Hematopathology / FLOW CYTOMETRIC RETICULOCYTE COUNTING

Flow Cytometric Reticulocyte Counting

Parallel Evaluation of Five Fully Automated Analyzers: An NCCLSICSH Approach
Mauro Buttarello, MD, Pietro Bulian, MD, Giorgio Farina, MD, Valeria Temporin, MD,
Lucia Toffolo, MD, Ernesto Trabuio, MD, and Paolo Rizzotti, MD
Key Words: Reticulocyte counting; Automated analyzers

Abstract
We performed a parallel evaluation of 5 automated
reticulocyte analyzers. The guidelines were those
proposed by the National Committee for Clinical
Laboratory Standards and the International Council for
Standardisation in Haematology. Duplicate analyses
were performed for 225 healthy subjects and 115
patients affected by various diseases.
The reference intervals were different for each
method (ADVIA 120, 27-125 103/L [27-125
109/L]; CELL DYN 4000, 25-108 103/L [25-108
109/L]; GEN-S, 20-85 103/L [20-85 109/L]; SE
9500 RET, 23-95 103/L [23-95 109/L]; and VEGA
RETIC, 30-130 103/L [30-130 109/L]). The
comparisons of percentage counts with the microscopic
reference method were satisfactory for all automated
methods. However, a tendency to overestimate at low
counts was noted. This progressively increased from the
SE 9500 RET to the VEGA RETIC. The imprecision
was excellent for all the methods at normal and high
concentrations. This was higher at low concentrations.
When compared with the microscopic reference, the
analyzers showed satisfactory sensitivity at low counts
and excellent sensitivity at high counts. The overall
agreement varied from 74.8% for the GEN-S to 86.1%
for the SE 9500 RET.

100

Am J Clin Pathol 2001;115:100-111

Automated reticulocyte counting, made possible several

years ago by means of multipurpose or dedicated flow
cytometers, recently has been extended to common hematology analyzers, initially in semiautomation and later in
complete automation. The advantages derived from automation are high throughput, reduced imprecision (essentially
owing to the high number of cells counted), and the production of further clinically useful parameters, such as the immature reticulocyte fraction and reticulocyte indices.
Problems still exist that essentially depend on the
differing sensitivities of the dyes used to stain the RNA of
reticulocytes, on the technology used to identify positive cells
(fluorescence, light scattering, absorbance), and on the software that is more or less capable of separating reticulocytes
from erythrocytes (since there is a physiologic continuum
between these populations) and from other cells, such as
platelets or nucleated RBCs (NRBCs). All this results in
counts that sometimes disagree with one another depending
on the method, with the consequent need for reference intervals that are method specific.1 A further limit is due to the
different performances on samples with severe reticulocytopenia.1,2 Differences also have been noted in reticulocytosis,3 but rarely do these have implications for clinical decision making, as in patients with reticulocytopenia. The lack
of calibration material that is stable over time and that can
be used by all methods contributes to the interpretive uncertainties regarding the comparison of methods and the agreement of the results of the various commercially available
systems.
The use of fresh blood, the ideal material for reticulocyte
counts, as a calibrator has some limitations such as the
following: (1) the poor stability over time of the concentration of reticulocytes, which undergo progressive maturation;
American Society of Clinical Pathologists

Hematopathology / ORIGINAL ARTICLE

(2) the lack of a universally accepted reference method of

measuring reticulocyte concentrations; and (3) the exacting
calibration procedures that are strictly manufacturer dependent and that cannot easily be modified by the user.
Two standards have been published: (1) H44-A,4 jointly
prepared by the National Committee for Clinical Laboratory
Standards (NCCLS) and the International Council for Standardisation in Haematology (ICSH), proposes guidelines for
the evaluation of automated methods of reticulocyte counts.
(2) The ICSH standard5 proposes a microscopic reference
method for reticulocyte counts.
The aim of our study was, through the use of the aforementioned guidelines, to evaluate the analytic performance of
the 5 most recently available automated hematology analyzers
by a parallel study in a single institution. Five analyzers were
studied. Particular attention was given to the performance on
samples with very low reticulocyte concentrations.

Materials and Methods

Analyzers
Five hematology analyzers with the option of reticulocyte counts in complete automation were evaluated over 6
weeks, in a parallel study at the Clinical Pathology Laboratory of the Geriatric Hospital of Padua, Padua, Italy. The
analyzers studied were the following: CELL DYN (CD)
4000 (software version 7.5, Abbott, Santa Clara, CA);
VEGA RETIC (currently called PENTRA 120; software
version 3.16, ABX, Montpellier, France); ADVIA 120 (software version 1.16, Bayer Diagnostic Division, Tarrytown,
NY); GEN-S (software version 1D, Coulter-Beckman,
Hialeah, FL); and SE 9500 RET (previously called SEAVANTE; software version B00-04, Sysmex, Kobe, Japan).
These instruments analyze tens of thousands of cells per
sample and, with stains that selectively bind the RNA of
reticulocytes, they can separate stained cells (reticulocytes)
from nonstained (erythrocytes) and from those whose size or
level at staining place them outside the reticulocyte gate
(platelets, NRBCs, leukocytes). The stains and general principles used are reported in Table 1. The instruments were
calibrated according to the specifications of the manufacturers by specialists from Italian representatives of companies at the time of installation. During evaluation they were
checked 3 times a day, using a multilevel control supplied by
each manufacturer.
Samples
A total of 225 healthy subjects, according to NCCLS
H44-A criteria, were selected for calculation of the reference
intervals for tested analyzers. The subjects were males and
American Society of Clinical Pathologists

