster next closest to 1.0 is determined to be the second candidate. Ranking by similarity in this manner is advantageous as it provides a basis for determining the accuracy of classification. For example, if the first candidate is close to 1.0 and the difference between it and the second candidate is large, the unknown sample is likely to belong to the first candidate cluster. Conversely, if the difference between the first candidate and the second candidate and subsequent candidates is small, it can be inferred that the chromatograms are unreliable or of a completely different type. Displaying the results graphically, as in Figure 13(b), allows for clear identification of differences in cosine distance.

[0052] The cosine distance CDi in each cluster is calculated using the following formula (2).

number

[0053] FIG. 14(a) shows an example of a cluster center waveform (chromatogram), and FIG. 14(b) shows an example of the cluster center waveform (chromatogram) squared. FIG. 15(a) shows an example of a chromatogram to be verified, and FIG. 15(b) shows an example of the chromatogram to be verified squared. FIG. 16 shows an example of the result of multiplying the cluster center waveform (chromatogram) and the chromatogram to be verified. FIG. 17 shows an example of the distance between the cluster center waveform (chromatogram) and the chromatogram to be verified.

[0054] FIG. 18 is a diagram showing an example of a glycated hemoglobin analyzer for carrying out a glycated hemoglobin analysis method.

[0055] The glycated hemoglobin analyzer 100 shown in Fig. 18 is broadly composed of a mechanism unit 101, a control unit 102, an analysis unit 103, and a similarity calculation / estimation unit 104. The control unit 101 includes a CPU (Central Processing Unit), an FPGA (Field-Programmable Gate Array), etc., and is also composed of a display unit 106 composed of a display, etc., an input / operation unit 105 composed of a keyboard, mouse, touch panel, etc., and an output unit 107 composed of various printers, etc. The mechanism unit 102 is composed of a sample processing unit 108 that transports and pre-processes samples and injects the pre-processed samples, a separation unit 109 that separates and fractionates the processed samples, and a detection unit 110 that detects separated fractions (peaks), etc. The analysis unit 103 is composed of a memory unit 111 that stores data including chromatograms obtained by measurement, and a chromatogram analysis unit 112 that performs general quantitative and qualitative calculations.

[0056] The similarity calculation / estimation unit 104 performs the second stage of processing shown in Figure 10 on the chromatogram processed by the analysis unit, and stores a judgment unit 114 for determining the shape of the first specific component peak A0 in the chromatogram based on the patient sample, an estimation unit 115 for calculating the similarity with the cluster center group, and a cluster center group for height normalization processing (see 708 in Figure 7) and a cluster center group for no height normalization 113 (see 709 in Figure 7) created based on a large number of accumulated chromatogram groups.

[0057] The same effect can be obtained even if the similarity calculation / estimation unit 104 is not included in the glycated hemoglobin analyzer 100 but is an independent configuration. In this case, data including chromatograms can be transmitted and received between the similarity calculation / estimation unit 104 and other devices via some kind of communication mechanism, such as RS-232c. Data including chromatograms can also be transferred using a storage medium such as a USB memory. In other words, the similarity calculation / estimation unit 104 may be a device separate from other components, and the entire system may constitute a glycated hemoglobin analysis system. [Example]

[0058] Figure 19 shows the flow path of the glycohemoglobin analyzer GHbVIII manufactured by Tosoh Corporation used in this example. In Figure 19, a fixed amount of a patient's blood sample (10) loaded on a sample rack (11) is dispensed using a sampling needle (9) and a measuring syringe (B8), and then hemolyzed and diluted with a hemolysis / dilution solution (2). A fixed amount of the diluted sample is injected into an analytical column (ion exchange column) (12) using a measuring syringe (A7) and a sample injection mechanism 6, where separation is performed.

