Abstract
Life-course epidemiologic studies frequently obtain data on perinatal and childhood factors through adult self-report. Data from a prospective US birth cohort of 262 women, born between 1959 and 1963 (average age at adult follow-up, 41.8 years), were used to assess the validity of self-reported birth weight category by sociodemographic characteristics. The effect of reporting error on the associations of birth weight with childhood and adult body mass index was evaluated by comparing the estimates of associations from linear regression analyses. The level of agreement between the birth weight category reported by 85% of the participants and the birth weight recorded at the time of birth was moderate to good (sensitivity = 73%, weighted kappa = 0.67). The validity varied by birth weight category (sensitivity range = 58%–81% for the lowest and highest birth weight category, respectively) and was highest for participants who were white, of lower childhood family income, and born to older mothers. Despite this moderate to good validity, the associations of birth weight with childhood and adult body mass index were attenuated and no longer statistically significant when self-reported birth weight was used. In conclusion, birth weight reported in middle adult life is measured with error, limiting its utility for detecting modest associations with health in later life periods.
Keywords: birth weight, body mass index, ethnic groups, mental recall, reproducibility of results, socioeconomic factors
Growing evidence suggests that factors encountered in perinatal and early childhood periods are associated with health outcomes in later life, including chronic diseases developing in middle to late adult life (1–4). Long-term prospective studies offer the optimal design for collecting reliable and valid data on early life factors but have many practical challenges, including a high cost of follow-up and selective attrition by various sociodemographic characteristics (5–8). An alternative method for ascertaining early life factors is through self-reported data collected retrospectively with considerably more ease and less expense but greater potential for bias arising from recall of past experiences. The validity of self-reported retrospective data is often unknown and may vary by sociodemographic factors and type of information recalled and reported. Given the greater efficiency and cost-effectiveness of retrospective data collection, further examination of the potential reporting error in important early life factors is needed.
One of the easiest early life factors to collect retrospectively is birth weight. The most accurate or “gold standard” measurement of birth weight is the weight recorded in medical files and sometimes on birth certificates by clinical staff immediately after delivery. Only a few studies have evaluated the validity of adult self-reported birth weight (9–14). In several studies, validation analyses were restricted to substantially smaller subsets of the original samples who had both adult self-reported and original birth weight records and who differed from the original sample in recorded birth weight and by socioeconomic (e.g., educational level), demographic (e.g., age at reporting, nulliparity), and parental (e.g., maternal age and vital status) characteristics (9–12).
Using data from a multiethnic birth cohort of US women with complete birth weight data, we investigated the validity of adult self-reported birth weight and whether the validity was influenced by childhood and adult socioeconomic status and ethnic background. We also evaluated the effect of reporting error on the associations of birth weight with childhood and adult body mass index.
MATERIALS AND METHODS
Study populations
The New York Women's Birth Cohort Adult Follow-up is a cohort of 262 women who were enrolled at birth in the New York City site of the National Collaborative Perinatal Project, followed up prospectively until age 7 years, and traced and enrolled in adult follow-up studies. Women who were traced and participated in the adult follow-up survey (44% of 841 eligible women were successfully traced, and 70% of those traced participated) did not differ from those who were not traced and enrolled in terms of birth weight, childhood growth measures, and maternal characteristics (e.g., age at pregnancy, prepregnancy body mass index, weight gain during pregnancy), but they were significantly more likely to be white and of higher childhood socioeconomic status (for more information, refer to references 15–17). All participants were born between 1959 and 1963 and were, on average, aged 41.8 years (standard deviation (SD), 1.8; range, 38–46) at the time of the adult follow-up survey.
Variables
Birth weight.
Birth weight in grams was recorded at the time of delivery for each participant (“recorded birth weight”). Participants were requested to provide their birth weight on the adult follow-up survey by choosing one of the following response options: don't know, less than 5.5 pounds, 5.5–6.9 pounds, 7–8.4 pounds, 8.5–9.9 pounds, and 10 pounds or more (“self-reported birth weight”) (1 pound = 0.454 kg). Because of the small number of participants weighing 10 pounds or more at birth, the 2 highest birth weight categories were combined. Birth weight categories were converted into grams for comparison with recorded birth weight. Those with missing self-reported birth weight or a “don't know” response were categorized as nonreporters.
Socioeconomic status.
