Results

The Bayesian meta-analysis was performed initially to determine the prevalence of sickle cell disease in the various racial and/or ethnic groups. This analysis is based on evidence shown in the evidence tables in the Guideline Report (Sickle Cell Disease Guideline Panel, in press). Meta-analysis was used to compute the rates with the assumption that Hb SS, Hb SC, and Hb S/B-thalassemia were the hemoglobin phenotypes to be found. The results given in Table 1 show the prevalences of these three phenotypes for each ethnic group. In the second column is the mean value of the prevalence; the third column shows the 95 percent confidence interval. Hispanics were separated into two groups: the first group included Hispanics from the Eastern States, where the population is primarily from the Caribbean Islands. The second group of Hispanics included those from the Western States where the Hispanic population is primarily MexicanAmerican. There are no usable CORN data for Hispanics and Native Americans.

Table 1. Prevalence per 100,000 population of sickle cell disease (Hb SS, Hb SC, Hb S/ẞ-thalassemia) by race and/or ethnic group

A second meta-analysis was performed to determine the prevalence of sickle cell trait (Table 2). Only the AS phenotype was included in this analysis. The subcommittee also analyzed other disease states, including children who were homozygous or doubly heterozygous for other abnormal hemoglobins, as shown in Table 3. Finally, a meta-analysis of

hemoglobinopathy traits other than S was performed as shown in Table 4. Because some articles did not indicate the frequencies of non-sickle hemoglobinopathies and trait conditions, the numbers are probably low. When articles did not indicate the hemoglobin phenotype of the "other" category, it was classified as "other trait" for this analysis.

Table 2. Prevalence per 100,000 population of sickle cell trait (AS) by race and/or ethnic group

Table 3. Prevalence per 100,000 population of doubly heterozygous or homozygous hemoglobinopathies other than SS, SC, or S beta-thalassemia by race and/or ethnic group

Table 4. Prevalence per 100,000 population of other hemoglobinopathy traits by race and/or ethnic group

Discussion

The data presented in this section of the guideline are from sickle cell screening programs. Although the results were obtained by different screening methods, the results are remarkably consistent for most races and ethnic groups. The data, however, show a marked difference between Hispanics from the Eastern United States (primarily Caribbean or Latin American origin) and those who live in the Western States (primarily of Mexican origin). Data for the Western States are from Texas and California; data for the Eastern States are from Florida and New York. Our analysis of these data revealed that Hispanics from the Eastern States have sickle cell disease rates approaching rates in the black population, while Hispanics from the Western States have lower rates, approximating those in the white population. This difference is not apparent when Hispanics are viewed as a single group.

Although no cases of sickle cell disease were encountered in Native Americans, the data for Native Americans are too sparse to yield

conclusions about the prevalence of sickle cell disease in this group. The data also do not permit assessment of the prevalence of sickle cell disease among different Asian groups.

The data listed for other traits (Table 4) and children doubly heterozygous or homozygous for conditions other than Hb AA, Hb SS, Hb SC, or Hb S B-thalassemia (Table 3) should be viewed with considerable skepticism. Recordkeeping was erratic, and in most cases, it was not clear whether unusual hemoglobins were included. In some cases, the phenotype was listed only as "other," which did not permit distinction between traits and disease conditions; trait was assumed in those cases. In addition, several studies showed anomalous rates of other traits and conditions, with the homozygous or doubly heterozygous exceeding the number of traits. The CORN data do not match with all the data for whites in Table 3, indicating these data are questionable and may be biased due to the grouping of Hispanics with whites.

