Clinical Chemistry 45: 1495-1500, 1999;
(Clinical Chemistry. 1999;45:1495-1500.)
© 1999 American Association for Clinical Chemistry, Inc.
Predictive Value of Cord Blood Hematological Indices and Hemoglobin Barts for the Detection of Heterozygous 
-Thalassemia-2 in an African-Caribbean Population
Fey P.L. van der Dijs1,
Marcel Volmer2,
Dieuwke G. van Gijssel-Wiersma2,
Jan W. Smit3,
Reind van Veen3 and
Frits A.J. Muskiet2,a
1
Public Health Laboratory, Curaçao, The Netherlands Antilles.
2
Central Laboratory for Clinical Chemistry and
3
Central Laboratory for Hematology, Groningen University
Hospital, 9700 RB Groningen, The Netherlands.
a Address correspondence to this author at: Central Laboratory for Clinical Chemistry, Room Y 1.147, Groningen University Hospital, P.O. Box 30.001, 9700 RB Groningen, The Netherlands. Fax 31-50-3612290; e-mail f.a.j.muskiet{at}lab.azg.nl
 |
Abstract
|
|---|
Background: Cord blood hemoglobin Barts (HbBarts) and
hemocytometric indices may be used for classification of newborns into
those without 
-thalassemia-2 (/) and with heterozygous
-thalassemia-2 (-3.7/). We investigated by
logistic regression analysis whether the combination of HbBarts and
hemocytometric indices improves classification compared with
classification based on a single analyte.
Methods: HbBarts percentages and hemocytometric indices were
determined in cord blood of 208 consecutive newborns in Curaçao
(Netherlands Antilles). Of these, 157 had / and 51 had
-3.7/, as established by DNA analysis.
Results: Between-group differences were significant for
erythrocytes, mean cell volume, mean cell hemoglobin (MCH), mean cell
hemoglobin concentration, platelets, hemoglobin F0
(HbF0), and HbBarts. The Logit equation of the
logistic regression model, using MCH (pg) and HbBarts (%), was:
42.7164 + 5.7916(HbBarts) - 1.3110(MCH). A sensitivity of 100%
was reached at a Logit value of -3.70. The corresponding specificity
was 62.2%, and the predictive value of a positive test (PV+) was
46.3% (95% confidence interval, 37.0–55.7%). The relative
information gains were as follows: 88% for the HbBarts-MCH
combination, 26% for MCH (not significant), and 0% for HbBarts
compared with the 24.6% -3.7/ prevalence.
Conclusion: Combined use of cord blood HbBarts and MCH improves
classification compared with classification based on single
hemocytometric indices.
|
Introduction
|
|---|
HPLC is a powerful tool for the simultaneous screening of
newborns for hemoglobinopathies [e.g., hemoglobin
(Hb)1
S, HbC, and HbE] and -thalassemia by measurement of the
percentage of hemoglobin Barts (HbBarts) (1)(2).
Types 1 and 2 -thalassemia are the commonest -thalassemias. They
are caused by partial (type-2; -) or total (type-1; - -) -gene
deletion, which gives rise to various degrees of impaired (-/,
- -/, - -/-) or even completely absent (- -/- -)
hemoglobin -chain synthesis as well as abnormally low hemocytometric
indices [mean cell volume (MCV), mean cell hemoglobin (MCH), mean cell
hemoglobin concentration (MCHC), Hb] (3)(4)(5). HbBarts is
composed of four hemoglobin 
chains (). It occurs at
higher percentages in cord blood of newborns with -thalassemia
because of the self-assembly of the accumulating unpaired chains.
-Thalassemia-1, which is caused by an ~20-kb deletion involving
both genes (commonly denoted as - -SEA) is
prevalent in Southeast Asian populations, whereas -thalassemia-2,
which is caused by a 3.7-kb deletion involving the 3' part of the 2
gene and the 5' part of the 1 gene (commonly denoted as
-3.7) is prevalent among African populations,
including those in the US and the Caribbean.
Subdivision of cord blood HbBarts percentages into HbBarts 
0.5%,
0.5% < HbBarts 2.0%, and 2.0%< HbBarts 10%
corresponds to a large extent with classification into
non--thalassemic newborns (/), and those with
heterozygous (-/) and homozygous (-/-)
-thalassemia-2, respectively (6). However, in a previous
study comprising 211 consecutive spontaneous live births in the
Caribbean island of Curaçao (The Netherlands Antilles), we found
that this classification is far from perfect (van der Dijs et al.,
submitted for publication). Using the above HbBarts cutoff
values, we found that 4.5% of the 158 newborns with / were
misclassified into the -/ group. From the 51 newborns with
-/, 47.1% were misclassified into the / group,
whereas 7.8% were misclassified into the -/- group. The two
newborns with -/- were correctly classified.
