Laboratory Assay Reproducibility of Serum Estrogens in Umbilical Cord Blood Samples¹

There is emerging evidence suggesting that the intrauterine environment may influence the future risk of breast cancer in the female offspring in humans (1, 2) as well as experimental animals (3, 4). A variable level of estrogen exposure in utero has been suggested to explain this hypothesized relationship between perinatal experiences and breast cancer risk (2). However, data on perinatal estrogen levels in humans to examine the potential involvement of estrogens in the hypothesized link are limited (5, 6, 7, 8, 9).

Umbilical cord blood provides a source of measuring relative levels of perinatal estrogen exposure without testing the mother or the newborn invasively. To make a valid comparison of the cord blood estrogen levels among populations in epidemiological studies, the laboratory reproducibility of sex hormone assays must be assessed for the sources of variation in measurements. Although several studies have addressed the reproducibility of serum and plasma estrogen assays for blood samples from adult women (10, 11, 12, 13, 14), such data on cord blood samples available to date are sparse. Data for adult samples may not be applicable to cord blood measurements because estrogen levels in the cord blood are extremely high compared to levels in the peripheral blood of adults.

As part of the pilot study that we conducted to examine ethnic differences in cord blood estrogen levels, we evaluated the reproducibility of laboratory assays and its implications on sample size estimation. Specifically, we addressed the following two issues: (a) examination of the laboratory reproducibility of cord blood estrogen levels (i.e., correlation between duplicate measurements of the same blood samples); and (b) estimation the relative contribution of variability due to study subject and assay batch to the overall variation in measured hormone levels.

A total of 25 female babies from full-term pregnancies (i.e., ≥37 weeks gestation) were included in this study. These subjects were recruited between March and December 1997 at two study sites; 15 Caucasian babies at Stanford Medical Center and 10 Chinese-American babies at Chinese Hospital in San Francisco. There were no exclusion criteria other than race and gestational age.

Approximately 25 ml of whole blood (mostly venous) was collected directly from the umbilical cord by draining blood from the free end of the cord into a plastic centrifuge tube containing no additives after the cord was clamped and cut. Collected cord blood was refrigerated at 4°C until serum was separated by centrifugation. Serum was then mixed well, divided into 2-ml aliquots in plastic vials, and stored at −70°C. All of the samples were processed within 24 h from the time of blood collection (median, 6 h; range, 2–24 h).

The study protocol has been reviewed and approved by the Institutional Review Board of each participating institution.

Two aliquots of cord blood serum from each baby were sent to the laboratory of Endocrine Sciences (Calabasas, CA) in either of two batches shipped in August and December 1997, according to the schedule summarized in Table 1. Thus, duplicates were assayed about 4 months apart for 14 subjects, whereas the other 11 subjects had their duplicate sera assayed in the same batch. In addition, an aliquot of the two pooled sera controls, which was prepared by mixing four Caucasian and three Chinese sera, was included in each batch to monitor the laboratory drift. Identity of duplicate serum aliquots from the same baby and each baby’s race were concealed from the laboratory by labeling vials with randomly selected numbers.

Each serum aliquot was assayed for estrone, total estradiol, weakly bound estradiol (equivalent to “bioavailable” estradiol, consisting of unbound estradiol and albumin-bound estradiol), and SHBG³ using standard laboratory methods at Endocrine Sciences (see below). Because the serum estrogen level in cord blood samples is much higher than typical clinical samples, the measurements were confirmed by assay on multiple dilutions when needed.

Serum estrone and total estradiol were measured by extraction with hexane:ethyl acetate and chromatography on Sephadex LH-20 columns with benzene:methanol, followed by RIA with specific antiserum, using a modification of the method of Wu and Lundy (15). Coefficient of variation for the interassay variation from routine clinical assays at the laboratory was 8.2% for estrone (based on 45 assays; mean, 20.8 ng/dl; SD, 1.7 ng/dl) and 6.9% for total estradiol (based on 44 assays; mean, 30.5 ng/dl; SD, 2.1 ng/dl).

Percentage of weakly bound estradiol was measured by mixing a sample with tritiated estradiol and precipitating bound estradiol with ammonium sulfate. Serum concentration of weakly bound estradiol was computed as the percentage multiplied by the total serum estradiol level of the sample.

SHBG was measured by an immunoradiometric assay using monoclonal anti-SHBG antibodies coated onto plastic beads. Coefficient of variation for the interassay precision from routine assays was 10.6% (based on 19 assays; mean, 177 nmol/liter; SD, 18.8 nmol/liter). Data on the stability testing results at the laboratory showed that percentage SHBG remaining after storage as whole blood for 24 h at 4°C (similar to the blood processing conditions in this study) was 97.2%.

