HEMATOPATHOLOGY
Original Article
T h e
Erythrocyte
S e g r e g a t e s
Cell H e m o g l o b i n
T h a l a s s e m i a
Traits F r o m
C o n d i t i o n s
W i t h
D i s t r i b u t i o n
O t h e r
W i d t h
N o n t h a l a s s e m i c
M i c r o c y t o s i s
TE-CHIH LIU, MRCPath,1 PECK-SHEONG SEONG,1 AND TING-KWONG LIN, PhD2
Red cell heterogeneity, as represented by the red cell distribution
width (RDW), can be used to distinguish thalassemia traits from
iron deficiency. Two other indices of heterogeneity, the hemoglobin distribution width and the cell hemoglobin distribution
width (CHDW), are also available. In addition, the CHDW may
reflect the process of cell hemoglobinization more accurately
than does the RDW. In this study, recursive partitioning methods
were used to compare the ability of these three indices to discriminate between thalassemia traits and other nonthalassemic conditions among hospital patients who had microcytosis. The data
indicate that the CHDW can segregate patients who have either
iron replete or iron deficient nonthalassemic conditions from
those who have thalassemia traits. A CHDW level of less than
3.05 correctly discriminated 78.4% of patients in a mixed hospital
sample. A CHDW/RBC ratio of 0.57 improved the segregation
further, with a sensitivity of 79.2% and a specificity of 88.5% for
the identification of a thalassemia trait. (Key words: Cell hemoglobin distribution width; Discriminant function; Thalassemia
trait; Microcytosis; Iron deficiency) Am J Clin Pathol
1997;107:601-607.
RDW in the segregation of pl-TH from ID have consequently been shown to be more accurate than some others described. 3 As the mean corpuscular hemoglobin
(MCH) often varies directly with the mean cell volume
(MCV), a similar heterogeneity in the distribution of
RBC concentration of hemoglobin (Hb) and total Hb
content is not unexpected. The variation in concentration
of Hb is difficult to appreciate, however, on light microscopic examination. Nevertheless, the concentration can
be reliably quantified with hematology analyzers that
are based on the principles of laser light scattering. As a
result, two additional measures of heterogeneity are now
available on some of the newer analyzers. Individual cell
volumes and cell Hb (CH) concentrations can be directly
measured for each RBC, and the product of the two
gives the CH content. The dispersion of the CH content
and the concentration of Hb in RBCs can subsequently
be quantified from the distribution histograms as the CH
From the ^Division of Haematology, Department of Laboratory
distribution width (CHDW) and Hb distribution width
Medicine, National University Hospital, Singapore; and the ^Department
(HDW), respectively.
of Economics and Statistics, Faculty of Arts and Social Science, National
University of Singapore, Singapore.
The CHDW, in particular, may have a possible
advantage over the RDW because CH, unlike cell volPresented in part at the 37th Annual Meeting of the American
ume, is largely unchanged during reticulocyte matuSociety of Hematology, Seattle, Washington, December 1995.
ration 4 and may better reflect the conditions that preManuscript received August 16, 1996; revision accepted
October 30,1996.
vail during erythropoiesis. This advantage, in turn,
Address reprint requests to Dr Liu: Division of Haematology,
may translate into a greater accuracy when applied to
Department of Laboratory Medicine, National University Hospital,
the segregation process.
5 Lower Kent Ridge Rd, Singapore 119074.
Certain differences exist between the RBCs of individuals who have thalassemia (TH) traits and those
of individuals who have other microcytic conditions.
Many attempts have been made to use these differences to distinguish patients who have |3-TH from
those who have iron deficiency (ID). Although these
efforts have been promising, the success has been
modest, 1 especially when mixed populations that
have different causes for the microcytosis are studied.
RBCs from patients who have TH appear more
homogeneous when compared with those from patients
who have ID. The quantified heterogeneity of RBC volume, the red cell distribution width (RDW), is considered one of the more discriminatory RBC parameters
available. 2 Discriminant functions associated with the
601
HEMATOPATHOLOGY
OriginalArticle
602
We compared the CHDW with the RDW and HDW to
determine which is the better index of heterogeneity. We
then tested the CHDW in combination with other indices
to determine its potential clinical utility in segregating
TH from other nonthalassemic (NT) conditions, including ID, among hospital patients who had microcytosis.
