Software for prenatal Down syndrome risk calculation: a comparative study of six software packages.
To evaluate the influence of software design on risk calculation, we compared six software packages [Prenatal Interpretive Software (Maciel Inc., distributed in France by Abbott), Prisca (DPC France), DIANASoft (BioChem ImmunoSystemes France), T21 (Chiron), PrenatScreen (CIS bio international), and MultiCalc (Wallac)] in two populations: 529 control patients (ages 18-37 years) selected randomly from 90 000 patients screened in our laboratory, in accordance with the national maternal age structure of pregnant patients in France (Institut National des Statistiques et des Etudes Economiques,1993), and all 125 Down syndrome-affected pregnancies (patient ages, 20-37 years). AFP (SFRI) and total hCG (13) were expressed in multiples of the median (MoM) for five software packages and in IU/L corresponding to the manufacturer's MoM for the Abbott package. Gestational age was determined in 99% of cases from ultrasonographic crown-rump length. The cutoff defining the at-risk group was 1 in 250 for all software packages and was calculated at term and not at sampling. The efficacy of the screening tests was evaluated by determining the number of true positives [detection rate for Down syndrome in the Down syndrome-affected population (sensitivity)], and false-positive rate [number of patients above the cutoff in the control population (specificity)].
Risk calculation for Down syndrome combines at least two factors in all cases: the risk of Down syndrome related to maternal age and the risk indicated by maternal serum markers. The maternal age-related risk at term used in each of the six software packages is given in Fig. 1. Although all six software packages yielded a similar risk, substantial variation was apparent, particularly for patients 37 years (1 in 215 to 1 in 273) and 38 years (1 in 167 to 1 in 214) of age.
The percentages of control and Down syndrome-affected pregnancies included in the at-risk group are shown in Table 1. The percentage of cases in which amniocentesis was performed was 2.4-6.8%, and the Down syndrome detection rate was 54.4-66.4%. Software packages that yielded the lowest amniocentesis rate also gave the lowest detection rate, and differences in sensitivity and specificity between the two least sensitive software packages and the four others were significant (P <0.01). For all software packages, the proportion of patients at risk depended on maternal age. The main differences were observed for patients <30, for whom the detection rate was 20-40% and the amniocentesis rate was 1.2-3.1%.
To define how these variations influence individual risk estimates, we determined the number of patients below or above the cutoff as a function of the software. In the control population, all six software packages classified 491 (93%) of the 529 patients in the not-at-risk group and 5 (0.9%) in the at-risk group. In the fetal Down syndrome population, 63 of the 125 cases (50.4%) were detected by all six software packages, and 38 cases (30.4%) were detected by none (data not shown). Discordances were therefore observed in 6% of cases in the control population and in a striking 20% of Down syndrome cases.
The cause of these differences can be analyzed by comparing the likelihood ratio, i.e., the ratio of prior risk (age-specific) to final risk. The percentage of patients with a likelihood ratio greater than an arbitrary value of 1.5 varied from 10% to 20% in the control population and from 62% to 73% in the Down syndrome population (data not shown). These large variations demonstrated that factors other than maternal age are responsible for these differences.
Down syndrome risk factor is derived from two statistical functions (14): the age-related risk and a risk derived from the biochemical indices. In all mathematical models tested here, the risk of Down syndrome related to maternal age is taken into account, according to published values (15-18). However, we observed that for a patient 37 years of age, the risk ranged from 1 in 215 to 1 in 273, thereby changing the decision concerning amniocentesis if the cutoff is 1 in 250. These differences could be minimized by considering maternal age in 1-month and not 1-year intervals (19).