Using the data of the same study group, we investigated by logistic
regression analysis whether the combination of HbBarts and
hemocytometric indices improves the classification of newborns into
those with / and -/, compared with classification
based on each of these analytes separately. Because HbBarts decreases
with increasing gestation in newborns with both / and
-/ (van der Dijs et al., submitted for publication), we also
entered the length of gestation into the model.
|
Materials and Methods
|
|---|
study design and study group
Cord blood was collected from consecutive and spontaneously
liveborn babies in Curaçao in the period of October 1992 to
January 1993. The babies were born either in the Maternity Clinic
"Rio Canario" or the St. Elisabeth Hospital, where the vast
majority of Curaçao births take place. Gestational age, based on
the time of last menstruation, was recorded, and cord blood was
analyzed for hematological indices, hemoglobin profiles, and
-thalassemia genotypes (see below). From the total number of 251
newborns, we had data on length of gestation and -thalassemia
genotype for 210. Of these, 158 had /, 51 had -/,
and 2 had -/-; there were no newborns with -thalassemia-1. The
distribution was in Hardy-Weinberg equilibrium, with expected
-thalassemia-2 genotype frequencies of 75.6% for /
(found, 74.9%), 22.7% for -/ (found, 24.2%), and 1.7% for
-/- (found, 0.9%). The hemoglobin profiles indicated that 190
(90.5%) had HbAA, 12 had HbAS (5.7%), and 8 had HbAC (3.8%).
One of the newborns with / (HbAA; length of gestation, 38
weeks) exhibited a HbBarts percentage of 2.0%. Its MCV (104 fL) and
MCH (34.8 pg) were in the low ranges of the respective 95% confidence
intervals for the MCV and MCH of the entire group of /
newborns. This newborn was suspected to have a form of -thalassemia
other than -thalassemia-1 or -2 and therefore was excluded from the
study.
From the finally selected 157 newborns with /, 8 had HbAS
and 4 had HbAC. We had no information on the hemoglobin profile of one.
From the selected 51 newborns with -/, 4 had HbAS and 4 had
HbAC. Both newborns with -/- had HbAA. We considered the finally
selected group as a representative sample of the Curaçao newborn
population because of the consecutive sampling design used, the
agreement of the present hemoglobinopathy incidence with that of our
previous screening study (1), and the encountered
Hardy-Weinberg equilibrium for the -thalassemia-2 genotypes. The
study protocol was in agreement with local ethics standards and the
Helsinki Declaration of 1975, as revised in 1989.
samples and methods
EDTA-anticoagulated cord blood was collected by midwives and one
of the authors. After clamping, cord blood was sampled into a
Vacutainer Tube containing EDTA (Becton Dickinson Vacutainer Systems
Europe) and transported to the Public Health Laboratory in melting ice.
Part of the blood was used for the measurement of hemocytometric
indices [white blood cells, red blood cells (RBCs), hemoglobin,
hematocrit, MCV, MCH, MCHC, and platelets (PLTs)] with a
Coulter Counter S 550 (Coulter International). Another part of the
whole blood was used immediately for hemoglobin profiling by HPLC with
spectrophotometric detection according to our previously described
method (1). White blood cells for DNA analyses were isolated
from the rest of the EDTA blood by centrifugation at 1500g
for 15 min at 24 °C followed by collection of the buffy coat. The
buffy coat was subsequently washed three times with phosphate-buffered
saline, pH 7.2, and resuspended in lysis buffer. The samples were
frozen at -20 °C and transported to The Netherlands in dry ice for
the subsequent identification of cases with -thalassemia types 1 and
2 in the Central Laboratory for Hematology of the Groningen University
Hospital by molecular biological methods. Briefly, DNA was isolated by
standard phenol/chloroform extraction. -Thalassemia-2 (a 3.7-kb
deletion involving the 3' part of the 2 gene and the 5' part of the
1 gene, commonly denoted as -3.7) was
detected by two PCR amplifications of parts of the -globin genes,
using primers C10 and C2 and C10 and C3, respectively, followed by
detection of PCR product presence and size after agarose gel
electrophoresis and ethidium bromide staining (7).