[0059] Three eluents (1) with different ionic strengths pass through a degasser (3) to remove dissolved gases from the eluents, and are then sent to an analytical column (ion exchange column) (12) via a solenoid valve (4) and a liquid delivery mechanism (5). The eluents (1) to be delivered are selected by opening and closing the solenoid valve (4). The components separated in the analytical column are quantified (415 nm) by a visible light detector (13). The analytical column and detector are placed in a thermostatic bath (14) maintained at a constant temperature to ensure analytical accuracy. The visible light detector (13) corresponds to the detection unit 110 in FIG. 18.

[0060] Figure 20 shows the processing flow in this example. In this example, a "hierarchical cluster analysis" was first performed in the processing device 300 on approximately 500 chromatograms, including abnormal hemoglobin species, previously obtained at multiple facilities using multiple analyzers. The "hierarchical cluster analysis" was performed using the multivariate analysis function of Lightstone's data analysis software, OriginPro2021b. The analyzer used was a GHbVIII glycohemoglobin analyzer manufactured by Tosoh Corporation. The chromatograms used were those obtained using the GHbVIII glycohemoglobin analyzer in variant mode. The GHbVIII glycohemoglobin analyzer separates patient blood samples into fractions using liquid chromatography and calculates HbA1c% from the area ratio of the s-A1c peak. The processing device 300 is comprised of a PC (personal computer) including a processor, memory, etc., a server, etc.

[0061] Next, the obtained results of the "hierarchical cluster analysis" were stored in advance in the memory unit 102 of the glycated hemoglobin analyzer 100, and the similarity of the unknown specimens was estimated using the chromatogram analyzer in accordance with the procedure shown in Figure 10.

[0062] Figure 21 shows a measurement example (printed report) using the GHbVIII glycohemoglobin analyzer. As shown, in healthy individuals, the HbA1c% is approximately 5-6%, so the S-A1c peak (blacked out) is small and is reported in an enlarged scale. Therefore, the A0 peak, which is the main component, is overscaled, making it difficult to discern the overall shape of the peak. In healthy samples, the A0 peak is normally distributed and unimodal, which is not a particular problem. However, in samples containing abnormal hemoglobin species, the A0 peak may be multimodal, and in this case, the overscaled peak may make it difficult to determine its shape. [Example]

[0063] First, as a first step, offset processing was performed on the approximately 500 chromatograms mentioned above. Processing was performed so that the output value at time zero was zero. Note that if offset processing has been performed on the chromatogram data obtained from the measuring instrument, this step is not necessary.

[0064] Next, in the second step, the chromatogram was subjected to height normalization. Here, the height (output) of the S-A1c peak obtained in the qualitative results was processed to be "10" and at most "100" (i.e., M = 100). If the height (output) of the S-A1c peak is s, the correction coefficient is f = 10 / s, and the output value at each time is multiplied by this coefficient. If the result exceeds 100, the output value is set to "100", and if it is 100 or less, the multiplied value is used as the corrected output value (see formula (1)).

[0065] Fig. 22 shows the process of height normalization. Fig. 22(a) shows an example of a chromatogram after offset, Fig. 22(b) shows an example of a chromatogram after height correction, Fig. 22(c) shows an example of a final chromatogram (full scale), and Fig. 22(d) shows an example of a final chromatogram (enlarged). Fig. 23 shows the correction values ​​of the height normalization.

[0066] In Figure 22, column (1) is the time, and column (2) is the output value after offset processing (the output at time 0 is subtracted), which corresponds to Figure 22(a). *1 (see Figure 23) is the peak top of S-A1c. Since the height (output) is 12.775, the correction coefficient f is 10 / 12.775 = 0.7828. Column (3) is the output value obtained by multiplying column (2) by the correction coefficient f, which corresponds to Figure 22(b). In other words, it is the output value obtained by multiplying (2) by 0.7828. *2 (see Figure 23) indicates data for which the multiplication result exceeded 100. Column (4) shows the final result, where data exceeding "100" are replaced with 100, which corresponds to Figures 22(c) and (d). The final result is a combination of time: (1) and output: (4), and the peak height (output) of S-A1c is unified to "10" regardless of the chromatogram. This process was performed on approximately 500 chromatograms containing abnormal hemoglobin species that had been previously obtained, and these were used as the source data for the "hierarchical cluster analysis."

