Second-Trimester Ultrasound to Detect Fetuses With Down Syndrome: Title and subTitle BreakA Meta-analysis FREE

Chromosomal abnormalities occur in 0.1% to 0.2% of live births,^{1 - 4} and the most common clinically significant chromosomal abnormality among liveborn infants is Down syndrome (trisomy 21). Aside from mental retardation, infants with Down syndrome are at high risk of having associated structural defects, including congenital heart disease, craniofacial abnormalities, and gastrointestinal abnormalities.⁵ Having a child with Down syndrome can be traumatizing and disruptive to families, and the financial burden, including health care expenditures, physical therapy, and special education, can be substantial.^{6 - 7} There has been a growing interest in the prenatal detection of affected fetuses so that parents can be prepared for the birth of an affected child or consider pregnancy termination.

The incidence of Down syndrome increases with maternal age,^{1 ,8} and until 1984, advanced maternal age was the only factor usually considered to identify those at sufficiently high enough risk to justify amniocentesis and fetal karyotyping. Subsequently, the levels of 4 maternal serum biochemical markers in the second trimester of pregnancy—alpha fetoprotein, human chorionic gonadotropin, estriol, and inhibin A—have been found to be associated with Down syndrome.^{9 - 11} Second-trimester maternal biochemical serum screening for abnormal levels of some or all of these markers is now part of routine obstetrical care in the United States and allows the detection of approximately 60% of cases of Down syndrome, with a false-positive rate of 7%.^{12 - 14} Since there is considerable overlap of maternal serum values for unaffected and chromosomally abnormal fetuses, a positive screening test result needs to be followed by a diagnostic test. Amniocentesis can reliably determine fetal karyotype, but there is a 0.5% to 1.0% fetal mortality rate associated with this procedure.^{15 - 20}

Most pregnant women undergo prenatal ultrasound during the second trimester as a routine part of antenatal care, and the majority are not at elevated risk of having a fetus with Down syndrome based on age or serum testing results. A high percentage of chromosomally abnormal fetuses have structural abnormalities that might be recognized on prenatal ultrasound, although these have typically been difficult to detect.^{21 - 40} Several sonographic "markers" have also been reported to be associated with chromosomal abnormalities, including choroid plexus cysts and nuchal fold thickening. While not pathologic themselves, these markers have been used to screen for or adjust the risk for Down syndrome.^{41 - 47} Two previous quantitative summaries of second-trimester ultrasound for the detection of Down syndrome have been published^{48 - 49} ; however, both pooled the results of all studies despite profoundly inconsistent results, and neither study stratified the results by whether the markers were seen in conjunction with fetal structural abnormalities. Thus, the true accuracy of the markers is unknown.

While ultrasound has potential to improve the performance of a Down syndrome screening program, it can also cause harm by prompting unnecessary medical intervention, anxiety related to false-positive findings, and false reassurance to women with affected pregnancies who may be dissuaded from undergoing a diagnostic test because of a normal ultrasound result.^{49 - 50} We reviewed the literature and used meta-analytic techniques to estimate the accuracy of prenatal ultrasound in screening for Down syndrome.

We performed a MEDLINE search for articles published between January 1980 and February 1999 and manually searched bibliographies of relevant published articles. The MEDLINE search included the following second-trimester ultrasonographic markers that have been reported to be associated with chromosomal abnormalities: choroid plexus cyst, nuchal fold thickening, echogenic intracardiac focus, echogenic bowel, renal pyelectasis, shortened humerus, shortened femur, and fetal structural malformations. Additional MEDLINE search terms included Down syndrome, trisomy 21, prenatal ultrasound, chromosomal abnormality, diagnostic tests, biochemical testing, and genetic ultrasound. Review articles, letters, case reports, comments, and non–English-language articles were excluded.

Articles were independently selected and reviewed, and their data were extracted by 2 investigators. Studies were included that recorded second-trimester prenatal ultrasonographic markers and outcome information on a well-described sample of women, and from which estimates of sensitivity and specificity could be calculated (n = 220). Retrospective studies were included provided that the original ultrasound interpretation was used. Active ascertainment of all pregnancy outcomes by chromosomal analysis or visual inspection was required for study inclusion. Studies were excluded if they reported fewer than 5 fetuses with a chromosomal abnormality (n = 15); obtained outcome data only on fetuses with a specific ultrasound finding (n = 68); had incomplete follow-up (n = 11); performed the ultrasound following genetic testing (n = 7); or were primarily focused on a different topic (eg, only looked at major structural abnormalities or had no information on clinical outcomes) (n = 45). Studies that were based on first-trimester ultrasound were not included (n = 18). When relevant data could not be extracted from published articles, the corresponding author was contacted to obtain additional information (n = 5, 1 of whom responded.) For studies that resulted in multiple publications, data from the most recent publication were used. A full list of excluded studies is available from the authors with reasons for their exclusion.

