METHOD FOR DETECTING A DENGUE INFECTION

Abstract

The invention relates to a method for detecting a dengue infection in a patient blood sample, comprising the steps: a) Performing an analysis of prespecified parameters of blood platelets and prespecified types of blood cells in the sample and determining parameter values for the prespecified parameters of the platelets and the prespecified types of cells; b) Obtaining sample parameters from the values determined in step a); and c) Evaluating the sample parameters in relation to a prespecified criterion, wherein, if the criterion is fulfilled, a dengue infection is present.

Claims

1. A method for detecting a dengue infection in a patient blood sample, comprising the steps of: a) performing a valid perox cells analysis comprising determining the number of cells detected passing through a flowcell from a perox channel (valid perox cells); b) performing a red blood cells analysis comprising determining a red blood cell count (RBC); c) performing a platelet distribution analysis in the sample comprising determining the platelet distribution, its standard deviation and its percentage mean and determining the platelet distribution width by dividing the standard deviation by the percentage mean (PDW); d) performing a neutrophil cluster analysis comprising determining the located mean of the neutrophil cluster on the x-axis and determining the located mean of the neutrophil cluster on the y-axis; e) performing a raw red cell analysis comprising determining a raw red cell count on a Red Blood Cell (RBC) channel; f) determining the percentage of Red Blood Cells (RBCs) with volumes less than 60 fL; g) obtaining sample parameters from the values determined in steps a)-f); and h) evaluating the sample parameters in relation to a prespecified criterion, wherein, if the criterion is fulfilled, a dengue infection is present.

2. The method as claimed in claim 1, wherein lower values of the parameters determined in c) of claim 1 are predicative for the presence of a dengue infection.

3. The method as claimed in claim 1, wherein lower values of the parameters determined in d) of claim 1 on the y-axis are predicative for the presence of a dengue infection.

4. The method as claimed in claim 1, wherein higher values of the parameters determined in f) of claim 1 are predicative for the presence of a dengue infection.

5. The method as claimed in claim 1, wherein: values of the parameter determined in a) of claim 1 smaller than 3723.5 cell count and values of the parameter determined in b) of claim 1 larger than 3.77389×10.sup.6 cells per μL and values of the parameter determined in c) of claim 1 smaller than 64.1605 percent are predicative for the presence of a dengue infection.

6. The method as claimed in claim 1, wherein: values of the parameter determined in a) of claim 1 larger than 3723.5 cell count and values of the parameter determined in d) of claim 1 on the x-axis smaller than 63.8149 if scaled from 0 to 99 and values of the parameter determined in e) of claim 1 smaller than 75 cell count and values of the parameter determined in b) of claim 1 larger than 3.94312×10.sup.6 cells per μL and values of the parameter determined in d) of claim 1 on the y-axis smaller than 78.6959 if scaled from 0 to 99 are predicative for the presence of a dengue infection.

7. The method as claimed in claim 1, wherein: values of the parameter determined in a) of claim 1 larger than 3723.5 cell count and values of the parameter determined in d) of claim 1 on the x-axis larger than 63.8149 if scaled from 0 to 99 and values of the parameter determined in f) of claim 1 larger than 29.1512 percent are predicative for the presence of a dengue infection.

8. The method as claimed in claim 1, wherein the parameters are determined by scattered light measurements.

9. The method as claimed in claim 8, wherein: cell volume information is determined from low angle scattered light measurements ranging from 2° to 3° deviation from a laser light axis; and cell density information is determined from high angle scattered light measurements ranging from 5° to 15° deviation from the laser light axis.

10. The method as claimed in claim 1, wherein the dengue infection involves an infection with dengue virus serotype DEN-1, DEN-2, DEN-3, or DEN-4.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Embodiments of the inventive method are also described in greater detail below, which refer to the following figures, in which:

[0058] FIG. 1 shows a schematic diagram of a flow cytometry facility 1 for performing the inventive method.

[0059] FIG. 2 shows a schematic diagram of the large angle scattered light on the basis of a red blood cell 11, which correlates with the granularity and density of the cell and is measured in a typical range of 5° to 15°. This corresponds to the determination of the cell density.

[0060] FIG. 3 shows the determination of the cell volume (or of the cell size) of the cell 11 in the low angle scattered light range (2° to 3°).

[0061] FIG. 4 shows a process by which a prespecified criterion is determined which is used for evaluating sample parameters in a method for detecting a dengue infection in a patient blood sample according to the invention.

[0062] FIG. 5 shows a decision tree representing a hematology parameter and a prespecified criterion according to the invention.

[0063] FIG. 6 shows a decision tree representing a hematology parameter and a prespecified criterion according to the invention.

DETAILED DESCRIPTION

[0064] First, training data (20) is obtained. Here, patient samples which were positive for any serotype of the dengue virus were obtained from laboratories in India. Data were not subdivided according to serotype as the purpose of this invention is not to discriminate between strains of dengue, but rather to alert to the presence of the virus.