Childhood socioeconomic status was measured by maternal education (in number of years) and family income (in dollar amount and adjusted for family size), both assessed around the time of birth. Adult socioeconomic status measures included participants’ highest completed educational degree and current family income (in 12 ordinal categories ranging from <$5,000 to >$129,999).
Ethnicity.
Participants reported their ethnicity in the follow-up survey and were categorized into 3 mutually exclusive groups of Hispanics, non-Hispanic African Americans (African Americans), and non-Hispanic whites (whites).
Body mass index.
Participants’ weight and height at age 7 years were obtained on the metric system by trained clinical researchers. In the adult follow-up survey, participants reported their height and current weight; these data were converted from the imperial to the metric system (1 pound = 0.454 kg; 1 inch = 0.0254 m) and used to calculate body mass index (weight (kg)/height (m)2).
Statistical analysis
We first compared the participants who reported their birth weights (“reporters”) with those who did not provide a birth weight report (“nonreporters”) on sociodemographic factors using the chi-square test. We calculated the sensitivity or the proportion of participants whose self-reported and recorded birth weights placed them in the same birth weight category and the weighted kappa statistic to measure the chance-corrected agreement between self-reported and recorded birth weight. We examined the influence of sociodemographic variables on the validity of self-reported birth weight by comparing the sensitivity and weighted kappa for each stratum of the sociodemographic variables. Finally, we conducted linear regression analyses to compare the estimates of the associations of each measure of birth weight with childhood and adult body mass index, first in univariate models and then in models adjusting for sociodemographic variables. The percent difference between the estimates of associations of body mass index with the 2 methods of birth weight measurement was calculated to assess the direction and extent of bias.
RESULTS
Of the 262 participants, 15% (n = 39) did not report their birth weight. Nonreporters and reporters did not differ on average recorded birth weight (3,130.1 g vs. 3,139.2 g, respectively), but nonreporters were less likely than reporters to have a college or higher educational degree (31% vs. 46%) and to be white (15% vs. 29%); these and other differences between the 2 groups did not reach statistical significance.
Table 1 provides an assessment of the agreement between recorded and self-reported birth weight categories among the 223 women who reported a birth weight category. The overall sensitivity and weighted kappa were moderate to good with approximately 73% of participants correctly reporting their birth weight category (weighted kappa = 0.67, 95% confidence interval: 0.60, 0.74). Sensitivity varied across birth weight categories, being lowest for the lowest birth weight group of less than 5.5 pounds (58%) and highest for the weight groups 5.5–6.9 pounds (80%) and 8.5 pounds or more (81%). Among the 60 participants who misclassified their birth weight category, 95% (n = 57) were incorrect by 1 category, and 65% (n = 39) reported a heavier birth weight category than the recorded category.
Table 1.
Overall Agreement Between Birth Weight Data According to Adult Self-Report and Birth Records, New York Women's Birth Cohort Adult Follow-up, 2001–2006 (n = 223)
| Recoded Birth Weight, pounds | No. by Self-reported Birth Weight in Poundsa |
Total no. | Sensitivity, % | |||
| <5.5 | 5.5–6.9 | 7.0–8.4 | ≥8.5 | |||
| <5.5 | 14 | 9 | 1 | 0 | 24 | 58 |
| 5.5–6.9 | 4 | 70 | 13 | 1 | 88 | 80 |
| 7.0–8.4 | 1 | 13 | 66 | 15 | 95 | 69 |
| ≥8.5 | 0 | 0 | 3 | 13 | 16 | 81 |
| Total | 19 | 92 | 83 | 29 | 223 | 73 |
One pound = 0.454 kg.
