Comparison Study of the Rates of Manual Peripheral Blood Smear Review From 3 Automated Hematology Analyzers, Unicel DxH 800, ADVIA 2120i, and XE 2100, Using International Consensus Group Guidelines Sue Jung Kim, MD, MS; Yoonjung Kim, MD; Saeam Shin, MD; Jaewoo Song, MD, MS; Jong Rak Choi, MD, PhD
the clinical laboratory, it is important both NtoContext.—In reduce the number of peripheral blood slide reviews to save time and money and to avoid reporting false results. Objective.—To determine differences in the slide review rates of 3 widely used automated hematologic analyzers, the Unicel DxH 800 (Beckman Coulter Inc, Fullerton, California), ADVIA 2120i (Siemens Diagnostics, Tarrytown, New York), and XE 2100 (Sysmex, Kobe, Japan), using International Consensus Group for Hematology Review guidelines. Design.—A total of 1485 samples were tested, and 300 were manually reviewed. Slide review rates, sensitivity, specificity, and false-positive and false-negative rates were estimated using consensus group rules and compared using x2 tests, Fisher exact tests, or generalized estimating equations. Results.—Unicel DxH 800, ADVIA 2120i, and XE 2100 showed 22.8%, 20.2%, and 28.6% slide review rates; ost clinical laboratories rely on automated hematoM logic analyzers for complete blood cell count (CBC). Peripheral blood smear review is still required, however, 1
to assess findings that the devices’ algorithms cannot classify and to confirm cell morphology or cell numbers when the algorithms detect abnormalities.2 The rate of manual slide review greatly affects laboratory costs, productivity, and turnaround time, because slide review is laborious, time-consuming, and demanding even for experienced examiners.3 In addition to the overall rates of slide review, the false-positive and false-negative rates of laboratory protocols triggering slide review are also important for both efficiency and patient safety. The rate of slide review differs according to the type of hematology analyzer and criteria for slide review, as well as patient characteristics including age, disease, and Accepted for publication January 11, 2012. From the Department of Laboratory Medicine, Yonsei University College of Medicine, Seoul, Korea. This study was funded by Beckman Coulter Inc, Siemens Diagnostics, and Sysmex. The authors have no relevant financial interest in the products or companies described in this article. Reprints: Jong Rak Choi, MD, PhD, Department of Laboratory Medicine, Yonsei University College of Medicine, 250 Seongsanno, Seodaemun-gu, Seoul 120-752, Korea (e-mail:
[email protected]). 1408
Arch Pathol Lab Med—Vol 136, November 2012
14.3%, 14.3%, and 9.7% false-negative rates; and 13.7, 11.3%, and 17.3% false-positive rates, respectively. All analyzers showed significantly higher false-negative rates than that of the consensus group (2.9%). Conclusions.—False-negative rates were higher than the recommended levels. Among 3 automated hematologic analyzers, XE 2100 showed the highest rate of slide review. Because the present study clearly shows that the slide review rates have distinct characteristics among the studied analyzers, each individual laboratory should consider selecting the most appropriate analyzer according to clinical characteristics. Analyzers with high sensitivity may be advantageous in outpatient settings for screening patients, whereas analyzers with