Measures of central location (arithmetic mean and geometric mean) and of spread (SD) across babies were calculated, using the baby-specific average of duplicate measurements for each hormone or protein measured. Pearson’s correlation coefficients between duplicate measurements were calculated to examine the level of concordance between assay results of two aliquots. All of the Ps reported are two-sided.

A linear model was used to estimate the components of variance for each hormone or protein associated with variation among subjects and variation between assay batches (August versus December 1997). The model is a random effects model in which the measurement (on the logarithmic scale with base 10) for aliquot k (k = 1 or 2) from subject i (i = 1–25) measured in batch j [j = 1 (August) or 2 (December)] is the sum of four terms as shown below:

where μ, α_i, β_j, and e_ijk denote the overall mean, the deviation between the overall mean and that of the ith baby (i = 1–25; random effect), the deviation for the ith baby between her mean and the mean of her measurement in batch 1 (versus batch 2; random effect), and residual error, respectively. The model assumes that α_i, β_j, and e_ijk are independently and normally distributed with mean zero and variance σ_a², σ_b², and σ_e² respectively. Variance components were estimated by the restricted maximum likelihood method, using PROC VARCOMP in the SAS statistical package (16). ICC was calculated as a proportion of the total variance that was due to between-subject (baby) variation. SE of the ICC was estimated by the δ method (17).

Data were analyzed on the logarithmic scale (base 10) for Pearson’s r and variance component analysis, as the deviation from normality, as measured by the Shapiro-Wilk test (16), was smaller with log-transformed data than with those in the original scale. All of the statistical analyses were performed using the SAS statistical package (16).

Results of the laboratory assays for estrone, total estradiol, weakly bound estradiol, and SHBG from 25 babies are summarized in Table 2.

A graphic illustration of correlation between duplicate measurements of total estradiol is shown in Fig. 1. The sample that gave the highest value in the assay of aliquot 1 was much farther off the diagonal line (indicating perfect concordance between the two measurements) than were the other samples.

Quantitative assessment of concordance between the duplicate measurements was made by calculating Pearson’s correlation coefficients (Table 3). These coefficients were estimated for all 25 babies as well as for the two subgroups defined with respect to the synchronicity of two measurements (i.e., whether duplicates were measured in the same batch). Pearson’s correlation coefficients were >0.90 for all of the compounds except for total estradiol when all of the subjects were included. These correlation coefficients were lower for the samples whose duplicates were measured in different batches than for the samples whose duplicates were measured in the same batch with an exception of estrone. All of the correlation coefficients shown were statistically significant at the α level of 0.05. Exclusion of the outlier shown in Fig. 1 did not make much difference in Pearson’s correlation coefficient for total estradiol or weakly bound estradiol.

We also estimated variance components to examine relative contributions of between-subject (baby), between-batch, and residual variations to the total variability of assay results. Variance component estimates based on the model A (see “Materials and Methods”) are shown in Table 4. Using the variance component estimates, we also calculated ICC, defined as a proportion of the total variance due to between-subject (baby) variation. The ICC was >90% for estrone and SHBG and between 80 and 90% for total estradiol and weakly bound estradiol. Inclusion of an additional parameter accounting for the variation between race (Caucasian versus Chinese) in the statistical model did not change the other parameter estimates (data not shown).

To demonstrate the effect of measurement error on sample size required for detecting a statistically significant difference between the mean levels in two populations, we calculated a sample size for some hypothetical scenarios using the variance component estimates obtained from this study (18). Table 5 shows the estimated sample size (per sample of equal size) for total estradiol and SHBG that had the smallest and largest ICC, respectively, in this study. The sample size was calculated for a two-sided test with an α level of 0.05 and 80% statistical power. We assumed that log-transformed values are approximately normally distributed and that the levels in the two populations have equal variance. We used two values for the variance (of log-transformed values) based on our variance component estimates (Table 4); the variance component estimate for subject (baby) effect only for the scenario with no measurement error and the estimated total variance for the scenario with measurement error. This calculation also assumes that measurement error is random and does not differ between two sets of measurements being compared. Because the sample size is proportional to the variance, ≈25% more (i.e., 0.0453/0.0362 = 1.25) subjects would be needed for total estradiol comparison if the level of measurement error is similar to what we observed in this study than what would be required were there no measurement error. In contrast, the sample size would increase only minimally (0.0204/0.0198 = 1.03, or a 3% increase) for SHBG.

We observed a high correlation between duplicate measurements of cord blood serum estrogen and SHBG levels. Variance component analysis showed that >80% of the variation in assay results could be explained by the variability between babies. There has been only one study that presented the assay reproducibility of cord blood estrogen levels to our knowledge. In a study of 256 male and female babies by Maccoby et al. (19), Pearson’s correlation coefficients between duplicate measurements conducted in three samples of babies ranged from 0.98 to 0.99.

