Evaluation of the platelet count using a new method and an existing method Materials and methodsInstrumentsThe two instruments tested were the new instrument the Bayer Diagnostics ADVIA 2120i (Bayer Diagnostics) and an existing instrument the Coulter LH 750 (Beckman-Coulter).The Coulter LH 750 method counts and sizes cells by detecting and measuring changes in electrical resistance when a cell in a conductive liquid goes through a small aperture. Each cell suspended in a conductive liquid acts as an insulator. As each cell goes through the aperture, it momentarily increases the resistance of the electrical path between two submerged electrodes, one located on each side of the aperture. This causes an electrical pulse that can be counted and sized. While the number of pulses indicates particle count, the size of the electrical pulse is proportional to the cell volume. Pulses representing cells from 2 to 20 fL are classified as platelets (Beckman-Coulter, 2018).The Bayer Diagnostics ADVIA 2120i analyses the platelet count along with the red blood cell count with a single optical cytometer after appropriate dilution of the blood sample with ADVIA 2120i RBC/PLT reagent. The red blood cells are isovolumetrically sphered and lightly fixed with glutaraldehyde to preserve the spherical shape. Red blood cells and platelets are counted from the signals from a common detector with two different gain settings.On the ADVIA 2120i Haematology system the platelet signals are amplified considerably more than the red blood signals. Coincidence correction is made to each of the counts so that the accurate counts are made over a wide range of each cell type.The method of sizing the platelets and red blood cells uses the simultaneous measurement of laser light scattered at two different angular intervals, which eliminates the adverse effect of variation of cellular haemoglobin concentration on the determination of cell volume (Siemens, 2018).Blood samples50 different patient whole blood samples were used after obtaining the patient’s consent. All samples were taken in evacuated 2.5ml tubes containing ethylendiaminotetracetic acid (EDTA) (VacutainerTM, Becton Dickinson) and tested in parallel on both instruments. Specimens were selected to span the full range of concentrations expected in clinical practice from low platelet counts to high platelet counts. The samples were analysed within four hours of the sample been taken. Quality control samples were also used which were supplied with the new analyser by the manufacturer. The control samples supplied were low level controls and high level controls with target ranges included as well. ImprecisionThe imprecision of the platelet counts on the ADVIA 2120i instrument was determined using quality control samples that were tested five times a day for 5 consecutive days. Two different control samples, low and high platelet counts, were used. The within run precision was measured using the mean, standard deviation and coefficient of variation (CV%).  The between run precision was measured using the Analysis of variance (ANOVA).AccuracyThe accuracy of the new ADVIA 2120i instrument was measured using a method comparison study. The platelet count of 50 patient samples were measured on both the new method and the existing method. The Bland Altman plot, Correlation plot, the Paired T Test and the Quality controls were used to measure the accuracy of these results.ResultsImprecision The mean, standard deviation and CV% was measured for the within run precision using the controls and the results are recorded in the table below:Table 1. Within run precision at low and high platelet count controls on the ADVIA 2120iTitle: ADVIA 2120i within run precision resultsLow level controlHigh level controlMean (x109/l)SDWithin batch(CV %)Mean(x109/l)SDWithin batch (CV %)24.21.35.4563.09.31.624.41.14.7557.06.01.125.22.28.6561.47.41.325.41.14.5562.28.11.424.41.56.2564.26.81.2Mean24.721.45835.9561.567.38291.3 *CV = coefficient of variation, SD = the standard deviationA moderate imprecision for the ADVIA 2120i was observed in the within run for the low level control and a low imprecision was observed in the within run for the high level control.The ANOVA was used to measure the between run precision for the control results measured over five consecutive days. The results are recorded in the table below:Table 2. The between run precision results using the ANOVA for the low platelet controlTitle: The between run precision results using the ANOVA for the low platelet controlSourceSSdFMSBetween runs5.8441.46F = 0.55408Within runs45.2202.26Error42.16162.635*SS = Sum of Squares, dF = deviations from the means, MS = mean of squares, F = F statistic.The F-ratio value is 0.55408. The p-value is .698997. The result is not significant at p < .05.Table 3. The between run precision results using the ANOVA for the high platelet controlTitle: The between run precision results using the ANOVA for the high platelet controlSourceSSdFMSBetween runs151.36437.84F = 0.61891Within runs1156.82057.84Error978.241661.14*SS = Sum of Squares, dF = deviations from the means, MS = mean of squares, F = F statistic.The F-ratio value is 0.61891. The p-value is .65546. The result is not significant at p < .05.AccuracyThe Bland Altman was used to assess the agreement between the two methods which is displayed in figure one below: Fig.1: Altman–Bland