[0067] Here, the S-A1c peak height was specified as 10, with a maximum of 100, but other values ​​are also acceptable. If the maximum value is 100, it is preferable to set the S-A1c peak height to approximately 10 to 50. If the S-A1c peak height is too high, it may become difficult to distinguish the shape of the A0 peak, so a value of 10 to 20 is more preferable.

[0068] FIG. 24 shows a "tree diagram" obtained by performing "hierarchical cluster analysis" on a group of chromatograms that have undergone height normalization processing, and FIG. 25 shows the cluster analysis conditions. In FIG. 24, the horizontal axis represents data and the vertical axis represents cosine distance. The "tree diagram" shown in FIG. 24 was obtained by performing "hierarchical cluster analysis" on a group of chromatograms normalized by the correction process shown in FIGS. 22 and 23 under the conditions shown in FIG. 25. As can be seen from this, several clusters (classifications) are formed. For example, when the cosine distance is set to 0.5, five clusters (classifications) are formed (shown by dashed lines in the figure). Here, the cosine distance was set to approximately 0.1, and verification was performed with 26 clusters (classifications). FIG. 26 shows the number of chromatograms (genera) classified into the 26 clusters.

[0069] Figures 27, 28, and 29 are diagrams in which chromatograms classified into each cluster are overlaid. Figure 27 shows clusters 1 to 10, Figure 28 shows clusters 11 to 20, and Figure 29 shows clusters 21 to 26. The number of genus in the figures indicates the number of chromatograms (number of data) contained in each cluster. As can be seen from these figures, a group of chromatograms with similar shapes is formed.

[0070] Clusters 17 and beyond are groups with other peaks present after the standard hemoglobin A0 peak (approximately 0.9 minutes). The groups are also separated by the position (time) of the peak eluting after A0. Roughly speaking, clusters 17-23 have a peak at approximately 1.17 minutes, while clusters 24-26 are groups with a peak at approximately 1.33 minutes.

[0071] Other characteristic clusters are also observed. For example, clusters 12 and 13 are presumed to be groups with many unstable LA1c peaks at approximately 0.4 minutes, cluster 16 is presumed to be a group with a bimodal A0 peak at approximately 0.9 minutes with a larger front peak, and cluster 14 is presumed to be a group with a bimodal A0 peak at approximately 0.9 minutes with a larger rear peak. Furthermore, clusters 9 and 10 have a large number of genera (number of corresponding chromatograms) and are very similar to the chromatograms of healthy subject samples, so these two clusters can be presumed to be groups of "healthy subjects."

[0072] Conversely, cluster 16 contains a mixture of chromatograms in which the A0 peak is bimodal and chromatograms in which the output value exceeds 100 and is flat. This indicates that, due to the normalization of the output, it is difficult to distinguish between chromatograms in which the A0 peak is slightly split near the peak apex. As for cluster 15, it is a group in which all A0 peaks are flat, but considering that it is a cluster close to the previous cluster, it is possible that the A0 peak near the peak is slightly bimodal, making it difficult to determine.

[0073] Figure 30 is a diagram illustrating the central chromatogram. Figure 30(a) shows a group of chromatograms (9 types) classified into cluster 14, and Figure 30(b) shows the central chromatogram of cluster 14. As shown in Figures 30(a) and (b), the average value of the group of chromatograms classified into each cluster was calculated as the "cluster central chromatogram," and this was used as an index for determining the similarity of unknown samples.

[0074] Figure 31 shows a "dendrogram" obtained by "hierarchical cluster analysis" performed on a group of chromatograms without height normalization. In Figure 31, the horizontal axis represents data and the vertical axis represents cosine distance. As can be seen from Figure 31, several clusters (hierarchical divisions) are formed. For example, when the cosine distance is set to 0.45, five clusters (hierarchical divisions) are formed (shown by the dashed lines in the figure). Here, the cosine distance was set to approximately 0.1, and verification was performed with 26 clusters (hierarchical divisions). Figure 32 is a table listing the number of chromatograms (number of genera) classified into the 26 clusters.