MATERIALS AND METHODS
(AVIV Biomedical, Lakewood, NJ). Serum levels of ferritin, if available, were also recorded. Iron stores were
considered low if the serum level of ferritin was less
than 20 ug/L. Among the patients who had NT, those
who were iron replete (IR), with normal or high concentrations of ferritin, constituted a mixed group. Some of
these patients might have had undiagnosed a-2 TH.
Most, however, especially those who had anemia, probably had some degree of anemia of chronic disease.
Patients and Materials
Statistics
We analyzed routine blood samples received for
hemoglobinopathy screening in the hematology laboratory of our tertiary care hospital. In most cases, the
screening was ordered to investigate the possibility of
anemia or microcytosis. A full set of RBC indices, including the RDW, HDW, and CHDW, from the Technicon
H*3 hematology analyzer (Technicon Instruments,
Tarrytown, NY) was obtained prospectively from all
nonpregnant patients who had microcytosis with an
MCV of less than 80 fL. The H*3 analyzer provided a
direct measurement of individual RBC size and Hb concentration. It then calculated the individual CH as the
product of the two measurements (CH = volume x Hb
concentration). The CHDW and HDW are, respectively,
the standard deviations of the CH and Hb concentration
distribution histograms present on the research 1 screen
of the H*3. Hb electrophoresis, quantitation of Hb A 2
and Hb F, and brilliant cresyl blue stain for Hb H inclusion bodies were performed on all samples.
P-TH was diagnosed if the concentration of Hb A 2
was greater than 3.5%, as measured by high-perform a n c e liquid c h r o m a t o g r a p h y (BioRad Variant,
Hercules, Calif). a-TH was considered present if, in
addition to microcytosis, the concentration of Hb A 2
was not increased and the occasional Hb H inclusion
body could be detected. An average of 100 oil fields
or 20,000 RBCs were screened for each sample. Our
evaluation, along with reports from other investigators, 5 indicated that this method can reliably detect
a-1 TH (deletion of two a genes), although some cases
of a-2 TH may be missed. Because the search for Hb
H inclusion bodies was tedious and the results were
largely operator dependent, the test was performed
by a few designated technologists. Patients who had
other, less c o m m o n l y e n c o u n t e r e d h e m o g l o binopathies, including Hb H disease, Hb E trait, and
homozygous |3-TH, were excluded. Cases were considered NT if they did not fit the diagnostic criteria
for either a- or P-TH.
The level of zinc protoporphyrin was also measured
in each patient with the use of hematofluorometry
Patients who had TH, either a or p, were grouped into
a common TH category for analysis. To minimize the
possibility that a chance occurrence would be reported as
significant, the accumulated database was randomly
allocated into two data sets that included similar numbers of TH and NT patients for separate hypothesis
derivation and verification. The first data set constituted
the learning set from which the observations were
drawn. The various test parameters (RDW, CHDW, and
HDW) were compared with the use of recursive partitioning methods 6 on a computer program (S-Plus,
Statistical Science, Seattle, Wash). This procedure
involved two stages. During the first, or reiterative, stage,
the cutoff point or decision level for each parameter was
determined. The optimal decision level was the value
that separated the learning data set into two groups so
that the distribution of NT and TH patients between the
two partitions was as different as possible. During the
second stage, the parameter that produced the best segregation was determined. An index of heterogeneity, the
deviance, was calculated for each test parameter in the
unpartitioned and post-partition groups in the learning
data set. The deviance is given by the formula
[D = - 2 n 0 N T log(p 0NT ) - 2 iv™ log(p 0TH )]
where n 0 N T and p 0 T H are, respectively, the numbers
and proportions of NT and TH patients in the unpartitioned group. If the group was homogeneous and was
composed only of patients who had either TH or NT,
the formula would have a minimum value of zero.