-Thalassemia-1 (an ~20-kb deletion involving both genes,
commonly denoted as - -SEA) was also detected by
two PCR amplifications using primer sets 7 and 8 and 7 and 9,
respectively, followed by the same detection procedures (8).
data analysis and statistics
Between-group (univariate) differences in the gestational age,
hemocytometric indices, and hemoglobin percentages of newborns with
/ and -/ were investigated with the Mann–Whitney
U-test (9) at P <0.05. ROC
curves were constructed with a computer program (ROC 2.1; University
Hospital Groningen) to examine the diagnostic value of each analyte.
Newborns with / served as the reference group, whereas
those with -/ were the patient group. The AUCs of the ROC
curves and their 95% confidence intervals (CIs) were evaluated as
measures of diagnostic accuracy (10). The percentage of
relative information gain was calculated from the difference between
the predictive value of a positive test (PV+,
also called posttest probability) and the prevalence (P, also
called pre-test probability) as follows: 100 x
(PV+ - P)/P.
A (multivariate) logistic regression analysis (SYSTAT-7; SPSS) was
performed to identify the combination of those analytes that yielded
optimal separation between newborns with / as the reference
group and those with -/ as the patient group. Stepwise
selection was used to obtain the best subset of analytes in the
logistic regression model (probability to enter and to remove, 0.15).
The following analytes were initially used for the stepwise procedure:
gestational age, RBCs, MCV, MCH, MCHC, PLTs, HbF0
and HbBarts. Hosmer-Lemeshow statistics were used to test the goodness
of fit of the logistic regression. The estimated Logit 
(X) = constant + aX1 +
bX2 + ...
pXp of the final logistic regression
model was used as a new variable for the ROC curve method
(11). In this formula X1,
X2 ...
Xp represent the various analytes,
whereas a, b, ... p are the coefficients of the regression equation.
|
Results
|
|---|
The lengths of gestation, hemocytometric indices, and hemoglobin
percentages, together with the outcome of the Logit equation (see
below) for the 157 newborns with / and the 51 with
-/ are shown in Table 1
. Using the Mann–Whitney U-test, we found
significant between-group differences for RBCs, MCV, MCH, MCHC, PLTs,
HbF0, HbBarts, and the Logit outcome. These
differences were confirmed by evaluation of the 95% CIs of the
respective AUCs (Table 1
, column 4), using the criterion AUC >0.5. The
two analytes that entered the logistic regression model, using the
stepwise selection procedure, were MCH and HbBarts. These were among
the analytes with the highest AUCs (Table 1
). The Logit equation of the
logistic regression model was: 42.7164 + 5.7916(HbBarts) -
1.3110(MCH), in which HbBarts is in percentage and MCH is in picograms.
The equation implies that at a given MCH, the chance of having
-/ increases by a factor 32.8 (95% CI of this odds ratio,
2.8–377.8) compared with having / for each 0.1% increase
in HbBarts. Analogously, at a given HbBarts, the chance of
having -/ increases with a factor 3.7 (95% CI, 2.3–5.9) for
each 1-pg reduction of the MCH. The Hosmer-Lemeshow statistics, using
the deciles of risk strategy, indicated excellent goodness of fit of
the logistic regression equation (P = 0.997). Fig. 1
shows a comparison between the ROC curves of the MCH (left
panel), HbBarts (middle panel), and the outcome of the Logit equation,
using the MCH and HbBarts (right panel). The MCV had no additive value
in the multivariate approach.
View larger version (10K):
[in this window]
[in a new window]
|
Figure 1. ROC curves for the diagnosis of heterozygous
-thalassemia-2, based on cord blood MCH (left),
HbBarts percentage (middle), and the outcome of the
Logit equation for MCH and HbBarts (right).
For areas under the ROC curves, see Table 1
. The Logit equation of the
logistic regression model was: 42.7164 + 5.7916(HbBarts) -
1.3110(MCH), in which HbBarts is in percentage and MCH is in
picograms.
|
|
The frequency distributions of the Logit (MCH, HbBarts) outcomes for
newborns with / and -/ are shown in Fig. 2
. A sensitivity of 100% was reached at a Logit value of -3.70.