Table 1
Analytic Methods of Reticulocyte Counters*
Analyzer
ADVIA 120
CELL DYN 4000
GEN-S
SE 9500 RET
VEGA RETIC
*

Dye

Technology

Oxazine 750
CD4K 530
New methylene blue
Auramine O
Thiazole orange

Collected
Events

Absorbance
~40,000
Fluorescence
50,000
Light scattering
32,000
Fluorescence
~30,000
Fluorescence
32,000

See text for proprietary information.

females 3 to 50 years old. From these subjects, only 126

samples were used for the manual microscopic reference
method. An additional 115 patients affected by various
diseases or conditions (bone marrow transplantation,
hemolytic and posthemorrhagic anemias, bone marrow
aplasia, acute and chronic inflammations, neoplastic
diseases, nutritional anemias before and during treatment,
myeloproliferative and lymphoproliferative diseases) were
examined.
To study the behavior of the analyzers on samples with
reticulocyte counts near zero, samples were collected from
46 patients who had undergone intensive chemotherapy for
acute leukemia or were in the aplastic phase after bone
marrow transplantation. None of these patients had received
transfusions during the 5 days before sample collection.
Since some potential interfering substances on reticulocyte
counts are known to exist,4 the behavior of the analyzers was
verified in the presence of the following substances: thrombocytes (n = 6; range, 850-1,310 103/L [850-1,310
109/L]); giant platelets (n = 8); microcytic RBCs (n = 9;
range, 58-67 fL); macrocytic RBCs (n = 5, range, 108-120
fL); dimorphic RBC population (n = 4); lymphocytes (n = 3;
range, 54,000-93,000/L [54.0-93.0 109/L]); NRBCs (n =
4; range, 400-5,000/L [0.4-5.0 109/L]). All samples were
collected by venipuncture using 5-mL tripotassium
EDTAevacuated tubes (Vacutainer, Becton Dickinson,
Rutherford, NJ). For patients with bone marrow aplasia,
samples were taken from leftover blood collected for routine
CBC counts. The wedge films were prepared within 2 hours
after collection for microscopic assessment. All samples
were processed within 4 hours using the various analyzers.
Microscopic Assessment
The new blue methylene method was used (Sigma Diagnostics, St Louis, MO), and reticulocytes were counted
according to the indications of the ICSH.5 Briefly, 2 sets of
trained clinical pathologists proceeded with the counts,
blinded, on each of 2 different films for each sample. With a
2-headed microscope, all the erythrocytes and reticulocytes
observed in several fields (with edge rule) were counted for
each film until a count of at least 1,000 RBCs was reached.
Am J Clin Pathol 2001;115:100-111

101

Buttarello et al / FLOW CYTOMETRIC RETICULOCYTE COUNTING

The mean value obtained by both sets of observers was used

for comparison with the automated methods.
Statistical Analysis
Every sample was analyzed in duplicate and in random
order. The absolute value ( 109/L) was provided by the
various analyzers, while for the reference method it was
obtained by multiplying the percentage values by the RBC
count given by the routine hematology analyzer. The mean
values obtained for each method were compared for a total
of 256 samples comprising healthy subjects and patients.
The comparison with the manual reference method was
performed for all samples and separately for the subgroup of
the reticulocytopenic samples. This comparison was
presented graphically with 95% confidence intervals, which
represent the standard error of the parameter, and with linear
regression statistics. 4 The results obtained also were
compared by nonparametric statistics based on ranks: first a
Friedman test and subsequently a Wilcoxon 2-tailed test.6
To verify interchangeability of the various automated
methods, the reciprocal comparison was summarized in the
regression and correlation statistics.7
The imprecision was estimated with 2 methods: (1) with
1-way analysis of variance4,8 between duplicates of all
samples analyzed divided into 3 subgroups (low, normal,
high) selected after having established reference intervals for
each method; and (2) with the imprecision profile 9-11
extended to a wide range of values (1.5-250 103/L [1.5 to
250 109/L]) by performing repeated analysis (8-10 counts)
on several samples (15-20 samples) for each analyzer.
Clinical Usefulness
The clinical usefulness (ability to distinguish between
normal and abnormal conditions) was evaluated on samples
from the 115 patients using the criteria proposed by the
NCCLS H20-A12 and by the ICSH.8 The absolute count was
classified by reference and tests into normal, abnormal high,
and abnormal low according to whether the results were
above or below the reference interval.

