We have several concerns related to the report “30-Year Trends in Serum Lipids Among United States Adults: Results from the National Health and Nutrition Examination Surveys II, III, and 1999–2006” by Cohen et al. Our major concern involves the conclusions presented in this report regarding trends in serum triglyceride and in high-density lipoprotein (HDL) cholesterol in adults from 1976 to 2006 and their clinical implications. We are concerned about the validity of the statistical techniques on which the conclusions concerning trends in serum triglyceride are based and the effect of a method change in HDL cholesterol between 1999 to 2002 and 2003 to 2006 on trends in HDL cholesterol, both of which could alter the conclusions regarding these 2 lipid fractions presented by Cohen et al. Other concerns involve the effect of this HDL cholesterol method change on low-density lipoprotein (LDL) cholesterol and the lack of sound statistical techniques in estimating and analyzing trends in the percentage of adults by total cholesterol and lipid fraction cut points.

Cohen et al. concluded that the age-adjusted arithmetic mean serum triglyceride level of adults aged 20 to 74 years decreased between 1976 to 1980 and 1988 to 1994 and then increased between 1988 to 1994 and 1999 to 2006. They also concluded that the age-specific arithmetic mean serum triglyceride level of adults aged 60 to 74 years increased between 1976 to 1980 and 1999 to 2006. The investigators based these conclusions on the assumption that by the Central Limit Theorem (applied to finite populations), the sample size was sufficiently large so that the distribution of the arithmetic means of serum triglyceride was approximately normally distributed.

However, in instances in which the population distribution is highly skewed, the convergence of the sampling distribution of the sample mean to normality is expected to be slow, and its normality depends on the skewness of the population distribution. The greater the skewness of the distribution, the larger is the required sample size.

For the National Health and Nutrition Examination Survey (NHANES) II (1976 to 1980), NHANES III (1988 to 1994), and NHANES 1999 to 2006, the distribution of serum triglyceride was highly, positively skewed. Log-transforming the distribution of triglyceride yields a distribution closer to normality, so that the Central Limit Theorem may be applied, and the distribution of the sample mean of the log-transformed values of serum triglyceride can be assumed to be approximately normal. The geometric mean (obtained by back-transforming the mean of the log-transformed values of serum triglyceride) would have been more appropriate than the arithmetic mean for estimating triglyceride means. The geometric mean has been used in analyzing serum triglyceride from the NHANES in many reports, including that of Carroll et al. Cohen et al reported age-adjusted arithmetic mean ± SEM triglyceride levels of adults aged 20 to 74 years for 1976 to 1980, 1988 to 1994, and 1999 to 2006 of 136.8 ± 1.8, 129.9 ± 2.6, and 145.8 ± 2.3 mg/dL, respectively, whereas the geometric mean ± SEM levels are 113.6 ± 1.55, 116.2 ± 1.90, and 120.5 ± 1.15 mg/dL, respectively. Using the arithmetic mean, Cohen et al report an increase in serum triglyceride in adults aged 20 to 74 years between 1988 to 1994 and 1999 to 2006 in their summary and in their results and an increase in adults aged 60 to 74 years between 1976 to 1980 and 1999 to 2006. However, neither the age-adjusted geometric mean of adults aged 20 to 74 years nor the age-specific geometric mean of adults aged 60 to 74 years increased significantly during the respective time periods. The investigators acknowledged as a weakness of their study that it included the non-normality of the distributions of serum total cholesterol and lipid fractions in some of the subgroups, but they did not attempt to apply a transformation to serum triglyceride.

Another issue relates to the combined 8-year estimates reported for NHANES 1999 to 2006. Cohen et al assumed that there were no changes in serum triglyceride during these years. Between 1999 to 2002 and 2003 to 2006, there was a gradual decrease in the geometric mean serum triglyceride level. For adults aged 20 to 74 years, the geometric mean serum triglyceride level decreased from 122.2 ± 1.9 mg/dL from 1999 to 2002 to 118.9 ± 1.4 mg/dL from 2003 to 2006. Also, the serum triglyceride data from NHANES 2007 to 2008 shows a further decrease to 111.7 ± 2.1 mg/dL.

