|Home | About | Journals | Submit | Contact Us | Français|
Objective: Platelet counts generated by automated analyzers on blood specimens that contain platelet clumps are often inaccurate and require verification by blood-smear review. In this study, we assessed the reliability of the Sysmex XE-5000 instrument to detect platelet clumps.
Method: We reviewed automated complete blood count (CBC) results and the findings of the microscopic review of corresponding blood smears of 600 blood specimens specifically selected from the routine laboratory workload. The sensitivity, specificity, efficiency, positive predictive value (PPV), and negative predictive value (NPV) of its 2 platelet-associated flags (abnormal platelet-size distribution [PAD] flag and platelet-clumps [CLP] flag) were determined.
Results: The respective values for the sensitivity, specificity, efficiency, and PPV were 42%, 83%, 63%, and 1% for the PAD flag and 57%, 99%, 78%, and 37% for the CLP flag. The NPV was 100%.
Conclusion: The overall reliability of the CLP flag is superior than that of the PAD flag but there is room for further improvement.
Clinical laboratories worldwide use the XE-5000, an automated hematology analyzer from Sysmex Corporation, to perform complete blood counts (CBCs) and differential leukocyte counts (DIFFs) on ethylenediamineteraacetic acid (EDTA)–anticoagulated blood specimens. The overall reliability of the results generated by this analyzer has been assessed and found to be acceptable for clinical use.1 However, the CBC results generated on some of the blood specimens are flagged by the analyzer for verification of the result of the flagged parameter by other means. One such parameter of clinical significance is the automated platelet count, which is often unreliable and consequently unreportable if the blood specimen contains platelet clumps.
We have assessed the reliability of the Sysmex XE-5000 analyzer to detect platelet clumps by determining the sensitivity, specificity, efficiency, positive predictive value (PPV), and negative predictive value (NPV) of its 2 relatively frequently generated platelet-associated flags, namely, the abnormal platelet size distribution (PAD) flag and the platelet clumps (CLP) flag. The PAD flag is generated by the analyzer based on automated analysis of the platelet histogram-associated parameters. The CLP flag is generated based on automated analysis of the specific area (upper right area of the ghost population) in the DIFF, immature cell information (IMI), and nucleated red blood cell (NRBC) scattergrams.2 These flags may be generated by the analyzer individually or together, with or without accompanying white blood cell and/or red blood cell-associated flags.
Our study focused on the selected platelet-associated flags and their variant combinations. The latter category included 2 scenarios, one represented by PAD and/or CLP and the other by PAD and CLP together. In our laboratory, which processes, on average, 634 specimens for CBC per day (based on 1 week of data), the number of specimens flagged in either or both flag categories averages 48 (7.6%) per day. Among these flags, the PAD is the most prevalent, accounting for 5.7% of total specimens; the CLP flag accounts for 1.4%. The remaining 0.5% account for generation of the 2 flags (PAD and CLP) together by the analyzer.
In this study, we included automated CBC results generated by the analyzer and morphologic findings of microscopic review of the corresponding blood smears of a total of 600 EDTA-anticoagulated blood specimens, specifically selected as described later herein, from the routine workload of our laboratory during a 5-month period. The analyzer-generated acceptable platelet counts of these specimens spanned a range of 5 × 103 per μL to 1102 × 103 per μL. The categorical distribution of the platelet counts among the entire group of specimens included 35.5% decreased (ie, <140 × 103/μL, of which 72.4% were <100 × 103/μL), 6.2% increased (ie, >440 × 103/μL), and 58.3% normal (ie, within the range of 140-440 × 103/μL).
Our laboratory routinely performs CBCs on 2 Sysmex XE-5000 analyzers, which are calibrated and quality controlled per manufacturer instructions.2 The platelet count is routinely performed via the impedance method by this analyzer. The optical method for platelet count is also available on this analyzer, but we did not consider it to be appropriate or necessary for our study. Blood smears are routinely prepared and stained with a Romanowsky stain (Wright stain) on the Sysmex SP-1000 instrument. One of the coauthors (G.G.) reviewed all smears microscopically for the presence of platelet clumps, fibrin strands, and large/giant platelets. The smear review process was unbiased, despite the fact that the reviewer was aware of the flagging status of each specimen included in the determination of PPV and NPV by virtue of the specimen-selection process.
For determination of sensitivity, we retrospectively reviewed the CBC results of 100 blood specimens, for which smears revealed significant platelet clumping and/or the presence of fibrin strands by microscopic review, for the presence of PAD and CLP flags. We defined significant clumping and the presence of fibrin strands as the degree of clumping and/or the presence of fibrin strands that rendered the automated platelet count unreliable and consequently unreportable. Similarly, for determination of specificity, we retrospectively reviewed the CBC results of an equal number of blood specimens (ie, 100), for which smears did not reveal any platelet clumps and/or fibrin strands via microscopic review, for the presence of PAD and CLP flags. The sensitivity was defined as the percentage of specimens, among the morphologically positive cases, for which test results revealed either or both of the analyzer-generated flags. The specificity was defined as the percentage of specimens revealing neither of the 2 flags among the morphologically negative cases. The efficiency was defined as the percentage of specimens among this group of 200 specimens that were correctly identified by the analyzer as having true positive or true negative results.