Linearity and Carryover

Linearity was evaluated by mixing (in distinct experiments) a sample with a moderately elevated reticulocyte
concentration (mean of 4 counts from 94.4 103/L [94.4
109/L; 6.3%] for the SE 9500 RET to 164 103/L [164
109/L; 8.7%] for the CD 4000) in decreasing proportion with
an ABO-compatible reticulocytopenic sample (mean of 4
counts from 1.4 103/L [1.4 109/L; 0.06%] for the SE
9500 RET to 10.5 103/L [10.5 109/L; 0.6%] for the
GEN-S). The counts were performed in duplicate for each
dilution, and the mean value was used. The calculation of
carryover was done in separate experiments for each instrument by analyzing a pair of samples: 1 sample at a high
reticulocyte concentration (from 250 103/L [250 109/L]
for the GEN-S to 117 103/L [117 109/L] for the VEGA
RETIC) analyzed 3 times (i1, i2, i3), immediately followed by
1 sample at a low concentration (from 3.5 103/L [3.5
109/L] for the CD 4000 to 12.4 103/L [12.4 109/L] for
the VEGA RETIC), also analyzed 3 times consecutively (j1,
j2, j3), and using the formula [(j1j3)/(i3j3)] 100.8

Results
Reference Intervals
The reference intervals in percentage and absolute
values of the 5 tested methods and the reference method are
given in Table 2. Since the distributions are approximately
log-normal, the intervals were calculated with a nonparametric method (the central 95% of the distribution). Note
that it is possible to distinguish 2 groups: one consisting of
the microscopic reference method, SE 9500 RET, and GENS with median values between 43 and 46 103/L (43-46
109/L), and the other consisting of the ADVIA 120, the CD
4000, and the VEGA RETIC with median values between 57
and 60 103/L (57-60 109/L).
The results agree with published results obtained
with the microscopic method1,2 and with previous results

Table 2
Reticulocyte Reference Intervals*
Reticulocyte Count, %
Median
Reference (new methylene blue) (n = 126)
ADVIA 120 (oxazine 750) (n = 221)
CELL DYN 4000 (CD4K 530) (n = 216)
GEN-S (new methylene blue) (n = 218)
SE 9500 RET (auramine O) (n = 213)
VEGA RETIC (thiazole orange) (n = 217)
* See

1.0
1.2
1.3
1.0
0.9
1.3

95% Range
0.4-2.3
0.6-2.5
0.4-2.2
0.5-1.8
0.5-1.9
0.6-2.6

Reticulocyte Count, 103/L ( 109/L)

Median

95% Range

46 (46)
58 (58)
57 (57)
43 (43)
44 (44)
60 (60)

19-111 (19-111)
27-125 (27-125)
25-108 (25-109)
20-85 (20-85)
23-95 (23-95)
30-130 (30-130)

text for proprietary information.

102

Am J Clin Pathol 2001;115:100-111

American Society of Clinical Pathologists

Hematopathology / ORIGINAL ARTICLE

for the SE 9500 RET (data referred to the R 1000

or 3000), 2,4 the ADVIA 120 (formerly the H3), 1,4 the
GEN-S (previous data referred to MAXM or STKS),2 and
the CD 4000. 13 Data are not available for the VEGA
RETIC.

This behavior also was evident on the bias plot in

Figure 2, which shows the comparison of the reticulocy-

topenic samples only (<10 103/L [<10 109/L]). The

tendency was toward overestimation, with average differences of 4.1 103/L (4.1 109/L) for the SE 9500 RET, 6.4
103/L (6.4 109/L) for the CD 4000, 6.6 103/L (6.6
109/L) for the ADVIA 120, 10.8 103/L (10.8 109/L) for
the GEN-S, and 11.4 103/L (11.4 109/L) for the VEGA
RETIC.
Even if the overestimation could be due to difference
between the discrete manual method, which measures the
amount of a precipitate, and the automated methods, which
use a continuum measure of RNA content, this cannot
completely justify the difference between instruments. In
fact, independent of the results obtained by the microscopic
reference method, the concentrations obtained on the 10
lowest counts of aplastic marrow samples by each analyzer
were included between 1.0 and 2.0 103/L (1.0-2.0
109/L; mean, 1.7 103/L [1.7 109/L]) for the SE 9500
RET; between 0.9 and 2.6 103/L (0.9-2.6 109/L; mean,
1.7 103/L [1.7 109/L]) for the CD 4000; between 1.2
and 3.3 103/L (1.2-3.3 109/L; mean, 2.3 103/L [2.3
109/L]) for the ADVIA 120; between 3.1 and 6.0 103/L
(3.1-6.0 109/L; mean, 5.1 103/L [5.1 109/L]) for the
GEN-S; and between 2.9 and 7.1 103/L (2.9-7.1 109/L;
mean, 5.7 103/L [5.7 109/L]) for the VEGA RETIC.
The analytic sensitivity was, therefore, particularly noteworthy for the SE 9500 RET and the CD 4000.

Comparison Between Methods

The comparison between the manual reference method
and tests according to the NCCLS H44-A recommendations4
is shown in Figure 1A, Figure 1B, Figure 1C, Figure
1D, and Figure 1E. The linear regression statistics are
summarized in Table 3 . Figure 1F shows the corresponding regression line of the 5 methods.
Table 4 shows the results obtained with the Wilcoxon
test in percentages and absolute counts.
Note that all methods have a tendency to overestimate,
both in percentage and absolute values with respect to the
microscopic method. This overestimation becomes negligible for the SE 9500 RET (in fact, it is the only method that
gave differences that were not statistically significant) and
increases progressively to its highest value (0.39%, 17.41
103/L [17.41 109/L]) for the VEGA RETIC.
Even regression statistics confirm the existence of a
constant positive intercept (from 11.15 103/L [11.15
109/L] for the SE 9500 RET to 23.42 103/L [23.42
109/L] for the VEGA RETIC) with a slope less than 1. This
agrees with the tendency to overestimate at low values and to
underestimate at normal and high values.