The corresponding specificity was 62.2%, and the corresponding PV+ was
46.3% (95% binominal CI, 37.0–55.7%). At this PV+, the relative
information gain was 88% when compared with the 24.6% -/
prevalence of the present study population. Fig. 3
shows for the Logit the influence of the -/ prevalence
on PV+ at the fixed sensitivity of 100% and the fixed specificity of
62.2%. The PV+ was significantly different from the prevalence in a
prevalence range of 5–90%. Comparison with classification based on
MCH and HbBarts separately gave the following results. A sensitivity of
100% for the MCH was reached at a cutoff value of 37.2 pg, at which
the PV+ was 31.1% (95% CI, 24.0–38.2%). A sensitivity of 100% for
HbBarts was reached at a value of 0.0%. The PV+ at this cutoff value
was 24.6% (95% CI, 18.8–30.5%). These results imply that neither
MCH nor HbBarts exhibited a significant information gain compared with
the 24.6% -/ prevalence.
|
Discussion
|
|---|
We measured cord blood hemocytometric indices and hemoglobin
profiles of 208 spontaneously born infants on the Caribbean island of
Curaçao. After their genotypic classification into those without
-thalassemia-2 (/; n = 157) and those with
heterozygous -thalassemia-2 (-/; n = 51), we made
between-group comparisons of the hemocytometric indices and hemoglobin
profiles by two univariate tests (Mann–Whitney U-test and
ROC curve method) and by multivariate analysis (logistic regression).
The aim of the study was to investigate whether we could improve the
predictive value for the establishment of heterozygous
-thalassemia-2 by using a combination of hemocytometric indices and
hemoglobin percentages in a multivariate model. Because at least one of
the analytes of importance, i.e., HbBarts, decreases with gestation in
both newborns with / and -/ (van der Dijs et al.,
submitted for publication), we also entered the length of gestation
into the model.
Most importantly, we found statistically significant univariate
differences for RBCs, MCV, MCH, MCHC, PLTs,
HbF0, and HbBarts (Table 1
). MCH, MCV, and
HbBarts exhibited the biggest differences (Fig. 1
). Logistic regression
analysis identified MCH and HbBarts as the analytes that caused maximum
separation between the / and -/ groups. MCV did
not enter the model. The Logit equation was 42.7164 +
5.7916(HbBarts) - 1.3110(MCH), and a sensitivity of 100% (i.e.,
100% predictive value of a negative test) was reached at a Logit
cutoff value of -3.70. The corresponding specificity was 62.2%, and
the PV+ was 46.3% with a 95% CI of 37.0–55.7%. These results imply
that newborns with Logit values below -3.70 do not have heterozygous
-thalassemia-2 and that those with Logit values equal to or above
-3.70 have a 46.3% chance to carry the -/ genotype. In other
words, the 46.3% posttest chance calculated with the logistic
regression analysis is statistically significant and gives rise to a
88% relative information gain when compared with the 24.6% pre-test
probability of heterozygous -thalassemia in the Curaçao
population. The multivariate approach is a considerable improvement
compared with univariate evaluation because the corresponding posttest
chance for the MCH was 31.1% (when MCH is 37.2 pg), whereas that for
HbBarts, it was 24.6% (when HbBarts is 
0.0%). These posttest
chances were not significantly different from the prevalence, implying
that neither MCH nor HbBarts contribute to the information gain when
evaluated separately. We conclude that the combination of MCH and
HbBarts is more useful for the screening of -thalassemia in
Curaçao compared with the use of each of the
hemocytometric indices and hemoglobin percentages separately.
The question arose whether the above conclusion also applies for
populations with different -thalassemia-2 prevalences or with more
complex mixtures of -thalassemias, such as mixtures with the other
common -thalassemia (i.e., type 1) and the less frequently occurring
non-deletion -thalassemias (4). It also remains to be
seen whether the same Logit equation and cutoff value can be adopted by
other laboratories. Newborns with heterozygous -thalassemia-1 and
those who are double heterozygous for -thalassemia-1 and
-thalassemia-2 (HbH disease) have lower MCH concentrations and
higher percentages of HbBarts in cord blood compared with those with
heterozygous -thalassemia-2 (4)(6). The
present Logit cutoff value is, therefore, likely to classify them into
the group with -/. Subsequent DNA analyses will be needed to
identify the molecular biologic defects of the underlying
-thalassemia. The most critical factors in adopting the Logit
equation and its cutoff value in other laboratories seem to be the
inter- and intralaboratory accuracy and precision of the MCH and
HbBarts measurements and the prevalence of -thalassemia-2 in the
population being studied. With respect to accuracy and precision, it is
imperative to have similar MCH and HbBarts reference values, as given
in Table 1
. With respect to the prevalence of -thalassemia-2, our
data indicate (Fig. 3
) that there is statistically significant
information gain when one is dealing with an -thalassemia-2
prevalence in the 5–90% range. A more relevant criterion, however, is
the magnitude of the desired information gain. For example, the
-thalassemia prevalence range becomes restricted to 5% to ~20%
when the relative information gain is set at 100% (Fig. 3
). We
conclude that newborns with heterozygous -thalassemia-1 and double
heterozygous -thalassemia-1 and -2 are very likely to be codetected
with the present Logit cutoff value. The most important factors for
adopting the present Logit equation and cutoff value by others is the
interlaboratory comparability of the MCH and HbBarts results in the
sense of accuracy, precision, and reference values. The reference
values obviously should have been established by the same method
[i.e., according to IFCC recommendations (12)]. Other
considerations are the prevalence of -thalassemia in the study
population and the required information gain.
