Analytical errors in nucleated red blood cell enumeration

Abstract

BackgroundEnumeration of nucleated red blood cells (NRBCs) in peripheral blood of dogs and cats is performed by manual counting during blood film evaluation. Automated methods have increased precision and accuracy; however, most analyzers cannot distinguish leukocytes and NRBCs. The Sysmex XN‐V Series may distinguish NRBCs and leukocytes; however, analytical errors occur.ObjectivesWe aimed to investigate cases with discrepant automated and manual NRBC counts, and to evaluate reasons for analytical errors.MethodsData from samples with increased NRBCs were collected retrospectively and compared with manual counts performed on blood films using Spearman's correlation, Passing‐Bablok agreement analysis, and Bland–Altman comparisons. Precision of the automated method and interobserver agreement of manual counts were evaluated. Cases with discrepant results were investigated.ResultsAgreement between the methods was good at ≤1NRBC ×109/L in dogs and cats, and inadequate at ≥1NRBC ×109/L. The automated method demonstrated a negative proportional difference to the manual method. Precision was good for the automated method (overall CV 7.1%) and interobserver agreement for the manual method was poor overall (mean CV 27.3%, range 0%‐106.1%). Inaccuracies in NRBC enumeration by the automated method occurred with high hematocrits, the mergence of the cell fragments and leukocyte clouds, and the presence of earlier erythroid precursors.ConclusionsNRBC enumeration by the WNR channel on the Sysmex XN‐1000 V is precise and has good agreement with manual counts in canine and feline blood samples at ≤1NRBC ×109/L. Manual film review is indicated for samples with ≥1NRBC ×109/L, earlier erythroid precursors, samples from greyhounds and dehydrated patients, and if gating errors are noted.

Validation of the Sysmex XN-V Automated Nucleated Red Blood Cell Enumeration for Canine and Feline EDTA-Anticoagulated Blood

“…This procedure resulted in a good agreement between manual and automated nRBC counts and reliable values could be obtained. Brown et al previously reported a between a high or high normal hematocrit and the misclassification of WBC as nRBCs in the WNR channel [29]. This phenomenon can also be confirmed for our investigated cases including the case of Cat 51.…”

“…Manual nRBC counting was used as the reference method for method comparison, as it represents the most commonly used method for the enumeration of nRBCs in veterinary medicine. However, manual counting may miss low numbers of nRBCs and shows high imprecision and interobserver variation [16,17,20,29]. A high imprecision of manual counts was also noted in this study, with a mean CV of 19.6%.…”

“…Precision analysis of automated nRBC counts has shown low within-and between-run imprecision, with a mean CV of 4.5% and 5.4%, respectively. The imprecision of Sysmex XN-V determined in this study was found to be lower than previously reported [29]. Given the higher imprecision of manual counts compared to automated counts, data from method comparisons using manual counting as the reference method need to be interpreted in light of the possible unfavorable performance of the reference method (manual count).…”

“…However, the negative proportional bias may be also driven by an underestimation of the automated counts in patients with severe normoblastosis. An underlying reason for this may be a poorer identification of early precursors (e.g., rubriblasts and prorubricytes), as previously hypothesized by Brown et al, where a negative proportional bias was also observed [29]. This hypothesis was not further investigated in the frame of the present study.…”

“…Additionally, good agreement has been found for feline and canine specimens with absolute nRBC counts of <1000 nRBCs/µL, excluding samples without circulating nRBCs in data analysis. In cases with higher nRBC counts, blood smear evaluation is recommended for confirmation due to a wider 95% LoA [29]. Correlation between the Sysmex XN-V and manual nRBC counts was found to be fair (r = 0.57) in a study using 63 canine specimens with a median automated absolute count of 10 nRBCs/µL (ranging from 0 to 1120 nRBCs/µL).…”