The most discriminatory parameter would be the one
that resulted in the greatest reduction in deviance; this
parameter should yield the best segregation.
The efficacy of the discrimination obtainable for
each parameter was finally verified by testing the
decision levels obtained on the second, or test data
set. The reported result was the diagnostic efficiency,
or percentage of correct classifications, from the test
data set.
A]CP» :ay 1997
LIU ET AL
Cell Hemoglobin Distribution Width
603
We used a similar process for each of the discriminant functions: MCV - (5 x Hb) - RBC/ MCV/RBC, 8
M C H / R B C , 9 (MCV 2 x R D W ) / ( 1 0 0 x H b ) , 3 a n d
%microcytic/%hypochromic RBCs. 10 The results of
segregation when these functions were used were
then compared with the results of discrimination
when the three measures of heterogeneity were used.
in Table 1. The patients who had ID NT conditions in
the total database had mean values of Hb, RBC, and
MCV of 91 g/L, 4.5 x 10 9 /L, and 69.2 fL, whereas the
patients who had IR NT conditions had mean values of
100 g/L, 4.45 x 10 9 /L and 73.4 fL, respectively. Serum
levels of ferritin were determined for 133 patients who
had TH; 16 of these patients (12%) had ID.
Quality
Comparison of Main TH and NT Groups
Control
In-house replicate studies of 20 other patients who
had microcytosis showed good precision and stability
for the CH and CHDW. The coefficient of variation of
6-8 replicate measurements of the CHDW was 0.75%,
compared with 0.56% for the RDW and 1.08% for the
HDW. There was also no significant change in the
CHDW when the samples were stored up to 5 days at
4°C. The magnitude of the fluctuations for the CHDW
did not exceed ± 1.2% of the mean day 0 value. The
RDW was similarly stable, although fluctuations of the
HDW ranged from 3% to 5% during the same period.
All samples were analyzed on the day of phlebotomy;
the interval was often less than 4 hours. Stabilized blood
cells from the manufacturer (Test-Point Hematology
Control, Bayer, Tarrytown, NY) were tested at four different times each day. In addition, a blood sample from a
healthy donor was obtained as a secondary control early
each morning and was included with each processing
batch every 2 hours. The performance of the machine
was further monitored through enrollment in two external quality control programs for the entire study.
RESULTS
Patient
Groups
A total of 510 patients (250 TH and 260 NT) were
included. The numbers of patients in the various subgroups and the two randomized data sets are detailed
There was no demonstrable difference between
patients who had a-TH and those who had [3-TH in
the mean values and standard deviations of CHDW,
RDW, or HDW. We could therefore combine the two
types of TH for analysis. The optimal decision levels
for CHDW, RDW, and HDW are shown in Table 2. Of
the three indicators of RBC heterogeneity, the best
segregation b e t w e e n TH and NT conditions was
obtained with a CHDW level of 3.05 pg, followed by
the RDW and then the HDW. The CHDW had an
overall diagnostic efficiency of 78.4%; 88% of patients
who had TH and 69% of those who had NT conditions were correctly identified in the test data set.
