Of the 5,683 children and adolescents in NHANES III who were 6–16 years of age, 4,619 (81.3%) completed at least one test of cognitive abilities and had available serum cotinine values. Children with serum cotinine levels > 15 ng/mL (n = 155) or who reported active smoking during the 5 days before testing (n = 65) were excluded from this analysis. This yielded a final sample of 4,399 children (77.4% of all 6- to 16-year-olds) in the primary analysis. Children who were excluded from the final sample had lower scores on tests of math (p < 0.001), reading (p = 0.003), and block design (p = 0.02) than did children in the analysis. Excluded children were also more likely to live in households with lower marriage rates (p = 0.02). Because of our eligibility criteria, excluded children also had significantly higher levels of serum cotinine (p < 0.001).
Serum cotinine was detectable in 84% of children in the final sample, with 16% having cotinine levels below the limit of detection (< 0.05 ng/mL). After imputing randomly selected values from the left tail of a log-normal distribution, the geometric mean serum cotinine level for the sample was 0.23 ng/mL (SE = 0.01). Serum cotinine concentrations varied by children’s characteristics (). Serum cotinine concentrations were significantly higher among African Americans than Hispanics or non-Hispanic whites, among children of parents with a lower household income or lower educational achievement, among children living in the Midwest United States, and among those with higher blood lead concentrations. Children exposed to both prenatal and postnatal smoke and those exposed to postnatal smoke alone also had higher serum cotinine levels.
Mean serum cotinine concentrations and cognitive test scores for children and adolescents, 6–16 years of age, NHANES III (1988–1994).
Consistent with previous research (Jordaan et al. 1999
), mean serum cotinine levels were significantly higher among children who had at least one smoker living in their home (p
< 0.001). Children’s serum cotinine levels also increased as the number of smokers in a household increased (p
< 0.001) and as the number of cigarette packs smoked per day in a household increased (p
Overall mean scores for math, reading, block design, and digit span are presented in . In unadjusted analyses, cognitive performance scores differed significantly by sex, race or ethnicity, poverty status, parent marital status and educational level, and blood lead concentration. There was also a significant inverse relationship between serum cotinine and cognitive test scores. Children with the highest serum cotinine levels received significantly lower performance scores on all four tests than did children in the lowest cotinine level.
In multiple regression analyses using the log-transformation of cotinine and adjusting for covariates, serum cotinine was significantly associated with lower scores for reading, math, and visuospatial skills (). An increase in the log serum cotinine from level 1 to 10 ng/mL was associated with a 1.93-point loss in math scores (p ≤0.001) and a 2.7-point loss in reading scores (p ≤0.001) for tests with a standardized mean of 100. The same change in the log serum cotinine level was associated with a 0.55-point loss in block design scores (p ≤0.001) and a 0.08-point loss in digit span scores for tests with a standardized mean of 10.
Log-linear effect of cotinine ≤15 ng/mL [β-coefficient (SE β)] and potential covariates on cognitive test scores at 6- to 16 years of age, NHANES III (1988–1994).
There was a significant inverse relationship between the log of serum cotinine and cognitive abilities at lower levels of exposure. Children with serum cotinine levels < 0.1 ng/mL had an adjusted average reading score of 94.7. Children with cotinine levels between 0.1 and 1 ng/mL had an average 2.6-point drop in reading scores, children with levels 1–3 ng/mL had an additional 0.2-point drop in reading, and children with cotinine values > 3 ng/mL had an additional 4.8-point drop in reading score. Although the greatest decrease in reading scores was observed among children with higher cotinine levels (range, 3–15 ng/mL), there was a greater proportional change in reading scores per unit of cotinine exposure at levels in the range of 0.1–1 ng/mL. According to population estimates employing the appropriate sampling weights, we estimated that > 33.3 million children are at risk for ETS-related reading deficits (i.e., children with cotinine levels ≤15 ng/mL). Math and block design scores showed similar trends in decreasing cognitive scores with increasing cotinine levels ().
Adjusted mean cognitive test scores (mean ± SE) at increasing log cotinine levels for children 6–16 years of age, controlling for potential covariates, NHANES III (1988–1994).
Questionnaire data on maternal smoking during pregnancy, birth weight, and NICU stay were available for a subsample of 2,738 children 6–11 years of age. Among this sub-sample of children, the covariate-adjusted relationship between the log of serum cotinine and cognitive scores indicates that an increase in the log serum cotinine from 1 to 10 ng/mL was associated with a related 2.4-point decrease in reading scores (p ≤0.05), a 1.6-point decrease in math scores, and a 0.42-point decrease in block design scores (p ≤0.05). In secondary analyses, inclusion of prenatal tobacco smoke exposure, birth weight, and NICU stay had little effect on the relationship between ETS exposure and reading scores (p ≤0.05). In contrast, the association of ETS exposure with block design was attenuated ().
Adjusted slope of log cotinine [β-coefficient (SE β)] on cognitive test scores between the full sample of 6–16-year-olds and subsample of 6–11-year-olds.
The ETS-associated decrements in reading scores appeared to be greater at lower levels of serum cotinine (). Math and block design also showed a steeper decline in scores at lower cotinine values than at higher values. To test whether the difference in the slopes was statistically significant, we conducted a stratified adjusted analysis including linear (non-transformed) cotinine values above and below 1 ng/mL (~80th percentile). Children with cotinine values ≤1 ng/mL (< 80th percentile) had an average 5.0-point decrease in reading scores for each 1-ng/mL increase in serum cotinine compared with an average 0.8-point decrease in reading scores for children with cotinine values above 1 ng/mL (> 80th percentile) (). The difference between these averages was statistically significant (t = 2.38, p = 0.02).
Log-linear regression line for reading scores by serum cotinine levels. Dashed lines indicate 95% confidence interval.
Log-linear model for cotinine (solid line) versus linear models for cotinine among children with cotinine above and below 1 ng/mL (dashed lines; ~ 80th percentile).