► Click here to download the original poster (PDF)
Albert J. de Graaf1,2, Arjan de Mare1,3, Shubham Rastogi4, Jennita Slomp1,5
1Unilabs Diagnostics B.V., Enschede, Netherlands 2Saxenburgh Medisch Centrum, Hardenberg, Netherlands 3Ziekenhuisgroep Twente, Almelo, Netherlands 4Horiba Medical, Montpellier, France 5Medisch Spectrum Twente, Enschede, Netherlands
Clinical laboratories heavily rely on the sensitivity and specificity of their hematology analyzer’s flagging algorithm to select samples for manual slide review. When evaluating a new hematology analyzer, in addition to the verification of its numerical results in the widest possible range a laboratory should therefore assess which are known to be pathological; the diagnostic accuracy of its flagging algorithm. The two main issues for such an assessment are: (i) the availability of fresh samples, (ii) selection bias in smears reviewed manually. To overcome these issues, we present an approach for a fair comparison of analyzers based on side-by-side routine operation with data analysis afterwards.
In this study the automated hematology analyzers Sysmex XN-9000 and HORIBA Yumizen H2500 were compared. 18,420 consecutive daytime samples were routinely measured on our XN as well as on a Yumizen H2500 analyzer. Numerical values were compared by selecting, for each parameter, 300 equally-spaced data points.
Our approach provided a wide range of numerical parameters for evaluation, e.g. hemoglobin 2.2-12.2 mmol/L, MCV 55-142 fL. Correlations were excellent. In addition, our big data approach allowed to identify rarely-occurring discrepancies between the analyzers e.g. due to cold agglutinins to which the Yumizen H2500 is less sensitive. (Fig .1)
Of the 407 included smears, 193 were classified as positive i.e. containing blasts, myeloid progenitor cells, (suspected) malignant lymphocytes, plasma cells, >20% large granular lymphocytes (LGL), or >20% atypical (reactive) lymphocytes. The calculated sensitivities and specificities are given in the table below. (Fig .2)
Hb: XN
MCV: XN
MCH: XN
Fig. 1
| Slide review | n | H2500 | XN |
|---|---|---|---|
| Normal | 214 | 96 (spec. 84%) | 67 (spec. 97%) |
| Abnormal | 193 | 163 (sens. 45%) | 153 (sens. 31%) |
| Blasts | 15 | 15 (100%) | 15 (100%) |
| Malignant Lymphocytes | 36 | 34 (94%) | 36 (100%) |
| Reactive Lymphocytes | 59 | 45 (76%) | 23 (39%) |
| Immature Granulocytes | 75 | 66 (88%) | 72 (96%) |
| LGL | 3 | 2 (67%) | 22 (67%) |
Fig. 2
We have demonstrated a feasible way of performing a comparison between hematology analyzers based on real-world use. Our method provides a wide range of numerical parameters with the opportunity to identify and investigate rarely-occurring discrepancies on fresh samples. Furthermore, to evaluate the flagging algorithms a large number of positive and relevant negative blood smears can be obtained at very little sample selection effort. Only a modest number of extra smear reviews was required to overcome the selection bias of the routine analyzer triggering manual counts.
Vous avez des questions ou des demandes ? Utilisez ce formulaire pour contacter nos spécialistes.
