Evaluation of the analytical performance of the PC100 platelet counter

Venous blood was collected from 119 volunteers after informed consent was obtained in accordance with the Declaration of Helsinki. Blood was collected on K₂-EDTA (Vacuette®, Greiner Bio, Kremsmünster, Austria) for platelet count determination in whole blood. All measurements were performed within 3 h after blood collection to ensure platelet count stability [13].

Furthermore, whole blood from 74 out of 119 volunteers was also collected on 1:6 (v/v) acid citrate dextrose (ACDA; Vacuette®, Greiner Bio, Kremsmünster, Austria) and 3.2%(w/v) trisodium citrate (Vacuette®, Greiner Bio, Kremsmünster, Austria) for PRP sample preparations. PRP was prepared from ACDA whole blood by centrifuging it at 200 x g for 10 min and then further concentrated by an extra centrifugation step at 900 x g for 10 min to obtain the desired platelet concentration (3600 × 10³/μl). The sodium citrate anticoagulated plasma was used to prepare platelet poor plasma (PPP) by double centrifugation (2750 x g for 5 min and 10,000 x g for 10 min). Combining PRP and PPP, a series of different concentrations of validation samples were prepared with the platelet count between 250 and 3600 × 10³/μl.

The Sysmex® XN-9000 (Sysmex Corporation, Kobe, Japan) is a 5-part differential hematology analyzer, designed for high-throughput hematology laboratories. Platelet count determination is based on direct impedance with hydrodynamic focusing [14]. The analysis range for counting platelets in whole blood is between 0 and 5000 × 10³/μl.

The Sysmex® XP-300 (Sysmex Corporation, Kobe, Japan) is an automated 3-part differential hematology analyzer that was designed for measuring complete blood cell counts using the direct current detection method with coincidence correction. This device is considered as a POC, showing a great precision compared to other high-throughput hematology analyzers [15]. This device is suitable for measuring platelet count in the whole blood between 10 and 999 × 10⁹/L [16].

The PC100 platelet counter (Dutch Medical Devices, Valkenswaard, the Netherlands) is a POC device using a patented digital 3D image-based system to identify and count platelets [17]. Briefly, images of the blood sample are recorded through a glass slide covered by a plate. The mounted camera captures a unique optical pattern, indicative of platelet at a given location, which is then further processed by the software, resulting in the platelet count of the blood sample [17]. The device uses a unique disposable glass slide, containing two adjacent separate chambers which facilitates measurement of technical replicates. The platelet concentrations are determined at multiple locations to ensure high accuracy and measurement consistency. The platelet concentrations are determined at multiple locations. The analyzer requires 20 μl sample/measurement and provides the platelet count per nanoliter (plt/nL ) [12]. The results in plt/nL were converted into international units (× 10³/μl).

EDTA anticoagulated whole blood was measured on Sysmex®XN-9000 and Sysmex® XP-300 according to the instruction manual. In short, undiluted whole blood was aspirated into the devices and platelet count was obtained. Whole blood platelet count was determined with the PC100 platelet counter after manual dilution (1:25) of the whole blood with ammonium oxalate solution (ThromboCount Pur; Bioanalytic GmbH, Umkirch/Freiburg, Germany). Platelet enumeration was performed in two technical replicates by using single glass slides with two adjacent chambers (chamber A and B).

Prior to platelet count determination in PRP, the validation samples were diluted to 250 × 10³/μl with homologous PPP and subsequently were measured by Sysmex® XP-300 in a single measurement. The PRP validation samples were diluted (1:100) with ThromboCount Pur and subsequently measured with PC100 platelet counters as described above for the whole blood samples.

Statistical analysis was carried out by following the Clinical and Laboratory Standards Institute (CLSI) guideline for Measurement Procedure Comparison and Bias Estimation Using Patient Samples (EP09c) [18]. The closeness of agreement (trueness) between candidate method, the PC100 platelet counter and the reference methods, Sysmex® XN-9000 and Sysmex® XP-300 was assessed with Passing-Bablok linear regression and Bland-Altman analysis. Slope and intercept of Passing-Bablok regression were calculated with their 95% confidence interval (CI). Significant proportional or constant bias was considered when the 95% CIs of slope and intercept did not include 1 and 0, respectively. Furthermore, the clinical significance of differences between methods was determined by assessing the bias and its 95% CI calculated from the regression equation at medical decision points of 50, 100 and 600 × 10³/μl for whole blood or at extremes of the extremes of the measuring interval in PRP samples. The Bland-Altman’s method was applied to identify the mean of differences values between the candidate and reference methods and to visualize the underlying variability characteristics of relationship between the methods. Plotting the percentage difference values was preferred where the differences between methods were proportional to concentration of the reference methods. The equivalence between the methods is concluded when the bias was within a predefined acceptance criterion of ±10% or ± 10 × 10³/μl.

All samples were independently quality controlled, and all measurements were performed within 3 h after blood collection to reduce the risk of preanalytical variation due to different time-intervals that may cause changes in platelet count [13, 19]. The platelet count obtained in the whole blood samples or in PRP was excluded upon the occurrence of any of the following criteria: i) either one of the reference methods or the PC100 platelet counter failed to obtain a value; ii) an assignable cause occurred; iii) the results were outside of the measuring interval. After data control and exclusion of non-eligible donors, platelet counts obtained from 102 and 60 donors were used for analysis of whole blood and PRP platelet counts, respectively.

