OBJECTIVE: A 2008 report from the American Academy of Pediatrics recommended both population and individual approaches (including pharmacologic interventions) for adolescents who had low-density lipoprotein (LDL) cholesterol levels above various cutoff points (130, 160, and 190 mg/dL). However, the tracking and variability of these very high levels have not been investigated.
METHODS: A total of 6827 subjects underwent multiple LDL cholesterol determinations in childhood and adulthood in the Bogalusa Heart Study. The total number of determinations was 26748, and the median interval between examinations was 3 years.
RESULTS: Correlations between initial and subsequent LDL cholesterol levels ranged from r ∼ 0.8 for measurements made within the same year to r ∼ 0.5 for periods of ≥20 years. Most children who had very high LDL cholesterol levels, however, had substantially lower levels at the next examination. LDL cholesterol levels between 160 and 189 mg/dL (n = 201) decreased, on average, by 21 mg/dL at the next examination, whereas levels of ≥190 mg/dL (n = 44) decreased by 34 mg/dL. In contrast, the mean increase for LDL cholesterol levels of <70 mg/dL was 13 mg/dL. These changes were equal to those expected on the basis of regression to the mean.
CONCLUSIONS: There can be large changes in extreme levels of LDL cholesterol because of regression to the mean, and practitioners should be aware that very high levels may decrease substantially in the absence of any intervention.
WHAT'S KNOWN ON THIS SUBJECT:
Children with high levels of LDL cholesterol are likely to have elevated levels in adulthood.
WHAT THIS STUDY ADDS:
Despite the tracking of LDL cholesterol levels, children with very high levels (eg, ≥160 mg/dL) are likely to have lower levels at the next measurement. These decreases may be entirely attributable to regression to the mean.
The development of atherosclerosis begins in childhood1 and is associated with lipid and lipoprotein levels.2,3 Although adverse lipid levels among children predict subsequent levels in adolescence and adulthood,4,–,8 a substantial proportion of children with initially high levels have lower levels on reexamination.7 This tendency, in which subsequent measurements of extreme (high or low) biological levels are closer to the mean than was the initial determination, was recognized by Galton in the 1800s and has been termed “regression to the mean.”9
Because of the importance of low-density lipoprotein (LDL) cholesterol levels in early life, a 2008 report from the American Academy of Pediatrics10 identified several LDL cholesterol cutoff points for dietary and pharmacologic interventions for children and adolescents, with levels of <110 mg/dL being considered acceptable. A population approach was recommended, and changes in diet and physical activity were recommended for children and adolescents with LDL cholesterol levels of ≥130 mg/dL, a level that is between the 90th and 95th percentiles for adolescents.11 Pharmacologic interventions were considered for 3 groups of adolescents (≥10 years of age), that is, (1) those with LDL cholesterol levels of ≥190 mg/dL despite dietary therapy, (2) those with LDL cholesterol levels of ≥160 mg/dL and other risk factors, and (3) those with LDL cholesterol levels of ≥130 mg/dL and diabetes mellitus.10 It has been estimated that 200 000 adolescents in the United States might be eligible for pharmacologic treatment.11
Despite the tracking of elevated LDL cholesterol levels, the persistence of very high levels among children and adolescents, such as those of ≥160 mg/dL or ≥190 mg/dL, has not been examined. Because LDL cholesterol levels show substantial biological variability,12,–,14 it is possible that, even without any intervention, a large proportion of children and adolescents with LDL cholesterol levels of >160 mg/dL would have much lower levels at reexamination. The objective of the current study was to examine the persistence of very high levels of LDL cholesterol among children and adolescents.
Bogalusa is a biracial (one-third black) community in Louisiana. The objective of the Bogalusa Heart Study was to examine the natural history of cardiovascular disease and risk factors among school-aged children,15 and the first of 7 cross-sectional studies was conducted in 1973–1974.16 On average, each cross-sectional study examined ∼3500 children 5 to 17 years of age; the final examination of school-aged children was performed in 1992–1994. Several substudies of selected groups of children were conducted between 1973 and 2002, and those LDL cholesterol determinations were included in the current analyses. For example, children who were 5, 8, 11, or 14 years of age at the 1973–1974 examination were reexamined in each of the subsequent 3 years. Young adults (up to 44 years of age) were examined in various studies conducted in 1982–2002.17,18
The panel design and the participation of subjects in various examinations resulted in widely varying intervals between consecutive LDL cholesterol determinations. We excluded 4190 of ∼36000 LDL cholesterol determinations because the subject was not fasting, and we excluded another 71 determinations because of missing data. Because we were interested in examining the stability of high LDL cholesterol levels, we further restricted the analyses to subjects who were examined initially before 18 years of age and who had ≥2 LDL cholesterol determinations. These exclusions resulted in a sample of 6827 subjects, who were examined an average of 5 times (range: 2–13 times). Of those subjects, 3410 were examined at least once in adulthood (age of ≥18 years). The current analyses were based on 26 748 LDL cholesterol determinations.