As seen in Table 2, the greatest variations in the validity of self-reported birth weight were observed for maternal age at pregnancy, with accuracy improving with increasing maternal age (sensitivity of 65%, 72%, and 81% for maternal age at pregnancy of 16–20, 21–30, and 31–45 years, respectively; one-sided Ptrend = 0.03), and for childhood family income, with accuracy diminishing with increasing income levels (sensitivity of 79%, 75%, and 67% for the lowest, middle, and highest income levels, respectively; one-sided Ptrend = 0.05). White participants also provided a moderately more accurate birth weight category than did Hispanic and African-American participants (sensitivity of 79% for whites vs. 70% and 71% for African-American and Hispanic groups, respectively). The average maternal age at pregnancy was similar across racial groups (mean = 26.4 years (SD, 6.2) for African Americans, 25.6 years (SD, 5.7) for Hispanics, and 26.9 years (SD, 6.5) for whites; P = 0.42) but differed by childhood family income levels (mean = 28.6 years (SD, 6.2) for the lowest, 25.6 years (SD, 6.4) for the middle, and 24.9 years (SD, 5.0) for the highest income level; P = 0.001). Furthermore, the average recorded birth weight did not differ by maternal age (mean = 3,148.2 g for ages 16–20 years, 3,153.0 g for ages 21–30 years, and 3,099.5 g for ages 31–45 years; P = 0.80) or childhood family income (mean = 3,069.0 g, 3,156.4 g, and 3,178.8 g for the lowest, middle, and high income levels, respectively; P = 0.41), but it was significantly lower among African-American participants (2,980.0 g (SD, 493.3)) than among Hispanic (3,183.5 g (SD, 494.1)) and white (3,271.9 g (SD, 467.0)) participants (P = 0.001). There were a slight pattern toward decreasing accuracy by increasing maternal education and an improvement in accuracy with higher personal educational attainment, but the sample size for certain categories (e.g., maternal education, greater than high school) was small, and these differences were not statistically significant (P = 0.21 and 0.17, respectively).
Table 2.
Agreement Between Birth Weight Data According to Birth Records and Adult Self-Reports, by Sociodemographic Factors, New York Women's Birth Cohort Adult Follow-up, 2001–2006 (n = 223)
| Characteristic | No. of Participants | Sensitivity, % | Weighted Kappa | 95% Confidence Interval |
| Maternal age at pregnancy, years | ||||
| 16–20 | 40 | 65 | 0.59 | 0.42, 0.77 |
| 21–30 | 129 | 72 | 0.63 | 0.53, 0.74 |
| 31–45 | 54 | 81 | 0.79 | 0.67, 0.92 |
| Participant's age at adult survey, years | ||||
| 38–40 | 81 | 69 | 0.63 | 0.51, 0.75 |
| 41–42 | 89 | 74 | 0.69 | 0.58, 0.81 |
| 43–46 | 53 | 77 | 0.69 | 0.54, 0.85 |
| Participant's ethnicity | ||||
| White | 63 | 79 | 0.75 | 0.63, 0.86 |
| African American | 76 | 70 | 0.61 | 0.47, 0.75 |
| Hispanic | 84 | 71 | 0.64 | 0.51, 0.76 |
| Maternal education | ||||
| Less than high school | 100 | 75 | 0.70 | 0.59, 0.80 |
| High school | 89 | 72 | 0.65 | 0.53, 0.77 |
| Greater than high school | 31 | 68 | 0.52 | 0.28, 0.77 |
| Childhood family income | ||||
| Lowest | 63 | 79 | 0.74 | 0.61, 0.87 |
| Middle | 80 | 75 | 0.72 | 0.61, 0.84 |
| Highest | 69 | 67 | 0.55 | 0.41, 0.68 |
| Participant's education | ||||
| High school graduate or less | 33 | 67 | 0.57 | 0.37, 0.76 |
| Technical school graduate or some college | 88 | 73 | 0.68 | 0.56, 0.79 |
| College graduate or more | 102 | 75 | 0.70 | 0.59, 0.81 |
| Participant's current income | ||||
| Lowest | 76 | 74 | 0.68 | 0.55, 0.80 |
| Middle | 72 | 68 | 0.58 | 0.44, 0.72 |
| Highest | 71 | 77 | 0.72 | 0.60, 0.85 |
Childhood and adult body mass indexes were positively associated in a graded fashion with birth weight category by using either measure of birth weight, but the associations were statistically significant only for recorded birth weight and childhood body mass index (Tables 3 and 4; Figures 1 and 2). The estimates of the associations were smaller for self-reported birth weight compared with recorded birth weight, with the exception of estimates adjusted for adult family income, for which the estimates of associations relying on self-reported birth weight were slightly larger for birth weight categories of 5.5–6.9 pounds and 8.5 pounds or more relative to less than 5.5 pounds as compared with estimates using recorded birth weight. The degree of attenuation or percent of bias in the estimate of associations varied from 43% to 85% for childhood body mass index and from 9% to 56% for adult body mass index, and it was generally smaller for the highest birth weight category. Reporting error in birth weight attenuated the associations with childhood body mass index more than with adult body mass index, with an average percent bias of 60% for multivariate associations of birth weight with childhood body mass index as compared with 38% for adult body mass index.