The direct comparison of PC100 platelet counters with the Sysmex® XP-300 and XN-9000 were performed using a Bland-Altman plot and Passing-Bablok regression method. The Bland-Altman difference plot revealed a heterogeneous distribution of data around the zero difference line with more than 95% of the points located within the 95% limits of agreement (LoA) (Fig. 1a,c). Given that Sysmex® XP-300 and XN-9000 are considered as representatives of the true value, the differences were plotted against the reference methods on the X axis. The median was chosen as an estimate of the bias for both comparisons due to the skewness in distribution of differences. The Passing-Bablok regression fit (Fig.1b,d) provided the estimate of slopes and intercept for both comparisons.

The median bias of the PC100 platelet counter against Sysmex® XP-300 and Sysmex® XN-9000 was − 1.35% (95%CI, − 3.56 to 0.12%) and − 2.98% (95%CI, − 4.82% to − 1.31%), respectively (Table 1.). A strong linear correlation (r ≥ 0.98) was found between the PC100 platelet counter and the reference methods, confirming the reliability of the regression results estimating a perfect slope of 1 for both comparisons.

The 95% CI of slope was 0.98 to 1.03 (− 2 to 3%) and 0.97 to 1.03 (− 3 to 3%) when PC100 platelet counter was compared with the Sysmex® XP-300 or the Sysmex® XN-9000, respectively. In both comparisons, the predefined acceptance criterion (±10%) covered the 95% CI of proportional bias (slope) and median bias obtained from the Passing-Bablok algorithm and difference plot.

Given the minor differences seen between the comparison of PC100 platelet counters against Sysmex® XP-300 and Sysmex® XN-9000, direct comparison of the two reference methods was conducted (Fig. 2, Table 2). To visualize the differences between the two methods, the Bland Altman plot was generated, showing the differences between the two methods against Sysmex® XN-9000 on the X axis (Fig. 2a). This plot indicated that difference values were heterogeneously scattered around zero difference line with more than 95% of the points falling within the 95% LoA. The median was chosen as the best estimate of central tendency due to presence of a potential outlier, leading to skewness in distribution of differences. The Passing-Bablok regression analysis was applied to provide an estimation of slope and intercepts (Fig. 2b).

In order to test the higher and the lower limits of the PC100 platelet counter, platelet count was measured in PRP validation samples consisting of 250–3600 × 10³/μl. Given the perfect agreement between the two reference methods validated using whole blood samples, platelet count in PRP was measured only with Sysmex® XP-300 for further analyses. The platelet count in PRP was measured using PC100 platelet counter and Sysmex® XP-300, and subsequently the differences were compared.

The direct comparison of PC100 platelet counter with the Sysmex® XP-300 was performed using Bland-Altman and Passing-Bablok regression analysis (Fig. 3.). The Bland-Altman plot showed heterogeneous distribution of data around the zero-difference line with more than 95% of the points being within the 95% LoA. The median difference between the PC100 platelet counter and Sysmex® XP-300 was − 1.42% (95% CI, 2.17 to 0.93%) (Fig. 3a, Table 3).

The Passing-Bablok algorithm was applied to estimate the slope and intercept, governing the linear relationship between the two methods (Fig. 3b.). It revealed that a slope of 0.98 (− 2%) with 95% CI of 0.97 to 0.99 (− 3 to 1%) (Table 3.). While 95% CI of the slope did not include 1, the 2% proportional bias and its 95% CI were within the 10% predefined acceptance criterion. Moreover, a correlation coefficient of 0.99 revealed a strong linear relationship between the two methods (Table 3).

Following the estimated differences in platelet counts between the PC100 platelet counter and the reference methods, clinical significance of the bias was assessed in the whole blood and PRP samples. The clinical significance of bias was assessed in the whole blood samples at medical decision points of 50, 100 and 600 × 10³/μl. For the platelet counts measured in PRP samples, the clinical significance of bias was assessed at the extremes of the measurement interval (250–3600 × 10³/μl) since no reference interval or medical decision points have been reported for platelet counts in PRP.

The Passing-Bablok regression analysis was used to estimate the bias between the PC100 platelet counter and the reference methods at the predefined medical decision points (50, 100 and 600 × 10³/μl) (Table 4.). In the whole blood samples, the PC100 platelet counter showed the largest bias at the lowest medical decision point (50 × 10³/μl) regardless of the comparative method. Notably, the estimated bias in platelet count of whole blood was considerably reduced in both comparisons, as the platelet concentration was increased towards the higher medical decision points. Comparing with Sysmex® XP-300, the 95% CI of bias at all medical decision points was covered within the predefined acceptance criterion (±10% or ± 10 × 10³/μl). At all medical decision points, the predefined acceptance criterion (±10% or ± 10 × 10³/μl) covered the 95% CI of bias resulting from comparison of the PC100 platelet counter with Sysmex® XP-300. However, when compared to Sysmex® XN-9000, the predefined acceptance criterion only included the estimated 95% CI of bias at the highest medical decision point (600 × 10³/μl). Nevertheless, the estimated bias obtained from both comparisons were covered by the acceptance criterion (±10% or ± 10 × 10³/μl).

Regarding the PRP samples, the bias between the PC100 platelet counter and Sysmex® XP-300 were 1.56 (0.6%) and − 67.46 (− 1.9%) at the lower and upper limits of the measurement interval, respectively. The 95%CIs of bias were reasonably narrow considering the broad range of measuring interval and fell within the predefined acceptance criterion (10% or ± 10 × 10³/μl).