[0075] Figures 33, 34, and 35 are diagrams in which chromatograms classified into each cluster are overlaid. Figure 33 shows clusters 1 to 10, Figure 34 shows clusters 11 to 20, and Figure 35 shows clusters 21 to 26. The number of genus in the figures indicates the number of chromatograms (number of data) contained in each cluster. As can be seen from these figures, a group of chromatograms with similar shapes is formed.

[0076] Figure 36 shows the characteristics of the peaks in clusters 11 to 26. Clusters 11 and beyond are groups in which other peaks exist after the standard hemoglobin A0 peak (approximately 0.8 minutes). The groups are also separated by the position (time) and intensity of the peak eluting after A0.

[0077] Other distinctive clusters were also observed. For example, clusters 4, 5, and 8-10 are clusters with a slightly bimodal hemoglobin A0 peak (approximately 0.88 min). Clusters 1 and 2, for example, are clusters with an extremely small hemoglobin A0 peak (approximately 0.9 min). Furthermore, clusters 7 and 8 have a large number of genera (number of corresponding chromatograms) and are very similar to the chromatograms of healthy volunteers, suggesting that these two clusters represent "healthy volunteers."

[0078] Figure 37 is a diagram illustrating the central chromatogram. Figure 37(a) shows a group of chromatograms (11 types) classified into cluster 8, and Figure 37(b) shows the central chromatogram of cluster 8. As shown in Figures 37(a) and (b), the average value of the group of chromatograms classified into each cluster was calculated as the "cluster central chromatogram," and this was used as an index for determining the similarity of unknown samples.

[0079] Figure 38 shows the characteristics of chromatograms with and without height normalization. As explained above, we found that the results and trends of "hierarchical cluster analysis" differ when height normalization is performed on the chromatogram and when it is not. As shown in Figure 38, each method has its strengths and weaknesses. Therefore, we found that using both methods appropriately can enable more reliable judgments. After various studies, we found that, since data showing large values ​​have a significant impact on the results of "hierarchical cluster analysis," a more accurate judgment can be achieved by determining the shape of the large A0 peak in advance and then determining which method is more appropriate based on the results. [Example]

[0080] Next, unknown patient blood samples (12 samples) were measured, and the shape of the A0 peak was determined using the second derivative value. Then, it was determined which of the results of the "hierarchical cluster analysis" obtained in Example 1 (1) by performing chromatogram height normalization processing, or the results of the "hierarchical cluster analysis" obtained without performing chromatogram height normalization processing, was more appropriate to use. The similarity between the determined (1) or (2) cluster centers (chromatograms) was calculated, and a verification was performed to determine whether the most similar group could be identified.

[0081] 39-41 show the chromatograms of 12 samples. In Figures 39-41, the left-hand figures show the chromatograms after offset processing, the center figures show the second-order derivative curves, and the right-hand figures show the chromatograms normalized so that the S-A1c peak height is 10 and a maximum of 100. Here, the threshold for determining the shape of the A0 peak was set to -300,000. As a result, samples Nos. 1, 2, 3, 4, 8, 9, 10, and 11 were determined to have one minimum point in the second-order derivative, while samples Nos. 5, 6, 7, and 12 were determined to have two minimum points in the second-order derivative. Therefore, it was determined that it would be appropriate to evaluate the similarity of the former group of samples with height normalization and the latter group of samples without height normalization.

[0082] Figures 42 to 45 show the similarity evaluation results for 12 samples. Figures 46 to 57 show the similarity evaluation results for individual samples. In Figures 46 to 57, the left figure shows the chromatogram of the sample (after offset processing), the center figure shows the similarity evaluation results (software screen), and the right figure shows the chromatogram of the sample from the selected cluster (after offset processing).

