We analyzed data from the NHANES (2003–2008), which uses a stratified multistage probability design to obtain a nationally representative sample of the civilian, noninstitutionalized US population and collects data from survey participants via household interviews and physical examinations. Detailed information about NHANES procedures is available elsewhere.²² For our analyses, we used data collected from participants aged 8 to 18 years during 2003–2004, 2005–2006, and 2007–2008. Of 7199 potential study subjects, we excluded 72 who were pregnant, 432 whose data did not meet the minimum dietary recall data quality standards (http://www.cdc.gov/nchs/nhanes/nhanes2003-2004/diet03_04.htm), 289 with missing BP data or height or weight measurements, and 171 who were under-weight (<5th percentile in BMI), leaving 6235 children for analyses.

All subjects provided at least one 24-hour dietary recall at the mobile examination center, and 91% also provided a second recall via telephone interview 3 to 10 days later. NHANES estimated participants’ nutrient intake from foods by using the US Department of Agriculture Food and Nutrient Databases for Dietary Studies for each 2-year NHANES cycle.²³

Dietary data from a single 24-hour recall may not represent a respondent’s usual dietary intake because of day-to-day variations in diet, and use of such data may bias estimates of the association between nutrient intake and health outcomes due to the measurement errors.^24,25 To limit such possible bias, we used a 2-step method developed by the National Cancer Institute (NCI) to estimate participants’ usual sodium intake.²⁶ Because sodium was consumed daily by nearly every participant, we used only the amount part of the model in the first step. The second step involves calculating the individual’s estimated usual intake with parameters from the first step.²⁷ The NCI method requires that some of the participants have multiple days of nutrient values to estimate the within- and between-individual variations.^26,27 The models for estimating usual intake included the following covariates: an indicator of first- versus second-day recall, day of the week recall (weekday vs weekends), race/ethnicity (non-Hispanic white, non-Hispanic black, Mexican American, and others), and age groups (8–12, 13–15, and ≥16 years).²⁸ We used the similar approach to estimate the usual potassium and total energy intakes and estimated these usual intakes for male and female subjects separately.

NHANES measured participants’ BP up to 3 times during their visits to the mobile examination center. All BP measurements were taken with a sphygmomanometer by certified examiners with a proper cuff size after children rested quietly in a sitting position for 5 minutes.²⁹ The average of 3 (78%), 2 (11%), and 1 (11%) BP measurements per individual were used in the study. The guidelines recommend multiple BP measurements at different times to define persistent prehypertension and hypertension in children and adolescents.³⁰ To differentiate the BP measurements in the current study (up to 3 measurements in a single mobile examination center visit) from the recommended definitions, we used the terms of pre-high BP (pre-HBP) and HBP with the same cutoff points that are used to define prehypertension and hypertension.^30,31 We classified children’s (8–17 years) BP as normal, pre-HBP, or HBP on the basis of age-, gender-, and height-specific BP percentiles derived from the reference population for the 2000 Centers for Disease Control and Prevention growth charts.³⁰ Normal BP is defined as systolic blood pressure (SBP) and diastolic blood pressure (DBP) below 90th percentile for their age, gender, and height; pre-HBP as an average SBP or DBP ≥90th but <95th percentile or with observed BP levels ≥120/80 mm Hg; and HBP as an SBP or DBP ≥95th percentile. For adolescents aged 18 years, pre-HBP is defined as either 120<SBP<140 mm Hg or 80 < DBP<90 mm Hg and HBP as either the SBP ≥140 mm Hg, the DBP ≥90 mm Hg, or was taking antihypertensive medication.³² To produce more stable estimates of the association, we combined subjects classified as having pre-HBP or HBP into 1 group, which we defined as having pre-HBP/HBP.

Our baseline covariates include age, gender, race-ethnicity, table salt use (never/rarely, sometimes, or often), and BMI calculated as kilograms of weight divided by meters of height squared. To account for variability by age and gender, BMI in children is compared with age- and gender-specific reference values from the 2000 Centers for Disease Control and Prevention growth charts to define the weight status.³³ We defined overweight as a BMI-for-age/gender between the 85th and 95th percentiles and obesity as a BMI-for-age/gender ≥95th percentile.³⁴ In our analyses, however, we collapsed the overweight and obese categories into “overweight/obese” to produce more stable estimates.

Apart from the weight status, several studies have suggested that the association between sodium intake and BP among children might be affected by their physical activity level and by family history of hypertension.^12,14,35,36 However, NHANES 2003–2008 did not collect family hypertension data from children or physical activity data from children <12. In our sensitivity analyses, we assessed the effect of physical activity (inactive, active but at below recommended levels, and active at or above recommended levels) only among adolescents aged 12 to 18 years (Supplemental Tables 4 and 5). For sensitivity analyses of BP measurements, we restricted our analyses among participants with 3 BP measurements (Supplemental Tables 6 and 7).