high specificity may be beneficial in inpatient settings for efficient patient care. (Arch Pathol Lab Med. 2012;136:1408–1413; doi: 10.5858/arpa.2010-0757-OA) disease severity.4–7 In general, analyzer flags constitute the major proportion of criteria for manual review. The International Consensus Group for Hematology Review3 developed guidelines for action following CBC results, including the manual review criteria dubbed ‘‘rules’’ in 2005. The rules suggested by the consensus group are also largely dependent on analyzer flags. Therefore, the rates of slide review may be different depending on the analyzer, because of different flagging algorithms. We compared the rates of slide review of 3 widely used automated hematologic analyzers, the Unicel DxH 800 (Beckman Coulter Inc, Fullerton, California), the ADVIA 2120i (Siemens Diagnostics, Tarrytown, New York), and the XE 2100 (Sysmex, Kobe, Japan), following the rules suggested by the consensus group. In addition, we also compared the false-positive and false-negative rates of each instrument using the rules of the consensus group. MATERIALS AND METHODS The evaluation was carried out with K2EDTA whole blood specimens sent for CBC testing to the clinical laboratory in Severance Hospital in Seoul, Korea. Severance Hospital is a 2075bed tertiary care hospital and one of the largest hospitals in Korea. In 2009, the laboratory handled an average of 1758 CBC tests per day. We selected 300 consecutive CBC samples per day for 5 days. The samples were selected at 1420 hours on the first day, 1000 hours on the second day, and 0830 hours on the third to Comparison Study on the Rates of Slide Review—Kim et al
Table 1. Parameters
Neonates WBC count Platelet count Hemoglobin MCV MCHC RDW No diff or incomplete diff Neutrophil Lymphocyte Monocyte Eosinophil Basophil RBC fragment Dimorphic RBC Platelet clumping Large platelet Immature granulocyte Left shift Atypical lymphocyte Blast NRBC Other suspect flagf Total
Rates of Slide Review of Unicel DxH 800, ADVIA 2120i, and XE 2100 Rulesa
First sample 1. ,4000/mL or . 30 000/mL, and 2. First time or delta failed within 3 days 1. ,100 000/mL or . 1 000 000/mL, and 2. First time or delta failed 1. ,7 g/dL or .2 g/dL above upper reference range for age and sex, and 2. First time 1. ,75 mm3 or .105 mm3 (adult), and 2. First time $2 units above upper limit of reference range 1. .22, and 2. First time No differential count or incomplete differential count 1. ,1000/mL or . 20 000/mL, and 2. First time 1. .5000/mL (adult) or . 70 000/mL (age ,12 y), and 2. First time 1. .1500/mL (adult) or .3000/mL (age ,12 y), and 2. First time 1. .2000/mL, and 2. First time 1. .500/mL, and 2. First time Relevant flag 1. Relevant flag, and 2. First time Relevant flag 1. Relevant flag, and 2. First time 1. Relevant flag, and 2. First time or previously confirmed data with positive delta failed for WBC count 1. Relevant flag, and 2. Age ,12 y 1. Relevant flag, and 2. First time or previously confirmed data with positive delta failed for WBC count Relevant flag Relevant flag Relevant flag
Unicel DxH 800, No. (%)
ADVIA 2120i, No. (%)
XE 2100, No. (%)
0 (0.0) 70 (4.7)
0 (0.0) 79 (5.3)
0 (0.0) 74 (5.0)
... .75
104 (7.0)
92 (6.2)
110 (7.4)
.41
13 (0.9)
11 (0.7)
13 (0.9)
.90
20 (1.3)
15 (1.0)
19 (1.3)
.68
0 (0.0) 24 (1.6)
0 (0.0) 5 (0.3)
0 (0.0) 9 (0.6)
... ,.001c,d
0 (0.0)
0 (0.0)
0 (0.0)
...