Although previous analyses from the Bogalusa Heart Study examined the tracking of LDL cholesterol levels, those studies were based on LDL cholesterol levels for various subgroups that ranged in size from 273 to 1169 children.4,8,19 In contrast, the current study uses all of the longitudinal data from the Bogalusa Heart Study.
Examinations and Laboratory Determinations
Height was measured to the nearest 0.1 cm and weight to the nearest 0.1 kg.20 BMI was calculated as kilograms per square meter, and BMI-for-age z scores and percentiles were calculated from the Centers for Disease Control and Prevention growth charts.21,22
All chemical analyses were performed in the Bogalusa Heart Study Core Laboratory. Serum concentrations of total cholesterol and triglycerides were determined by using enzymatic procedures (Abbott VP [Abbott, North Chicago, IL]). After heparin-calcium precipitation of LDL cholesterol and very low-density lipoprotein cholesterol, the LDL cholesterol concentration was determined from the densitometric (electrophoretic) ratio and cholesterol contents of the 2 lipoproteins.16
Data management was performed by using SAS (SAS Institute, Cary, NC), and R23 was used for the analyses. The reproducibility of the LDL cholesterol determinations was assessed in a 10% sample of blind duplicates from various studies of children and adults, and we used the intraclass correlation coefficient, the median absolute difference, and other statistics to describe the observed laboratory measurement error.
Various characteristics of the sample were summarized by using means and SDs, and lowess24 was used to illustrate the association between age and LDL cholesterol levels within each race/gender group. We examined the correlation between an individual's initial LDL cholesterol level and subsequent levels in analyses that were stratified according to the interval between measurements. Because subjects were examined in multiple examinations (mean: 5 examinations), with unequal intervals between examinations, most subjects contributed multiple pairs for this analysis. For example, a subject with 3 LDL cholesterol measurements would be included in the analysis of the first and second determinations, as well as the first and third determinations. We also classified LDL cholesterol levels among children by using cutoff points of 70, 110, 130, 160, and 190 mg/dL for the 6 categories, and then we examined the relationships of initial and subsequent LDL cholesterol level categories. LDL cholesterol levels of <70 mg/dL have been suggested as a treatment goal for adults with cardiovascular disease.25
To examine LDL cholesterol levels during follow-up monitoring, we fit multilevel regression models with the lme package (linear mixed effects),26 to account for within-subject correlations. Predictors in these models included age at initial examination, initial LDL cholesterol level, number of years of follow-up monitoring, race, gender, and interaction between initial LDL cholesterol level and years of follow-up monitoring. Subjects and years of follow-up monitoring were treated as random effects, and polynomials were used to model age and LDL cholesterol levels. We calculated predicted LDL cholesterol levels according to the number of years after the initial determination and the initial LDL cholesterol level.
Table 1 shows the reproducibility of LDL cholesterol measurements for children and adults who had duplicate venipuncture samples in any of the cross-sectional studies. Among those 3044 persons, the intraclass correlation coefficient between replicate measurements was 0.956, and the median absolute difference between determinations was 3.6 mg/dL. Stratified analyses indicated that the median absolute difference between replicates varied only slightly (3–5 mg/dL) with the LDL cholesterol level. Of the 92 children with LDL cholesterol determinations of ≥160 mg/dL, approximately one-fifth had absolute differences between replicate determinations of ≥10 mg/dL, and 4% had differences of ≥20 mg/dL.
Approximately one-half of the 6827 children were reexamined as adults, and Table 2 presents data on various characteristics. Subjects were examined 5 times, on average, and 1397 individuals were examined ≥6 times. Approximately 37% of the examined subjects were black. Participants were relatively heavy, with ∼21% of the children and ∼46% of the adults considered to be overweight or obese. Among children, the mean LDL cholesterol level was 92 mg/dL, and the prevalence of elevated levels was as follows: levels of ≥130 mg/dL, 8%; levels of ≥160 mg/dL, 1%; levels of ≥190 mg/dL, 0.3%. Approximately 30% of adults had LDL cholesterol levels of ≥130 mg/dL, and 2% had levels of ≥190 mg/dL.