In summary, we conclude that the combined use of the cord blood MCH
concentration and percentage of HbBarts improves the predictive value
of a positive test for heterozygous -thalassemia-2 from 24.6% to
46.3%, with no false negatives, in a Caribbean population of
predominantly West African descent. The 88% relative information gain
is higher than that for the use of the MCH and HbBarts separately.
Codetection of homozygous -thalassemia-2, heterozygous
-thalassemia-1, and double heterozygous -thalassemia-1 and -2
(HbH disease) is very likely because these conditions are generally
characterized by lower MCH and higher HbBarts compared with
heterozygous -thalassemia-2. The Logit equation of the logistic
regression model and the Logit cutoff value may be of use to other
laboratories, provided that they ensure comparability with the present
reference values for the MCH and HbBarts. The multivariate approach is
only meaningful when both the prevalence of -thalassemia-2 in the
study population and the required information gain are taken into
account. Finally, it should be pointed out that predictive models might
give overoptimistic results and that the present model should as yet be
validated by independent data to establish its usefulness in daily
practice.
|
Acknowledgments
|
|---|
We thank Etienne Winklaar and Celeste Rosado for performing the
hemoglobin profiling, and Herman J.R. Velvis for valuable aid in sample
collection.
|
Footnotes
|
|---|
1 Nonstandard abbreviations: Hb, hemoglobin; HbBarts, hemoglobin Barts; MCV, mean cell volume; MCH, mean cell hemoglobin; MCHC, mean cell hemoglobin concentration; RBC, red blood cell; PLT, platelet; PV+, predictive value of a positive test; AUC, area under the curve; and CI, confidence interval. 
|
References
|
|---|
-
van der Dijs FPL, van den Berg GA, Schermer JG, Muskiet FD, Landman H, Muskiet FAJ. Screening cord blood for hemoglobinopathies and thalassemia by HPLC. Clin Chem 1992;38:1864-1869.
[Abstract/Free Full Text]
-
Fucharoen S, Winichagoon P, Wisedpanichkij R, Sae-Ngow B, Sriphanich R, Oncoung W, et al. Prenatal and postnatal diagnoses of thalassemias and hemoglobinopathies by HPLC. Clin Chem 1998;44:740-748.
[Abstract/Free Full Text]
-
Bunn HF, Forget BG. Hemoglobin: molecular, genetic and
clinical aspect. Philadelphia: WB Saunders, 1986:677pp..
-
Higgs DR. -Thalassaemia. In: Higgs DR Weatherall DJ,
eds. Bailliere's clinical haematology. The haemoglobinopathies.
London: Baillière Tindall, 1993:117–50..
-
Weatherall DJ. The thalassaemias. Br Med J 1997;314:1675-1678.
[Free Full Text]
-
Lie-Injo LE, Solai A, Herrera AR, Nicolaisen L, Kan YW, Wan WP, Hasen K. Hb Barts's level in cord blood and deletions of -globin-genes. Blood 1982;59:370-376.
[Abstract/Free Full Text]
-
Dodé C, Krishnamoorthy R, Lamb J, Rochette J. Rapid analysis of -3.7 thalassemia and anti 3.7 triplication by enzymatic amplification analysis. Br J Haematol 1992;82:105-111.
-
Bowden DK, Vickers MA, Higgs DR. A PCR-based strategy to detect the common severe determinants of thalassaemia. Br J Haematol 1992;81:104-108.
[Web of Science][Medline]
[Order article via Infotrieve]
-
Conover WJ. Practical nonparametric statistics, 2nd ed.
New York: John Wiley & Sons, 1980:493pp..
-
Zweig MH, Campbell G. Receiver-operating characteristic (ROC) plots: A fundamental evaluation tool in clinical medicine. Clin Chem 1993;39:561-577.
[Abstract/Free Full Text]
-
Hosmer DW, Lemeshow S. Applied logistic regression. New
York: John Wiley & Sons, 1989:307pp..
-
Gräsbeck R. Reference values, why and how. Scand J Clin Lab Investig 1990;50(Suppl 201):45-53.
Top
Abstract
Introduction