23 (1.5)
20 (1.3)
25 (1.7)
.75
7 (0.5)
2 (0.1)
8 (0.5)
.16
21 (1.4)
3 (0.2)
21 (1.4)
,.001c,e
2 (0.1)
2 (0.1)
2 (0.1)
..99
1 (0.1)
0 (0.0)
1 (0.1)
..99
10 (0.7) 10 (0.7)
1 (0.1) 0 (0.0)
27 (1.8) 2 (0.1)
,.001c,d,e .001c
4 (0.3) 2 (0.1)
9 (0.6) 18 (1.2)
18 (1.2) 49 (3.3)
.01d ,.001c,d,e
80 (5.4)
63 (4.2)
172 (11.6)
,.001d,e
2 (0.1)
1 (0.1)
1 (0.1)
23 (1.5)
34 (2.3)
60 (4.0)
35 41 10 339
(2.4) (2.8) (0.7) (22.8)
42 35 0 300
(2.8) (2.4) (0.0) (20.2)
23 19 23 425
(1.5) (1.3) (1.5) (28.6)
Pb
.0.99 ,.001d,e .06 .02d ,.001c,e ,.001d,e
Abbreviations: MCHC, mean corpuscular hemoglobin concentration; MCV, mean corpuscular volume; NRBC, nucleated red blood cell; RBC, red blood cell; RDW, red blood cell distribution width; WBC, white blood cell. a
Data derived from Barnes et al3 and included with permission from Cargen Jennings Publishing Co Ltd. P value was calculated by x2 test or Fisher exact test. c Unicel DxH 800 showed significantly different frequency from ADVIA 2120i using adjusted P value. d Unicel DxH 800 showed significantly different frequency from XE 2100 using adjusted P value. e ADVIA 2120i showed significantly different frequency from XE 2100 using adjusted P value. f Cellular Inter flag in Unicel DxH 800, MPO_deficiency flag in ADVIA 2120i, and IP Abn(WBC)WBC Abn Scattergram flag in XE 2100 comprised other suspect flags. b
fifth days. Because the majority of the samples from the inpatient department were collected during the morning, morning samples were included from 3 of the 5 days to obtain samples from both inpatient and outpatient departments. Samples of less than 2 mL volume were excluded. A total of 1500 samples were processed throughout the 5-day period with 3 hematologic analyzers, the Unicel DxH 800, ADVIA 2120i, and XE 2100, in October 2009. All 3 instruments were calibrated following the manufacturers’ instructions before testing. During the entire evaluation, quality control of the instruments was carried out using commercial 2-level controls. All samples were tested according to the standard operating procedures of the laboratory. The rates of slide review, including both differentials and smear scans, were determined by the guidelines given by the International Consensus Group for Hematology Review (Table 1).3 All Arch Pathol Lab Med—Vol 136, November 2012
the rules of the consensus group, which are related to slide review, were applied to the results of CBC tests. The slide review rates of each analyzer were calculated as the number of samples that triggered slide review by at least 1 rule divided by the total number of samples tested. The number of slide reviews triggered by a single rule was also counted, and the rates were calculated to evaluate the level of contribution of each of the rules to total slide review rates. Rules related to numerical parameters were delineated by specific ranges, and rules for morphologic parameters consisted of relevant suspect flags, which were signals for abnormal morphology displayed by the analyzers. For example, rules for blast were applied when ‘‘blast_cond’’ flag greater than or equal to 1+ for ADVIA 2120i, ‘‘IP SUS(WBC)Blasts?’’ flag greater than or equal to 100 for XE 2100, or ‘‘Blast3’’ flag positive for Unicel DxH 800 was displayed. Most Comparison Study on the Rates of Slide Review—Kim et al
1409
Table 2. The Agreements Among Unicel DxH 800, ADVIA 2120i, and XE 2100 in Triggering Slide Review Compared Analyzers
Unicel DxH 800 versus ADVIA 2120i Unicel DxH 800 versus XE 2100 ADVIA 2120i versus XE 2100
k
P
0.739 0.674 0.658
,.001 ,.001 ,.001
rules contained the term of ‘‘first time’’ to prevent repetitive slide review for the same conditions. The cutoffs of suspect flag values were conservatively set for estimating total review rates; level 3 was applied to the Unicel DxH 800, 1+ to the ADVIA 2120i, and 100 to the XE 2100. We selected all 300 samples from 1 of the 5 days on which samples were collected (i.e., 300 of 1500 samples) to estimate false-negative and false-positive rates. Unlike other days, all of those 300 samples, including those that did not trigger slide review, were reviewed manually. The samples were WrightGiemsa stained using an automated slide maker/stainer, SP1000i (Sysmex). Slides were reviewed with 200–white blood cell (WBC) counting by 2 experienced senior medical technicians blinded to each other’s results.8 Positive smears were determined following the criteria of the consensus group: red blood cell (RBC) morphology of either 2+/moderate or greater; any positive