Data were abstracted for each of the following ultrasonographic markers: choroid plexus cyst, nuchal fold thickening, echogenic intracardiac focus, echogenic bowel, renal pyelectasis, shortened humerus, shortened femur, and fetal structural malformations. Data are not presented for the following: shortened ear length, shortened middle phalanx, sandal gap deformity, and increased iliac angle (other markers of Down syndrome), because fewer than 5 articles were found for each and most did not meet the inclusion criteria. The types of fetal structural abnormalities were recorded by organ system (central nervous system, neck, heart, lung, intestine, renal, face, other, and unspecified) and included a range of malformations.

For each article, 2 of the authors abstracted and recorded the number of true-positive, false-positive, true-negative, and false-negative results for each of the markers. Whenever possible, data were abstracted separately for the markers seen as an isolated abnormality or in combination with fetal structural malformations. The following definitions of the markers were used: choroid plexus cysts, cysts of any size or number in the cerebral ventricles (Figure 1A); nuchal fold thickening, thickness of 6 mm or greater (Figure 1B); echogenic intracardiac focus, punctate intracardiac echogenic focus within either ventricle (Figure 1C); renal pyelectasis, anterior-posterior diameter of the renal pelvis of 4 mm or greater (Figure 1D); and echogenic bowel, echogenicity (brightness) equal to or greater than bone (Figure 1E). Data were abstracted for more than 90% of the studies using these definitions. Femur and humerus shortening were less consistently defined (not shown). They were most often defined as an observed-to-expected length ratio of less than 0.90, and the expected length was based on the biparietal diameter using an equation generated from a normal population. Seven of the 11 humerus and 19 of the 29 femur studies used these definitions. In 8 of the femur studies, the definition of a shortened femur was a ratio of biparietal diameter to femur length of greater than 1.5 SDs from the mean for the normal population.

Because some of the abnormal ultrasound findings have been described in association with other chromosomal anomalies as well as Down syndrome, 3 fetal outcomes were considered: unaffected, Down syndrome, or all chromosomal abnormalities when data for Down syndrome alone were unavailable. All chromosomal abnormalities was the outcome used in 50% of the articles relating to the finding of choroid plexus cyst. For the other ultrasonographic markers, we were able to abstract data specifically for Down syndrome from the vast majority of articles.

For each article, the study date, maternal risk of chromosomal abnormality, and the presence and type of structural abnormalities were recorded. Discrepancies in data abstraction between investigators were resolved by reaching a consensus with a third author.

For each study, the sensitivity, specificity, and exact 95% confidence intervals (CIs) were calculated for each of the ultrasound findings. Although we used narrow definitions for each of the markers, studies may have used varying implicit definitions for an abnormal result.⁵¹ Before we combined the results across studies, we sought evidence of a trade-off between sensitivity and specificity introduced through varying thresholds by testing for a correlation between true-positive and false-positive rates.⁵² Such a trade-off would argue against the appropriateness of summarizing the accuracy measures as single point estimates, favoring the use of more complex summary receiver operating characteristic curve analysis.

Pooled estimates of sensitivity and specificity were calculated for each ultrasound finding using a single-term logistic regression model, and the unit of analysis was the study. This method produces results similar to those obtained from calculating a weighted average of the sensitivities and specificities, weighting by sample size. The consistency of the study results was assessed by considering the "goodness-of-fit" deviance statistic. The dispersion of the data was calculated by dividing the deviance statistic by its df: dispersions >1 correspond to variability greater than that expected by chance. Where the fit was poor, the model was adjusted to account for the overdispersion by multiplying the SE by the square root of the dispersion statistic.

The sensitivity and specificity estimates were calculated stratified by whether the finding was seen as an isolated abnormality or in conjunction with other structural abnormalities. When studies did not explicitly state either, they were labeled as unknown and analyzed separately. Results were also stratified by study design and size. To identify whether the results might be influenced by study design, sample size, or by whether the markers were reported in isolation, terms for these factors were introduced into the logistic regression model.