The CORN data demonstrate different prevalences of sickle cell disease among the black populations in different States. Figure 1 shows the density function for sickle cell for each State and the effect of using metaanalysis to combine these results. Louisiana and California (curves for these States peak on the left side of the figure) show lower prevalences of sickle cell than Texas, Virginia, and Michigan (curves peak on the right). Wisconsin (the broader peak on the left) lies in the middle. The dotted line represents the results of the meta-analytic combination of the other lines. The reason for these differences is not known, but they may stem from a difference in the admixture between racial and ethnic groups in these areas. It should be emphasized that the selection of racial and ethnic categories for use in this analysis does not imply that any of these groups are pure or even consistently identified. For example, our analysis shows a higher prevalence of sickle cell disease among whites than the analysis by Tsevat

0

Figure 1. Meta-analysis of CORN data on prevalence of sickle cell disease among black populations

Note: CORN = Council of Regional Networks for Genetic Services. This figure shows the probability density functions for prevalence of newborn sickle cell disease in the States for which CORN data are available. The dotted line (central peak) indicates the results of combining all the other curves. The sharper peak means less uncertainty about the results (see text).

and colleagues (1991). This is likely due to the inexact determination (or even definition) of "white." However, to the extent that the methods used to determine race or ethnicity in our analysis are the same as those used in a future study, the same results should be obtained.

Although there are significant reasons to believe that the populations studied do differ by location, the use of the hierarchic approach biased the results into higher prevalences. This occurred because many of the samples were of small to moderate size and had zero occurrences. When a large study with a low prevalence is combined with a small study with zero prevalence, the hierarchic approach yields a prevalence estimate that is higher than the prevalence of the large study. As a result, it was felt that straight Bayesian meta-analysis would yield a more accurate estimate of the mean. However, the use of Bayesian meta-analysis resulted in unreasonably tight confidence intervals that are difficult to believe. There is greater (but not measured) uncertainty about the estimates than indicated by the confidence intervals.

Cost-Effectiveness of Screening

The cost-effectiveness of universal neonatal hemoglobinopathy screening is a complicated issue. Tsevat and colleagues (1991) found that the cost per life saved by universal screening would vary greatly among those populations with mixed racial composition. They equated the costeffectiveness of universal screening solely with the prevention of death from pneumococcal sepsis by the administration of prophylactic penicillin to infants whose sickle cell disease would otherwise have been unrecognized prior to age 3 years. Their adaptation of risk-reduction data from the Prophylactic Penicillin Study (PROPS), however, biased their results against a policy of early diagnosis (Gaston, Verter, Woods, et al., 1986). All patients in PROPS were known to have sickle cell disease prior to entry, and all were followed carefully; almost two-thirds of the patients were older than 12 months when enrolled. Further, the theoretical population on which Tsevat's most striking result was based does not correspond to any existing U.S. screening jurisdiction.

Lane and colleagues (1992), using a computerized decision model to analyze their experience in Colorado, found that easily overlooked procedural and administrative costs associated with targeted screening could be high enough to make universal screening less expensive. Hidden costs included "loss of economy of scale in the screening laboratory and additional personnel costs for determining each infant's ethnic background."

Sprinkle, Hynes, and Konrad (in press) projected the costs of finding cases of sickle cell disease in 53 U.S. jurisdictions (the 50 States, the District of Columbia, Puerto Rico, and the Virgin Islands) through universal neonatal hemoglobinopathy screening. They compared these costs to the costs projected for finding cases of phenylketonuria (PKU) through the universal neonatal screening practices long established for PKU in the same jurisdictions. Their estimates suggested that 35 jurisdictions would be able to find a case of sickle cell disease for less, often far less, than onehalf the cost of finding a case of PKU. The remaining 18 jurisdictions could substantially reduce the per-case costs, typically for finding both diseases, by combining efforts with other States. In fact, many States already have used this approach to reduce costs. States with relatively few African-Americans tended to be States with small populations in which the efficiency of screening for PKU and other metabolic diseases also could be enhanced by such combination, whether or not they decided to screen for hemoglobinopathies. States deciding to simplify the screening of neonates at high risk for sickle cell disease by testing all neonates, regardless of racial classification, would have little trouble finding demographically complementary screening "partners" with which to form low-cost screening composites.