finding related to malaria; platelet morphology (giant platelets) of either 2+/moderate or greater; platelet clumps more frequent than rare/occasional; Do¨hle bodies, toxic granulation, or vacuoles of either 2+/moderate or greater; and presence of abnormal cell types (blast $1, metamyelocyte .2, myelocyte/ promyelocyte $1, atypical lymphocytes .5, nucleated RBC $1, and plasma cells $1).3 When no positive finding was found in slide review, although a manual review was triggered, the sample was labeled as false-positive, and in the opposite case, it was labeled as false-negative. False-positive and false-negative rates were defined as in the study of the consensus group: falsepositive rate was calculated as the ratio of number of falsepositive cases to total number of cases, and false-negative rate as the ratio of number of false-negative cases to total number of cases.3 We also evaluated the effect of changes in the cutoff values of suspect flags for the rates of slide review and false-positive and false-negative rates by reanalyzing the same samples using different cutoff values. Unicel DxH 800 and ADVIA 2120i expressed some of the suspect flags quantitatively by numbers from 1 to 3. In light of data provided by the analyzers, the cutoff values were defined for 3 levels, that is, low, medium, and high, in the present study. For the Unicel DxH 800 analyzer, a low level corresponded to 3, a medium level to 2, and a high level to 1. For the ADVIA 2120i analyzer, a low level corresponded to 1+, a medium level to 2+, and a high level to 3+. Meanwhile, XE 2100 provided confidence levels of suspect flags as Q-flags using numbers ranging from 0 to 300. For the XE 2100 analyzer, a low level corresponded to 100, a medium level to 200, and a high level to 300. For each analyzer, users could choose a cutoff level for suspect flags, that is, set the analyzer for when to give a flag signal. Then, slide review rates, false-negative rates, and falsepositive rates were calculated according to each respective cutoff level and the results thereof were compared with each other. Statistical analysis was performed by using either PASW Statistics 18 (SPSS Inc, Chicago, Illinois) or SAS 9.13 (SAS Institute, Cary, North Carolina). x2 tests or Fisher exact tests were performed to compare the rates of slide review and falsenegative and false-positive rates. Generalized estimating equations were used to evaluate the differences in sensitivity and specificity among the analyzers. k statistics were used to evaluate the agreement of the analyzers in triggering slide review. k greater than 0.75 represented excellent agreement beyond chance, k below 0.40 represented poor agreement, and k of 0.40 to 0.75 represented fair to good agreement.9 A P value less than 1410
Arch Pathol Lab Med—Vol 136, November 2012
.05 was regarded as significant. This study was approved by the Institutional Review Board of Severance Hospital.
RESULTS A total of 1500 samples were processed; 15 were excluded because of insufficient volume of the samples. Results of 1485 specimens from 1400 patients were analyzed to estimate slide review rates. Of these patients, 755 (53.9%) were male, and the median age was 57 years (range, 0–91 years). Of the CBC samples, 945 (63.6%) were submitted from outpatient departments, and 540 (36.4%) from inpatient departments. The majority of CBC tests were requested from medical departments (71.3%), followed by surgical departments (26.6%), pediatric departments (1.4%), and dental departments (0.7%). The rate of slide review was 22.8% for the Unicel DxH 800, 20.2% for the ADVIA 2120i, and 28.6% for the XE 2100 (Table 1). The individual parameter that most contributed to the number of slide reviews was platelet count, followed by immature granulocytes. The XE 2100 showed a significantly higher rate of slide review than the other 2 analyzers (P , .001). The XE 2100 showed almost double the number of cases triggered by immature granulocytes, atypical lymphocytes, large platelets, and platelet clumping compared with the other analyzers. The Unicel DxH 800 showed significantly higher rates in the red blood cell distribution width (RDW) parameter than the other 2 analyzers (P , .001). The ADVIA 2120i showed significantly lower rates of slide review triggered by monocytes (P , .001) and RBC fragment parameters (P , .001) than the other analyzers. All analyzers showed good agreement with one another in triggered slide review; the k values were 0.739 for the Unicel DxH 800 versus the ADVIA 