Table 3
Comparison of Percentage Results for Reticulocyte Count: Linear Regression Statistics*
No. of Samples

Intercept

Slope

r2

256
256
256
256
256

0.38
0.39
0.36
0.20
0.49

0.89
0.96
0.86
0.84
0.91

0.76
0.82
0.73
0.82
0.81

ADVIA 120
CELL DYN 4000
GEN-S
SE 9500 RET
VEGA RETIC
*

See text for proprietary information.

Table 4
Comparison of Methods: Analysis of Data by the Wilcoxon Test*
ADVIA 120

Mean
Minimum
Maximum
Mean difference
SD of differences
P
*

CELL DYN 4000

109/L

1.18
0.02
8.66
0.25
0.65
.01

58.51
1.20
251.60
11.12
24.76
.01

1.52
0.04
8.25
0.35
0.57
.01

109/L
58.85
0.90
269.00
11.46
21.89
.01

GEN-S
%
1.38
0.11
6.65
0.20
0.69
.01

109/L
52.76
3.07
220.49
5.37
27.22
.01

SE 9500 RET

VEGA RETIC

109/L

109/L

1.20
0.04
6.86
0.02
0.54
.50

47.80
1.00
237.40
0.42
20.65
.75

1.56
0.13
7.23
0.39
0.57
.01

64.80
2.90
278.55
17.41
25.03
.01

The values 109/L are Systme International units; traditional units are 103/L. See text for proprietary information.

American Society of Clinical Pathologists

Am J Clin Pathol 2001;115:100-111

103

Buttarello et al / FLOW CYTOMETRIC RETICULOCYTE COUNTING

6
CD 4000

10

ADVIA 120

10

5
4

5
4

0
0

10

Reference Method

10

10

Reference Method

D
10

10

SE 9500 RET

GEN-S

5
4

5
4

1
0

0
0

10

Reference Method

Reference Method

Figure 1 Comparison of tested instruments (percentage count) against microscopic reference method. A, ADVIA 120. B,
CELL DYN (CD) 4000. C, GEN-S. D, SE 9500 RET.

Table 5 gives the regression statistics when methods

were compared reciprocally to verify possible interchange.

The results were better when the various automated systems
were compared among themselves rather than against the
microscopic reference method.
104

Am J Clin Pathol 2001;115:100-111

Imprecision
The coefficients of variation (CVs) are given in Table
6. The imprecision of all the methods decreased as the
reticulocyte concentration increased and did not differ from
results obtained in similar studies.1,2,14-16 It should be noted
American Society of Clinical Pathologists

Hematopathology / ORIGINAL ARTICLE

F
10

10

Test Method

VEGA RETIC

5
4

5
4

ADVIA
CD 4000
GEN-S
SE 9500
VEGA

0
0

3
4
5
6
Reference Method

10

4
5
6
Reference Method

10

E, VEGA RETIC. The lines represent the 95% binomial confidence interval. F, Comparison of regression lines. See text for
proprietary information.
Table 5
Simple Regression Analysis of Absolute Reticulocyte Count From 256 Samples*

ADVIA 120
Intercept
Slope
r2
CELL DYN 4000
Intercept
Slope
r2
GEN-S
Intercept
Slope
r2
SE 9500 RET
Intercept
Slope
r2
VEGA RETIC
Intercept
Slope
r2
Reference
Intercept
Slope
r2

ADVIA 120

CELL DYN 4000

GEN-S

SE 9500 RET

VEGA RETIC

Reference

3.758
0.930
0.816

7.076
0.975
0.817

8.312
1.050
0.828

3.064
0.856
0.767

18.141
0.852
0.704

7.562
0.877
0.816

9.586
0.934
0.796

6.563
1.094
0.954

0.299
0.904
0.908

18.202
0.858
0.758

3.714
0.838
0.817

2.609
0.852
0.796

7.134
0.954
0.796

0.839
0.801
0.783

17.429
0.746
0.627

1.641
0.789
0.828

3.527
0.872
0.954

3.804
0.834
0.796

3.122
0.786
0.862

11.151
0.773
0.773

12.329
0.897
0.767

5.649
1.005
0.908

13.273
0.977
0.783

12.393
1.096
0.862

23.429
0.873
0.706

0.994
0.827
0.704

4.618
0.884
0.758

2.984
0.842
0.627

0.380
0.999
0.773

5.006
0.809
0.706

* See text for proprietary information.

that all methods at all concentrations showed an imprecision less than that of the microscopic method. 1,2,17
The results generally indicated a lower imprecision for
the SE 9500 RET and the CD 4000, followed by the
American Society of Clinical Pathologists

GEN-S and then the VEGA RETIC and ADVIA 120. The
imprecision profile also showed analogous behavior, and,
as seen in Figure 3, near moderate reticulocytopenia (10
103/L [10 109/L]); the CV was a minimum of 11% for
Am J Clin Pathol 2001;115:100-111

105

Buttarello et al / FLOW CYTOMETRIC RETICULOCYTE COUNTING

the SE 9500 RET to a maximum of 25% for the VEGA

RETIC.