analysis of results obtained by Bayer Diagnostics ADVIA 2120i versus Coulter LH 750 automated haematology analyser: the difference of the platelet values is plotted against mean value (x-axis). Blue and green lines represent the lines of agreement (mean ± 2 SD).UpperMeanLowerA correlation plot was generated to measure of how well the data fell on a straight line, which is seen in the figure below: Fig.2: Comparison of the two instruments, the old method Coulter LH 750 and the new method ADVIA 2120i.A paired t test was carried out to show was there good correlation between the two methods. The results are recorded in the table below:Table 1.4: The Paired T Test results between the two methodsTitle: The Paired T test results between  the two methodsDifference Scores Calculations Mean13.9×109/Lµ0S2 = SS/df= 21570.50/(50-1) = 440.21S2M = S2/N= 440.21/50 = 8.80SM = ?S2M=?8.80 = 2.97T – value Calculation t = (M -µ)/SM= (13.9 – 0)/2.97 = 4.68*S² = sample variance, SS = Sum of Squares, df = deviations from the means, N = sample size.The value of t is 4.684548. The value of p is 2.3E-05. The result is significant at p ? 0.05.    The accuracy of the method was also compared with the control results. Where the results of the controls were analysed to determine if they fell within the target range. The results are recorded in the two figures below:Day 3                 Day 4                        Day 5UpperLowerFig. 3: The low level control results for each day over 5 consecutive days with the target range 20 – 28 x109/L. Day 1                         Day 2                      Day 3                        Day 4                      Day 5UpperLowerFig.4: The high level control results for each day over 5 consecutive days with the target range 532 – 588 x109/L.DiscussionIn this validation study the imprecision and accuracy of the new method, the Bayer Diagnostics ADVIA 2120i haematology analyser was measured only using the platelet count. Patient samples and control samples were used.The imprecision was determined using the coefficient of variance (CV %) and the Analysis of variance (ANOVA). The within run precision was measured using the CV%. The between run precision was measured using the ANOVA.On the new method, the ADVIA 2120i, control samples at two different levels of platelet counts, were measured five times a day, over five consecutive days. The within run precision was measured using the CV%. The results of the within run precision recorded in table 1, showed that there was better precision for the high level control with an average CV% of 1.3%. The average CV% for the low level control was 5.9%. this could be due to the fact that the imprecision rises with lower platelet concentrations because fewer platelets are counted. The acceptable CV% is less than 10% which is the case for the low level controls. The ideal CV% is less than 5% which is the case for the high level controls.The CV% is a more valuable result than the standard deviation, as the standard deviation increases with concentration, whereas the CV% is the ratio of standard deviation to the concentration, so it therefore provides a better estimate of the performance over a wide range of concentrations or platelet counts, according to CLSI guidelines. It is recommend that precision should be established for the full reportable range of platelet counts especially at the transfusion threshold, as an imprecise measurement can influence whether a patient receives a transfusion or not. The within run precision in this study was carried out on low and high level controls that were measured five times for five days. Ideally the controls should have been measured 10 times each day and a third level control at the normal level should have been included as well. The high and low level controls used were outside the normal reference range which is 150 – 450 x109/L, which are important results which can for example influence transfusion needs (Briggs et al., 2014).The between run precision over the five days was measured using the ANOVA. The ONE- ANOVA compared the control results run on different days, assuming that the results are Gaussian. The p value result obtained tests the null hypothesis that the data from the different days are all identical. If the p value is large, then the data from the different days are identical. If the p value is small, then the data from the different days are not identical. The p value is computed from the F value which is obtained from the ANOVA. If the null hypothesis is true, the F value is expected to be close to one. A large F value means that there is a lot of variation among the values. Applying this knowledge in interpreting the ANOVA results, the F-ratio value was 0.61891 and the p-value was 0.65546. The result is not significant as p < .05, so the null hypothesis is not rejected, the control results obtained from the different days are almost identical. The null hypothesis is also true as the F value is close to one. The between run precision was carried out over five consecutive day results, however, 20 to 30 consecutive day results would be better.The accuracy was measured using the Bland Altman plot, a correlation plot, the paired t test and the controls.The Bland Altman plot is a method for comparing two measurements of the same variable which in this study is the platelet count, measured by two different methods. The