Introduction
During the last few years an increasing tendency to favor a modified schedule for
performing the tests carried out in routine first-trimester screening could be observed.
According to this modified schedule, each patient should be seen at 2 different visits.
The first one should take place between 9+0 and 11+0 weeks of gestation and be used
for drawing the blood to determine the biochemical marker concentrations. The sonographical
examinations should be carried out at a second visit with the usual timing, i. e.,
between 11+1 and 14+0 weeks. Several authors have argued that preponing the date of
blood sampling leads to better diagnostic accuracy of the screening procedure [2]
[3]
[4].
The major challenge which had to be addressed in adapting the statistical approach
behind the previous versions of the first-trimester screening algorithm of the FMF-Germany
to this modified time schedule was to recruit sufficiently large samples of patients
with blood sampling performed before 11+1 weeks of gestation and extend the reference
bands to be used for assessing the degree of outlyingness of test results.
Materials and Methods
Patients
In developing a new algorithm for the assessment of findings from first-trimester
screening allowing for early blood sampling a database comprising n=186 215 euploid
pregnancies and n=925 pregnancies with chromosomal defects (trisomies 21, 18 and 13)
was established. In the reference group with negative outcome n=898 patients provided
blood samples drawn before 11+1 weeks of gestation. In the group with a pathological
outcome n=726 fetuses had a trisomy 21, and in the remaining 199 fetuses either a
trisomy 18 or 13 was found. All these samples consisted of data taken in singleton
pregnancies with well-documented outcome.
Ultrasound and biochemical procedures
The basic parameters analyzed for the purpose of first-trimester screening were maternal
age and weight, the sonographic parameters crown-rump length (CRL) and nuchal translucency
(NT) of the fetus, and the 2 biochemical markers PAPP-A (Pregnancy associated plasma
protein A) and free ß-hCG (free beta-human Chorionic Gonadotropin) whose concentrations
were measured in maternal serum.
For the sake of quality assurance of the sonographical methods employed, only data
provided by level-II or level-III certified ultrasound investigators were included.
The biochemical investigations were carried out in certified laboratories using the
Brahms Kryptor® system.
Statistical Methods
Adjusting the measured concentrations of PAPP-A and free ß-hCG for maternal weight
In a previous paper [1], the problem of eliminating a systematic influence of maternal weight on the false-positive
rates of the diagnostic procedures to be based on the posterior risks calculated for
the respective type of trisomy was successfully addressed by fitting the following
exponential regression model to the measured concentrations of PAPP-A and free ß-hCG: the following exponential regression model:
In this equation, Y represents the concentration of PAPP-A or free ß-hCG measured in a randomly selected
patient, and W denotes maternal weight in kg. Furthermore, β
0 and β
1 are unknown constants which have to be estimated from the data, and ε is an error term assumed to be stochastically independent of Y. For our present purposes, it will be more convenient to replace the above model with
the log-linear regression model
As soon as the parameters ̃β0 and ̃β1 have been estimated by means of standard least squares fit, weight-corrected log-concentrations
̃Y
corr
are obtained by means of the formula
Performing this correction in any individual case simply requires to add the difference
between individually observed and average maternal body weight multiplied by the regression
coefficient ̃β1 to the natural logarithm ̃Y of the measured concentration Y. As a fraction of the deviation of a pregnant woman’s actual weight from the average,
the amount of this correction equals the slope of the regression line describing the
relationship between log-concentration and maternal weight.
Modelling the regression of NT and the weight-corrected concentrations of the biochemical
parameters on gestational age
As previous experience has clearly shown, the form of the regression of the quantities
measured in the context of first-trimester screening on gestational age cannot simply
be inferred from a well-established biological theory and varies considerably between
the 3 parameters under assessment. Therefore, appropriate modelling of this relationship
requires to refer to a class of regression functions which is sufficiently rich in
its coverage of different shapes and can be specified without relying on theoretical
assumptions. A good choice that takes both aspects into account, and which was introduced
from the beginning of the FMF-Germany project, is the class of all polynomials of
degree up to 10. What degree is really needed was determined automatically in each
of the 3 cases (i. e., for NT, PAPP-A and free ß-hCG) by means of stepwise multivariable
regression treating the first 10 powers of CRL as potentially relevant regressors.
For fitting a regression model of this form to NT, the directly measured quantity
was taken as the dependent variable. In contrast, in the regression analysis of the
biochemical parameters, the originally measured concentrations were replaced by their
weight-corrected log-transforms. Furthermore, in analyzing NT, gestational age as
measured in terms of CRL was allowed to vary over the traditional range of 45–84 mm.