ADVIA

50
0

Linearity and Carryover

The linearity was excellent for all systems, both in
absolute values (intercept between 0.585 and 1.09 103/L
[0.585-1.09 109/L]; slope between 0.993 and 1.005, and r2
>0.99) and in percentages. Figure 4 shows the corresponding linearity graph for the CD 4000 and for the GENS. The carryover calculated on the absolute concentrations
was negligible and varied from 0.8% for the CD 4000 and
the GEN-S to 1.8% for the SE 9500 RET, 1.9% for the
ADVIA 120, and 3.5% for the VEGA RETIC.

50

CD 4000

50
0
50

GEN-S

50
0

SE 9500 RET

50

50
0
50

VEGA RETIC

50
0
50
0

10

Reticulocytes ( 10 9 /L)

Figure 2 Comparison of methods: plot of paired differences on reticulocytopenic samples. ADVIA 120 minus
reference method; mean difference, 6.6 103/L (6.6
109/L). CELL DYN (CD) 4000 minus reference method;
mean difference, 6.4 103/L (6.4 109/L). GEN-S minus
reference method; mean difference, 10.8 103/L (10.8
109/L). SE 9500 RET minus reference method; mean difference, 4.1 103/L (4.1 109/L). VEGA RETIC minus reference method; mean difference, 11.4 103/L (11.4 109/L).
Values shown in the figure are given in Systme International units; traditional units are 103/L. See text for proprietary information.

Clinical Usefulness
Table 7 and Table 8 give the results for clinical
usefulness for 115 patients by using only absolute concentrations. The values of sensitivity and the predictive value of a
positive result were calculated separately for low and
high samples; the calculation of specificity, of the predictive value of a negative test, and of concordance refer to the
entirety of samples. The incorrect classification of patients
concerns adjacent groups (low vs normal or high vs normal)
with the exception of only 1 patient with beta-thalassemic
trait and microcytic RBCs (mean corpuscular volume, 59
m3 [59 fL]) classified as high with the GEN-S (reticulocyte
count, 137 103/L [137 109/L]) and low (reticulocyte
count, 7.1 103/L [7.1 109/L]) with the reference method.
Even the other methods showed a somewhat contained overestimation in this sample, placing the result in the normal
group (SE 9500 RET, 51.5 103/L [51.5 109/L]; CD
4000, 69 103/L [69 109/L]; ADVIA 120, 80 103/L
[80 109/L]; VEGA RETIC, 88.6 103/L [88.6 109/L]).

Discussion
The popularity that automated methods have obtained
depends on the possibility of replacing manual visual counts,
which in routine are subjective, highly imprecise, and
tedious, and on the advantages that derive from a precise and
objective count mainly at low values, necessary for the

Table 6
Imprecision: Duplicate Determinations*
ADVIA 120
Concentration
Low
Normal
High

No.
48
260
8

Mean
6.9
60.1
164.8

CELL DYN 4000

CV

No.

24.1
10.9
6.5

47
257
24

Mean
9.7
57.8
141.6

GEN-S

CV

No.

18.1
6.5
3.5

39
261
32

Mean
12.9
45.4
130.0

SE 9500 RET

VEGA RETIC

CV

No.

Mean

CV

No.

Mean

CV

15.1
10.3
5.1

48
256
24

9.7
46.7
137.1

15.0
6.7
4.3

42
257
19

13.4
64.9
164.

18.9
11.7
6.9

CV, coefficient of variation (given as percentage); Mean, absolute counts.; No., number of samples.
* See text for proprietary information.

106

Am J Clin Pathol 2001;115:100-111

American Society of Clinical Pathologists

Hematopathology / ORIGINAL ARTICLE

60

ADVIA

SE 9500

CD 4000

VEGA RETIC

GEN-S

50

CV (%)

40

ADVIA
CD 4000
GEN-S
SE 9500
VEGA RETIC

30

20

10

0
0

50

100

150

200

250

Reticulocytes ( 10 9/L)

Figure 3 Imprecision profile of reticulocyte absolute count. Values in the figure are given in Systme International units;
traditional units are 103/L. The vertical line represents the CV of all instruments at 10 109/L. See text for proprietary information.

diagnosis of hypoplastic anemias or for the early monitoring

of erythropoietic regeneration of bone marrow. A further
advantage of automated methods, not considered in this
study, is the identification of other reticulocyte parameters,
such as the immature reticulocyte fraction or the reticulocyte
indices, such as the mean reticulocyte volume or the mean
hemoglobin content, which have been shown to be useful in
several clinical conditions.18,19
The increase in the number of methods available (the 5
instruments evaluated represent the most recent and sophisticated evolution of these methods) requires, however, that the
results from the various methods agree with one another.