Figure 1 shows smoothed (lowess) levels of LDL cholesterol according to age for each race/gender group. This analysis, which did not account for the longitudinal structure of the data, is shown only to illustrate age-related changes. Levels of LDL cholesterol tended to decrease between the ages of 10 and 15 years and subsequently showed large increases (∼30–40 mg/dL) into adulthood. The observed gender and race differences varied according to age. Among children, girls had higher LDL cholesterol levels than did boys; among adults, however, LDL cholesterol levels were higher among white men than among white women.
Correlations between initial and subsequent levels of LDL cholesterol, stratified according to initial age and the interval between measurements, are shown in Table 3. The magnitude of the association between LDL cholesterol levels decreased, but at a fairly slow rate, as the follow-up period increased. For example, correlations between LDL cholesterol levels were r = 0.79 (in both age groups) for measurements performed within 1 year of each other and r = 0.62 for measurements performed 7 years apart. There were, however, moderate correlations (r = 0.46–0.48) between LDL cholesterol determinations that were separated by ≥20 years.
Despite the moderate/high correlations between serial LDL cholesterol determinations, a large proportion of children with very high LDL cholesterol levels had substantially lower levels at the next examination (Table 4). (Table 4 is based on all consecutive LDL cholesterol determinations.) Among those determinations, 95% of determinations (2859 of 2996 determinations) that were initially below 70 mg/dL remained below 110 mg/dL at the next examination. In contrast, of the 44 determinations that were ≥190 mg/dL, only 39% remained above that cutoff point at reexamination. Only 4 of those 44 very elevated LDL cholesterol determinations were below 130 mg/dL at the next examination. Of the 201 determinations that were initially between 160 and 189 mg/dL, 132 (66%) were below 160 mg/dL and 25% were below 130 mg/dL at the next examination.
Table 4 also shows the mean changes in LDL cholesterol levels from the initial examination to the subsequent examination, as well as the mean levels at the subsequent examination. There was a strong inverse association between the initial LDL cholesterol levels and the observed changes between examinations. Persons with low LDL cholesterol levels (levels of <70 mg/dL) showed a mean increase of 13 mg/dL, but there were mean decreases of 21 mg/dL (levels of 160–189 mg/dL) and 34 mg/dL (levels of ≥190 mg/dL) among persons in the 2 upper LDL cholesterol level categories,. In addition, the variability of the LDL cholesterol changes among persons with initial LDL cholesterol levels of ≥190 mg/dL was very large; the SD was 56 mg/dL, and the 10th and 90th percentiles of the LDL cholesterol level changes were −79 mg/dL and 16 mg/dL, respectively.
Figure 2 shows predicted LDL cholesterol levels, on the basis of multilevel models, according to baseline LDL cholesterol determinations and the number of years after the initial measurement. Predicted levels are shown for children who had initial LDL cholesterol levels that were at the mean of each of the 6 categories. LDL cholesterol levels among children in the lowest category (levels of <70 mg/dL) increased almost linearly over the 20 years of follow-up monitoring, from ∼60 mg/dL to 96 mg/dL. Despite the tendency for LDL cholesterol levels to increase with age, however, mean levels among children in the 2 upper LDL cholesterol level categories (levels of 160–189 mg/dL and ≥190 mg/dL) were higher at the initial examination than after 20 years of follow-up monitoring. For example, the predicted LDL cholesterol level after 20 years for children who had initial LDL cholesterol levels of 212 mg/dL (the mean for the category of LDL cholesterol levels of ≥190 mg/dL) was 183 mg/dL. In general, LDL cholesterol level differences between the 6 categories narrowed substantially during the follow-up period. There was a difference of 161 mg/dL (212 mg/dL [highest category] − 61 mg/dL [lowest category]) between the 2 extreme LDL cholesterol categories at baseline, but the predicted difference in the mean LDL cholesterol levels between these categories was 87 mg/dL (183 mg/dL − 96 mg/dL) 20 years later.
A single LDL cholesterol measurement among children and adolescents predicts subsequent levels,4,–,8,19,27,28 and we found moderately strong tracking for LDL cholesterol levels. Intraindividual correlations between LDL cholesterol determinations ranged from r ∼ 0.8 for measurements made within the same year to r ∼ 0.5 for intervals of >20 years. However, most children who were identified as having extremely high (≥160 or ≥190 mg/dL) LDL cholesterol levels had substantially lower levels at the next determination. During the 4-year interval, on average, between examinations, mean LDL cholesterol levels decreased by 21 mg/dL (initial LDL cholesterol levels between 160 and 189 mg/dL) or 34 mg/dL (initial levels of ≥190 mg/dL). Although for the most part children with extremely high LDL cholesterol levels had high levels at reexamination, the changes in LDL cholesterol levels that occurred between examinations would have altered the recommended10 dietary or pharmacologic interventions for many children.