The segregation obtained with the RDW appeared
poor when compared with the CHDW; almost twice
the number of misclassifications occurred with the
RDW. This finding is at odds with those of earlier
reports, 2 which showed that the RDW can segregate
ID from TH. One explanation for these 'discrepant'
observations is that the patients who had NT conditions were stratified according to their serum level of
ferritin. The RDW is mainly effective in segregating
ID NT conditions from TH; the RDW values for IR NT
conditions show a big overlap with TH (Fig 1). The
level of iron stores therefore appears to have a significant effect on the RDW values of patients who have
NT conditions. Consistently higher values of RDW
are usually associated with ID, and extrapolation of
the results of segregation by the RDW to include other
TABLE 1. DISTRIBUTION OF PATIENT SUBGROUPS IN THE COMBINED AND RANDOMIZED DATA SETS
Data Set 2
Data Set 1
Nonthalassemia
Thalassemia
Total
130
130
Low ferritin
Normal ferritin
Unknown ferritin
125
125(48.1%)
51 (19.6%)
84 (32.3%)
250
58 (46.4%)
67 (53.6%)
60 (48%)
65 (52%)
255
260
64 (49.2%)
20 (15.4%)
46 (35.4%)
61 (46.9%)
31 (23.8%)
38 (29.2%)
125
a
P
Combined
255
Vol. 107>No.5
•
118 (47.2%)
132(52.8%)
510
HEMATOPATHOLOGY
Original Article
604
TABLE 2. COMPARISON OF SEGREGATING CAPABILITY AMONG THE THREE INDICES OF RED CELL HETEROGENEITY
Decision level of thalassemia
Decrease in deviance (set 1)
Total number in set 1 correctly
classified (%)
No. of thalassemia cases in set
2 correctly classified (%)
No. of nonthalassemia cases in set
2 correctly classified (%)
Total no. in set 2 correctly
classified (%)
CHDW
RDW
HDW
<3.05
115.3
209 (82)
< 18.25
19.9
153 (60)
>2.44
6.83
130 (51)
110 (88)
112 (89.6)
123 (98.4)
90 (69.2)
42 (32.3)
8 (6.2)
200 (78.4)
154 (60.4)
131 (51.4)
CHDW = cell hemoglobin distribution width; RDW = red cell distribution width; HDW = hemoglobin distribution width.
(A)
.
is
o
„
0
o
g
0
i
I
1
o
o
1
±
•
N
.
NT
TH
•
NT-link
NT-IR
NT-ID
THAL
NT-unk
NT-IR
NT-ID
THAL
NT-unk
NT-IR
NT-ID
THAL
Group
Group
Group
Group
FIG 1. Box and whisker plots of the cell hemoglobin distribution width (CHDW; A and B), the red cell distribution width (RDW; C), and the
discriminant function of Green and King (D). Each box marks the median and the interquartile range for the group. Whiskers indicate a distance equivalent to 2 times the interquartile range from the median. Outliers greater than 2 and 3.5 times the interquartile range from the
median are indicated by an asterisk and open circle, respectively. N = normal; NT = nonthalassemic; TH = thalassemic trait; unk = unknown
ferritin status; ID = iron deficient; IR = iron replete.
microcytic NT conditions is likely to cause a high
error rate. In contrast, the level of CHDW is similar in
all NT patient groups, regardless of their levels of ferritin. The levels of CHDW in patients who have NT
conditions are, in fact, similar to those in healthy individuals. The finding of a low value of CHDW for TH
could be a characteristic of the TH phenotype that
allows these patients to be distinguished from those
who have NT conditions. The CHDW is therefore the
preferred discriminator when there are mixed ID and
IR groups or when the level of ferritin is uncertain.
The HDW, however, gives poor results at segregation.
Combinations of RBC parameters in discriminant
functions usually demonstrate higher levels of accuracy at segregation than do single parameters. The
optimal decision level and discrimination achieved
for some of these functions are listed in Table 3. The
discriminant function with the best result in segrega-
tion is the formula reported by Green and King 3 :
(MCV2 x RDW)/(100 x Hb). Although the CHDW is a
single parameter, it compares favorably with these
more complex discriminant functions and is at least as
accurate as this formula. When the RBC count alone is
used as a discriminator, it also gives credible results. 11
A combination of the RBC with the CHDW could theoretically achieve a better segregation. We therefore
tested this combination. A CHDW/RBC ratio of 0.57
improved the efficiency obtained with the CHDW
from 78.4% to 83.9%. Although good results were
obtained with the CHDW in the identification of
patients who had TH, poor results were obtained for
those who had NT conditions; 31% were misclassified. The sensitivity of the C H D W / R B C ratio was
79.2%, and the specificity was 88.5%. The inclusion of
other p a r a m e t e r s did n o t r e s u l t in a significant
increase in diagnostic yield.