The likelihood ratio method (14) is used in all six software packages tested. However, the detection rate and false-positive rate were markedly different. The relative weight given to hCG and AFP and to maternal age can explain these variations. We tested the weight of each biochemical marker by varying hCG from 1 MoM to 2.5 MoM and AFP from 1 MoM to 0.5 MoM when maternal age, gestational age, weight, and ethnic background were fixed. When hCG was within the reference interval (1 MoM), four software packages (MultiCalc, PrenatScreen, Prenatal Interpretive Software, and T21) gave the same weight to AFP and two software packages (DIANASoft and Prisca) gave a lower weight to AFP. However, a low value for AFP (0.5 MoM) did not assign the patient to the risk group, the risk varying from 1 in 559 to 1 in 800. When hCG was at 2.5 MoM, and AFP was within the reference interval (1 MoM), this high value of hCG was not sufficient for all six software packages to place the patient above the cutoff, the risk varying from 1 in 260 to 1 in 375. To place the patient at a risk of 1 in 250, the necessary AFP value was 0.7 MoM for PrenatScreen, 0.78 MoM for MultiCalc, and 0.95 MoM for the four others. Therefore, the relative weight given to markers depends on the software and markedly affects the risk estimation.
The weight of maternal age in risk calculation was analyzed for a patient with AFP at 0.80 MoM and hCG at 2.2 MoM and a maternal age of 20-38 years. For all software packages, the risk progressively increased with maternal age, but large variations were observed. For example, at 20 years, the risk varied from 1 in 710 (PrenatScreen) to 1 in 378 (DIANASoft and Prisca). Depending on the software, a patient with these marker values would be in the at-risk population at 30 or 34 years.
Another important factor is the choice of marker distribution parameters used in the statistical model. Parameter sets used in software are calculated from relatively few Down syndrome pregnancies and usually are based on women from a single center. Cuckle (20) published variances and covariances obtained by metaanalysis of all studies published to 1995. If all software packages were to use these parameters, differences will be diminished.
Biochemical markers are gestational age-dependent, and the results are expressed in MoM. This supposes that median values are well defined in large populations and that gestational age is given with a high degree of precision (21).
In addition to these main factors, other parameters can be taken into account by the software: choice and number of markers, whether values are corrected for maternal weight, smoking, ethnic background, or diabetes. Risk may be calculated at term or during the second trimester. The fetal death rate in Down syndrome between 15 weeks and term has been estimated as 18-23% (16, 22), but this rate is not always taken into account (Chiron T21). Risk estimation also depends on the extent to which ultrasound is used to estimate gestational age.
Laboratory-related differences must be added to the discrepancies attributable to software. This can lead to great variation in individual risk: Cavalli (23) observed a risk of 1 in 502 and 1 in 80 for the same patient with a Down syndrome-affected pregnancy. The notion of equality between patients, therefore, is clearly flawed.
[FIGURE 1 OMITTED]
In conclusion, this study demonstrates that with the same maternal serum markers, variations are observed between software packages, with a mean detection rate of 54.4-66.4.% and a false-positive rate of 2.4-6.8%. In practice, in a population of 100 000 patients, including 143 cases of Down syndrome, the least sensitive software will detect 78 cases of Down syndrome through 2400 amnio-centeses, whereas the most sensitive will detect 95 cases through 6800 amniocenteses. These differences will have an impact on public health policy (24) and should be minimized. This may be achieved in different ways: use of the same maternal age-related risk, definition for each country of the risk at term or at sampling, use of daily medians, and use of the parameter sets defined by Cuckle (20). However, if we obtain the same value, this does not necessarily mean that it is the most accurate one. We thank the Association pour la Recherche en Medecine Foetale for financial support and all of the manufacturers who contributed software packages. We also thank Laurence Bussiere, who was involved in the early stages of this work.
(1.) Benn PA, Horne D, Briganti S, Greenstein RM. Prenatal diagnosis of diverse chromosome abnormalities in a population of patients identified by triple-marker testing as screen positive for Down syndrome. Am J Obstet Gynecol 1995;173:496-501.