Positive and negative likelihood ratios (LRs) were calculated for each ultrasound finding, stratified by whether it was seen in isolation or in association with fetal structural malformations, and were pooled using a DerSimonian and Laird random effects method for pooling risk ratios.⁵³ All analyses were performed with STATA statistical software (release 6.0, STATA Corp, College Station, Tex).

Positive and negative predictive values were calculated at a disease prevalence of 1:700 (the population prevalence of Down syndrome) and 1:300 (equivalent to the mid second-trimester prevalence in a 35-year-old woman and generally considered high-risk for Down syndrome) by application of Bayes theorem to the estimated LRs. For each ultrasound finding, the number of women that would need to be screened to detect a case of Down syndrome was calculated as the inverse of the product of the prevalence of Down syndrome and the sensitivity of the ultrasound finding. The number of normal fetuses that would be lost for each case of Down syndrome identified was calculated assuming that all screen-positive ultrasound examinations were followed by invasive diagnostic testing, and that the rate of fetal loss per amniocentesis procedure is 0.8%.^{16 - 20}

Fifty-six studies met inclusion criteria and described 130 365 unaffected fetuses and 1930 cases of Down syndrome (Table 1).^{22 ,26 - 27 ,38 ,40 - 46 ,54 - 98} The mean maternal age was 34 years, and 88% of the studies (n = 49) included women at increased risk of chromosomal abnormality based on age (n = 39), serum biochemical testing (n = 36), or family history of chromosomal abnormality (n = 16). The number of women who were at increased risk for each indication was generally not provided. The overall prevalence of Down syndrome was 1.5% (compared with 0.1% in the general population). Down syndrome was present in 6% of the women in case-control studies (n = 24) compared with 1% in the prospective studies. Outcome ascertainment included fetal karyotyping in 53 studies (95%). The number of studies that evaluated each marker ranged from 5 to 29. For studies that included fetal structural malformations, the types of abnormalities specified varied widely, ranging from mild, such as cleft lip, to lethal, such as anencephaly.

The sensitivity and 1 − specificity (false-positive rate) reported by each study for the ability to detect cases of Down syndrome are presented for each ultrasonographic marker in Figure 2. The results were markedly heterogeneous for all findings. For example, the percentage of fetuses with Down syndrome correctly identified using a thickened nuchal fold varied from 7%⁶³ to 75%.^{41 ,68} These reported differences in detection were not the result of a trade-off between the sensitivity and specificity (Spearman correlations between the sensitivity and the specificity were not significant, and there is no obvious trend in the false-positive rates in Figure 2 when ordered by the sensitivity rates).

We explored possible reasons for the inconsistency across studies, illustrated for nuchal fold in Table 2. There were no differences in the sensitivity based on study size (test for difference between groups, P = .23). However, we found significant differences in the reported accuracy based on study design (test for difference, P = .008) and whether the marker was seen as an isolated abnormality or in association with fetal structural abnormalities (test for difference, P<.001).

The sensitivity was significantly lower among studies that reported the results of the markers as isolated abnormalities, illustrated for thickened nuchal fold in Figure 3. For example, a thickened nuchal fold observed in isolation was seen in 4% of cases of Down syndrome, compared with 26% when observed in addition to other abnormalities. Additionally, the results were more consistent among the studies that reported on the isolated findings: most of the CIs for these studies overlap the summary estimate (Figure 3).⁹⁹

For each of the ultrasonographic markers, the sensitivity for Down syndrome was low when the marker was seen without associated structural malformations or other markers, ranging from 1% for choroid plexus cyst (95% CI, 0%-3%) to 16% for shortened femur (95% CI, 5%-40%) (Table 3). The specificity for each of the markers was greater than 95% when the finding was seen as an isolated abnormality.

If nuchal fold thickening is identified in the second trimester, the odds of Down syndrome increase by approximately 17-fold (positive LR, 17; 95% CI, 8-38) (Table 4). For the other 6 markers, the positive LRs were significantly lower, and for choroid plexus cyst and renal pyelectasis, the positive LRs were not significant. Because the markers were seen in only a minority of abnormal fetuses, a normal finding did not substantially decrease the risk of a fetus having Down syndrome. None of the negative LRs were significant.