AJCP • May 1997
LIU ET AL
Cell Hemoglobin Distribution Width
605
TABLE 3. RESULTS OF SEGREGATION BY VARIOUS DISCRIMINANT FUNCTIONS
MCV2 x RDWI
(100 x Hb)
Decision level for TH
Decrease in deviance (set 1)
Total no. in set 1 correctly classified (%)
No. TH in set 2 correctly classified (%)
No. NT in set 2 correctly classified (%)
Total no. in Set 2 correctly classified (%)
MCV-(5xHb)
-RBC
< 14.35
67.9
190 (74.5)
104 (83.2)
88 (67.7)
192 (75.3)
<79.6
78.6
196 (76.9)
105 (84.0)
95 (73.1)
200 (78.4)*
MCVI
RBC
MCHI
RBC
Micro!
Hypo
CHDWI
RBC
< 13.05
<4.25
55.2
185 (72.5)
99 (79.2)
84 (64.6)
183 (71.8)
>1.45
<0.57
119.5
210 (82.4)
99 (79.2)
115 (88.5)
214 (83.9)*
77.7
195 (76.5)
79 (63.2)
113 (86.9)
192 (75.3)
7.6
149 (58.4)
56 (44.8)
107 (82.3)
163 (63.9)
MCV = mean cell volume; RDW = red cell distribution width; Hb *= hemoglobin; MCH = mean corpuscular hemoglobin; Micro = microcytic; Hypo = Hypochromic; CHDW - cell hemo
globin distribution width; TH =2thalassemia; NT = nonthalassemia.
V of CHDW/RBC and (MCV x RDW)/(100 x Hb) gave P < .001.
DISCUSSION
Current Problems With
Discrimination
D i s c r i m i n a n t functions h a v e p r i m a r i l y b e e n
described for highly selected patients with TH or ID
who do not have complicating conditions. Two problems, however, are associated with the use of these
functions in clinical practice. First, it is uncertain
whether discriminant functions can be used to segregate patients who have IR NT microcytosis from those
who have TH. Second, the results of discrimination in
different populations have been variable. Follow-up
reports on the use of these discriminant functions have
given optimal cutoff points that differ significantly
from the original decision levels. 1 ' 10,12 Better and possibly more robust results may be obtained from discriminant functions when nonselected patients are
included, similar to the patients in our study.
Decision Levels and Deviance
A t t e m p t s to a d d r e s s the issue of result reproducibility have led investigators to require separate
databases for the reporting and testing of hypotheses. 13 We have taken the process further by adding a
statistical yardstick for assessment. Han and Fung 1 4
previously attempted to use discriminant analysis
for segregation. Their method, however, had two
major problems that limited its applicability to clinical u s e . First, the m e t h o d m a d e a n u m b e r of
assumptions about the sample distribution that may
not hold true in clinical practice. Second, the discriminant formula derived from analysis is difficult
to understand and apply clinically. The recursive
partitioning method overcomes these two problems.
This method makes no prior assumptions about the
distribution of the sample population. In addition, it
reports a result that is immediately understandable
and applicable to clinical practice. When test values
are reported along a continuous scale, the determination of the optimal decision level is somewhat
arbitrary. Use of the reiterative comparison of proportions after partitioning makes the determination
of optimal cutoff point more objective. In addition,
the calculation of the deviance minimizes the likelihood that a chance occurrence will be interpreted as
significant, because this method provides a statistical basis for the choice of a particular parameter.
Although the deviance, which is also known as the
likelihood ratio statistic, is infrequently encountered
in the biomedical literature, it is widely used in the
social sciences. The size of reduction in the deviance
does not always correlate with the m a g n i t u d e of
improvement in the accuracy of the discrimination
but it does provide a good indication of the process,
as can be seen by the verification obtained with our
test data set. Recursive partitioning removes any
possible subjectivity from the process because it
allows different discriminant functions to be compared more equitably.
Why the CHDW Is Better Than the RDW
Recent findings suggest that the erythrocyte CH
content is an important cellular attribute. Although
an individual normally has RBCs of varying sizes,
the erythrocyte CH content is constant through different sizes of RBCs and concentrations of Hb for a
particular i n d i v i d u a l , as d e m o n s t r a t e d by direct
measurement. 1 5 In addition, d'Onofrio et al 4 have
shown that the CH is largely unchanged through
r e t i c u l o c y t e m a t u r a t i o n ; the MCV, h o w e v e r ,
decreases by approximately 20%. We believe that the
Vol. 107 • No. 5
HEMATOPATHOLOGY
Original Article
606
CH is a better reflection of the state of hemoglobinization during erythropoiesis than is the MCV;
correspondingly, the CHDW is a better indicator of
variability than is the RDW.
