Is the Hemocytometer Obsolete for Body Fluid Cell Counting?

Automated body fluid (BF) cell counts performed on hematology analyzers have been steadily replacing manual hemocytometer (counting chamber) cell counts in clinical laboratories. Hematology analyzers with dedicated BF modes are capable of producing an RBC count, total nucleated cell (TNC) count, WBC count, and a limited WBC differential count. Distinguishing TNCs from WBCs is particularly useful for serous fluids, where body cavity lining cells may be present in significant numbers and can interfere with the accuracy of manual WBC counts if they are not excluded. Although lining cells rarely interfere with WBC counts in cerebrospinal fluid (CSF), the extremely low CSF WBC concentrations relative to blood have limited the use of automated hematology analyzers for CSF cell counts. Achieving the accuracy and precision required to reliably distinguish pathologic CSF specimens from normal samples has posed a challenge to automated hematology analyzer manufacturers. In this issue, Fleming et al1 present an evaluation of a new high-sensitivity analysis (hsA) BF mode on the Sysmex XN hematology analyzer (Sysmex, Kobe, Japan) for CSF cell counts. The hsA BF mode is not yet commercially available but can be accessed for research purposes on the XN analyzers. Their results suggest that Sysmex has met the challenge of automated enumeration of WBCs and RBCs in CSF samples. The limit of quantitation (LoQ) expresses the precision (functional sensitivity) of the analyzer and is useful for comparing the accuracy of different analyzers for low cell counts, such as exist in CSF.2 These investigators defined LoQ as the lowest cell count that can be obtained with a coefficient of variation (CV) of 20% or less. This CV is reasonable, given that the CVs for manual BF cell counts on proficiency testing surveys typically vary from 20% to 40%.3 Previous studies in the past several years, using a variety of hematology analyzers, have shown encouraging results, suggesting that they can replace manual cell counts for most CSF samples.3 However, these studies either did not include rigorous determination of the LoQ or showed a LoQ for WBCs and/or RBCs that exceeded the upper limits of the reference ranges. This limitation has posed a barrier to wide acceptance of automation for CSF cell counts. This obstacle persisted despite the fact that the comparator “gold-standard” manual method was known to be very imprecise and the fact that the reference ranges in clinical practice were established by this imprecise method. At University Hospitals Case Medical Center, BF cell counts were validated and have been performed for years on the Sysmex XE-5000 and, more recently, on Sysmex XN analyzers, except for clear CSF samples, which are still performed manually. The investigators from Erasmus University Medical Center in Rotterdam previously evaluated the Sysmex XE-5000 BF mode and Sysmex XN BF mode and reported an improved LoQ for WBCs of 5 cells/μL on the XN.4,5 The LoQ for RBCs was not evaluated, because neither of these BF modes report RBC counts below 1,000 cells/μL. The XN hsA BF research mode is a software enhancement of the XN BF mode that reports RBC concentration to more decimal places, provides a four-cell WBC differential count, and has a flagging algorithm for detecting abnormal cell populations. In their evaluation of the XN hsA research mode, they demonstrated a LoQ of 2 cells/μL for WBCs and 10 cells/μL for RBCs. This high degree of accuracy and precision for WBCs is sufficient to reliably distinguish a normal CSF WBC count from a pathologically elevated WBC count. The LoQ for RBC counts is also sufficient to exclude even minimal peripheral blood contamination from a traumatic tap. The ability to confidently exclude contamination of CSF by peripheral blood is essential for the accurate diagnosis or exclusion of central nervous system (CNS) leukemia. The study compared WBC and RBC counts by the hsA research mode and the manual method on 111 CSF samples. Passing-Bablok regression analysis showed excellent agreement of WBCs and RBCs by the two methods over the entire range of data for the 111 samples (rWBC = 0.99 and rRBC = 0.98). To address low cell counts specifically, subset analyses of 64 samples with WBCs less than 10/μL and 69 samples with RBCs less than 50/μL were done and yielded comparable measures of agreement (rWBC<10 = 0.94 and rRBC <50 = 0.91). Only three of 50 samples with normal WBC counts by the manual method had WBC counts that were more than 5 cells/μL using the XN hsA method, yielding a specificity of 94%. In comparison to the two-cell differential count (neutrophils and mononuclear cells) provided by the XN BF mode, the XN hsA research mode provides a four-part WBC differential count that enumerates neutrophils, lymphocytes, monocytes, and eosinophils and uses gating algorithms to flag abnormal cells. In total, 179 samples were used to compare manual differential counts performed on cytospin smears with the automated differential cell counts. An interesting aspect of the study was the use of one (semi)-automated technology, CellaVison Digital Cell Morphology (DM96; Sysmex) to validate another automated technology. The DM96 was used to capture the images and preclassify 200 leukocytes on the cytospin smears. The images were reviewed for final classification by an experienced technologist. The correlations for lymphocytes, monocytes, and neutrophils were excellent, but there were not enough samples with eosinophils to evaluate this parameter. The small positive bias observed for neutrophils by the hsA research mode has been previously observed by these investigators and others and is unlikely to be clinically significant. As the authors explain, the apparent bias is probably due to cellular degeneration over time and cytospin preparation, both of which cause disproportionate loss of the more labile neutrophils and would affect the manual results more than the automated results.1,5 The flagging algorithm of the hsA mode was evaluated to a very limited extent. The authors report a 70% specificity for flagging of atypical lymphocytes; however, it is difficult to interpret exactly what this means without more specific information about the lymphocyte morphology observed on cytospin smears and the patient diagnoses. The potential benefit of this flag for CSF samples would be for the detection of leukemic blasts or lymphoma cells, but none of the cases in this study had malignant cells. It will be useful to determine the sensitivity of the hsA mode for detection of these malignant cells, as this is an important clinical indication for CSF examination. Only 20 CSF samples in this study were from infants younger than 1 year, and the rest were from patients older than 1 year, so it is not clear exactly how samples were from pediatric patients. Additional studies that are specifically focused on neonates and other pediatric age groups and published in pediatric journals would help promote acceptance of automated cell counts by pediatricians, particularly pediatric oncologists, who are guided by cooperative group clinical treatment protocols that define criteria for CSF disease based on manual cell counts.6 The accuracy and precision for CSF cell counts reported by Fleming et al1 suggest that the “glass ceiling” for automated BF cellular analysis has been broken. Hemocytometers, long since abandoned for manual platelet counting, may soon disappear from clinical laboratories altogether. For today’s “lean and mean,” minimally staffed, highly automated laboratories, this will be welcome news. To be clear, the main advantage of automated CSF cell counting is not the technologist time savings but rather the consistency and reliability of the results. However, microscopic review of cytospin smears, perhaps assisted by digital image analysis, will still be indicated for patients with possible CNS malignancy and for the detection of micro-organisms. Can the combination of flow-through cellular analysis with digital image analysis in the same “box” be far behind?