Among women at average risk of having a fetus with Down syndrome, if a thickened nuchal fold is identified, the risk is 2% (Table 4), and among women at high risk, the risk is 5%. The positive predictive values were significantly lower for the 6 other markers (≤1% in low-risk women, and ≤2% in high-risk women for each of the markers). The negative predictive value for each of the markers was greater than 99%, reflecting the rarity of Down syndrome.

If ultrasound screening is used in women who are at average risk of having an affected fetus, the number needed to be screened to detect a case of Down syndrome is high (range, 4454-87 413) depending on the marker used (Table 4). For each case of Down syndrome correctly identified, there will be false-positive diagnoses, ranging from 79 (for nuchal fold) to 611 (for choroid plexus cysts). Because an abnormal finding is followed by amniocentesis to make a definitive diagnosis, there will be inevitable losses of unaffected fetuses as a complication of amniocentesis. For example, if the presence of an echogenic intracardiac focus is used as a basis to offer amniocentesis, 2 fetal losses in low-risk women and 1 fetal loss in high-risk women will occur as complications for each case of Down syndrome identified.

One study contributed approximately 25% of the unaffected fetuses to our meta-analysis (n = 33 000).⁶³ To determine the impact of this study on the overall summary measures, the sensitivity and specificity rates were recalculated excluding that study's data, and there was no change in the sensitivity rates for any of the markers. On the other hand, the false-positive rates doubled when this study was deleted (for choroid plexus cyst, the false-positive rate increased from 1% to 3%, for nuchal fold from 0.5% to 0.9%, and for echogenic bowel from 1% to 3%). Within the studies that considered isolated markers, there was typically no evidence of a trade-off between the true-positive and false-positive rates (eg, for nuchal fold the correlation between sensitivity and specificity was not significant [P = .27]), and thus summary statistics could be calculated and are valid.

When seen as isolated findings, the second-trimester ultrasonographic markers of choroid plexus cyst, echogenic intracardiac focus, echogenic bowel, renal pyelectasis, shortened humerus, and shortened femur are not helpful in either confirming or excluding the presence of Down syndrome.¹⁰⁰ Only 1 marker, thickened nuchal fold, may be useful at distinguishing between unaffected and affected fetuses. If nuchal fold thickening is identified, the risk of Down syndrome increases by approximately 17-fold. However, this marker is present in only a minority of fetuses with Down syndrome, and many fetuses would need to be screened for nuchal fold thickening to make a single diagnosis of Down syndrome.

Moreover, because the sensitivity of the ultrasonographic markers is so low, and because Down syndrome is rare, the vast majority (>99%) of fetuses with an isolated marker will be unaffected. The use of the ultrasonographic markers as an indicator for invasive testing with amniocentesis will lead to an increase in the number of unaffected fetuses lost as a complication of the procedure. Although women may place different values on the balance between the identification of a fetus with Down syndrome and the loss of an unaffected fetus due to a complication of amniocentesis,¹⁰¹ this highlights an important potential harm associated with a test often considered risk-free. If, for example, identification of an echogenic intracardiac focus is used as a basis for offering amniocentesis to pregnant women at low risk of carrying an affected fetus, 2 unaffected fetuses will be lost as a complication of amniocentesis for each correctly identified Down syndrome case. Additionally, because the false-positive rate is 1% or greater for most of the markers, when all of these markers are used in aggregate, the false-positive rate may approach 10% or more, leading to much needless anxiety throughout pregnancy and beyond.^{50 ,102 - 104}

Furthermore, if women who are at elevated risk of carrying a fetus with Down syndrome based on maternal age or serum testing results are dissuaded from amniocentesis due to the absence of ultrasonographic markers, this will reduce the prenatal detection of fetuses with Down syndrome as well as the effectiveness of biochemical screening.⁴⁹ Finally, there are significant costs associated with ultrasound screening for Down syndrome, particularly those that relate to the further evaluation of detected findings. Even for nuchal fold thickening, more than 15 000 scans would need to be performed in average-risk women and more than 6000 scans in high-risk women to detect a single case of Down syndrome. Based on available data, the accuracy of a second-trimester ultrasound screening program would be far less, and the cost far greater, than that achieved by current biochemical screening.¹⁰⁵ There are no data on whether serum markers and these ultrasonographic markers are independent indicators of Down syndrome risk, and thus it is unclear whether the use of the ultrasonographic markers contributes any additional detection to that achievable with serum testing.