(2.) Goodburn SF, Wates JRW, Raggatt PR, Carr C, Ferguson-Smith ME, Kershaw AJ, et al. Second trimester maternal serum screening using alpha-fetoprotein, human chorionic gonadotrophin, and unconjugated oestriol: experience of a regional programme. Prenat Diagn 1994;14:391-402.
(3.) Kellner LH, Weiss RR, Weiner Z, Neuer M, Martin G, Schulman H, Lipper S. The advantages of using triple-marker screening for chromosomal abnormalities. Am J Obstet Gynecol 1995;172:831-6.
(4.) Mooney RA, Peterson J, French CA, Saller DN, Arvan DA. Effectiveness of combining maternal serum a-fetoprotein and hCG in a second-trimester screening program for Down syndrome. Obstet Gynecol 1994;84:298-303.
(5.) Herrou M, Leporrier N, Leymarie P. Screening for fetal Down syndrome with maternal serum hCG and oestriol: a prospective study. Prenat Diagn 1992;12:887-92.
(6.) Muller F, Aegerter P, Boue A. Prospective maternal serum human chorionic gonadotrophin screening for the risk of fetal chromosome anomalies and of subsequent fetal and neonatal deaths. Prenat Diagn 1993;13:29-43.
(7.) Palomaki GE, Knight GJ, McCarthy JE, Haddow JE, Donhowe JM. Maternal serum screening for Down syndrome in the United State: a 1995 survey. Am J Obstet Gynecol 1997;176:1046-51.
(8.) Spencer K, Carpenter P. Prospective study of prenatal screening for Down's syndrome with free R human chorionic gonadotrophin. Br Med J 1993;307: 764-8.
(9.) Wald NJ, Kennard A, Densem JW, Cuckle HS, Chard T, Butler L. Antenatal maternal serum screening for Down's syndrome: results of a demonstration project. Br Med J 1992;305:391-4.
(10.) Wald NJ, Densem J, Smith D, Klee GG. Four-marker serum screening for Down syndrome. Prenat Diagn 1994;14:707-16.
(11.) Wenstrom KID, Williamson RA, Grant SS, Hudsen JD, Getchell JP. Evaluation of multiple-marker screening for Down syndrome in a statewide population. Am J Obstet Gynecol 1993;169:793-7.
(12.) Wald NJ, Cuckle H, Densem J, Nanchahal K, Royston P, Chard T, et al. Maternal serum screening for the Down's syndrome in early pregnancy. Br Med J 1988;297:883-7.
(13.) Muller F, Boue A. A single chorionic gonadotrophin assay for maternal serum screening for Down's syndrome. Prenat Diagn 1990;10:389-98.
(14.) Palomaki GE, Haddow JE. Maternal serum a fetoprotein, age and Down syndrome risk. Am J Obstet Gynecol 1987;156:460-3.
(15.) Hook EB. Rates of chromosome abnormalities at different maternal ages. Obstet Gynecol 1981;58:282-5.
(16.) Hook EB, Mutton DE, Ide R, Alberman E, Bobrow M. The natural history of Down syndrome conceptuses diagnosed prenatally that are not electively terminated. Am J Hum Genet 1995;57:875-81.
(17.) Hecht CA, Hook EB. Rates of Down syndrome at livebirth by one-year maternal age intervals in studies with apparent close to complete ascertainment in populations of European origin: a proposed revised rate schedule for use in genetic and prenatal screening. Am J Med Genet 1996;62:376-85.
(18.) Cuckle H, Wald N, Thompson S. Estimating a woman's risk of having a pregnancy associated with Down's syndrome using her age and serum AFP. Br J Obstet Gynaecol 1987;94:387-402.
(19.) Hecht CA, Hook EB. The imprecision in rates of Down syndrome by one-year maternal age intervals: a critical analysis of rates used in biochemical screening. Prenat Diagn 1994;14:729-38.
(20.) Cuckle H. Improved parameters for the risk estimation in Down's syndrome screening. Prenat Diagn 1995;15:1057-65.