Extension of this range to a range of 16–84 mm in constructing reference bands for
the biochemical parameters was the crucial step in re-designing the screening algorithms
in a way making them suitable for evaluating constellations with early blood sampling.
Typically, preponing the date of blood sampling will entail temporal dissociation
between the sonographical and the biochemical part of the underlying clinical investigation.
In accordance with this, all analyses have to deal with 2 different values of CRL
for each individual patient, denoted by CRL and CRLBE in the sequel.
Construction of reference bands and computation of DoE’s
In constructing reference bands for NT and the weight-corrected, logarithmically transformed
concentrations of PAPP-A and free ß-hCG, we adhered to the same general principles
as have made up the statistical basis behind PRC from the beginning (for details see
Merz et al., 2008 [5]). Again, all 3 reference bands were to provide an overall coverage proportion of
90%, and through allowing for asymmetry of the conditional distributions of the measurements
it was ensured that the proportions of euploid patients with values below the lower
and above the upper limit both equal 5%. Reference bands exhibiting these coverage
properties were generated by means of the method originally introduced by Wellek and
Merz, 1995 [6] which essentially consists of computing smallest admissible vertical distances between
the bounds of the band and the regression line around which it is spanned.
Upon determing CRL-dependent reference limits for the quantity under consideration,
each individual data point was evaluated in terms of its degree of outlyingness through
computing the corresponding DoE. This measure is defined as the signed vertical distance
of a point in the respective scatterplot from the regression line as expressed as
a multiple of the width of the upper or lower part of the reference band at the actual
value of CRL as abscissa (for a complete formal definition see Merz et al., 2008 [5]).
Establishing the diagnostic decision rule
Like all widely-used procedures for evaluating standard first-trimester screening
data [5]
[7]
[8]
[9]
[10]
[11]
[12], the newly developed version of PRC uses the Bayesian posterior risk of trisomy
as the basis for deciding between euploidy and (potential) presence of the chromosomal
anomaly under investigation. In order to calculate the posterior probability of a
pregnant woman to carry a fetus exhibiting the currently considered form of trisomy,
we apply Bayes’ theorem with the following specifications:
-
Parametric estimate of the joint likelihood of the DoE’s for the 3 markers (i. e.,
NT, weight-corrected logarithms of PAPP-A and free ß-hCG) among euploids and pregnancies
with positive outcome, respectively.
-
Prior probability of carrying a trisomy-fetus among pregnants of specified age in
the given week of gestation.
The components of the likelihood are obtained as values of densities of Gaussian normal
distributions with population parameters substituted by sample means and variances.
The DoE’s of the 3 markers are assumed to be mutually independent so that the joint
likelihood can be computed as the product of the likelihood of the individual DoE’s.
The prior probabilities were determined through interpolating the values tabulated
by Snijders et al., 1999 [13] and Snijders et al., 1995 [14] for trisomy 21 and trisomy 13/18 respectively.
As usual, the final step of the diagnostic screening procedure consists of comparing
the posterior risk calculated for the given patient to the predefined cutoff, i. e.,
to the number 1/150≈0.67%. If the cutoff is exceeded, the decision is to recommend
amniocentesis or performance of some other procedure providing very high diagnostic
accuracy. Otherwise, the findings from first-trimester screening are classed as free
of evidence of trisomy.
Investigating the robustness of the false positive rates against changes in maternal
weight
In order to ascertain sufficient robustness of the algorithm against changes in maternal
weight, the interval over which maternal weight ranged within the reference sample
of euploids was partitioned into 18 adjacent classes. The lengths of these sub-intervals
were chosen depending on the density of the weight-distribution in the respective
part of the total range. In the upper tail of the distribution, groups of length 10
or even 15 kg were formed. All other classes were defined to have length 2.5 kg. For
each of the 18 strata, the group-wise false positive rate was determined and the variability
of the FPR over the strata measured in terms of an ordinary coefficient of variation.
In addition to quantifying the lack of robustness in that way, a graphical check was
performed by means of bar graphs providing direct visualization of the amount of fluctuation
between the weight groups.
Incorporating add-in markers
In addition to the parameters maternal weight, ethnicity and smoking status which
we already included in our risk calculation program PRC 2.0 [1], we have now included as further parameters Ductus venosus flow and Tricuspid regurgitation
[15]
[16] for a multi-variate analysis.