A few studies have been conducted to examine the laboratory reproducibility of serum and plasma estrogen levels in adult women. Bolelli et al. (10) evaluated the effects of long-term preservation of frozen plasma and serum samples on the sex hormone assay results including estradiol (10). When assays were repeated 3 years after baseline, Pearson’s correlation coefficient between the two measurements for both serum and plasma estradiol was 0.99 for postmenopausal women. Hankinson et al. (12) assessed the laboratory reproducibility of endogenous hormone levels by sending four to seven replicate plasma samples to each of the four laboratories on one or two separate occasions. Their estimates of ICC for plasma estrone levels ranged between 0.12 and 0.90 among the laboratories. The authors also found that laboratory reproducibility varies among hormones and emphasized the need for evaluating laboratory performance before sending samples for hormone analysis. A study of six female volunteers by Potischman et al. (13) to evaluate the reproducibility of hormone measurements among days of assay and laboratories gave rather discouraging results. Ten serum aliquots per woman (five sets of duplicate aliquots) were prepared and each of the three laboratories analyzed six aliquots per day for 10 consecutive working days for each hormone. The variability of assay procedures currently used at the participating laboratories to measure estrogens and androgens was large enough for the authors to suspect that it might hamper efforts to study the association between disease and endogenous hormones in epidemiological studies. Falk et al. (14) investigated the consistency of laboratory assays in one laboratory, using serum from three postmenopausal female volunteers. Their analyses showed that ICCs for some compounds including SHBG were high, suggesting that these assays were suitable for population-based studies, whereas other hormones measured in the study (e.g., estradiol) did not perform as well.

Assay reproducibility of estrogen levels in cord blood samples may not necessarily be comparable to that in peripheral blood samples from adult women. For example, Pearson’s correlation coefficient is influenced by the range of measured values such that data with a wider range result in a larger correlation coefficient estimate. Because estrogen levels in cord blood are up to 100-fold higher than levels in adult women, a high correlation coefficient in our study may not be directly comparable with data on adult women. However, it may be worth comparing the relative contribution of various sources of variation to the overall variability in measured estrogen levels by the variance component analysis. Gail et al. (11) measured sources of assay variability (i.e., day of assay, laboratory, aliquots derived from the same blood sample, and study subject) for estrogens and progesterone, using plasma samples from 15 female volunteers. The authors estimated variance components for these parameters and ICC, using a statistical model similar to what we used in this study. ICCs for both estrone and estradiol in sample from postmenopausal women ranged substantially among three laboratories in the study by Gail et al. (Ref. 11; 61.1–83.6% for estrone and 41.5–91.3% for estradiol). ICC estimates for cord blood samples in this study were 91.8% for estrone and 80.0% for estradiol. SHBG showed the highest ICC (97.1%) among the four compounds measured in this study. This magnitude of measurement error would increase the sample size required for detecting a difference between two populations for total estradiol and SHBG in cord blood by 25 and 3%, respectively.

Estrogen levels in the cord blood are not routinely measured for clinical purposes. However, a few studies have investigated cord blood estrogen concentrations in relation to various covariates, such as the baby’s sex and birth rank (19, 20, 21, 22, 23). Mean and SD (as a measure of spread across babies) of estrone, total estradiol, and SHBG levels in those reports and this study are summarized in Table 6. Results of estrone and estradiol concentrations in the present study were higher than those observed in other reports. This could be due to: the real difference among different populations studied in different studies; differences in conditions for collecting, processing, and storing cord blood specimens; assay variability among laboratories; or chance. Given the relatively large SDs in these studies, however, our results may not be totally incompatible with those of previous reports.

In summary, our data demonstrated an overall high reproducibility of serum estrogen and SHBG assays for cord blood samples. A nonsignificant batch effect (i.e., variation due to the timing of laboratory measurements) in this study suggests that allocation of serum samples to multiple assay dates, at least up to 4 months apart, may not be a serious concern. However, effects of long-term storage of frozen serum specimens and laboratory drift over a longer study period on cord blood hormone measurements must be assessed in future studies.

We thank the medical and nursing staff in obstetrics at Chinese Hospital and Stanford Medical Center for their cooperation in subject recruitment and cord blood sample collection; Maria Isabel Garcia, Shirley Huang, Tien Tien Woo, and Sandy Yeung for technical assistance; Dr. David Simmons for sharing his protocol of cord blood collection and processing; Dr. Carol Nagy Jacklin for bringing the earlier work on cord blood sex hormones to our attention; Drs. Barry Behr and Donald Chandler for comments on sex steroid hormone assays; Drs. Byron Brown, Philip Lavori, and Alice Whittemore for suggestions on statistical analysis; and anonymous reviewers for helpful comments on the previous version of this manuscript.