The correlations between serial LDL cholesterol determinations in the current study are similar to those reported by others.4,5,7,27 For example, a previous analysis of data for children (2–10 years of age at baseline) in the Bogalusa Heart Study4 reported correlations of r = 0.70 over a 3-year interval and r = 0.60 over an 8-year interval. Several studies also found that large proportions of children who initially had high LDL cholesterol levels had lower levels at reexamination.4,–,8,19,27 The cutoff points for classifying children's LDL cholesterol levels as high varied across these studies, being based on levels in the upper quintile,4,6,27 in the upper decile,5,7 above 130 mg/dL,8 or above cutoff points that varied from 135 to 155 mg/dL across gender and age groups.19
The substantial decreases in LDL cholesterol levels that we observed among children who had very high LDL cholesterol levels were attributable in part to the extremely high cutoff points identified in the 2008 report by the American Academy of Pediatrics.10 Only 1.4% of the children in the current study had LDL cholesterol levels of ≥160 mg/dL, and only 0.3% had levels of ≥190 mg/dL. (LDL cholesterol levels of 160 mg/dL were in the ∼99th percentile for 12- to 17-year-old subjects in the National Health and Nutrition Examination Survey in 1999–2004.29) The magnitude of regression to the mean, in which a subsequent determination is closer to the mean than is the initial value, depends on intraindividual variations (including both biological and laboratory variations) and the distance of the initial measurement from the mean. There is more regression if the initial measurement is far from the mean and if the correlation between serial measurements is weak.
The mean ± SD for LDL cholesterol levels among children in the current study was 92 ± 25 mg/dL, and regression to the mean could account for the longitudinal changes we observed. On the basis of regression to the mean, a child with an initial LDL cholesterol level of 167 mg/dL (3 SDs above the mean) would be expected to have a level at reexamination of 92 [overall mean] + (3 [number of SDs from the mean for the initial measurement] × 25 [SD] × r [correlation coefficient]) mg/dL.30 If the child was reexamined after a period of 4 years, than r would be 0.64 (Table 3) and the expected LDL cholesterol level at reexamination would be 140 mg/dL, a decrease of 27 mg/dL. It also can be estimated31,32 that the mean decrease among children who initially had LDL cholesterol levels of ≥160 mg/dL would be 22 mg/dL, a value similar to the decrease we observed among children who initially had LDL cholesterol levels of 160 to 189 mg/dL (Table 4). Regression to the mean also applies to low levels of biological characteristics, and we found that children who had initial LDL cholesterol levels of <70 mg/dL showed a mean increase of 13 mg/dL at reexamination.
There is substantial intraindividual variability in lipid and lipoprotein concentrations,13 with measured levels varying from day to day12,33,34 and even from hour to hour.35 Most of this variability is attributable to biological and not analytical variation.33,36 It also seems that the magnitude of LDL cholesterol tracking among children decreases most rapidly within days after the initial determination, with observed correlations among children decreasing from r = 0.996 for duplicate samples to r = 0.97 at 2 days and r = 0.85 at 7 days.34 Because of this biological variability, several investigators have emphasized the need to obtain multiple determinations to characterize a person's true, long-term, LDL cholesterol status.13,36,37 Although measurement errors can be reduced with more-accurate laboratory methods, the inherent biological variability in LDL cholesterol determinations will always be present. Additional characteristics, such as family history and BMI, may help to identify children who are likely to have high LDL cholesterol levels in adulthood.8
A limitation of our study is that the median period between examinations was ∼3 years, an interval that is longer than that between office visits for children and adolescents; a shorter interval would be associated with less regression to the mean. The correlation between LDL cholesterol determinations made within 1 year was r = 0.80 (Table 3), and it can be estimated that the mean LDL cholesterol level of a child with a level of 167 mg/dL would decrease by 15 mg/dL during a 1-year period, rather than by an estimated 27 mg/dL over 4 years. Although laboratory errors also contribute to regression to the mean, all LDL cholesterol determinations in our study were conducted in a single laboratory in which levels were obtained by using heparin-calcium precipitation and electrophoresis16 and the reproducibility was very high; the median difference between blind duplicate samples from children with initial determinations of ≥160 mg/dL was only −2 mg/dL (Table 1). It is possible, however, that there would be larger measurement errors among children examined at office visits. LDL cholesterol levels likely would be reported from many laboratories by using the Friedewald equation, in which triglyceride levels are divided by 5 to provide an estimate of very low-density lipoprotein cholesterol levels.38 Those LDL cholesterol measurements would be influenced by the large biological variability inherent in triglyceride levels, with levels fluctuating up to 40% around the mean value,39 which might result in extreme LDL cholesterol levels at office visits showing more regression to the mean than we observed.