Results
Weight correction for the log-concentrations of the biochemical parameters
[Fig. 1] shows the measurements taken on PAPP-A in the whole sample of patients with negative
outcomes as points in the plane, with maternal weight as abscissa and log-concentration
level as ordinate, together with the fitted regression line. The coefficients were
estimated by means of ordinary least squares fit to be β̃̂0=2.36735 and β̃̂1=0.01654.
Fig. 1 Scatterplot with linear regression relating log-concentrations of PAPP-A to maternal
weight.
Analogously, linear regression analysis with the values behind the log ß-hCG*WEIGHT
scattergram shown in [Fig. 2] gave the numbers β̃̂0=4.27126, β̃̂1=0.00972 as parameter estimates.
Fig. 2 Scatterplot with linear regression relating log-concentrations of free β-hCG to maternal
weight.
Since in the reference sample of all euploid pregnancies, mean maternal weight was
̅W=68.341242, we obtained the following formulae for computing weight-corrected log-concentrations
of PAPP-A and free ß-hCG:
Log PAPP-A
corr
=log PAPP-A+0.01654*(W−68.341242),
Log β-hCG
corr
=log β-hCG+0.00972*(W−68.341242).
In the above equations, W stands for the weight measured [in kg] in the constellation to be evaluated in terms
of trisomy risk. Back-transforming them to the original scale yields the correction
formulae
PAPP-A
corr
=PAPP-A*exp {0.01654*(W−68.341242)}
and
β-hCG
corr
=β-hCG *exp {0.00972*(W−68.341242)}.
The values of the correction factors to be used according to these formulae are tabulated
in [Table 1] for maternal weights W ranging between 60 and 120 in steps of 5 [kg].
Table 1 Multiplication factors for calculating weight-corrected concentrations of the biochemical
markers [mU/ml] for maternal weights W ranging between 60 and 120 in steps of 5 [kg].
|
W
|
PAPP-A
|
ß-hCG
|
|
60
|
0.87113
|
0.92212
|
|
65
|
0.94624
|
0.96804
|
|
70
|
1.02782
|
1.01625
|
|
75
|
1.11643
|
1.06686
|
|
80
|
1.21268
|
1.11999
|
|
85
|
1.31724
|
1.17577
|
|
90
|
1.43080
|
1.23432
|
|
95
|
1.55416
|
1.29579
|
|
100
|
1.68815
|
1.36032
|
|
105
|
1.83370
|
1.42807
|
|
110
|
1.99179
|
1.49919
|
|
115
|
2.16352
|
1.57385
|
|
120
|
2.35005
|
1.65223
|
Reference Bands and Transformation of Individual Measurements Into DOE’s
Nuchal translucency (NT)
Stepwise regression analysis with the originally measured values of NT lead to selecting the following polynomial (with coefficients rounded to 5 significant
digits) for modeling the dependence of the mean of that variable on gestational age
as expressed in terms of CRL:
In [Fig. 3], the graph of the curve given by this equation is plotted in cyan. The conditional
variance of NT turned out to be sufficiently stable for justifying the assumption of approximate
CRL-independence of the variability. Accordingly, the width of the reference band
constructed with the data represented in the diagram in the usual form of a scatterplot
is constant over the whole range of CRL values at which the measurements of NT were taken in the total sample of euploid pregnancies. More precisely speaking, the
vertical distance of the estimated upper and lower 5% reference limits (plotted in
red) from the regression curve was computed to be 0.65869 and 0.57159 mm, respectively.
Fig. 3 90% reference band for NT computed from the data of n=186 215 euploid pregnancies.
Weight-corrected logarithmic concentration of PAPP-A
Plotting against CRL for all n=186 215 euploidies gave the blue cloud of points shown
in [Fig. 4]. In order to indicate that the gestational age at which the blood samples for determining
the biochemical concentration levels were taken were allowed to differ from that at
which sonography was performed, the horizontal axis is labeled CRLBE instead of CRL.
Furthermore, the range of CRLBE extends to the left until 16 mm, and the density of
the cloud is much lower below the traditional limit of 45 mm. With these data, stepwise
least squares fit of polynomials of degree up to 10 resulted in a distinctly simpler
regression equation as compared with that obtained for NT, namely
Fig. 4 90% reference band for weight-corrected log-concentrations of PAPP-A computed in
a sample of n=186 215 euploid pregnancies.