(21.) Benn PA, Borgida A, Horne D, Briganti S, Collins R, Rodis JF. Down syndrome and neural tube defect screening: the value of using gestational age by ultrasonography. Am J Obstet Gynecol 1997;176:1056-61.
(22.) Halliday JL, Watson LF, Lumley J, Danks DM, Sheffield L. New estimates of Down syndrome risks at chorionic villus sampling, amniocentesis, and livebirth in women of advanced maternal age from a uniquely defined population. Prenat Diagn 1995;15:455-65.
(23.) Cavalli P. False-negative results in Down's syndrome screening. Lancet 1996;347:965-6.
(24.) Seror V, Muller F, Moatti JP, Le Gales C, Boue A. Economic assessment of maternal serum screening for Down's syndrome using human chorionic gonadotropin. Prenat Diagn 1993;13:281-92.
Francoise Muller,  * Philippe Aegerter,  Sandrine Ngo,  Agnes Fort,  Alain Beauchet,  Paul Giraudet,  and Marc Dommergues  ( Biochimie and  Biostatistiques, Hopital Ambroise Pare, 92104 Boulogne Cedex, France;  Medecine Foetale, Hopital Necker Enfants Malades, 75743 Paris, France; * author for correspondence: fax 33 149 09 58 63)
Table 1. Number of control patients (529 cases) and Down syndrome cases included in the at-risk group (cutoff, 1 in 250). Abbott DPC Prisca BioChem Prenatal DIANASoft Soft Control cases Patients <30 years (n = 321) 10 (3.1%) 7 (2.2%) 7 (2.2%) Patients 30-34 years (n = 163) 13 (7.9%) 7 (4.3%) 7 (4.3%) Patients 35-37 years (n = 45) 13 (28.8%) 9 (20%) 9 (20%) Total (%) (n = 529) 36 (6.8%) 23 (4.3%) 23 (4.3%) Down syndrome cases Patients <30 years (n = 25) 10 (40%) 9 (36%) 9 (36%) Patients 30-34 years (n = 45) 29 (64.4%) 29 (64.4%) 29 (64.4%) Patients 35-37 years (n = 55) 44 (80%) 39 (70.9%) 39 (70.9%) Total (%) (n = 125) 83 (66.4%) 77 (61.6%) 77 (61.6%) Chiron T21 CIS bio Prenatscreen Control cases Patients <30 years (n = 321) 6 (1.8%) 6 (1.87%) Patients 30-34 years (n = 163) 5 (3.0%) 4 (2.45%) Patients 35-37 years (n = 45) 9 (20%) 9 (20%) Total (%) (n = 529) 20 (3.8%) 19 (3.6%) Down syndrome cases Patients <30 years (n = 25) 8 (32%) 7 (28%) Patients 30-34 years (n = 45) 29 (64.4%) 27 (60%) Patients 35-37 years (n = 55) 40 (72.7%) 37 (67.2%) Total (%) (n = 125) 77 (61.6%) 71 (56.8%) Wallac MultiCalc Control cases Patients <30 years (n = 321) 4 (1.2%) Patients 30-34 years (n = 163) 2 (1.2%) Patients 35-37 years (n = 45) 7 (15.5%) Total (%) (n = 529) 13 (2.4%) Down syndrome cases Patients <30 years (n = 25) 5 (20%) Patients 30-34 years (n = 45) 26 (57.7%) Patients 35-37 years (n = 55) 37 (67.3%) Total (%) (n = 125) 68 (54.4%)
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Technical Briefs|
|Author:||Muller, Francoise; Aegerter, Philippe; Ngo, Sandrine; Fort, Agnes; Beauchet, Alain; Giraudet, Paul;|
|Date:||Aug 1, 1999|
|Previous Article:||Validation of accuracy-based amino acid reference materials in dried-blood spots by tandem mass spectrometry for newborn screening assays.|
|Next Article:||Day-to-day, postprandial, and orthostatic variation of total plasma homocysteine.|