SHORT COMMUNICATION

The variation of risk estimates through pregnancy in secondtrimester maternal serum screening for Down syndrome

Michael Christiansen1*, Estrid V. Høgdall2, Severin O. Larsen3 and Claus Høgdall4

1Department of Clinical Biochemistry, Statens Serum Institut, Copenhagen, Denmark2Institute of Cancer Epidemiology, Danish Cancer Society, Copenhagen, Denmark3Department of Biostatistics, Statens Serum Institut, Copenhagen, Denmark4The Gynaecologic Clinic, The Juliane Marie Centre, Rigshospitalet, Copenhagen, Denmark

The variation of risk estimates in second trimester maternal serum screening for Down’s syndrome has beenshown to be considerable in quality control schemes, i.e. UKNEQAS. We studied the biological variation ofrisk estimates in 16 women through pregnancy. The maternal serum markers alpha-fetoprotein (AFP),human chorionic gonadotrophin (hCG), unconjugated estriol (uE3) and b-hCG were determined six timesduring late first to late second trimester, and the associated likelihood ratios for Down syndrome werecalculated. The interpersonal variation of markers, as well as that of the likelihood ratio, was much greaterthan the intrapersonal variation. The average intrapersonal standard deviation (SD) of the triple test loglikelihood ratio was 0.2291, corresponding to a central 95-percentile interval 0.36–2.81 of the likelihoodratio. The interpersonal SD of the log likelihood ratio was 0.5482, corresponding to a central 95-percentileinterval 0.08–11.87 of the likelihood ratio. The large difference between the intra- and interpersonalvariation makes it unlikely that biological variation through pregnancy is a major contributor to thevariation of risk estimates obtained several times in the same pregnancy. Rather, improvements in analyticalquality and laboratory management must be expected to result in reduced variation and, in consequence,better performance of screening. Copyright # 2002 John Wiley & Sons, Ltd.

KEY WORDS: prenatal diagnosis; maternal serum screening; likelihood ratio

INTRODUCTION

The risk estimates for fetal Down’s syndrome insecond trimester maternal serum screening based onage, alpha-fetoprotein (AFP), human chorionic gona-dotrophin (hCG), and unconjugated estriol (uE3)(Wald et al., 1997) have been shown to vary con-siderably when the same serum samples are analysedin different laboratories as part of the UKNEQASquality control scheme (Seth et al., 1999), as well aswhen the same women are analysed several times indifferent laboratories through pregnancy (Cavalli,1996; Muller et al., 1999). This may be due to differ-ences in applied immunoassays, biological variationthrough pregnancy, the use of different risk algo-rithms, and variations in quality control systems usedat different centres (Cavalli, 1996; Muller et al., 1999).Analytical variation translates directly into the preci-sion of the individual risk estimate (Williams et al.,2000; Christiansen and Larsen, 2001), and changes inthe performance of immunoassays can be very costly,both financially (Arends et al., 1993), and in terms ofthe number of confirmatory invasive diagnostic pro-cedures that may result in the induction of abortions(Tabor et al., 1986).

Despite the inter- and intrapersonal variation ofrisk estimates reported, it has been shown that therisk estimate is a very good estimate of the actualrisk (Wald et al., 1996; Canick and Rish, 1998; Ondaet al., 1998). However, the reporting of different riskestimates based on different blood samples takenduring the course of the same pregnancy may requirean explanation to both the responsible physician andthe pregnant woman, and it becomes important toknow how a large part of the variation is inevitable asit is due to biological variation within the same womanand how large a part is due to analytical variation.

We examined the biological variation of the mater-nal serum markers AFP, hCG, uE3 and free b-hCG,both in terms of actual concentrations and log MoMvalues, from blood samples drawn at six appointmentsduring each pregnancy. Furthermore, we calculatedthe likelihood ratio for having a Down’s syndromefetus at each time point and used these data to cal-culate the intra- and interpersonal standard deviationof the markers and likelihood ratio.