[0065] Second, relevant feature vectors (21) from the patient samples are extracted. The samples were measured with an ADVIA 2120i Hematology System from SIEMENS HEALTHINEERS GMBH. The ADVIA 2120i instrument can produce over 400 results. Out of those features, statistical analysis was carried out to determine which features were significantly different between dengue and non-dengue patient samples. The features that showed a significant difference (p<0.05 after correction) were selected to train a classifier.

[0066] Third, the classifier is trained to distinguish between positive and negative patient samples based on the selected features. The classification method chosen here was the decision tree for classification. The classification tree model was trained by using a greedy algorithm that splits the dataset based on a feature and calculates how well the resultant split classified the data into positive and negative patient samples. The classification algorithm (22) runs over all features at different split values/combinations and only retains those features which provide the highest predictive capability. Those features that do not provide any additional predictive power are discarded. The classification algorithm (22) used in this case was the standard Classification and Regression Tree predictor-splitting algorithm as implemented in MATLAB.

[0067] Fourth, the resulting predictive model to diagnose future unknown patients' samples is stored. The predictive model (23) is the algorithm that will ultimately identify those samples that are dengue positive in unknown cases. A visualization of two different models (classification trees to identify dengue positive blood samples) is shown in FIG. 5 and FIG. 6, respectively. The model shown in FIG. 5 is based on such features only which are generally available from Hematology analyzers. The model shown in FIG. 6 uses features generally available from Hematology analyzers as well as features which may not be readily available from Hematology analyzers other than the ADVIA 2120i Hematology System. After training the model on those features described above, the features that ultimately provided the best classification accuracy referring to the model shown in FIG. 5 were those listed in Table 1, and those referring to the model shown in FIG. 6 were those listed in Table 2, below.

TABLE-US-00001 TABLE 1 Description of parameters and units. Parameter (Feature Name) Description Units Valid Perox Number of cells detected passing Cell count Cells through the flowcell from the perox channel. RBC The reported Red Blood Cell count. 10.sup.6 cells/μL PDW Platelet distribution width (standard % deviation divided by percentage mean). Neut X Mean The located mean of the neutrophil Scaled to 0-99 cluster on the x axis. Neut Y Mean The located mean of the neutrophil Scaled to 0-99 cluster on the y axis. rcount The raw Red Cell count on a Red Cell count Blood Cell channel. Micro Percent Percentage of RBC cells with % volume less than 60 fL.

TABLE-US-00002 TABLE 2 Description of parameters and units. Parameter (Feature Name) Description Units pltSigma Platelet volumes are fitted to a 10.sup.3 cells/μL log-normal distribution. Platelet Sigma (pltSigma) is the standard deviation of this distribution. Percent High Percent High Absorption % Absorption (Percent_high_absorption) is the percentage of (Perox) cells with absorption values greater than a predefined limit (97). PMN Y peak PMN Y peak (PMN_Y_peak) is the Scaled to 0-49 mean value in the Y direction of the Polymorphonucleocyte (PMN) cluster. MPC Mean Platelet Concentration grams/dL

[0068] The ADVIA 2120 produces a cytogram based on essentially how much a cell scatters light. Every cell has a scatter high value (the low angle scattered light, see, e.g., FIG. 3) and a scatter low value (the large angle scattered light, see, e.g., FIG. 2) which comes from how much light (intensity) hits the high or low detector on a scale of 0 to 49. The scatter high and low are then plotted as a scatter diagram (cytogram) with high on the x-axis, low on the y-axis. The PMN Y peak is the peak of the PMN cluster of cells in the Y direction.

[0069] The models generated to identify dengue virus are as follows: Each node in the decision trees shown in FIGS. 5 and 6 represents a hematology parameter (see Table 1 and Table 2, respectively) and the respective prespecified criterion. The terminating branches decide whether the new sample will be classified as dengue positive (P) or negative (N) depending on the corresponding measured parameter value of the sample to be classified.

[0070] These models were trained on a set of 1,585 patient samples. In order to prevent overfitting, cross validation was carried out by partitioning the data into 5 sets, 4 of which were used to train the model and 1 used to validate. This was carried out 5 times for each subset to test the model and the results averaged. Ten percent (10%) of the data was also held out of the analysis entirely, in order to judge the resulting model on data it had never seen before.

[0071] The performance of the models yielded a sensitivity of 90% and a specificity of 90% (model shown in FIG. 5) and a sensitivity of 96% and a specificity of 99% (model shown in FIG. 6), respectively. A combination of the two models can additionally be of advantage.

[0072] The inventive method enables on the basis of the prespecified criterions as determined above, the evaluation of the determined parameters, and a prediction of a dengue infection with high specificity and sensitivity using standard hematology systems.

LIST OF REFERENCE CHARACTERS

[0073] 1 Facility for flow cytometry

[0074] 2 Laser

[0075] 3 Sensor module

[0076] 4 Optical lenses

[0077] 5 Semi-transparent mirror

[0078] 6 Aperture

[0079] 7 Mirror

[0080] 8 Sensor

[0081] 9 Sensor

[0082] 10 Sensor

[0083] 11 Cell

[0084] 20 Training data

[0085] 21 Feature vectors

[0086] 22 Classification algorithm

[0087] 23 Model

[0088] P Positive

[0089] N Negative