Differences in counts are clearly evident in the College of

American Pathologists survey 3 in which the use of
stabilized control material could justify some unevenness,
especially for instruments based on light-scattering
techniques.
It should be noted that even previous comparisons based
on the use of fresh blood have shown some disagreement
(H3 vs R 1000, MAXM vs R 1000) with the need for
specific reference intervals.1,2,20 The present study confirms
the diversity of the reference intervals, which partially
overlap in 2 subgroups; the first consists of the microscopic
reference method, the SE 9500 RET, and the GEN-S, and the

Table 7
Clinical Usefulness: Distributional Classification (Absolute Counts)*
ADVIA 120
Reference
Method Low Normal
Low
Normal
High
Total
*

44
4
0
48

8 (2)
37 (5)
2 (1)
47

CELL DYN 4000

GEN-S

SE 9500 RET

High

Low

Normal

High

Low

Normal

High

0
4 (1)
16 (11)
20

45 (1)
3
0
48

7 (1)
35
0
42

0
7
18 (1)
25

38
1
0
39

13
30
0
43

1
14 (1)
18
33

Low Normal
45
3
0
48

7
36
0
43

VEGA RETIC

High
0
6
18
24

Low Normal
45 (7) 7
4
33 (3)
0
3
49
43

High

Total

0
8
15 (4)
23

52
45
18
115

Parentheses indicate number of samples with flag. See text for proprietary information.
Obtained by multiplying the percentage count of the reference method by the RBC count given by the routine hematology analyzer.

American Society of Clinical Pathologists

Am J Clin Pathol 2001;115:100-111

107

Buttarello et al / FLOW CYTOMETRIC RETICULOCYTE COUNTING

B
180

180
10

2
0

100

140

120

6
Measured Values ( 10 /L)

Percent

140
Measured Values ( 10 /L)

10
160

Percent

160

4 6 8 10
Percent

80
60

120

6
4
2
0

100

8 10

Percent

80
60

40

40

20

20
0

0
0

20

40

60

80

100

120

140

160

180

20

40

60

80

100

120

140

160

180

Theoretical Values ( 10 /L)

Theoretical Values ( 10 /L)

Figure 4 CELL DYN 4000 (A) and GEN-S (B) linearity. Portions of a sample with a high reticulocyte count were added in
different ratios to an ABO-compatible reticulocytopenic sample (the insets represent the percentage data). Values in the figure
are given in Systme International units; traditional units are 103/L divided by 1. See text for proprietary information.

second consists of the CD 4000, the ADVIA 120, and the

VEGA RETIC. Even more obvious is the difference in the
median between these 2 subgroups.
The comparison (percentage values) of the various
methods with the reference method showed generally satisfactory behavior (intercept between 0.2 and 0.49 103/L
[0.2-0.49 109/L], slope between 0.84 and 0.96, and determination coefficient between 0.73 and 0.82). Detailed
analysis, however, showed a tendency, which varied
according to the method, to overestimate at low levels and
underestimate, to a lesser extent, at high levels. These results
were confirmed by data from previous studies, even when
the comparison was based on different methods for

microscopic counting.1,2,20 When the various automated

methods were compared among themselves, better results
were obtained than when they were compared with the
microscopic reference method, without, however, strict
agreement.
For diagnostic purposes, at elevated concentrations, the
clinical sensitivity was excellent; the same was not true for
low concentrations in which an overestimation by several
systems reduced the sensitivity to nonoptimal conditions.
When the reticulocyte was used to verify the severe reticulocytopenia in patients with bone marrow aplasia or to monitor
early erythropoietic response, the interchange of the various
methods could be misleading owing to the differences in

Table 8
Assessment of Distributional Classification*

Sensitivity (%)
Low reticulocyte count
High reticulocyte count
Specificity (%)
Predictive value (%)
Positive result
Low reticulocyte count
High reticulocyte count
Negative result
Total agreement
* Numbers

108

ADVIA 120

CELL DYN 4000

GEN-S

SE 9500 RET

VEGA RETIC

84.6 (71.9-93.1)
88.9 (65.3-98.6)
82.2 (65.3-98.6)

86.5 (74.2-94.4)
100.0 (81.5-100.0)
77.8 (62.9-88.8)

73.1 (59.0-84.4)
100.0 (81.5-100.0)
66.7 (51.1-80.0)

86.5 (74.2-94.4)
100.0 (81.5-100)
80 (65.4-90.4)

86.5 (74.2-94.4)
83.3 (58.6-96.4)
73.3 (58.1-85.4)

91.6
80.0
78.7
83.4

93.8
72.0
83.3
85.2

97.4
54.5
69.8
74.8

93.8
75.0
83.7
86.1

91.8
65.2
76.7
80.8

in parentheses are 95% confidence intervals. See text for proprietary information.

Am J Clin Pathol 2001;115:100-111

American Society of Clinical Pathologists

Hematopathology / ORIGINAL ARTICLE

ADVIA
CD 4000
GEN-S
SE 9500
VEGA

Reticulocytes ( 10 9/L)

30

20

10

0
5

10

15

20

15

20

Days

ADVIA
CD 4000
GEN-S
SE 9500
VEGA

Reticulocytes ( 10 9/L)

30

20

10

0
5

10

Days

Figure 5 Reticulocyte response during bone marrow

aplasia. Data from patient 1 (A) and patient 2 (B) after
chemotherapy for acute leukemia. The erythropoietic
response was between days 14 and 20 for patient 1 and
between days 12 and 20 for patient 2. Values in the figure
are given in Systme International units; traditional units are
103/L. See text for proprietary information.