Our results provide additional s u p p o r t for the
premise that RBC homogeneity is a fundamental characteristic of RBCs in patients who have TH. In this
respect, the uniformity of Hb content is probably of
greater importance than is the uniformity of cell size.
The CHDW value of RBCs in patients who have
TH is lower than that in the RBCs of healthy individuals, which suggests that hemoglobinization is more
constrained in cases of TH. This difference in homogeneity between patients who have TH and healthy
individuals is evident from the RBC volume/Hb concentration scatterplots present in the Technicon H*
series of hematology analyzers (Fig 2). The results of
patients who have TH are more tightly clustered than
those of h e a l t h y i n d i v i d u a l s , which suggests an
inverse correlation between volume and Hb concentration. In contrast, the results of patients who have
ID are considerably more dispersed. Although visual
inspection of the scatterplots is useful in individual
cases, comparisons between groups are difficult. The
quantitation of these visual differences in the CHDW
allows for easier c o m p a r i s o n s b e t w e e n different
groups of patients.
Misclassifications
An overlap between the CHDW values of patients
who have TH and those who have NT conditions is
still present; 12% of patients who have TH and 31% of
those w h o have NT conditions are misclassified.
Misclassifications of TH are thought to occur because
of additional complicating factors associated with ID
or a concomitant illness. The contribution of ID to
misclassification in TH, however, may not be that significant. The prevalence of ID among our patients
who had TH was not high: 16 of 133 patients (12%) in
w h o m serum levels of ferritin were d e t e r m i n e d .
Nevertheless, 11 patients in this group were correctly
TABLE 4. MEAN CHARACTERISTICS OF PATIENTS CLASSIFIED ACCORDING TO A CELL HEMOGLOBIN DISTRIBUTION
WIDTH (CHDW) OF 3.05 PG
TH Correctly Classified
TH Misclassified
219
115
5.65
2.7
87.6*
31
106
4.90
3.4
149.4*
Number
Mean Hb (g/L)
Mean RBC (M1012/L)
Mean CHDW (pg)
Mean ZP (umol/mol Hb)
NT Correctly Classified
190
99
4.58
3.5
206.1
NT Misclassified
70
94
4.71
2.8
231.8
TH = thalassemia; NT = nonthalassemia; Hb = hemoglobin; ZP = zinc protoporphyrin.
'Student's (test comparison of levels of ZP for the two TH groups gave P = .001.
(B)V
(A)'.
(C)V
.
^
•
$L
% '•' \ * -
• ':JJHH
Bft^
:
" *§j
& : • .
555^5
'.. '
RBC
HC
HC
HC
FIG 2. Technicon H*3 (Technicon Instruments, Terrytown, NY) scatterplots of volume vs hemoglobin concentration for thalassemia trait (A), normal
individual (B), and iron deficiency (C). The progressively increased scatter of points reflects the respective cell hemoglobin distribution width values.
AJCP • May 1997
607
LIU ET AL
Cell Hemoglobin Distribution Width
classified as having TH. Patients with TH who were
misclassified usually had lower Hb and RBC counts
and higher zinc protoporphyrin levels; however, use
of the zinc protoporphyrin level, rather than the ferritin level, as an indicator of functional ID does not
improve the efficiency of the segregation. Although
p a t i e n t s w i t h TH w h o were misclassified h a d a
greater mean value of zinc protoporphyrin, patients
with a high overall value were not associated with a
significantly higher misclassification rate. Instances of
misclassification would certainly include cases of
undetected a-2 TH. Without the benefit of molecular
testing however, we could not determine the exact
size of this group.