2120i (P , .001), 0.674 for the Unicel DxH 800 versus the XE 2100 (P , .001), and 0.658 for the ADVIA 2120i versus the XE 2100 (P , .001) (Table 2). Three hundred cases selected from the 1485 samples were reviewed microscopically to estimate false-negative and false-positive rates, as well as sensitivity and specificity. For the 300 cases, the rate of slide review was 23.7% for the Unicel DxH 800, 21.3% for the ADVIA 2120i, and 32.0% for the XE 2100. False-negative rates for the Unicel DxH 800, ADVIA 2120i, and XE 2100 were 14.3%, 14.3%, and 9.7%, and false-positive rates were 13.7%, 11.3%, and 17.3%, respectively. The XE 2100 showed the lowest false-negative rates, but the difference was not significant (P 5 .14) (Table 3). There was no clinically significant diagnosis discovered among the false-negative cases. All 3 analyzers showed significantly higher false-negative rates than the consensus group (2.9%) (adjusted P , .001), and the DxH 800 and ADVIA 2120i showed significantly lower falsepositive rates than the consensus group (18.6%) (adjusted P 5 .02 for DxH 800 and adjusted P , .001 for ADVIA 2120i). The Unicel DxH 800 and the ADVIA 2120i showed lower false-positive rates than the XE 2100, but these results were also not significant (P 5 .10). The XE 2100 showed significantly higher sensitivity than the other 2 analyzers (XE 2100 versus Unicel DxH 800, adjusted P 5 .001, XE 2100 versus ADVIA 2120i, adjusted P 5 .002) and significantly lower specificity than the ADVIA 2120i (adjusted P 5 .001). Patients with positive smear findings had been diagnosed with nonhematologic malignant diseases (42.5%) or hematologic malignancies (16.4%), as well as inflammatory conditions including infections and autoimmune disorders (24.7%). In the analysis of the false-positive cases, rules Comparison Study on the Rates of Slide Review—Kim et al
16.0%, as well as an increase in false-negative rates from 9.7% to 12.0%.
Table 3. False-Negative Rates, False-Positive Rates, Sensitivity, and Specificity of Triggering Slide Review of Unicel DxH 800, ADVIA 2120i, and XE 2100 Unicel DxH 800
ADVIA 2120i
XE 2100
14.3 13.7 41.1 81.9
14.3 11.3 41.1 85.0
9.7 17.3 60.3a 77.1b
False-negative rate, % False-positive rate, % Sensitivity, % Specitivity, %
a XE 2100 showed significantly higher sensitivity than Unicel DxH 800 (adjusted P 5 .001) and ADVIA 2120i (adjusted P 5 .002) b XE 2100 showed significantly lower specificity than ADVIA 2120i (adjusted P 5 .001).
consisting of suspect flags showed the highest positive predictive values, and WBC count and differential rules presented lower positive predictive values (Table 4). Although the XE 2100 showed a higher positive predictive value for WBC count, platelet count, and differential parameters, its total positive predictive value was lower than that of the ADVIA 2120i. That was because the XE 2100 had more false-positive cases related to suspect flags. Abnormal RBC morphology accounted for one-half of the false-negative cases, followed by abnormal platelet morphology and immature granulocytes (Table 5). In the analysis of the effect of changes in the cutoff values of suspect flags, the different cutoff levels of left shift flag did not affect the slide review rates, false-negative rates, or falsepositive rates of any of the 3 analyzers (Table 6). The Unicel DxH 800 showed consistent results despite the changes in cutoff values of quantitative flags; only changes from low to medium level in the atypical lymphocytes flag cutoff caused a 0.4% decrease in slide review rate and false-positive rate. Meanwhile, a higher cutoff level of immature granulocytes caused more false negatives for the ADVIA 2120i analyzer, with no effect on false-positive rates; the slide review rates and false-negative rates were 21.3% and 14.3% with the low cutoff level, 19.0% and 16.7% with the medium cutoff level, and 18.0% and 17.7% with the high cutoff level. The XE 2100 presented the largest changes in review rate, false-negative rate, and false-positive rate according to changes in the cutoff values of quantitative flags. When cutoff of the atypical lymphocytes flag changed from low to high for the XE 2100 analyzer, slide review rates decreased from 32.0% to 29.3%, false-negative rates increased from 9.7% to 10.0%, and false-positive rates decreased from 17.3% to 15.0%. Changes of the immature granulocytes flag cutoff from low to high also resulted in decreases in slide review rates from 32.0% to 28.3% and in false-positive rates from 17.3% to Table 4.