[0083] The similarity was estimated in the glycated hemoglobin analyzer 100 based on the procedure shown in Fig. 10. As will be described later, the display unit 103 of the glycated hemoglobin analyzer 100 displays the data shown in Figs. 46 to 57. That is, by specifying the sample data to be evaluated using the operation unit 105, estimation is automatically performed based on the procedure shown in Fig. 10, the cosine distance of each cluster center waveform (chromatogram) is calculated, and the clusters are displayed as a bar graph in descending order of similarity (closest to 1.00). In addition, the verification data and the cluster center waveform (chromatogram) determined to have the highest similarity can be displayed superimposed on each other.

[0084] Figure 58 shows all the differences between the cluster (1st) determined to be the closest and the second and subsequent clusters (2nd) for each sample. In Figures 42 to 45, the difference between the cluster (1st) determined to be the closest and the cluster (2nd) determined to be the second closest is indicated by Δ.

[0085] Fig. 59 is a diagram showing an example of a screen displayed on the display unit 103 of the glycosylated hemoglobin analyzer 100. That is, Fig. 59 explains the contents of the screens in Figs. 46 to 57.

[0086] For sample 1, there is almost no difference between the closest cluster and subsequent clusters with and without height normalization. However, the cosine distance to the first cluster is over 0.99 in both cases, indicating that similarity can be determined accurately, demonstrating that both methods can accurately determine similarity. For sample 2, there is almost no difference between the closest cluster and subsequent clusters with and without height normalization. However, the cosine distance to the first cluster is closer to 1.00 with height normalization, indicating that this method provides a better determination. For sample 3, the cosine distance to the first cluster and the difference from the second cluster are both better with height normalization. For sample 4, there is almost no difference between the closest cluster and subsequent clusters with and without height normalization, and both cosine distances to the first cluster are similar, at around 0.97, indicating that both methods can accurately determine similarity. For sample 5, the difference between the nearest cluster and subsequent clusters was three times larger without height normalization, and the cosine distance to the first cluster was closer to 1.00 without height normalization, indicating that this method was the better choice. For sample 6, the difference between the nearest cluster and subsequent clusters was twice as large without height normalization, and the cosine distance to the first cluster was closer to 1.00 without height normalization, indicating that this method was the better choice. For sample 7, the difference between the nearest cluster and subsequent clusters was nearly twice as large without height normalization, and the cosine distance to the first cluster was closer to 1.00 without height normalization, indicating that this method was the better choice. For sample 8, the difference between the nearest cluster and subsequent clusters was slightly larger with height normalization, and the cosine distance to the first cluster was closer to 1.00 with height normalization, indicating that this method was the better choice.For sample 9, the difference between the nearest cluster and subsequent clusters was nearly six times larger with height normalization, and the cosine distance to the first cluster was closer to 1.00 with height normalization, indicating that this method was more suitable for the assessment. For samples 10 and 11, the difference between the nearest cluster and subsequent clusters and the cosine distance to the first cluster were superior with height normalization, which is the opposite of the second derivative result for this sample. For sample 12, the difference between the nearest cluster and subsequent clusters was nearly six times larger with no height normalization, and the cosine distance to the first cluster was almost the same for both, indicating that this method was more suitable for the assessment.

[0087] A useful method for identifying A0 peaks is to use the derivative value of the chromatogram. While the first derivative can be used, the zero point represents the peak's apex, the valley of a multimodal peak, or a baseline with no output fluctuation, making it difficult to determine the peak shape. The second derivative also represents the negative minimum point, making it difficult to determine the peak shape. In other words, if you want to determine the shape of the chromatogram of an unknown sample, if the A0 peak is determined to be unimodal using the above method, you can perform chromatogram height normalization and perform "hierarchical cluster analysis." If the A0 peak is determined to be multimodal, you can perform "hierarchical cluster analysis" without chromatogram height normalization to determine similarity.