Clinical
Application
The CHDW enabled us to achieve a good level of
segregation. The greater stringency required for classification by the CHDW/RBC ratio resulted in greater
specificity and made the discriminant function more
efficient. Alternative combinations, eg, (CHDW x
MCV)/RBC, were marginally more accurate but were
more complex. A simpler function is preferred to minimize the occurrence of errors, whether in the measurement or in the calculation. The overall efficiency
of 83.9% provided by the CHDW/RBC ratio, however, is not sufficient for it to be recommended as a
diagnostic test. Rather, a clinician is more likely to use
this function to indicate quickly the presence of an
underlying TH trait in a patient who has microcytosis.
In addition, unlike most discriminant functions, the
CHDW can be used in pregnant women; in most
other functions, the RBC indices are affected by the
expanded plasma volume that occurs during pregnancy. The CHDW, if not the CHDW/RBC, is unaffected by these changes; therefore, it has potential
importance in the development of TH screening programs in these patients.
CONCLUSION
The CHDW is an important distinguishing characteristic of TH vs NT RBCs. Use of this parameter
enables other NT causes of microcytosis, in addition
to ID, to be segregated from TH. The discrimination is
further enhanced if a combination of the CHDW and
RBC values is used. A CHDW/RBC ratio of 0.57 can
correctly identify 79% of patients who have TH and
89% of those who have NT microcytosis.
REFERENCES
1. Bentley SA, Ayscue LH, Watson JM, et al. The clinical utility of
discriminant functions for the differential diagnosis of
microcytic anemia. Blood Cells. 1989;15:575-582.
2. Bessman JD, Feinstein DI. Quantitative anisocytosis as a discriminant between iron deficiency and thalassemia minor.
Blood. 1979;53:288-293.
3. Green R, King R. A novel red cell discriminant incorporating
volume dispersion for differentiating iron deficiency anemia
from thalassemia minor. Blood Cells. 1989;15:481^91.
4. d'Onofrio G, Chirillo R, Zini G, et al. Simultaneous measurement of reticulocyte and red blood cell indices in healthy
subjects and patients with microcytic and macrocytic anemia. Blood. 1995;85:818-823.
5. Wasi P. Population screening. In: Weatherall DJ, ed. The
Thalassaemias, Vol 6 of Methods in Hematology. Edinburgh,
Scotland: Churchill Livingstone; 1983:134-144.
6. Ciampi A, Chang C-H, Hogg S, et al. Recursive partition: a versatile method for exploratory data analysis in biostatistics.
In: MacNeill IB, Umphrey GJ, eds. Biostatistics. New York,
NY: Reidel; 1987:23-50.
7. England JM, Fraser PM. Differentiation of iron deficiency from
thalassaemia trait by routine blood count. Lancet.
1973;1:449-452.
8. Mentzer WC. Differentiation of iron deficiency from thalassaemia trait. Lancet. 1973;1:882.
9. Srivistava PC. Differentiation of thalassaemia minor from iron
deficiency. Lancet. 1973;1:154-155.
10. d'Onofrio G, Zini G, Ricera BM, et al. Automated measurement
of red blood cell microcytosis and hypochromia in iron deficiency and B-thalassemia trait. Arch Pathol Lab Med.
1992;116:84-89.
11. Klee GG, Fairbanks VF, Pierre RV, et al. Routine erythrocyte
measurements in diagnosis of iron-deficiency anemia and
thalassemia minor. Am ] Clin Pathol. 1976;66:870-877.
12. Houwen B. The use of inference strategies in the differential
diagnosis of microcytic anemias. Blood Cells. 1989;15:509-532.
13. England JM. Discriminant functions. Blood Cells.
1989;15:463-473.
14. Han P, Fung KP. Discriminant analysis of iron deficiency
anaemia and heterozygous thalassaemia traits: a 3-dimensional selection of red cell indices. Clin Lab Haematol.
1991;13:351-362.
15. Lew VL, Raftos JE, Sorette M, et al. Generation of normal
human red cell volume, hemoglobin concentration and
membrane area distributions by "birth" or regulation? Blood.
1995;86:334-341.
Vol. 107 • No. 5