We calculated the weighted mean and SE of subjects’ estimated usual sodium and total calorie intakes by the categories described earlier and used the Satterthwaite adjusted F test to assess the differences in mean intakes. We calculated the SD score (SDS) of BP as observed minus expected BP divided by the SD³⁰ and used linear regression to estimate the adjusted β-coefficients of sodium intake (per 1000 mg/day) with SDS of SBP and DBP by weight status (normal weight vs overweight/obese). We calculated the middle value of each quartile of usual sodium intake in total population and by weight status (ie, 12.5th, 37.5th, 62.5th, and 87.5th percentiles), used the linear regression models to estimate the adjusted mean SBP and DBP associated with each of these values, and presented these adjusted means as SBP or DBP level associated with the each quartile of usual sodium intake in population.^37,38

We used logistic regression to estimate the adjusted odds ratio (OR) for the association between subjects’ sodium intake quartile and their risk for pre-HBP/HBP by comparing the risk among those in each of the top 3 quartiles with that among those in the lowest quartile (Q4, Q3, and Q2 vs Q1).^37,38 In both the linear and logistic regression analyses, we estimated age, gender, race-ethnicity, and total energy adjusted β-coefficient or ORs as well as the estimates adjusted for, in addition, weight status, table salt use, and usual potassium intake.

Many studies have suggested that assessment of interaction on an additive scale is more meaningful than an assessment of interaction on a multiplicative scale (a product term in the regression models) especially from a public health perspective.^39,40 We used the relative excess risk due to interaction (RERI) to examine the joint effect of sodium intake and weight status on risk for pre-HBP/HBP.⁴⁰ The RERI was calculated by using the coefficients from the multivariable logistic regression as follows: RERI=exp(β1+β2+β3)−exp(β1)−exp(β2)+1, where β1 is the logistic regression coefficient of sodium intake per 1000 mg per day, β2 is the coefficient of the weight status, and β3 is the coefficient of the product term of sodium intake and weight status.⁴¹ RERI = 0 indicates exact additivity of 2 exposure variables and no interaction, and RERI >0 indicates the synergistic interaction on additive scale. We used the rescaling bootstrap method that takes into account the complex survey design by changing the sampling weights for each resample.^42,43 We generated 1000 rescaled bootstrap weights and reported 2.5th and 97.5th percentile values as 95% CI of RERI. We used SUDAAN (version 9.2) and SAS (version 9.3) for all analyses.⁴⁴ All tests were 2-sided, and P values <.05 were considered statistically significant.

Strengths of our study included our analysis of dietary sodium intake data based on the multiple days of 24-hour recalls. Nutrient intake data based on a single 24-hour dietary recall are not reliable indicators of subjects’ usual nutrient intake because of large day-to-day variations in intake, which may bias the association between nutrient intake and disease.^25,37,56 We used a measurement error model developed by NCI to estimate subjects’ usual sodium intake for associations between sodium intake and both BP and risk for pre-HBP/HBP.^26,27,38 Several studies have indicated that the estimated usual intakes provided significant improvements in assessments of nutrient-disease associations.^37,38

Our study also had some notable limitations. First, our estimates of subjects’ “usual sodium intake” were based on self-reported dietary recalls rather than on the 24-hr urine collection, which is considered to be the most reliable method. Although studies suggested that 24-hr dietary recalls may underestimate the usual sodium intake,⁵⁷ sodium intake from the repeated dietary recalls correlated significantly with the 24-hr urinary excretion and provided a valid method for previous association studies.^57,58 Approximately 9% of subjects had no second-day dietary recalls in our study; however, the patterns of association remained unchanged by excluding those subjects (results not shown). Second, the 24-hr dietary recall underestimated total energy intake by ~11% in NHANES and likely underestimated sodium intake as well because the total energy and sodium intakes are highly correlated. Non-differential recall bias in total energy and sodium intakes by weight status (Table 1) may underestimate the effect of sodium intake on BP and risk for pre-HBP/HBP.⁵⁹ To assess the extent to which any underestimate of sodium intake may have affected our results, we calculated the total energy-adjusted sodium intake by the residual method⁶⁰ and found the patterns of association to be largely unchanged from those in our primary analyses. The adjusted β-coefficients of SDS for SBP per 1000 mg per day sodium intake were 0.078 (95% CI: −0.006 to 0.163), 0.028 (95% CI: −0.091 to 0.147), and 0.160 (95% CI: 0.006 to 0.315) for all, normal-weight and overweight/obese children, respectively. The corresponding adjusted ORs were 1.36 (95% CI: 1.00–1.89), 1.02 (95% CI: 0.67–1.55), and 1.72 (95% CI: 1.13–2.63), respectively. Third, because of underreporting of total energy and sodium intakes, the effect of weight status, which reflects several lifestyle factors such as physical activity and healthy diets, on BP or risk for pre-HBP/HBP may be overestimated. In the sensitivity analyses for children aged 12 to 18 years, we also adjusted for physical activity, and the patterns of association remain largely unchanged (Supplemental Tables 4 and 5). Fourth, we restricted our analyses to the 78% of children and adolescents with 3 BP measurements, and the patterns of association remained largely unchanged (Supplemental Tables 6 and 7). Finally, because our study data came from a cross-sectional survey, the associations we found should be interpreted with caution. A large randomized controlled trial would be needed to confirm our findings.