COMMENT The rate of slide review should be adequately set to detect hematologic abnormalities while being sensitive to laboratory costs and turnaround times for CBC test results.3,6 One of the major determinants of review rates is the particular type of analyzer because of differences among the analyzers in methodologies and flagging algorithms.4 For example, the ADVIA 2120i displays immature granulocyte flags by comparing parameters from the peroxidase and the basophil/lobularity chambers, whereas the XE 2100 uses the IMI channel to enumerate immature granulocytes. Among these 3 automated hematologic analyzers, the XE 2100 showed the highest rate of slide review in this study. As a consequence, the XE 2100 showed the highest sensitivity and lowest specificity. The rates of slide review, sensitivity, and specificity were comparable between the Unicel DxH 800 and the ADVIA 2120i. The XE 2100 showed almost 2 times as many cases whose review was triggered for immature granulocytes, atypical lymphocytes, large platelets, and platelet clumping in comparison with the other 2 analyzers. The results imply that the XE2100 is more sensitive to those parameters and might be more prone to false positives. For the 3 analyzers, both falsepositive and false-negative rates were higher than the recommended rates. The consensus group study revealed that the review rates varied from 5% to 95% among different laboratories.3 Meanwhile, in a study of 263 institutions, the 10th- to 90thpercentile smear review rates ranged from 9.9% to 50%.6 To standardize the variable rules for slide review between laboratories, the consensus group suggested criteria for action following automated CBC and WBC differential analysis.3 A group of investigators who participated in the consensus group later tested the consensus rules in 15 laboratories. In their study, the false-positive rate was 18.6% and the false-negative rate 2.9%. They recommended a false-negative rate of less than 5% and a falsepositive rate not much higher than that found by the consensus group of 18.6%. However, in the present study, we obtained 9.7% to 14.3% false-negative rates using the same guidelines and conservative quantitative flag setting. Such high false-negative rates may have resulted from the large proportion of cancer patients in our hospital. Illnesses of greater severity will lead to more triggering values for peripheral smears.6 Patients with
False-Positive Analysisa
Unicel DxH 800 Parameter
WBC count Platelet count Hemoglobin, MCV, RDW Differential Suspect flags Total
ADVIA 2120i
XE 2100
TP, No.
FP, No.
PPV, %
TP, No.
FP, No.
PPV, %
TP, No.
FP, No.
PPV, %
4 9 5 3 23 30
13 10 4 10 9 41
23.5 47.4 55.6 23.1 71.9 42.3
5 7 2 1 23 30
14 10 1 6 9 34
26.3 41.2 66.7 14.3 71.9 46.9
5 10 3 3 40 44
13 10 4 9 27 52
27.8 50.0 42.9 25.0 59.7 45.8
Abbreviations: FP, false-positive; MCV, mean corpuscular volume; PPV, positive predictive value; RDW, red blood cell distribution width; TP, true positive; WBC, white blood cell. a All numbers are the actual number of slide reviews. Therefore, the total number is not the sum of columns because multiple rules were triggered for some cases.
Arch Pathol Lab Med—Vol 136, November 2012
Comparison Study on the Rates of Slide Review—Kim et al
1411
Table 5.
False-Negative Analysisa
Unicel DxH 800 Parameter
RBC morphology Platelet morphology Metamyelocyte, myelocyte, promyelocyte Atypical lymphocyte WBC morphology NRBC Blast Total
ADVIA 2120i
XE 2100
No.
%
No.
%
No.