To determine the PPV, blood smears of 300 additional specimens flagged by the analyzer for PAD (n=100) or CLP (n=100), or both flags together (n=100) were reviewed for the presence of platelet clumps and fibrin strands. PPV was defined as the percentage of flagged specimens that, via microscopic review, revealed significant platelet clumping and/or the presence of fibrin strands.
The NPV was defined as the percentage of specimens not revealing significant clumping or any fibrin strands by microscopic review. To determine the NPV, we reviewed blood smears of an additional 100 specimens, which had not been flagged by the analyzer for any of the platelet-associated flags, for the presence of platelet clumps and fibrin strands.
Among the specimens that tested morphologically positive, the analyzer flagged 42 for the PAD and 57 for the CLP, yielding sensitivities of 42.0% and 57.0%, respectively (Table 1). The sensitivity increased to 73.0% when a specimen was considered flagged for platelet clumps if the PAD flag and/or the CLP flag was/were generated by the analyzer. It decreased, however, to 26.0% when a specimen was considered flagged for platelet clumps if the PAD and CLP flags were generated together by the analyzer. The reported platelet estimates of this group of 100 specimens ranged from 10 (10.0%) that were decreased with clumps to 76 (76.0%) that were normal with clumps to 14 (14.0%) that were increased with clumps.
Among the specimens that tested morphologically negative, the analyzer flagged 17 for PAD and 1 for CLP, yielding specificities of 83.0% and 99.0%, respectively (Table 2). The specificity dropped a bit, to 82.0%, when a specimen was considered flagged for platelet clumps, if either of the 2 flags or both flags were generated by the analyzer. The analyzer-generated platelet counts of this group of 100 specimens ranged from 41 × 103 per μL to 1102 × 103 per μL. Based on our laboratory’s reference range of 140 to 440 × 103 per μL, the categorical distribution of the platelet counts among this group included the following: decreased (ie, <140 × 103/μL), 36 (36.0%; 28 of 36 [77.8%] were <100 × 103/μL); increased (ie, >440 × 103/μL), 9 (9.0%); and normal (ie, within the range of 140-440 ×103/μL), 55 (55.0%) .
The respective efficiencies for the PAD and CLP flags, as determined from this data set from 200 specimens, were 63.0% and 78.0% (Table 3). The efficiency decreased to 54% when a specimen was considered flagged for platelet clumps if both flags were generated together by the analyzer. However, this value remained relatively unchanged at 77.5% when a specimen was considered flagged for platelet clumps if either of the 2 flags or both flags was/were generated by the analyzer. The categorical distribution of the platelet counts among this group of 200 specimens included the following: decreased, 46 (23.0%); increased, 23 (11.5%); and normal, 131 (65.5%).
The respective PPVs for the PAD and CLP flags for detecting platelet clumps were 1.0% and 37.0% (Table 4). The PPV increased minimally to 39.0% when a specimen was considered flagged for platelet clumps if both flags were generated together by the analyzer. The analyzer-generated platelet counts of this group of 300 specimens ranged from 5 × 103 per μL to 780 × 103 per μL. The categorical distribution of the platelet counts among this group included the following: decreased, 142 (47.3%; with 108 of the 142 [76.0%] having values of<100 × 103/μL); increased, 10 (3.3%); and normal, 148 (49.3%).
The NPV for the blood specimens not flagged by the analyzer for any of the platelet-associated flags was 100%. The analyzer-generated platelet counts of this group of 100 specimens ranged from 18 × 103 per μL to 771 × 103 per μL. The categorical distribution of the platelet counts among this group included the following: decreased, 25 (25.0%; with 14 of the 25 [56.0%] having a value of less than 100 × 103/μL); increased, 4 (4.0%); and normal, 71 (71.0%).
A summary of all data, including the sensitivity, specificity, efficiency, PPV, and NPV, is presented in Table 5.
Among the 2 platelet-associated flags and their variant combinations, which we assessed for their ability to detect platelet clumps, we found the CLP to be relatively more reliable, with sensitivity of 57.0%, specificity of 99.0%, PPV of 37.0%, and overall efficiency of 78.0%; the NPV was 100%. The PAD flag, compared with the CLP flag, had much lower values: sensitivity, 42.0%; specificity, 83.0%; efficiency, 62.5%; and PPV, 1.0%. When a specimen was considered correctly identified by the analyzer for the presence of platelet clumps because it generated a PAD flag, a CLP flag, or both (ie, the PAD and/or CLP combination), the sensitivity increased to 73.0%, but the specificity decreased to 82.0%, and efficiency remained relatively unchanged at 77.5%, compared with the corresponding values for the CLP flag. The variant combination of PAD and CLP flags together minimally increased the PPV to 39.0%, at the expense of a reduction in sensitivity to 26.0% and efficiency to 54.0%, with specificity remaining unchanged at 82.0%, compared with the corresponding values for the CLP flag.
