Decline in the negative association between low birth weight and cognitive ability

Data.

The 1958 NCDS is a longitudinal cohort study that followed the lives of around 17,500 individuals born during a single week of March in 1958 in Britain. We used data from the birth survey (98.8% response rate) and from the age at 11 y survey (response rate around 88%). The 1970 BCS is a longitudinal cohort study that followed the lives of around 17,000 individuals born in Britain during a single week of April in 1970. We used data from the birth survey (95.9% response rate) and the age at 10 y survey (response rate around 87%). The MCS is a longitudinal cohort study that followed the lives of around 19,000 children born in the United Kingdom from September in 2000 to January in 2002. The sample was selected from a random sample of electoral wards using a stratified sampling strategy to ensure the adequate representation of all four of the countries comprising the United Kingdom and of disadvantaged and ethnically diverse areas. In the analyses, we used data from sweep 1 (82% response rate), which were collected when the children were around 9 mo old, and from sweep 5, which were collected when the children were around 11 y old (response rate around 72%). For ease of exposition, we refer to the MCS as the 2001 cohort study, because the majority of the births in the sample occurred in 2001.

When the birth surveys of the 1970 BCS and 1958 NCDS cohort studies were carried out, consent to participate in surveys was gained by respondents agreeing to be interviewed or respondents returning the completed questionnaire to the study team. Parental consent was sought when the data at 11 y (1958 NCDS) and at 10 y (1970 BCS) were collected, but there is no evidence that written consent was obtained. Available records suggest that for the waves of data collection considered in this study, there was only internal ethnical review for the 1958 and 1970 cohort studies because they predated the establishment of the Multicentre Research Ethics Committee in 1997. The 2001 MCS sought written parental consent in each wave of data collection, and Multicentre Research Ethics Committee approval was given for all of the MCS waves of data collection.

Birth Weight.

In the NCDS and BCS, birth weight was recorded in the birth survey, which was completed by a midwife who attended the delivery. In the MCS, birth weight was reported by the main respondent in sweep 1 (when the cohort child was around 9 mo old), and evidence suggests that such reports are reliable and in accordance with registration data (18). Birth weight was divided into two categories: low (<2,500 g) and normal (2,500–4,500 g) (19). Heavy-weight (4,500 g and above) children (20) were excluded from the analyses because differences in cognitive outcomes between heavy-weight and normal-weight children in all cohorts were not statistically significant. In supplementary analyses, we divided the low-birth-weight group into two categories: very low birth weight (VLBW; <1,500 g) and moderately low birth weight (MLBW; 1,500 to <2,500 g).

Cognitive Ability.

We used a measure of verbal cognitive ability in childhood. One of the strengths of this study is that we relied on measures of cognitive ability that are comparable across the three cohorts because they were collected at around the same age in each cohort (age of 11 y in the 1958 NCDS and 2001 MCS and age of 10 y in the 1970 BCS); they tested the same dimension of cognitive well-being, namely, verbal ability; and they were found to serve as a good proxy for general cognitive ability scores (21⇓⇓–24). In the 1958 NCDS, cognition was assessed based on the verbal score of the General Ability Test (National Foundation for Educational Research), which consisted of 40 items (23). The children were tested individually by teachers, who recorded the answers for the tests. In the 1970 BCS, the cognition of each child was assessed using the 21-item Word Similarity subscale of the British Ability Scales (first edition), which was administered by a teacher (24). The scores were standardized from a normed pool of scores within 3-mo age ranges. In the 2001 MCS, the cognition of each child was assessed using the 37-item Verbal Similarity subscale from the British Ability Scales (second edition), which was administered by the interviewer (25). The scores were standardized from a normed pool of scores within 3-mo age ranges.

Because different tests were administered in each cohort, regression results based on the raw test scores would not be directly comparable across cohorts. We therefore standardized the cognitive scores to a mean of 0 and a SD of 1 within each cohort, which means that children were ranked within their respective cohorts. The standardization procedure ensured that the regression coefficients on the association between low birth weight and cognitive ability, obtained by estimating separate models within each cohort, could be compared across cohorts.

Family Characteristics.

We selected the maternal and the family characteristics that were hypothesized to attenuate the association between birth weight and cognition in childhood (19, 26), and that were collected in all three cohort studies. To capture the socioeconomic status of the family, we adjusted for the mother’s level of education (in the 1958 NCDS and the 1970 BCS, whether the mother stayed in school beyond the minimum age; in the 2001 MCS, whether the mother had a degree-level education) and for either the father’s (1958 NCDS) or the household’s (1970 BCS and 2001 MCS) registrar general social class. As sociodemographic variables, we considered the age of the mother at the birth of the cohort child and her marital status at conception or at the time of birth (in the 1958 NCDS and the 1970 BCS, whether she was or was not married; in the 2001 MCS, whether she was married, cohabiting, or single). We considered three markers of the mother’s health behavior: whether she smoked during pregnancy, whether she used antenatal care for the first time after 12 wk of pregnancy or did not use it at all, and whether she breastfed the child. Finally, we adjusted for the mother’s height, which is a marker of both socioeconomic status and health (27). Because the MCS provides a richer set of variables, we also showed in this cohort study how the results change after we controlled for the mother’s heavy drinking during pregnancy, for the quintile of household income, and for the mother’s ethnicity.

Inclusion Criteria and Exclusions.

We dropped observations with missing values on cognitive scores at the age of 10 or 11 y (1958 NCDS = 1,226, 1970 BCS = 2,267, 2001 MCS = 297). From this subsample, we dropped observations with missing values on birth weight (1958 NCDS = 1,128, 1970 BCS = 965, 2001 MCS = 488) and observations with missing values on any of the covariates measured around the time of the cohort child’s birth. We also excluded multiple births from the analytical samples. These exclusions reduced the 1958 NCDS sample to 10,341 observations, the 1970 BCS sample to 9,116 observations, and the 2001 MCS sample to 11,231 observations.

Statistical Analyses.

We used linear regression models to estimate, within each cohort, the association between the cohort children’s birth weight and verbal cognitive ability at the age of 10 or 11 y. Two sets of models were estimated for each cohort study. The baseline model included an adjustment for the basic characteristics of each child; namely, the child’s sex, and whether the child was a first or higher order birth. Model 1 included adjustments for the basic characteristics of the child, of the mother, and of the family. For the 2001 cohort, model 2 was adjusted for additional covariates that were not collected in the 1958 and 1970 cohort studies. To test directly whether the association between birth weight and cognition varied across cohorts, we pooled the three cohort studies and estimated a unique model in which we interacted each regressor by means of a cohort dummy.

All of the analyses were conducted in Stata, version 13. In the analyses of the 2001 MCS, we used weights to account for nonresponse and for the overrepresentation of disadvantaged and ethnically diverse areas and the survey command to account for the clustering of samples within strata. No weighting was used in the analyses of the 1958 NCDS and the 1970 BCS because these surveys were not based on a complex design and weights for nonresponse were not available.