%
22 11 9 2 1 0 0 43
51.2 25.6 20.9 4.7 2.3 0.0 0.0 100.0
23 12 7 1 2 1 0 43
53.5 27.9 16.3 2.3 4.7 2.3 0.0 100.0
17 9 2 1 0 0 0 29
58.6 31.0 6.9 3.4 0.0 0.0 0.0 100.0
Abbreviations: NRBC, nucleated red blood cell; RBC, red blood cell; WBC, white blood cell. a All numbers are the actual number of slide reviews. Therefore, the total number is not the sum of column entries because of the cases with multiple positive findings.
We evaluated the effect of changes in the cutoff values of suspect flags for the rates of slide review and falsepositive and false-negative rates by reanalyzing the same samples using different cutoff values. Whereas the review rates of the Unicel DxH 800 were relatively consistent, the XE 2100 presented a wide range of review rates, falsenegative rates, and false-positive rates according to the changes in cutoff levels of immature granulocytes and atypical lymphocytes. The ADVIA 2120i was sensitive only to change in the cutoff level of immature granulocytes. Details of optimization certainly depend largely on the patient population, laboratory need, and other local environments, and thus should be determined by each laboratory independently.11 The three hematologic analyzers showed different slide review rates. In a Q-Probes study by the College of American Pathologists, it was reported that the rates of manual review increased with more hospital beds.6 Hence, the appropriate rate of slide review may vary according to the clinical environment of the laboratory. In primary care clinics, high sensitivity might be needed for screening purposes. In addition, a low number of abnormal samples may save time and money despite high sensitivity. Meanwhile, in tertiary hospitals, high specificity might be helpful because most patients are screened at primary care clinics, and the proportion of severely ill patients who produce abnormal repetitive CBC results is high. Besides, because of the severity of diseases of patients in tertiary hospitals, a more rapid turnaround time may be required. Also, it is easy for a clinician to access peripheral blood smear tests, which can supple-
various malignancies under chemotherapy, radiotherapy, or multiple transfusions may show subtle to profound hematologic abnormalities. Some morphologic abnormalities, for example dimorphic RBCs after transfusion and presence of immature granulocytes during recovery after chemotherapy, are predictable and do not influence care.10 In our clinical setting, there was no clinically significant diagnosis discovered among the false-negative cases. In the study by the consensus group, the most common cause of false-negative results was the presence of immature granulocytes, whereas RBC morphologic abnormalities and platelet clumping were the most common in the present study.3 Except for the XE 2100, the major causes for false positives in this study were also different from those reported by the consensus group. Suspect flags were the most common causes of false positives for the consensus group. In this study, for the Unicel DxH 800 and the ADVIA 2120i, WBC count was the most common cause of false positives. These differences support the argument that different patient populations may produce different efficiencies in slide review. In this study, parameters linked to suspect flags showed the highest positive predictive values. Meanwhile, parameters related to WBC count and differential count presented lower positive predictive values. This clearly suggests the importance of suspect flags in detecting abnormal samples. Samples with low WBC, RBC, or platelet counts without morphologic abnormalities produced lower positive predictive values. Each laboratory can optimize its quantitative cutoff values of suspect flags based on its needs and priorities. Table 6.