Table 3.
Difference in Mean Childhood Body Mass Index According to Recorded and Self-reported Birth Weight by Using Linear Regression, New York Women's Birth Cohort Adult Follow-up, 2001–2006 (n = 221)
| Recorded Birth Weight |
Self-reported Birth Weight |
Bias, % | |||
| Coefficienta | 95% Confidence Interval | Coefficienta | 95% Confidence Interval | ||
| Unadjusted | |||||
| <5.5 poundsb | Referent | Referent | |||
| 5.5–6.9 pounds | 1.22 | 0.14, 2.30 | 0.21 | −0.99, 1.41 | −82.8 |
| 7.0–8.4 pounds | 1.60 | 0.52, 2.67 | 0.78 | −0.43, 2.00 | −51.3 |
| ≥8.5 pounds | 2.51 | 1.00, 4.03 | 1.40 | 0.00, 2.80 | −44.2 |
| Adjusted for maternal age at pregnancy | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 1.29 | 0.19, 2.39 | 0.19 | −1.01, 1.39 | −85.3 |
| 7.0–8.4 pounds | 1.67 | 0.59, 2.76 | 0.80 | −0.41, 2.02 | −52.1 |
| ≥8.5 pounds | 2.49 | 0.97, 4.00 | 1.41 | 0.01, 2.82 | −43.4 |
| Adjusted for participant's age at recall | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 1.27 | 0.18, 2.35 | 0.20 | −1.00, 1.40 | −84.3 |
| 7.0–8.4 pounds | 1.60 | 0.53, 2.67 | 0.77 | −0.44, 1.97 | −51.9 |
| ≥8.5 pounds | 2.56 | 1.04, 4.07 | 1.36 | −0.04, 2.76 | −46.9 |
| Adjusted for participant's ethnicity | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 1.09 | 0.00, 2.19 | 0.20 | −1.01, 1.41 | −81.7 |
| 7.0–8.4 pounds | 1.39 | 0.29, 2.48 | 0.61 | −0.61, 1.84 | −56.1 |
| ≥8.5 pounds | 2.31 | 0.76, 3.86 | 1.23 | −0.20, 2.65 | −46.8 |
| Adjusted for maternal education | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 1.23 | 0.13, 2.33 | 0.22 | −0.99, 1.43 | −82.1 |
| 7.0–8.4 pounds | 1.55 | 0.45, 2.66 | 0.76 | −0.47, 1.99 | −51.0 |
| ≥8.5 pounds | 2.55 | 0.98, 4.12 | 1.36 | −0.07, 2.80 | −46.7 |
| Adjusted for childhood family income | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 1.39 | 0.24, 2.53 | 0.29 | −0.93, 1.52 | −79.1 |
| 7.0–8.4 pounds | 1.69 | 0.57, 2.81 | 0.85 | −0.38, 2.08 | −49.7 |
| ≥8.5 pounds | 2.76 | 1.14, 4.38 | 1.54 | 0.09, 3.00 | −44.2 |
| Fully adjustedc | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 1.42 | 0.23, 2.61 | 0.28 | −0.96, 1.52 | −80.3 |
| 7.0–8.4 pounds | 1.57 | 0.40, 2.74 | 0.74 | −0.51, 1.99 | −52.9 |
| ≥8.5 pounds | 2.47 | 0.77, 4.17 | 1.21 | −0.29, 2.71 | −51.0 |
The coefficient represents the mean difference in body mass index (kg/m2) at age 7 years.
One pound = 0.454 kg.
Adjusted for maternal age at pregnancy, participant's age at recall of birth weight, participant's ethnicity, maternal education, and childhood family income.
Table 4.