counts at these concentrations. For example, the reticulocyte

concentrations in Figure 5 are for 2 patients during bone
marrow aplasia after chemotherapy and at the first evidence
of marrow response. While the increase in the values during
the response phase was emphasized by all analyzers (with
the exception of the VEGA RETIC for patient 1), strict
concordance could be found only between the SE 9500 RET
and the CD 4000; the ADVIA 120 showed intermediate
behavior, and the GEN-S and VEGA RETIC showed
discrete oscillations, even in the phase of complete aplasia,
with values (patient 2) that at times exceeded those indicated
by the SE 9500 RET and the CD 4000 at the time of early
erythropoietic response of bone marrow (patient 1). It
follows that even when monitoring during the early postaplasia phase, it is mandatory to use the same method (or at
American Society of Clinical Pathologists

least methods with overlapping values), possibly chosen

from those with the highest analytic sensitivity.
The imprecision of the automated methods was notably
less for the manual-visual method mainly at low concentrations. The analytic goal for imprecision, which currently is
widely shared, is based on biologic variability.21-23 Among
the various clinical conditions, monitoring has the most
restrictive criteria, with an imprecision goal of the CV
percentage of 0.5 CVb or less (where CVb represents withinsubjects biologic variability). In this case, the analytic imprecision adds 11.8% to the overall variability, but this value
increases to 41.4% when the CV percentage is equal to the
CVb. It therefore is obvious that the lower the imprecision
compared with biologic variability, the lower the difference
between successive measurements needed before the difference can be considered significant. The studies published on
within-subjects variability do not provide concordant results;
they range from 11%24 to 20%,25 therefore with different
analytic goals (5.5% and 10%, respectively). These goals are
reached or approximated only for high or normal reticulocyte
concentrations and with better results from the CD 4000 and
the SE 9500 RET. For concentrations below the reference
interval these results are still distant from the analytical
goals.
The need for low imprecision in tests used for reticulocytopenic patients can be justified in certain clinical applications, such as monitoring the early erythropoietic response
after bone marrow transplantation or spontaneous or
medically induced aplasia.26 In such cases, characterized by
reticulocyte concentrations less than 10 103/L (10
109/L), even small increases, with trends confirmed during
the following days, are an indication of recovery of erythropoiesis by the bone marrow. It is, however, questionable
whether similar analytic goals must be extended from
normal and high reticulocyte concentrations to severe reticulocytopenia. In fact, at a concentration of 5 103/L (5
109/L), the maximum variability allowed for a goal of 5.5%
would be 0.250 103/L (0.250 109/L), which is a negligible value when clinical decision making is concerned. It
should be pointed out that some systems, like the GEN-S and
the ADVIA 120, show an improvement over the previous
semiautomated methods proposed by the same manufacturers as MAXM1 and STKS20 or H3,2 respectively. The
precision profile, when examined over a wide range of
concentrations and with a sufficient number of samples,
presents a double advantage: (1) the possibility of knowing
the imprecision at every concentration, most important since
this is nonlinear; and (2) easy comparison because profiles of
different analyzers can be plotted on the same graph. With
this approach at normal or high concentrations the CV
percentage is less than 10% for all systems. Differences,
however, exist: the best results are from the SE 9500 RET
Am J Clin Pathol 2001;115:100-111

109

Buttarello et al / FLOW CYTOMETRIC RETICULOCYTE COUNTING

and CD 4000. At lower values, these differences are amplified further.

Regarding clinical usefulness, while behavior is satisfactory when the reticulocyte concentration increases, it is less
so when the concentration is low, with a sensitivity from
86.5% for the VEGA, the SE 9500 RET, and the CD 4000 to
84.6% for the ADVIA 120 and 73.1% for the GEN-S. Even
the exclusion of flagged samples (which would have to
undergo a careful microscopic count) does not improve the
sensitivity, causing a decrease in overall efficiency since
many correctly classified but flagged samples would be
excluded.
The potential interfering substances studied did not
demonstrate systematic effects on the counts. It should be
noted that in 3 samples with microcytosis and 1 with microcytosis with a dimorphic population, the GEN-S gave an
overestimation with respect to the microscopic count (from
33 to 130 103/L [33-130 109/L]) and the other instruments. In one of these samples, there also was overestimation by the VEGA RETIC (87 103/L [87 109/L]) and
the CD 4000 (60 103/L [60 109/L]). In one of the
samples containing NRBCs, there was an overestimation by
the VEGA RETIC (43 103/L [43 109/L]), the GEN-S
(43 103/L [43 109/L]), and the ADVIA 120 (131
103/L [131 109/L]). All instruments, however, always
flagged the result and indicated the suspicion of NRBCs.
Regarding the interference of NRBCs, our samples had
modest concentrations; therefore, further study is necessary
before reaching a final verdict. In conclusion, we can affirm
that total automation of reticulocyte counts represents a definite improvement over microscopic counts and, in certain
respects (greater precision), even over the previous semiautomated methods. Several problems remain unresolved: (1) the
incomplete agreement of the counts with the consequent
necessity of method-specific reference intervals; (2) the lessthan-optimal analytic sensitivity for some systems; and (3)
the difficulty using some analyzers to monitor early erythropoietic response when using only the reticulocyte count
owing to the imprecision at low concentrations.
From the Laboratory of Clinical Pathology, Geriatric Hospital,
Padua, Italy.
Address reprint requests to Dr Buttarello: Laboratorio Di
Patologia Clinica, Ospedale Geriatrico, Via Vendramini 7, 35100
Padova, Italia.

References
1. Buttarello M, Bulian P, De Pr M, et al. Reticulocyte
quantification by Coulter MAXM VCS (volume,
conductivity, light scatter) technology. Clin Chem.
1996;42:1930-1937.