Except for the changes concerning the labeling and spacing of both coordinate axes,
all other components of the graph have the same meaning as in [Fig. 3]. The points on the curves plotted in red have as ordinates the estimated age-specific
5th and 95th percentiles, respectively, and the regression curve is represented by the cyan line.
Weight-corrected logarithmic concentration of free ß-hCG
[Fig. 5] is the analogue of [Fig. 4] for the second biochemical marker, i. e., the log-serum concentration of free ß-hCG.
In contrast to PAPP-A, the mean levels of free ß-hCG decrease with gestational age.
Precisely, stepwise polynomial regression of the logarithmic free ß-hCG level on CRL yielded the model equation
Fig. 5 Analogue of [Fig. 4] for free β-hCG.
Again, no marked asymmetry of the distribution of the residuals around this curve
was found, and the bandwidth could be chosen nearly constant in time, due to the fact
that the variance around the regression curve changes only slightly from left to right.
Example illustrating the use of the reference bands for transforming individual measurements
to degrees of outlyingness (DoE’s)
[Table 2] shows the data obtained from a patient randomly selected from the reference sample
of pregnancies with an euploid fetus. The entries in the right-most column are obtained
through calculating the difference between the measured value and the ordinate of
the corresponding point on the regression curve making up the central line around
which the reference band is spanned, and dividing the result by the width of the upper
and lower part of the band, respectively. If the data point corresponding to the measured
value falls below the regression line, a negative sign has to be added in front of
the result. For a non-verbal, mathematically precise general formulation of the rule
for computing DoEs for arbitrary values of NT and the 2 biochemical markers, the reader
is referred to Merz et al., 2008 [5].
Table 2 Example illustrating the computation of DoE’s for given values of NT and corrected
log-concentrations of the 2 biochemical markers.
|
Parameter
|
Days of Gestation
|
Value
|
Reference Limit
|
DoE
|
|
|
Measured
|
Predicted †
|
Lower
|
Upper
|
|
|
NT
|
81
|
1.5
|
1.24928
|
0.67768
|
1.90797
|
0.38064
|
|
Log PAPP-A
corr
|
65
|
−1.15671
|
−0.27541
|
−1.20614
|
0.59482
|
−0.94689
|
|
Log β-hCG
corr
|
"
|
5.38482
|
4.25754
|
3.31495
|
5.25608
|
1.12893
|
† by means of the polynomial regression model obtained in § 5.2
Likelihood ratios to be entered in Bayes’ formula for calculating posterior risks
of trisomy
In determining the likelihood ratios required for calculating the posterior risks,
we relied on the following assumptions:
-
Except for some normalizing transformation, the distributions of the DoE’s have (approximately)
Gaussian form with both parameters potentially depending on a patient’s ploidy status.
-
Both for normal and aneuploid pregnancies, the DoE’s associated with the 3 markers
are mutually independent.
From (i), it follows that any observed value x of the DoE of one of the 3 markers under consideration has the likelihood ratio
where µ
eu and σ
eu denotes, respectively, the population mean and standard deviation for euploid pregnancies,
and the symbols µ
aneu and σ
aneu have the analogous meaning for the aneuploid population. Since the populations involved
are not known as a whole, the µ’s and σ’s must be replaced by the corresponding sample means and standard deviations, respectively.
The values which were obtained for the latter are displayed in [Table 3].
Table 3 Constants for be used in calculating the likelihood ratios according to formula for both types of trisomy.
|
Parameter
|
Transformation of DoE
|
X
eu
|
S
eu
|
X
aneu
|
S
aneu
|
|
|
|
|
T21
|
T13/18
|
T21
|
T13/18
|
|
NT
|
|
1.13263
|
0.04320
|
1.23730
|
1.24310
|
0.09251
|
0.11135
|
|
Log PAPP-A
corr
|
None
|
0.01629
|
0.61358
|
−0.85335
|
−1.70421
|
0.73057
|
0.85010
|
|
Log β-hCG
corr
|
" "
|
−0.01405
|
0.61453
|
0.78042
|
−1.30688
|
0.62854
|
0.94820
|
In the example of [Table 2], calculating the likelihood ratios for the DoEs of the individual markers gave the
results shown in [Table 4].