Difference in Mean Adult Body Mass Index According to Recorded and Self-reported Birth Weight by Using Linear Regression, New York Women's Birth Cohort Adult Follow-up, 2001–2006 (n = 218)
| Recorded Birth Weight |
Self-reported Birth Weight |
Bias, % | |||
| Coefficienta | 95% Confidence Interval | Coefficienta | 95% Confidence Interval | ||
| Unadjusted | |||||
| <5.5 poundsb | Referent | Referent | |||
| 5.5–6.9 pounds | 0.84 | −2.10, 3.79 | 0.55 | −2.64, 3.74 | −34.5 |
| 7.0–8.4 pounds | 1.21 | −1.72, 4.14 | 0.71 | −2.51, 3.93 | −41.3 |
| ≥8.5 pounds | 1.11 | −2.95, 5.16 | 0.95 | −2.76, 4.66 | −14.4 |
| Adjusted for maternal age at pregnancy | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 0.97 | −2.02, 3.96 | 0.53 | −2.67, 3.74 | −45.4 |
| 7.0–8.4 pounds | 1.35 | −1.62, 4.31 | 0.73 | −2.50, 3.97 | −45.9 |
| ≥8.5 pounds | 1.06 | −3.01, 5.13 | 0.96 | −2.76, 4.68 | −9.4 |
| Adjusted for participant's age at recall | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 0.64 | −2.32, 3.61 | 0.40 | −2.80, 3.61 | −37.5 |
| 7.0–8.4 pounds | 1.13 | −1.81, 4.06 | 0.61 | −2.62, 3.84 | −46.0 |
| ≥8.5 pounds | 0.87 | −3.21, 4.94 | 0.78 | −2.95, 4.51 | −10.3 |
| Adjusted for participant's ethnicity | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 1.56 | −1.35, 4.47 | 0.95 | −2.22, 4.11 | −39.1 |
| 7.0–8.4 pounds | 2.15 | −0.77, 5.07 | 1.56 | −1.63, 4.76 | −27.4 |
| ≥8.5 pounds | 2.47 | −1.58, 6.52 | 1.98 | −1.72, 5.67 | −19.8 |
| Adjusted for maternal education | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 1.31 | −1.65, 4.27 | 0.80 | −2.39, 3.99 | −38.9 |
| 7.0–8.4 pounds | 1.85 | −1.12, 4.81 | 1.14 | −2.09, 4.37 | −38.4 |
| ≥8.5 pounds | 1.71 | −2.43, 5.85 | 1.37 | −2.37, 5.11 | −19.9 |
| Adjusted for childhood family income | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 0.97 | −2.09, 4.03 | 0.61 | −2.56, 3.79 | −37.1 |
| 7.0–8.4 pounds | 1.02 | −1.98, 4.03 | 0.72 | −2.47, 3.91 | −29.4 |
| ≥8.5 pounds | 1.93 | −2.31, 6.17 | 0.85 | −2.90, 4.60 | −56.0 |
| Adjusted for participant's education | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 0.87 | −2.10, 3.82 | 0.71 | −2.53, 3.94 | −18.4 |
| 7.0–8.4 pounds | 1.20 | −1.74, 4.13 | 0.77 | −2.47, 4.00 | −35.8 |
| ≥8.5 pounds | 1.17 | −2.93, 5.27 | 1.06 | −2.68, 4.79 | −9.4 |
| Adjusted for participant's income | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 0.63 | −2.44, 3.70 | 0.69 | −2.58, 3.95 | 9.5 |
| 7.0–8.4 pounds | 1.15 | −1.91, 4.22 | 0.89 | −2.39, 4.18 | −22.6 |
| ≥8.5 pounds | 0.73 | −3.67, 5.12 | 0.94 | −2.90, 4.78 | 30.6 |
| Fully adjustedc | |||||
| <5.5 pounds | Referent | Referent | |||
| 5.5–6.9 pounds | 2.14 | −1.02, 5.30 | 1.07 | −2.17, 4.31 | −50.0 |
| 7.0–8.4 pounds | 2.71 | −0.45, 5.87 | 2.04 | −1.20, 5.29 | −24.7 |
| ≥8.5 pounds | 3.30 | −1.40, 8.00 | 2.01 | −1.93, 5.95 | −39.1 |
The coefficient represents the mean difference in current adult body mass index (kg/m2).
One pound = 0.454 kg.
Adjusted for maternal age at pregnancy, participant's age at recall of birth weight, participant's ethnicity, maternal education, childhood family income, participant's education, and current family income.
Figure 1.
Adjusted differences in mean childhood body mass index according to birth weight from birth records (A) and self-reported birth weight (B), from linear regression models adjusting for maternal age at pregnancy, participant's age at birth weight recall, participant's ethnicity, maternal education, and childhood family income, New York Women's Birth Cohort Adult Follow-up, 2001–2006. Filled squares, coefficient; error bars, 95% confidence interval. BMI, body mass index. One pound = 0.454 kg.
Figure 2.