PATIENTS AND METHODS

Sixteen pregnant women (median age 29 years, range24–36 years), all with a confirmed normal outcome,followed at the Juliane Marie Centre, Rigshospitalet,Copenhagen, were included in the study. Gestational

*Correspondence to: Michael Christiansen, Department of ClinicalBiochemistry, Statens Serum Institut, 5 Artillerivej, DK 2300 S,Denmark. E-mail: mic@ssi.dk

PRENATAL DIAGNOSIS

Prenat Diagn 2002; 22: 385–387.Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002 / pd.297

Copyright # 2002 John Wiley & Sons, Ltd. Received: 5 February 2001Revised: 26 September 2001Accepted: 26 October 2001

dating was based on ultrasound determination of thebiparietal diameter (BPD). The study was approved bythe Science Ethics Committee of Frederiksberg andCopenhagen and was conducted according to theHelsinki II declaration. At six visits during one month(24–39 days), each pregnant woman had blood drawnfrom the cubital vein by the same obstetrician. Theblood sample was left at 4uC overnight to clot, fol-lowed by centrifugation for 10 min at 2000 g. Serumwas aliquoted and stored at x20uC until analysis. Theconcentration of serum analytes was determined onthe DELFIA2 platform using equipment and reagentsfrom EG&G Life Sciences (Turku, Finland). The AFPkit was DELFIA code 1244-004, the hCG kit wasDELFIA code 1244-007, the uE3 kit was AutoDelfiaB083-201 and the b-hCG kit was AutoDelfia B067-101. All analytes of the same kind were analysed on thesame day to eliminate day-to-day variation. Gesta-tional age-dependent median values of analytes weredetermined by log-regression, and used to calculateindividual multiples of the median (MoMs) and theirlogarithmic distributions for each marker. For eachtime point the log likelihood ratio for the fetus beingnormal as opposed to trisomy 21 was calculated fromthe log MoM values of AFP, uE3 and hCG using amultivariate risk calculation in routine use at theDepartment of Clinical Biochemistry, Statens SerumInstitut, Copenhagen. For serum samples obtained atgestational ages earlier than Week 14 or later thanWeek 20 the risk calculation do not give clinicallysignificant values, but the values are reported forcomparison with later values in some women. Varia-tion was analysed by one-sided variance analysis anddifferences between variances were tested by the F-test.

RESULTS

The individual concentration profiles of maternalserum markers through pregnancy were characteristicand in accordance with the expected, namely decreas-ing concentrations of hCG and b-hCG and increasingconcentrations of AFP and uE3 (data not shown). Thecorresponding log MoM values for uE3, AFP andhCG were stable with the standard deviation (with 80degrees of freedom) within persons of 0.0670 forlog MoM AFP, 0.0917 for log MoM hCG, 0.0645 forlog MoM uE3, and 0.0739 for log MoM b-hCG. Thisintrapersonal variation is much less than the inter-personal variation of these markers as known from alarge number of studies (Wald et al., 1997).

The log likelihood ratios at the different time pointsare plotted in Figure 1. The intrapersonal standarddeviation (with 80 degrees of freedom) was 0.2291. Theinterpersonal standard deviation of the log likelihoodratio (with 15 degrees of freedom) was 0.5482. Thislarge difference (F-test, p<10x6) between inter- andintrapersonal variation is clearly reflected in the curvesshown in Figure 1. The calculated standard deviationscorrespond to a 95-percentile interval of 0.36–2.81for the intrapersonal variation of the likelihood ratio

and a 95-percentile interval: 0.08–11.87 for theinterpersonal variation of the likelihood ratio.

DISCUSSION

The biological variation of risk estimates throughpregnancy is much smaller than the interpersonaldifference in risk estimates from different women. Theratio between the lowest and highest values in the95-percentile interval of the likelihood ratio is 7.81 forthe intra-individual and 148.4 for the inter-individualvariation. This order of magnitude difference in theimportance of intra- and inter-individual variationunderlines the fact that intra-individual biologicalvariation is not an important contributor to thetotal variation. Thus, some of the discrepancies bet-ween risk estimates, determined in the same preg-nancy, must undoubtedly be referred to analyticalvariation. The more so, as a recent simulation studyhas shown that the precision of risk estimates in serumscreening is very dependent on the quality (precision)of the determination of the involved parameters(Christiansen and Larsen, 2001).