Table 4 Likelihood ratios for the DoE’s of the individual markers obtained in the example
of [Table 2].
|
Parameter
|
Type of trisomy to be diagnosed
|
|
T21
|
T13/18
|
|
NT
|
0.434902
|
0.384972
|
|
Log PAPP-A
corr
|
2.855852
|
1.663988
|
|
Log β-hCG
corr
|
4.727560
|
0.130494
|
Eventually needed is the joint likelihood ratio for the DoE’s of all 3 markers. Under
the independence assumption (ii), this is simply the product of the likelihood ratios
calculated for the individual markers. Combining the entries in [Table 4] in this way gave LR
joint
=0.434902×2.855837×4.727462=5.871551 and LR
joint=0.384972×1.663977×0.130494=0.083592 for trisomy 21 and trisomy 13/18, respectively.
The final decision rule and its properties
As soon as the formula for computing the joint likelihood ratios has been established,
completing the construction of the desired diagnostic decision rule requires only
to link the quantity LR
joint with the prior probability of finding an aneuploidy of the type under consideration.
Denoting the latter by π, Bayes’ rule states how this has to be done in the appropriate
way, namely by evaluating the expression
Given the type of trisomy, the prior probability π (also called “background risk”
in the usage of first-trimester screening) depends on both maternal age and gestational
week. There is no mathematical model which describes the relationship between π and
these baseline variables. Instead, one can recourse to tabulations of proportions
of aneuploidies observed for specific combinations of age and week of gestation in
large samples. The results shown in [Table 5] for a selection of exemplary findings from standard first trimester screening were
obtained with the background risks tabulated by Snijders et al. for trisomy 21 [13] and for trisomy 13/18 [14], and the same data are used throughout the implementation of the algorithm distributed
by the FMF-Germany.
Table 5 Results of trisomy risk calculations in a group of patients with euploid pregnancies
randomly selected from the database of FMF Germany. [Entries in columns 8–11: reciprocal
risks rounded to the nearest integer.]
|
Trisomy 21
|
Trisomy 13/18
|
|
Age
|
Weight
|
CRLBE
|
CRL
|
NT
|
PAPP_A
|
β-hCG
|
Backgr
|
Post
|
Backgr
|
Post
|
|
37
|
74.5
|
58
|
86
|
2.3
|
0.330
|
52.5
|
145
|
486
|
256
|
1 083
|
|
33
|
65.1
|
58
|
86
|
2.0
|
0.526
|
124.7
|
349
|
1 022
|
617
|
49 471
|
|
27
|
48.6
|
59
|
88
|
1.2
|
0.210
|
71.6
|
876
|
2 901
|
1 651
|
2 492
|
|
23
|
65.4
|
58
|
88
|
1.9
|
0.725
|
96.0
|
1 021
|
13 238
|
1 928
|
378 897
|
|
42
|
72.8
|
59
|
89
|
2.0
|
0.359
|
77.0
|
43
|
176
|
82
|
2 027
|
|
37
|
69.0
|
58
|
89
|
1.8
|
0.241
|
47.2
|
147
|
1 355
|
278
|
974
|
|
36
|
84.4
|
58
|
90
|
2.0
|
0.290
|
51.9
|
225
|
2 198
|
425
|
7 541
|
|
41
|
61.6
|
55
|
90
|
1.7
|
0.219
|
84.6
|
55
|
264
|
104
|
1 087
|
|
30
|
57.6
|
59
|
91
|
1.6
|
0.349
|
75.3
|
677
|
6 329
|
1 278
|
25 491
|
|
26
|
64.7
|
59
|
92
|
1.6
|
0.230
|
73.9
|
945
|
3 983
|
1 782
|
13 570
|
|
35
|
73.3
|
59
|
92
|
1.5
|
0.399
|
69.6
|
233
|
5 072
|
440
|
39 012
|
|
34
|
51.6
|
58
|
93
|
2.1
|
0.563
|
142.4
|
338
|
902
|
638
|
40 802
|
Which diagnostic decision has to be taken in an individual case depends on the cutoff
value C to which the calculated value of PPOST is compared. Speaking in general terms, the proportion of euploid pregnancies for
which PPOST turns out to exceed the prespecified value of C is the false positive rate (FPR) of the corresponding diagnostic procedure, and analogously, the proportion of trisomies
21 or 13/18 for which one finds that PPOST>C is its detection rate (DTR).