Cohen et al reported an increase in mean HDL cholesterol, a decrease in the percentage with low HDL cholesterol (<40 mg/dL), and an increase in the percentage with high HDL cholesterol between 1976 to 1980 and 1999 to 2006. Although they mentioned the change in HDL cholesterol methods between NHANES 1999 to 2002 and NHANES 2003 to 2006 in their methods and acknowledged in their comments that it may explain at least in part these changes, they did not correct for the methods change in HDL cholesterol either through modeling or average corrections. After applying the average correction of 3.1 mg/dL to HDL cholesterol data for adults aged 20 to 74 years from 2003 to 2006, age-adjusted and age-specific HDL cholesterol levels of adults did not change significantly over the 30-year time period from 1976 to 2006 (mean ± SEM HDL cholesterol levels were 50.1 ± 0.44 mg/dL for 1976 to 1980, 50.3 ± 0.42 mg/dL for 1988 to 1994, and 51.0 ± 0.22 mg/dL for 1999 to 2006). Also, the percentage with HDL cholesterol <40 mg/dL did not change significantly (25 ± 1.2% for 1976 to 1980, 25 ± 1.2% for 1988 to 1994, and 26 ± 0.6% for 1999 to 2006).

It is unlikely that the correlates of HDL cholesterol, which include body mass index, smoking, alcohol consumption, total carbohydrate and physical activity, and drugs that increase HDL cholesterol levels, could account for the increase in HDL cholesterol between 1999 to 2002 and 2003 to 2006. Neither body mass index nor smoking changed between these two time periods (age-adjusted body mass index of adults aged 20 to 74 years: 28.1 kg/m ² from 1999 to 2002 vs 28.5 kg/m ² from 2003 to 2006; age-adjusted percent serum cotinine >10 ng/L in adults aged 20 to 74 years: 29.7% from 1999 to 2002 vs 30.6% from 2003 to 2006; self-reported smoking 25.9% from 1999 to 2002 vs 26.4% from 2003 to 2006). Neither alcohol nor total carbohydrate, the two major dietary correlates of HDL cholesterol, changed between 1999 to 2002 and 2003 to 2006 (age-adjusted mean intake of alcohol: 11 g/day from 1999 to 2002 vs 12 g/day from 2003 to 2006; age-adjusted mean intake of total carbohydrate 278 g/day from 1999 to 2002 vs 272 g/day from 2003 to 2006). Also, the available evidence on physical activity from the National Health Interview Survey (on the basis of self-report) suggests that for adults aged 20 to 74 years, physical activity did not change substantially between 1999 to 2002 and 2003 to 2006 (age-adjusted percentage who met the 2008 guidelines for aerobic and strength: 16.4% from 1999 to 2002 vs 17.1% from 2003 to 2006). Furthermore, the use of the two main drugs that increase HDL cholesterol reported in NHANES, niacin and gemfibrozil, was rare. Therefore, the changes in HDL cholesterol reported by Cohen et al appear to be related to the change in HDL cholesterol methods between 1999 to 2002 and 2003 to 2006.

The change in HDL cholesterol methods also affects age-adjusted and age-specific mean LDL cholesterol levels because LDL cholesterol is calculated using the Friedewald equation (LDL cholesterol = total cholesterol − HDL cholesterol − serum triglyceride/5), which is valid for serum triglyceride levels ≤400 mg/dL. There is an average 1.6 mg/dL increase in LDL cholesterol levels of adults aged 20 to 74 years after applying the average correction of 3.1 mg/dl to HDL cholesterol. The age-adjusted mean LDL cholesterol level of adults aged 20 to 74 years was 119.5 mg/dL before the average correction and 121.1 mg/dL after the correction of HDL cholesterol for 2003 to 2006.

There were several other analytic issues in the report of Cohen et al. Using percentages rather than testing statistical hypotheses, the investigators concluded that the percentage of adults with high and very high (≥150 mg/dL) triglyceride levels decreased between 1976 to 1980 and 1988 to 1994 and then increased between 1988 to 1994 and 1999 to 2006 (see Table 2 in report ). Also in Table 2, the investigators included in the denominator adults for whom serum triglyceride levels were not available, they did not age-adjust triglycerides, and they did not present standard errors of the percentages.

In summary, we question the results of Cohen et al on trends in serum triglyceride and HDL cholesterol. Using the geometric rather than the arithmetic mean serum triglyceride level, we conclude that for adults aged 20 to 74 years, serum triglyceride did not change significantly between 1988 to 1994 and 1999 to 2006, nor did the age-specific mean of adults aged 60 to 74 years between 1976 to 1980 and 1999 to 2006, contrary to what was reported by Cohen et al. Furthermore, the percentage with low HDL cholesterol levels did not change significantly over this 30-year period.