Despite the moderate/strong tracking of LDL cholesterol levels, we found that children with very high LDL cholesterol levels (≥160 mg/dL) were likely to have substantially lower levels (mean decreases of 20 to 30 mg/dL) at reexamination. It would be important not to attribute these changes, which may be entirely attributable to regression to the mean, automatically to changes in diet or physical activity. Although the subsequent LDL cholesterol levels for most of these children would remain elevated, many children would be candidates for different interventions on the basis of their reexamination LDL cholesterol levels.10 Practitioners should be aware of the large effects of regression to the mean among children with very high LDL cholesterol levels and should understand that multiple determinations may be needed to characterize a child's true LDL cholesterol status, particularly if levels are ≥160 mg/dL. These extreme LDL cholesterol levels can be strongly influenced by regression to the mean.
This work was supported by National Institute on Aging Grant AG16592.
- Accepted April 23, 2010.
- Address correspondence to David S. Freedman, PhD, Centers for Disease Control and Prevention, Division of Nutrition, Physical Activity, and Obesity, K-26, 4770 Buford Highway, Atlanta, GA 30341-3724. E-mail:
The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
Funded by the National Institutes of Health (NIH).
- LDL =
- low-density lipoprotein
- McGill HC Jr.,
- McMahan CA,
- Malcom GT,
- Oalmann MC,
- Strong JP
- Porkka KV,
- Viikari JS,
- Taimela S,
- Dahl M,
- Akerblom HK
- Laskarzewski P,
- Morrison JA,
- deGroot I,
- et al
- Stigler SM
- Daniels SR,
- Greer FR
- Ford ES,
- Li C,
- Zhao G,
- Mokdad AH
- Smith SJ,
- Cooper GR,
- Myers GL,
- Sampson EJ
- Berenson GS,
- McMahan CA
- Srinivasan SR,
- Frerichs RR,
- Webber LS,
- Berenson GS
- Croft JB,
- Webber LS,
- Parker FC,
- Berenson GS
- Magnussen CG,
- Raitakari OT,
- Thomson R,
- et al
- Webber LS,
- Cresanta JL,
- Croft JB,
- Srinivasan SR,
- Berenson GS
- Kuczmarski RJ,
- Ogden CL,
- Grummer-Strawn LM,
- et al
- Kuczmarski RJ,
- Ogden CL,
- Guo SS,
- et al
- 23.↵R Development Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2010
- Cleveland WS
- Bates D,
- Maechler M
- Jolliffe CJ,
- Janssen I
- Cole TJ
- Davis CE
- Chuang-Stein C,
- Tong DM
- Bookstein L,
- Gidding SS,
- Donovan M,
- Smith FA
- Friedewald W,
- Levy R,
- Fredrickson D
- Rifai N,
- Warnick GR
- Rifai N,
- Dufour D,
- Cooper G
Regenerative Medicine: Still a Lung Way to Go: Recent articles in Science and Nature Medicine have described recent developments in the field of regenerative medicine. According to an article in The Wall Street Journal (Naik G, June 25, 2010) scientists at Yale dissolved all the cells in the lungs of dead adult rats and left behind a matrix of collagen and other proteins as well as underlying airway and vascular system structure. Lung cells from rat fetuses were then injected into this matrix and within days they had differentiated in this matrix into rat lung that was able to demonstrate good gas exchange for 45 to 120 minutes in live rats. A similar type or work was done at the Massachusetts General Hospital creating a rat liver, and in 2008 University of Minnesota researchers were able to create a beating heart from primitive fetal cells in the lab. While these experiments are exciting, especially given organ transplant shortages, investigators believe it could be years or even decades before such experiments will be tried on people. Nonetheless it appears the field of regenerative medicine is one undergoing substantial growth and development and is worth reading about as studies in this area increase.
Noted by JFL, MD
- Copyright © 2010 by the American Academy of Pediatrics