In order to keep consistent with the previous version of PRC, we determined both rates
using the same cutoff value for both types of trisomy. [Table 6] shows the results for 2 different choices of C, namely C=1/150=0.0067 and C=1/500=0.002. With the stricter specification of C to be considered as “red line” beyond which a patient should be advised to undergo
invasive diagnostics, the FPR turned out to be as small as 3.42 and 1.60% for trisomies
21 and 13/18, respectively. Both of the corresponding detection rates exceeded 86%,
and Youden’s index, a customary overall measure of diagnostic accuracy (defined as
the difference between DTR and FPR), came out greater than 83% for both types of trisomy.
Table 6 False positive and detections rates and Youden’s index for 2 different cutoff specifications.
|
Trisomy 21
|
Trisomy 13/18
|
|
1:150
|
3.42%
|
86.8%
|
83.4%
|
1.60%
|
86.4%
|
84.8%
|
|
1:500
|
9.25%
|
94.5%
|
85.3%
|
3.27%
|
89.5%
|
86.2%
|
Stability of the false positive rate under stratification with respect to maternal
weight
[Figs. 6]
[7] show the results of a stratified analysis of the false positive rates using maternal
weight as grouping variable. For trisomy 21, the weight-group-wise FPRs ranged from
2.74 to 4.50% with a coefficient of variation of 14.4%. The maximum was reached in
the class 87.5–95.0 kg whereas in terms of FPR, the heaviest patients (weighing more
than 120 kg) ranked only 7th from the bottom. Altogether, these results admit the conclusion that the FPR entailed
in diagnosing trisomy 21 by means of the new algorithm remains at least as stable
over the weight groups as held true for the previous version of PRC. The analysis
of the weight-group-wise FPR’s for trisomy 13/18 gave similar results: The range was
still narrower (0.76–1.89%), and for the trisomy 13/18 related part of the algorithm,
the lowest FPR was even found in the highest weight group. The coefficient of variation
over all 18 groups was again reasonably low (19.4%).
Fig. 6 False positive rate of the decision rule for the diagnosis of trisomy 21, by maternal
weight.
Fig. 7 Analogue of [Fig. 6] for trisomy 13/18.
Discussion
The objective of the project presented in this paper was to extend our approach to
developing a statistical algorithm for first-trimester screening in a way which makes
it suitable also for the evaluation of findings relating to blood samples taken before
the left-hand endpoint of the usual time window for performing sonography. The key
idea which lead to a satisfactory solution to this problem was to use the logarithmic
transformation not only for the purpose of correcting the biochemical marker concentrations
for maternal weight but at the same time for establishing a modified database for
centile estimation for PAPP-A and free ß-hCG. Working on the logarithmic scale the
reference bands for these parameters could be extended without affecting the smoothness
of the boundaries to a time window corresponding in terms of CRL to 16−84 mm rather
than 45−84 mm. Upon completion of the construction of these bands the same concepts
and computational procedures as for PRC 2.0 could be used for building the diagnostic
algorithms eventually needed. A crucial role was again assigned to measuring the degree
of outlyingness of an arbitrary data point by means of an index (called DoE) which
compares the vertical distance of the point from the center of the band to the width
of the latter. The criterion to be used for diagnostic decision making was derived
from the Bayesian posterior distribution obtained from modeling the joint distributions
of all 3 markers involved (i. e., NT and the concentrations of the 2 biochemical markers)
among euploid pregnancies on the one hand and pregnancies showing a trisomy of the
kind under consideration on the other.
Although the number of data points available in the left-most part of the extended
time window was rather small as compared with the traditional range 45−84 mm in which
the vast majority of the investigations were made, the overall diagnostic accuracy
eventually attained was even better than that provided by PRC 2.0 [1]. Furthermore, we were able to show that approximate independence of the false positive
rates of maternal weight can still be taken for granted.
In principle, the approach described in the previous sections can easily accommodate
additional markers like (i) smoking status, (ii) ethnicity, (iii) fetal nasal bone,
(iv) tricuspid regurgitation, and (v) ductus venosus Doppler. Likelihood ratios obtained
for these markers with the database of the FMF-Germany will be reported in a separate
paper.
It is important to note that the major results presented in this paper are specific
for the platform used for ascertaining the biochemical marker levels. Adapting the
algorithms for a different platform requires to repeat all steps of the construction
with data obtained in environments where this platform is in large-scale use. Recently,
we re-designed the complete diagnostic procedure for use in connection with the Perkin-Elmer
device and the Roche system. Thus PRC 3.0 can be used with results of the biochemical
markers measured with Brahms Kryptor®, Roche Elecsys®, or PerkinElmer DELFIA®, Xpress.