Although our results suggest poor accuracy for most of these ultrasonographic markers, it is common practice for clinicians to use them as evidence of an increased risk of Down syndrome.^{50 ,106} For example, in a large health maintenance organization in California, all pregnant women are offered second-trimester ultrasound to screen for fetal anomalies. If a choroid plexus cyst is identified, the patient literature reports that the fetus has a 1:100 risk of chromosomal abnormality and the woman may undergo amniocentesis. This is 7-fold her a priori risk of chromosomal abnormality, and this increased risk is not supported by our results. Similarly, Boyd et al⁶³ found that while the presence of ultrasonographic markers increased the prenatal detection of malformed fetuses by only 4%, identification of the markers was responsible for a 12-fold increase in the false-positive rate.

Although our review cannot exclude the possibility that there is benefit to identifying these markers, their validity is not supported by our findings. Furthermore, although we focused on the most common ultrasonographic markers used to screen for chromosomal abnormalities, there are many others described in the literature and evaluated in even fewer patients.

Our meta-analysis focused on the ultrasonographic markers that may be seen in isolation. Many of the studies examined these markers in conjunction with other markers and with fetal structural abnormalities. Using such a combination may improve the accuracy of screening with ultrasound. However, these studies reported statistically inconsistent results, and thus a summary measure of the accuracy of a composite score could not be determined. Furthermore, the low sensitivity of the isolated markers suggests that these may not contribute to the information provided by the presence of fetal structural malformations. Our analysis suggests that the high sensitivity for Down syndrome reported by many studies using ultrasonographic markers may be due to the detection of the associated structural abnormalities. None of the negative LRs were significant; thus, the absence of an individual marker could not be used to rule out Down syndrome. Therefore, it seems improbable that the absence of a combination of ultrasonographic markers would substantially decrease the risk of Down syndrome.

It is important to distinguish between the ultrasonographic markers, which are themselves harmless, and fetal structural abnormalities for several reasons. First, the prenatal identification of fetal structural malformations may have benefit to the mother unrelated to the risk of chromosomal abnormality (eg, the family can anticipate the special needs of the child).⁵⁰ Second, most clinicians believe that detection of a major structural abnormality is sufficient grounds to offer invasive testing. Finally, the vast majority of examinations demonstrating isolated sonographic markers are false-positives.

We focused on second-trimester ultrasound and did not include studies of nuchal translucency screening, a first-trimester ultrasound test used at approximately 10 to 14 weeks' gestation. Currently, this test has not been embraced into clinical practice in the United States, the implications of the accuracy of this test on clinical management are far different than for markers observed on routine ultrasound at 15 to 24 weeks' gestation, and prospective studies of nuchal translucency screening are under way.

Our analysis has several limitations. We may have missed unpublished work, particularly of smaller studies, but these are unlikely to substantially affect the results. Broadening the search to include other electronic databases and articles in languages other than English may have increased the number of studies included, but it seems unlikely that it would make our results any more favorable toward ultrasound. The results may predominantly reflect the findings of specialized centers, and this may inflate the accuracy in a general community setting. The majority of included studies were confined to high-risk pregnancies, such as women referred for prenatal diagnosis because of advanced maternal age or a positive serum screening test result, and the sensitivity of the markers may be different, and probably lower, in low-risk women.

We excluded many studies on the basis of methodologic criteria. It was not possible to determine if including these studies would have had a significant impact on the overall results because most of these excluded studies (n = 67) reported only a positive predictive value, without reporting the risk of Down syndrome among the population studied or the false-positive rate of the marker. The exclusion of the study that contributed 25% of the unaffected fetuses to this meta-analysis resulted in considerably lower estimates of the sensitivity, and worse estimates of screening performance of choroid plexus cyst, nuchal fold, and echogenic bowel.⁶³ That study collected results from an anomaly registry, and it is not clear if all of the cases with sonographic markers had been reported. In addition, we pooled the results from studies that used different designs, including case-control studies, which are likely to lead to overestimation of the sensitivity,¹⁰⁷ so that the true accuracy of these findings may be even lower than we report.

In summary, we found that a thickened nuchal fold may be useful in distinguishing between unaffected fetuses and those with Down syndrome, but that the overall sensitivity of this finding is low, and thus it is not practical for use as a screening test. The remaining ultrasonographic markers did not discern well between unaffected fetuses and those with Down syndrome. Because the use of these markers may be associated with more harm than benefit, clinicians should be very cautious about the use of these markers to counsel women about their risk of having a fetus with Down syndrome.