Adjusted differences in mean adult body mass index according to birth weight from birth records (A) and adult self-reported birth weight (B), from linear regression models adjusting for maternal age at pregnancy, participant's age at birth weight recall, participant's ethnicity, maternal education, childhood family income, participant's education, and participant's household income, New York Women's Birth Cohort Adult Follow-up, 2001–2006. Filled squares, coefficient; error bars, 95% confidence interval. BMI, body mass index. One pound = 0.454 kg.
DISCUSSION
In this study, the majority of women in their middle adult years were able to recall their birth weight category with moderate to good accuracy. Our results are comparable to the reporting rate and accuracy reported by Troy et al. (13) using data from the US Nurses’ Health Study II cohort (13) and the accuracy found in a case-control study of breast cancer (14), but they are higher than those reported in several other validation studies of adult self-reported birth weight, with reporting rates ranging from 24% to 53% and validity statistics falling in poor to moderate levels (9–11). One reason for the higher reporting and accuracy in our study as compared with other studies may be related to the question format and the level of details of self-reported birth weight. Rather than asking participants to report their birth weight in an open-ended format, participants in this study, as in the Nurses’ Health Study II cohort, were asked to choose 1 of 5 birth weight categories. This question format required participants to only approximate their birth weight. Additionally, birth weight categories of approximately 24 ounces (∼680 g) in our study and the Nurses’ Health Study II were larger than the definitions of accuracy used in studies with continuous self-reported birth weight data, which ranged from approximately less than 1 ounce to less than 18 ounces. No consensus on the definition of an accurate birth weight report exists, but less stringent definitions have been shown to increase the accuracy of self-reported birth weight. For example, among middle-aged and elderly women from the United Kingdom who reported a birth weight, sensitivity increased from 48% to 89% when the definition of an accurate report was expanded from ±4 ounces to ±16 ounces of the original birth weight record (11). Finally, as compared with several other validation studies, participants in our study and the Nurses’ Health Study II by Troy et al. were relatively younger (27–46 years vs. 43–78 years) and of later birth cohorts (1946–1964 vs. 1908–1957) (10, 11, 13). Both younger age at recall and later birth cohort have been shown to improve the validity of adult self-reported birth weight (10, 11).
Maternal age at the time of pregnancy and childhood family income were the most important sociodemographic factors related to reporting accuracy, with improvement in reporting accuracy among participants whose mothers were older at pregnancy and had lower family income at birth. These 2 early life factors were also related, with older mothers being of lower socioeconomic status. This finding suggests that birth weight information is likely to be communicated to offspring by their mothers, and factors that influence the validity of maternal reporting of their offspring's birth weight (e.g., pregnancy-related characteristics, maternal socioeconomic status) (18) may indirectly affect the offspring's self-report of their birth weight. Further support for this conclusion is provided by studies that have found birth weight reports by mothers and daughters to be highly correlated (14), and daughters with living mothers to be more likely than those with deceased mothers to recall and report their birth weight (11).
We observed relatively small differential misreporting of birth weight by adult socioeconomic status, as measured by the participant's own education and family income close to the time of adult birth weight self-reports. Two other studies also failed to find a significant impact of adult socioeconomic status, measured by participants’ own and/or spouse's occupational class, on the accuracy of adult self-reported birth weight (10, 11). Our findings of relatively less accurate reporting among African-American and Hispanic women as compared with white women are also in agreement with the only other validation study using a multiethnic cohort (13). Finally, the lowest reporting accuracy was found among low birth weight participants with recorded birth weight of less than 5.5 pounds (<2,500 g), a finding that is consistent with some reports of less accurate maternal report of birth weight for their low birth weight infants, as compared with normal weight infants (19, 20).
We illustrated the impact of the reporting error in self-reported birth weight by comparing the estimates of associations of recorded and self-reported birth weight with childhood and adult body mass index. Despite the moderate to good level of agreement, the results demonstrated similar patterns of increasing body mass index with higher categories of birth weight, but considerably weaker associations with self-reported birth weight than recorded birth weight. In the case of childhood body mass index, recorded birth weight was significantly associated with childhood body mass index, but the same association was no longer statistically significant when relying on self-reported birth weight. The main implication for investigating the associations between early life factors and health outcomes in later life periods is that even a relatively small amount of reporting error can lead to null associations when these associations do exist. This is particularly of concern in life-course studies examining the associations of early life factors with health in middle to late adulthood, the majority of which are of weak to modest magnitude. For example, the relative risk of breast cancer for high versus low birth weight groups from a meta-analysis of 16 studies with both pre- and postmenopausal women was only 1.23 (95% confidence interval: 1.13, 1.34) with the magnitude of associations generally larger in studies using birth records than in studies using self- or parental reported birth weight (21). Given the extent of reporting error in self-reported birth weight found in our study, it is reasonable to expect that other harder-to-recall early life factors (e.g., certain maternal or pregnancy-related conditions, postnatal growth measures) would be reported with even greater error.