With the development of modern automated ana-lytical equipment for high throughput screening inscreening laboratories (e.g. the Auto-Delfia platform),and formalised quality control measures (such as theUKNEQAS control program), the analytical andadministrative parts of a screening program may beoptimised, with the result that the performance of thescreening is improved.

The low intrapersonal variation through pregnancysuggests that repeat testing in pregnancy would be awaste of resources if the laboratory responsible forthe analytical work performs well. However, if ana-lytical variation is a problem it may be of benefit torepeat-test some pregnancies. That is because theanalytical variation is normally distributed around themean, why a more accurate estimate of the real valueis reached by using the mean of several determina-tions. This is most important with a risk estimate close

Figure 1—The distribution of log likelihood ratios for Down’ssyndrome in 16 pregnancies as a function of gestational age. Themeasurements from each pregnancy are connected with a line

M. CHRISTIANSEN ET AL.386

Copyright # 2002 John Wiley & Sons, Ltd. Prenat Diagn 2002; 22: 385–387.

to the cut-off separating screen-positive and screen-negative women. The potential usefulness of repeattesting in women with risk estimates close to the cut-off was recognised already in 1994 (Cuckle et al.,1994), but the impact of introducing this into screen-ing programs was considered to be so small that itwas not worthwhile. Precise knowledge on the relationbetween inter- and intra-individual variation may,however, also be important when counselling theindividual pregnant woman.

ACKNOWLEDGEMENT

Lone Rabøl is gratefully acknowledged for excellenttechnical assistance.

REFERENCES

Arends J, Petersen NH, Nørgaard-Pedersen B. 1993. Prenatalscreening for neural tube defects, quality specification of maternalserum alpha-fetoprotein analysis. Ups J Med Sci 98: 339–347.

Canick JA, Rish S. 1998. The accuracy of assigned risks in maternalserum screening. Prenat Diagn 18: 413–415.

Cavalli P. 1996. False-negative results in Down’s syndrome screen-ing. Lancet 347: 965–966.

Christiansen M, Larsen SO. 2001. Precision of risk estimates inmultidimensional likelihood ratio based screening with Down’ssyndrome as a special case. J Med Screen 8: 61–64.

Cuckle H, Densem J, Wald N. 1994. Repeat maternal serum testingin multiple marker Down’s syndrome screening programmes.Prenat Diagn 14: 603–607.

Muller F, Aegerter P, Ngo S, et al. 1999. Software for prenatalDown syndrome risk calculation: a comparative study of sixsoftware packages. Clin Chem 45: 1278–1280.

Onda T, Tanak T, Takeda O, et al. 1998. Agreement betweenpredicted risk and prevalence of Down syndrome in second-trimester triple-marker screening in Japan. Prenat Diagn18: 956–958.

Seth J, Sturgeson CM, Ellis AR, Al-Sadie R, Logan M. 2000.UKNEQAS For Peptide Hormones and Related Substances.Annual Review 1999. UKNEQAS: London.

Tabor A, Philip J, Madsen M, Bang J, Obel EB, Nørgaard-PedersenB. 1986. Randomised controlled trial of genetic amniocentesis in4606 low-risk women. Lancet 1: 1287–1293.

Wald NJ, Hacksaw AK, Huttly W, Kennard A. 1996. Empiricalvalidation of risk screening for Down’s syndrome. J Med Screen3: 185–187.

Wald NJ, Kennard A, Hacksaw A, McGuire A. 1997. Antenatalscreening for Down’s syndrome. J Med Screen 4: 181–246.

Williams KL, Nix BA, Dunstan FDJ. 2000. Effect of screeningalgorithm, parameter values and median smoothing on patient-specific risk estimates for Down’s syndrome screening. Ann ClinBiochem 37: 165–173.

VARIATION OF RISK ESTIMATES 387

Copyright # 2002 John Wiley & Sons, Ltd. Prenat Diagn 2002; 22: 385–387.