These low levels of sensitivity and efficiency for the combination of PAD and CLP point to its inferiority to the CLP flag in detecting platelet clumps. Among the 3 categories of platelet counts (decreased, increased, and normal), the flagging rate was fairly similar between the decreased (47.3%) and normal categories (49.3%), with the remaining 3.3% representing the increased category. However, among the flags, the PAD was the most prevalent (50.0%) in the decreased platelet count category, and the CLP flag was the most prevalent in the increased (70.0%) and normal (50.0%) platelet-count categories. The incidence of combined flagging (PAD and CLP together) was also fairly similar between the decreased (36.6%) and normal (31.1%) platelet count categories and a bit lower (20.0%) in the increased platelet count category. From these findings, one might infer that the analyzer generates platelet-associated flags at all levels of platelet counts but the type of generated flag varies somewhat with the level of the platelet count. The PAD flag, which has the lowest PPV of 1.0%, was more often associated with decreased platelet counts. In contrast, the CLP flag and the combination of the 2 flags (PAD and CLP together), which yield the relatively higher PPVs of 37.0% and 39.0%, respectively, were more often associated with the normal and decreased platelet counts.
To our knowledge, there are no published study results on the Sysmex XE-5000 with which we can compare our findings. However, comparable study results on other automated hematology analyzers have been published during the past several years. The most comparable observations are from a study on the evaluation of automated platelet counts generated by the XE-2100, an earlier model of automated hematology analyzer from Sysmex Corporation, the same manufacturer as the XE-5000.3 For detection of platelet clumps, this study reported a sensitivity of 1% for the PAD flag, 22.5% for the CLP flag, and 35.4% for the CLP+PAD flag when the blood specimens collected in lavender-top tubes were used to perform the CBCs. The sensitivity for the CLP flag was higher (57.1%) when blood specimens collected in Microtainer tubes (Becton, Dickinson and Company) were used to perform CBCs.3
Our findings regarding the XE-5000, based on the CBCs performed on specimens collected primarily (98.0%) in lavender-top tubes, show an improvement in the reliability of its automated platelet flagging system compared with that of the XE-2100, but the degree of improvement does not approach the desirable level of at or close to 100% for each analyzed statistical parameter. The overall efficiency of 78.0%, which we observed for the CLP flag generated by the XE-5000, also compares favorably with the previously reported efficiency in the range of 33.0% to 67.0% for platelet-clump flagging by 3 different hematology analyzers (Beckman Coulter LH750 [Beckman Coulter, Inc.], Sysmex XE-2100 [Sysmex Corporation], and Siemens ADVIA 120 [Siemens Healthcare GmbH]).4 A study on evaluation of the automated flagging system of an older hematology analyzer (GenS) from a different manufacturer (Beckman Coulter, Inc) reported a sensitivity of 12.0% to 44.0%, and specificity, efficiency, NPV, and PPV of more than 95.0% each for the CLP flag.5 Our findings on the XE-5000 reveal better sensitivity and similar or slightly lower NPV, PPV, and specificity compared with that of the GenS.
Another relatively recently published study reported sensitivity in the range of 40.4% to 82.8% for the detection of platelet clumps/fibrin strands in blood smears by the CellaVision DM96 (Siemens Healthcare GmbH), which is an automated image analysis system.6 These 2 systems may not be considered completely comparable because one (the CellaVision DM96) is semiautomated and uses stained blood smears for performing DIFFs and platelet scans, and the other (the Sysmex XE-5000) is fully automated and uses whole blood to perform the same analyses.
All of these previously published reports and our findings point towards the continued need for improvement in the reliability of the automated flagging system to detect platelet clumps. To maximize the sensitivity while maintaining the specificity, it is highly desirable that users of the XE-5000 analyzer include 1 or more additional criteria along with the CLP flag to verify the automated platelet count by smear review and that the manufacturer of the analyzer continue its efforts to enhance the automatic detectability of platelet clumps to the desirable level of at or close to 100%. The manufacturer of the analyzer recommends verification of the flagged results by other means before reporting; the verification method generally used by the clinical laboratories is blood-smear review. The clinical laboratory at our institution routinely verifies automated platelet counts by smear review if they are accompanied by the CLP flag, less than 100,000 per μL on initial encounter, and/or reveal delta failure with a50% or greater drop compared with the most recent past count. We chose these criteria to ensure patient safety in a cost-effective way, with the premise that the likelihood of the patient being subjected to unnecessary platelet transfusion, additional diagnostic work-up, and/or other undue measures is low in such circumstances. Our laboratory's policy on reporting the results of the platelet count based on the findings of the blood smear review, as outlined in Table 6, may be used as a guide by others for potential implementation in their laboratories.
We thank Nandita Patel, B.Sc., and Vikas Patel, B.Sc., for their help in obtaining the slides and the data from the analyzer that we needed for the study.