Review Rates, False-Negative Rates, and False-Positive Rates According to the Cutoff of Quantitative Flags Unicel DxH 800
Cutoff of Quantitative Flags
a
ADVIA 2120i
b
XE 2100
c
Left Shift
Immature Granulocytes
Atypical Lymphocytes
RR, %
FNR, %
FPR, %
RR, %
FNR, %
FPR, %
RR, %
FNR, %
FPR, %
Lowd Low Low Low Low Medium High
Lowd Low Low Medium High Low Low
Lowd Medium High Low Low Low Low
23.7 23.3 23.3 23.7 23.7 23.7 23.7
14.3 14.3 14.3 14.3 14.3 14.3 14.3
13.7 13.3 13.3 13.7 13.7 13.7 13.7
21.3 21.0 21.0 19.0 18.0 21.3 21.3
14.3 14.3 14.3 16.7 17.7 14.3 14.3
11.3 11.0 11.0 11.3 11.3 11.3 11.3
32.0 29.7 29.3 29.3 28.3 32.0 32.0
9.7 10.0 10.0 11.0 12.0 9.7 9.7
17.3 15.3 15.0 16.0 16.0 17.3 17.3
Abbreviations: FNR, false-negative rate; FPR, false-positive rate; RR, review rate. a
For the Unicel DxH 800, low level corresponds to 3, medium to 2, and high to 1. For the ADVIA 2120i, low level corresponds to 1+, medium to 2+, and high to 3+. For the XE 2100, low level corresponds to 100, medium to 200, and high to 300. d The original setting in this study. b c
1412
Arch Pathol Lab Med—Vol 136, November 2012
Comparison Study on the Rates of Slide Review—Kim et al
ment morphologic review, in large hospitals if hematologic abnormality is clinically suspected. In our laboratory, considering these circumstances, we perform CBC tests requested by inpatient departments mainly with the ADVIA 2120i and tests requested by the outpatient departments mainly with the XE 2100. The strength of the present study is that results from representative samples, as they appear in routine testing in this investigation, reflect real conditions in a tertiary hospital with many cancer patients. However, this study had a small number of samples with positive findings; studies with more samples may be needed to obtain more precise results. In summary, for all 3 hematology analyzers evaluated, false-negative rates were higher than those recommended by the consensus group. The XE 2100 triggered the most slide reviews among the 3 analyzers, with the highest sensitivity and the lowest specificity. The rates of slide review, sensitivity, and specificity were comparable between the Unicel DxH 800 and the ADVIA 2120i. Because the present study clearly shows that the rates of slide review have distinct characteristics among the studied analyzers, individual laboratories should consider selecting the most appropriate analyzer in accordance with clinical characteristics including clinic size and patient population. Analyzers with high sensitivity may be advantageous in outpatient settings for screening
Arch Pathol Lab Med—Vol 136, November 2012
patients, whereas analyzers with high specificity may be beneficial in inpatient settings for rapid and efficient patient care. References 1. Pierre RV. Peripheral blood film review: the demise of the eyecount leukocyte differential. Clin Lab Med. 2002;22(1):279–297. 2. Hoyer JD. Leukocyte differential. Mayo Clin Proc. 1993;68(10):1027–1028. 3. Barnes PW, McFadden SL, Machin SJ, Simson E. The international consensus group for hematology review: suggested criteria for action following automated CBC and WBC differential analysis. Lab Hematol. 2005;11(2):83–90. 4. Johnson M, Samuels C, Jozsa N, Gorney K. Three-way evaluation of highthroughput hematology analyzers—Beckman Coulter LH 750, Abbott Cell-Dyn 4000, and Sysmex XE-2100. Lab Hematol. 2002;8(4):230–238. 5. Galloway MJ, Osgerby JC. An audit of the indications for the reporting of blood films: results from the National Pathology Benchmarking Study. J Clin Pathol. 2006;59(5):479–481. 6. Novis DA, Walsh M, Wilkinson D, St Louis M, Ben-Ezra J. Laboratory productivity and the rate of manual peripheral blood smear review: a College of American Pathologists Q-Probes study of 95 141 complete blood count determinations performed in 263 institutions. Arch Pathol Lab Med. 2006;130(5):596–601. 7. Froom P, Havis R, Barak M. The rate of manual peripheral blood smear reviews in outpatients. Clin Chem Lab Med. 2009;47(11):1401–1405. 8. CLSI. Reference Leukocyte (WBC) Differential Count (Proportional) and Evaluation of Instrumental Methods; Approved Standard—Second Edition. Wayne, PA: USA Clinical and Laboratory Standards Institute; 2007. H20-A2. 9. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;33(1):159–174. 10. Hoffmann JJ. How useful are haematology analyser flags? Clin Chem Lab Med. 2004;42(4):357–358. 11. Sireci A, Schlaberg R, Kratz A. A method for optimizing and validating institution-specific flagging criteria for automated cell counters. Arch Pathol Lab Med. 2010;134(10):1528–1533.
Comparison Study on the Rates of Slide Review—Kim et al
1413