Strengths and limitations
In previous studies, the representativeness of the validation samples as compared with the initial or birth weight self-report sample was diminished because of the unavailability of original birth weight records and/or low response rate of self-reported birth weight (10). We had original birth weight records for all participants in the adult follow-up study, who did not differ from the nonparticipants from the initial cohort on birth weight or any other early childhood anthropometric measures (15). Additionally, the high response rate and lack of significant differences in birth weight between reporters and nonreporters of birth weight data in our study lower the probability that nonresponse might have led to bias in analyses involving self-reported birth weight. Nonetheless, we cannot rule out the possibility that differences in other characteristics between nonparticipants and participants in our adult follow-up study have influenced our results. For example, because both participation and reporting accuracy were associated with white race/ethnicity, it is possible that our results underestimate the extent of reporting error in self-reported birth weight data. The ethnic diversity and availability of data on multiple childhood and adult socioeconomic characteristics allowed for the investigation of accuracy of self-reported birth weight by these factors. In addition, we consider the influence of reporting error on the estimates of association for a health outcome measured at 2 life periods.
Comparison of our results with those from other validation studies suggests that reducing the level of detail required in self-reports of birth weight through categorization may improve both reporting rate and reporting validity. However, this may be viewed as a limitation and should be balanced against the disadvantages of birth weight categorization, such as reducing the statistical power for detection of associations and obscuring any differences within categories of birth weight.
Conclusion
Birth weight in middle adult life is reported with error, even when categories of birth weight are reported. The reporting error in birth weight showed variation by common sociodemographic factors, as well as by the birth weight itself. Inaccuracy in birth weight reporting more often occurred in the direction of exaggeration of the birth weight and weakened the positive associations between birth weight and childhood and adult body mass index. Adult self-reported birth weight should be used with caution in life-course studies.
Acknowledgments
Author affiliation: Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, New York (Parisa Tehranifar, Yuyan Liao, Julie D. Flom, Mary Beth Terry).
This research was supported by the Department of Defense Breast Cancer Research Program (grant DAMD 170210357) and the National Cancer Institute (grant KO7CA90685). Dr. Tehranifar was supported by a cancer training program from the US National Cancer Institute (NCI R25-CA94061).
The authors thank the following individuals for their contribution to the New York Women's Birth Cohort Adult Follow-up Study: Dr. Ezra Susser, Jennifer S. Ferris, Tara Kalra, Tamarra James, Lina Titievsky-Konikov, Dipal Shah, Shobana Ramachandran, Julia Meurling, Adey Tsega, Sujata Narayanan, and Summer Wright.
Conflict of interest: none declared.