Bland Altman plot plots the mean of the two measurements versus the difference between the measurements. If there is agreement between the two measurements then the results are clustered along the mean of differences and within two standard deviations. The Bland Altman plot of the two instruments in this study in figure one, showed that not all the results are scattered closely along the mean and that two of the results are outside the two standard deviation limit. Therefore the new and old methods are not in agreement with each other (Westgard, 2018).A correlation analysis was performed which quantified the association between the two methods. A correlation plot was generated with one method as the x axis and the other method as the y axis. The important value from this plot in figure 2 was the r2 value which is the correlation coefficient. The magnitude of the correlation coefficient indicates the strength of association. A correlation coefficient value of 0.9 indicates a strong and positive correlation, whereas a negative correlation value indicates a weak negative association. A zero value means there is no association between the two methods. The correlation coefficient value obtained from the correlation plot in this study was 0.9469 which means there is strong positive correlation between the two methods.The paired t test compares two paired groups i.e. the new and existing method. It calculates the differences between the groups and analyses that list of differences based on the assumption that the differences in the entire groups follow a Gaussian distribution. The t ratio is calculated by the mean of the differences divided by the standard error of the differences. If the t ratio is large, then the p value will be small. If the pairing of the two methods is ineffective, the p value will be small (less than 0.050). If the pairing of the two methods is effective, then the p value will be larger than 0.05. In this study referring to table 1.4, the value of p was 2.3E-05. This result is significant as p < 0.05 and there is statistically significant difference between the two methods. The difference between the two methods does not equal to zero.The controls were also used to demonstrate the accuracy of this method as well as establishing the method’s precision. The target ranges for the low and high level controls was supplied by the manufacturer and the control results obtained on the new instrument for each of the 5 days was compared to see did they fall within the target range, in figures 3 and 4. When the target range was applied to the results obtained for the each run, it was seen that all the control results fell within the target range (Westgard, 2018).A better way to determine the accuracy of the new analyser compared to performing a comparison study and using the manufacturer’s controls, is to use external quality controls supplied by external quality assurance schemes (EQA). Here, the sample’s platelet count is unknown until the laboratory submits its result and the correct result is then released. This is one aspect missing from this validation study that would be more useful in determining the instrument’s accuracy.There was no information provided regarding the training requirements for using this new instrument as in whether the operator carrying out this validation study had received training by the manufacturer and if a standard operating procedure (SOP) was followed. As operator errors can produce inaccurate and imprecise results (Briggs et al., 2014).Information regarding the time from when the sample was taken to when it was analysed on both instruments wasn’t provided. Also was there a delay between analyses from one analyser to another. A delay between testing time can cause the results to differ from each other such as was all the samples tested in the morning on one analyser and were they tested on the second analyser later on in the day (Briggs et al., 2014).The total number of samples used should be in the range from 250 to 300 for a full validation. In this validation study only 50 patient samples were used.In addition since samples are analysed at any time over the day, the effect of carry over from high to low specimens needs to be considered.To conclude this validation study, the new ADVIA 2120i analyser demonstrated good precision and accuracy over all when using controls and comparing it alongside the existing Coulter LH 750 analyser. However, the number of samples used was too little and other samples such as EQA samples should have been included. References: Beckman Coulter (2018). COULTER LH 750 System Reference. [online] Biobank.ctsu.ox.ac.uk. Available at: https://biobank.ctsu.ox.ac.uk/crystal/docs/lh750reference.pdf [Accessed 2 Nov. 2018].Briggs, C., Culp, N., Davis, B., d’Onofrio, G., Zini, G. and Machin, S. (2014). ICSH guidelines for the evaluation of blood cell analysers including those used for differential leucocyte and reticulocyte counting. International Journal of Laboratory Hematology, 36(6), pp.613-627.Siemens (2018). Manual – Advia 2120i Hematology System. [online] Manuall. Available at: https://manuall.co.uk/advia-2120i-hematology-system/ [Accessed 29 Oct. 2018].Westgard, S. (2018). CLSI EP15-A3: verification of precision and estimation of bias – Westgard. [online] Westgard.com. Available at: https://www.westgard.com/clsi-ep15a3.htm [Accessed 10 Nov. 2018].