[0088] It should be understood by those skilled in the art that various changes, substitutions, and alterations can be made to the present invention without departing from the scope of the present invention. For example, the above-described embodiments and modifications may be implemented in appropriate combinations within the scope of the present invention. [Explanation of symbols]

[0089] 1 Eluent A / B / C 2 Hemolysis / Dilution Solution 3 Degassing device 4. Solenoid valve 5 Liquid delivery mechanism 6 Sample injection mechanism 7 Measuring syringe A 8 Measuring syringe B 9 Sampling needle 10 specimens 11 Sample rack 12 Analytical columns 13 Detector 14 Constant temperature bath 15 Drain 100 Glycated Hemoglobin Analyzer 101 Control section 102 Mechanism 103 Analysis Department 104 Similarity calculation / estimation part

Claims

1. performing offset processing on the chromatogram data group of glycated hemoglobin obtained by liquid chromatography; For each of the offset-processed chromatogram data sets, determining whether a first component peak showing the maximum intensity of the chromatogram data is unimodal or multimodal; performing height normalization processing for normalizing the height of a second component peak different from the first component peak on a unimodal chromatogram data group determined to be unimodal among the offset-processed chromatogram data group; performing a first cluster analysis on each of the first chromatogram data groups after the height normalization process of the second component peaks, and classifying the data into first cluster groups; performing a second cluster analysis on each of the multimodal chromatogram data groups determined to be multimodal among the offset-processed chromatogram data groups, and classifying the data into second cluster groups; A method for analyzing glycated hemoglobin, comprising:

2. An offset process is performed on the chromatogram data of the unknown detected glycated hemoglobin obtained by liquid chromatography, determining whether a first component peak showing the maximum intensity of the chromatogram data after the offset processing is unimodal or multimodal; When the offset-processed chromatogram data is determined to be unimodal, height normalization of the second component peak is performed on the chromatogram data, and the chromatogram data after height normalization of the second component peak is compared with the first cluster group according to claim 1 to estimate a cluster having the highest similarity; When the offset-processed chromatogram data is determined to be multimodal, the offset-processed chromatogram data is compared with the second group of clusters according to claim 1 to estimate a cluster having the highest similarity. A method for analyzing glycated hemoglobin, comprising:

3. a first cluster group obtained by performing a first cluster analysis on each of unimodal chromatogram data groups in which a first component peak showing the maximum intensity of chromatogram data of glycated hemoglobin obtained by liquid chromatography is unimodal, and a second cluster group obtained by performing a second cluster analysis on each of multimodal chromatogram data groups in which the first component peak is multimodal, and storing the first cluster group; An offset process is performed on the chromatogram data of the glycated hemoglobin of the unknown specimen obtained by liquid chromatography, determining whether the first component peak is unimodal or multimodal for the chromatogram data after the offset processing; If it is determined that the chromatogram data is unimodal, a height normalization process is performed on the chromatogram data to normalize the height of a second component peak different from the first component peak, and the chromatogram data after the height normalization process of the second component peak is compared with the first cluster group to estimate a cluster having the highest similarity; If it is determined that the chromatogram data is multimodal, the chromatogram data after the offset processing is compared with the second cluster group to estimate the cluster with the highest similarity. A method for analyzing glycated hemoglobin, comprising:

4. a storage unit that stores a first cluster group classified by performing a first cluster analysis on each of unimodal chromatogram data groups in which a first component peak showing the maximum intensity of chromatogram data of glycated hemoglobin obtained by liquid chromatography is unimodal, and a second cluster group classified by performing a second cluster analysis on each of multimodal chromatogram data groups in which the first component peak is multimodal; an acquisition unit that acquires chromatogram data of glycated hemoglobin of an unknown sample obtained by liquid chromatography; a determination unit that determines whether a first component peak showing the maximum intensity of the chromatogram data is unimodal or multimodal; an estimation unit that, when the determination unit determines that the chromatogram data is unimodal, compares the chromatogram data with the first cluster group to estimate a cluster with the highest similarity, and, when the determination unit determines that the chromatogram data is multimodal, compares the chromatogram data with the second cluster group to estimate a cluster with the highest similarity; an output unit that outputs an estimation result by the estimation unit; A glycated hemoglobin analyzer comprising:

Images