110

Am J Clin Pathol 2001;115:100-111

2. Buttarello M, Bulian P, Venudo A, et al. Laboratory

evaluation of the Miles H3 automated reticulocyte counter: a
comparative study with manual reference method and Sysmex
R-1000. Arch Pathol Lab Med. 1995;119:1141-1148.
3. Davis BH, Spier CM, Kachin J, et al. College of American
Pathologists reticulocyte proficiency testing program using
surrogate blood: insights into contemporary clinical practice.
Lab Hematol. 1997;3:84-91.
4. National Committee for Clinical Laboratory Standards.
Methods for Reticulocyte Counting (Flow Cytometry and
Supravital Dyes). Approved Guideline. H44-A. Wayne, PA:
NCCLS; 1997.
5. International Council for Standardisation in Haematology
Expert Panel on Cytometry. Proposed reference method for
reticulocyte counting based on the determination of the reticulocyte to red cell ratio. Clin Lab Haematol. 1998;20:77-79.
6. Glantz SA. Primer of Biostatistics. New York, NY: McGrawHill; 1987:chap 10.
7. Davis BH, Bigelow NC, Koepke JA, et al. Flow cytometry
reticulocyte analysis: multiinstitutional interlaboratory
correlation study. Am J Clin Pathol. 1994;102:468-477.
8. International Council for Standardisation in Haematology
Expert Panel on Cytometry. Guidelines for the evaluation of
blood cell analysers including those used for differential
leucocyte and reticulocyte counting and cell marker
applications. Clin Lab Haematol. 1994;16:157-174.
9. Sadler WA, Smith MH. Use and abuse of imprecision profiles:
some pitfalls illustrated by computing and plotting confidence
intervals. Clin Chem. 1990;36:1346-1350.
10. Hbl W, Tlustos L, Bayer PM. Use of precision profiles to
evaluate precision of the automated leukocyte differential.
Clin Chem. 1996;42:1068-1073.
11. Buttarello M, Bulian P, Temporin V, et al. Sysmex SE-9000
Hematology Analyzer: performance evaluation on leukocyte
differential counts using an NCCLS H20-A protocol. Am J
Clin Pathol. 1997;108:674-686.
12. National Committee for Clinical Laboratory Standards.
Reference Leukocyte Differential Count (Proportional) and
Evaluation of Instrumental Methods. Approved Standard.
H20-A. Villanova, PA: NCCLS; 1992.
13. Kim YR, Kantor J, Landayan M, et al. A rapid and sensitive
reticulocyte method on a high-throughput hematology
instrument. Lab Hematol. 1997;3:19-26.
14. dOnofrio G, Kim YR, Schulze S, et al. Evaluation of the
Abbott Cell Dyn 4000 automated fluorescent reticulocyte
measurements: comparison with manual, FACScan and
Sysmex R1000 methods. Clin Lab Haematol. 1997;19:253-260.
15. Yu PH, So CC, Wong KF, et al. Automated reticulocyte
counting: an evaluation of GEN S, Cell-Dyn 3500 and CellDyn 4000 [letter]. Clin Lab Haematol. 1999;21:145-147.
16. Lacombe F, Lacoste L, Vial JP, et al. Automated reticulocyte
counting and immature reticulocyte fraction measurement:
comparison of ABX PENTRA 120 Retic, Sysmex R-2000,
flow cytometry, and manual counts. Am J Clin Pathol.
1999;112:677-686.
17. Peebles DA, Hochberg A, Clarke TD. Analysis of manual
reticulocyte counting. Am J Clin Pathol. 1981;76:713-717.
18. Davis BH. Immature reticulocyte fraction (IRF): by any name,
a useful clinical parameter of erythropoietic activity. Lab
Hematol. 1996;2:2-8.
19. Brugnara C, Zurakowski D, Di Canzio J, et al. Reticulocyte
hemoglobin content to diagnose iron deficiency in children.
JAMA. 1999;281:2225-2230.

American Society of Clinical Pathologists

Hematopathology / ORIGINAL ARTICLE

20. Rudensky B. Comparison of semi-automated new Coulter

methylene blue method with fluorescence flow cytometry in
reticulocyte counting. Scand J Clin Lab Invest. 1997;57:291296.
21. Fraser CG, Petersen PH. Desirable standards for laboratory
tests if they are to fulfill medical needs. Clin Chem.
1993;39:1447-1455.
22. Fraser CG. Analytical goals for haematology tests. Eur J
Haematol Suppl. 1990;45:2-5.
23. Klee G. A conceptual model for establishing tolerance limits
for analytic bias and imprecision based on variations in
population test distributions. Clin Chim Acta. 1997;260:175188.

American Society of Clinical Pathologists

24. Sandberg S, Rustad P, Johannesen B, et al. Within-subject

biological variation of reticulocytes and reticulocyte-derived
parameters. Eur J Haematol. 1998;61:42-48.
25. Jones AR, Twedt D, Swaim W, et al. Diurnal change of blood
count analytes in normal subjects. Am J Clin Pathol.
1996;106:723-727.
26. Lazarus HM, Chahine A, Lacerna K, et al. Kinetics of
erythrogenesis after bone marrow transplantation. Am J Clin
Pathol. 1992;97:574-583.

Am J Clin Pathol 2001;115:100-111

111