Glossary
Abbreviation
- SD
standard deviation
References
- 1.Barker DJ, Gluckman PD, Godfrey KM, et al. Fetal nutrition and cardiovascular disease in adult life. Lancet. 1993;341(8850):938–941. doi: 10.1016/0140-6736(93)91224-a. [DOI] [PubMed] [Google Scholar]
- 2.Law CM, de Swiet M, Osmond C, et al. Initiation of hypertension in utero and its amplification throughout life. BMJ. 1993;306(6869):24–27. doi: 10.1136/bmj.306.6869.24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Trichopoulos D, Adami HO, Ekbom A, et al. Early life events and conditions and breast cancer risk: from epidemiology to etiology. Int J Cancer. 2008;122(3):481–485. doi: 10.1002/ijc.23303. [DOI] [PubMed] [Google Scholar]
- 4.Gluckman PD, Hanson MA, Cooper C, et al. Effect of in utero and early-life conditions on adult health and disease. N Engl J Med. 2008;359(1):61–73. doi: 10.1056/NEJMra0708473. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Russell C, Palmer JR, Adams-Campbell LL, et al. Follow-up of a large cohort of black women. Am J Epidemiol. 2001;154(9):845–853. doi: 10.1093/aje/154.9.845. [DOI] [PubMed] [Google Scholar]
- 6.Psaty BM, Cheadle A, Koepsell TD, et al. Race- and ethnicity-specific characteristics of participants lost to follow-up in a telephone cohort. Am J Epidemiol. 1994;140(2):161–171. doi: 10.1093/oxfordjournals.aje.a117226. [DOI] [PubMed] [Google Scholar]
- 7.Klebanoff MA, Zemel BS, Buka S, et al. Long-term follow-up of participants in the Collaborative Perinatal Project: tracking the next generation. Paediatr Perinat Epidemiol. 1998;12(3):334–346. doi: 10.1046/j.1365-3016.1998.00125.x. [DOI] [PubMed] [Google Scholar]
- 8.Madans JH, Kleinman JC, Cox CS, et al. 10 years after NHANES I: report of initial followup, 1982–1984. Public Health Rep. 1986;101(5):465–473. [PMC free article] [PubMed] [Google Scholar]
- 9.Little RE. Birthweight and gestational age: mothers’ estimates compared with state and hospital records. Am J Public Health. 1986;76(11):1350–1351. doi: 10.2105/ajph.76.11.1350. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Andersson SW, Niklasson A, Lapidus L, et al. Poor agreement between self-reported birth weight and birth weight from original records in adult women. Am J Epidemiol. 2000;152(7):609–616. doi: 10.1093/aje/152.7.609. [DOI] [PubMed] [Google Scholar]
- 11.Allen DS, Ellison GT, dos Santos Silva I, et al. Determinants of the availability and accuracy of self-reported birth weight in middle-aged and elderly women. Am J Epidemiol. 2002;155(4):379–384. doi: 10.1093/aje/155.4.379. [DOI] [PubMed] [Google Scholar]
- 12.Kemp M, Gunnell D, Maynard M, et al. How accurate is self-reported birth weight among the elderly? J Epidemiol Community Health. 2000;54(8):639–640. doi: 10.1136/jech.54.8.639. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Troy LM, Michels KB, Hunter DJ, et al. Self-reported birth weight and history of having been breastfed among younger women: an assessment of validity. Int J Epidemiol. 1996;25(1):122–127. doi: 10.1093/ije/25.1.122. [DOI] [PubMed] [Google Scholar]
- 14.Sanderson M, Williams MA, White E, et al. Validity and reliability of subject and mother reporting of perinatal factors. Am J Epidemiol. 1998;147(2):136–140. doi: 10.1093/oxfordjournals.aje.a009425. [DOI] [PubMed] [Google Scholar]
- 15.Terry MB, Flom J, Tehranifar P, et al. The role of birth cohorts in studying adult health: the New York Women's Birth Cohort. Paediatr Perinat Epidemiol. 2009;23(5):431–445. doi: 10.1111/j.1365-3016.2009.01061.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Terry MB, Wei Y, Esserman D. Maternal, birth, and early-life influences on adult body size in women. Am J Epidemiol. 2007;166(1):5–13. doi: 10.1093/aje/kwm094. [DOI] [PubMed] [Google Scholar]
- 17.Terry MB, Ferris JS, Tehranifar P, et al. Birth weight, postnatal growth, and age at menarche. Am J Epidemiol. 2009;170(1):72–79. doi: 10.1093/aje/kwp095. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Olson JE, Shu XO, Rose JA, et al. Medical record validation of maternally reported birth characteristics and pregnancy-related events: a report from the Children's Cancer Group. Am J Epidemiol. 1997;145(1):58–67. doi: 10.1093/oxfordjournals.aje.a009032. [DOI] [PubMed] [Google Scholar]
- 19.Walton KA, Murray LJ, Gallagher AM, et al. Parental recall of birthweight: a good proxy for recorded birthweight? Eur J Epidemiol. 2000;16(9):793–796. doi: 10.1023/a:1007625030509. [DOI] [PubMed] [Google Scholar]
- 20.Gofin R, Neumark YD, Adler B. Birthweight recall by mothers of Israeli children. Public Health. 2000;114(3):161–163. [PubMed] [Google Scholar]
- 21.Michels KB, Xue F. Role of birthweight in the etiology of breast cancer. Int J Cancer. 2006;119(9):2007–2025. doi: 10.1002/ijc.22004. [DOI] [PubMed] [Google Scholar]