Insufficient levels of serum 25-hydroxyvitamin D (25(OH)D), the major circulating vitamin D metabolite, which is traditionally used to determine vitamin D status( Reference Holick 1 ), are a global phenomenon( Reference Hilger, Friedel and Herr 2 ). Possible adverse effects of low 25(OH)D levels on health are thus of great interest. While the role of vitamin D in the regulation of calcium, phosphorus and bone metabolism and its consequential importance for skeletal health have been known for a long time( Reference Holick 3 ), research in recent decades has raised the question about non-skeletal health effects of vitamin D. Several cross-sectional studies have reported an association between low levels of 25(OH)D and an unfavourable lipid profile( Reference Jorde and Grimnes 4 ). In two large cohorts, 25(OH)D levels have also been found to be significantly associated with a decrease in TAG levels over 14 years( Reference Jorde, Figenschau and Hutchinson 5 ) and with lower TAG and VLDL cholesterol levels as well as with a reduced odds for hypercholesterolaemia after 5 years( Reference Skaaby, Husemoen and Pisinger 6 ). The results of a Mendelian randomization study, in which genetically instrumented 25(OH)D levels were positively associated with HDL cholesterol (HDL-C) and inversely with TAG( Reference Skaaby, Husemoen and Martinussen 7 ), support the notion of a causal relationship between vitamin D and the lipid profile. By contrast, serum lipids are generally not significantly influenced by vitamin D supplementation in randomized controlled trials( Reference Wang, Xia and Yang 8 ). However, the trials conducted so far provide limited evidence, as they were insufficiently powered and not specifically designed to evaluate the effect of vitamin D on serum lipids( Reference Wang, Xia and Yang 8 ).
If a causal relationship exists, there are several potential mechanisms on how low vitamin D levels might directly or indirectly impact serum lipids( Reference Challoumas 9 ). Some of these pathways, such as decreased insulin sensitivity( Reference Sung, Liao and Lu 10 ), decreased activity of lipoprotein lipase (LPL)( Reference Huang, Li and Wang 11 ) and lower levels of adiponectin( Reference Vaidya, Forman and Underwood 12 , Reference Jung and Choi 13 ), are also related to obesity, a known risk factor for dyslipidaemia( Reference Jung and Choi 13 ). It is thus possible that the association between 25(OH)D and the lipid profile varies according to weight or waist circumference (WC) status. In a cohort of Chinese adults, the inverse association between 25(OH)D and the metabolic syndrome was significant only in overweight and obese individuals( Reference Li, Yin and Yao 14 ), but little is known about a possible effect modification of obesity on the association between 25(OH)D and serum lipids alone. 25(OH)D levels are generally low in obese individuals( Reference Earthman, Beckman and Masodkar 15 ) and should an inverse association between 25(OH)D and serum lipids be stronger in the obese, or occur primarily in this group, 25(OH)D deficiency could be regarded as a currently unaccounted risk factor for dyslipidaemia in obese individuals. Thus, the objective of the present study was to examine the interaction between WC and serum 25(OH)D levels in their associations with serum lipids in adult participants of the cross-sectional US National Health and Nutrition Examination Surveys (NHANES) 2001–2006.
Methods
Study population
NHANES are continuous cross-sectional surveys of the non-institutionalized civilian resident US population, which are conducted annually by the Centers for Disease Control and Prevention’s (CDC) National Center for Health Statistics (NCHS), Hyattsville, MD, USA. A nationally representative population sample is selected by means of a complex, four-stage probability sampling design. Certain subgroups (adolescents, adults aged 70 years or over, non-Hispanic blacks, Mexican Americans and persons with low income in the NHANES waves 1999–2006) are oversampled to increase reliability and precision. The surveys combine home-administered personal interviews with standardized physical examinations, interviews and laboratory tests conducted in specially equipped mobile examination centres. NHANES is conducted in accordance with the US Department of Health and Human Services’ Policy for Protection of Human Research Subjects. The data are released to the public in 2-year cycles by the NCHS( Reference Zipf, Chiappa and Porter 16 ). More details on sampling, interviews, examinations and laboratory measurements are given elsewhere( Reference Zipf, Chiappa and Porter 16 – 18 ).
For the present study, data from the NHANES waves 2001–2006 were pooled. In these waves, a total of 31 509 individuals completed the interview (response rates varied from 79 to 84 %) and 30 070 individuals completed the physical examinations (response rates varied from 76 to 80 %)( 18 ). The analytic sample was restricted to the 6164 non-pregnant participants aged ≥20 years, who had been randomly assigned to an examination in the morning session after an overnight fast. Further, all participants with missing information on any of the variables used for the analyses were excluded, leaving a final analytic sample of 4342 participants as the study population. A detailed description of the study population and the excluded participants is shown in a flow diagram in the online supplementary material, Supplemental Fig. 1.
Laboratory measurements
During the physical examination at the mobile examination centres, blood samples were drawn via venepuncture by certified phlebotomists( 18 ). The samples were collected using Vacutainer tubes (Becton-Dickinson, Franklin Lakes, NJ, USA) and subsequently centrifuged, aliquoted and frozen to −20°C, before being transported to laboratories across the USA for analysis( 18 ).
Serum lipids were measured at the Johns Hopkins Hospital, Baltimore, MD, USA. HDL-C was measured using the heparin manganese precipitation method in NHANES 2001–2002 and a direct HDL-C immunoassay in NHANES 2003–2006. To control for these differences in methods, the HDL-C values were corrected by NHANES using quality controls( 19 ). Total cholesterol and TAG were measured enzymatically. LDL cholesterol (LDL-C) was calculated according to the Friedewald equation: LDL-C=(total cholesterol)–(HDL-C)–(TAG/5)( 18 ). The ratio LDL-C:HDL-C was generated for the present analysis by dividing LDL-C levels by HDL-C levels.
Total serum 25(OH)D was measured at the National Center for Environmental Health, CDC, Atlanta, GA, USA using a RIA kit (DiaSorin, Stillwater, MN, USA)( 18 ). The sensitivity of this assay has been shown to be 1·5 ng/ml and the CV for the years 2001–2006 varied between 4 and 13 %( 20 – 22 ). From 2007–2008 onwards, 25(OH)D was measured using a standardized liquid chromatography–tandem mass spectrometry (LC-MS/MS) method and in October 2015, updated 25(OH)D values for 2001–2006 were released, which had been converted from RIA to LC-MS/MS equivalents using ordinary least squares regression( 23 ). As recommended by NHANES( 23 ), these LC-MS/MS equivalents were used in the present study.
Plasma glucose and serum insulin were measured at the University of Missouri-Columbia, Columbia, MO, USA from 2001 to 2004 and at the Fairview Medical Center Laboratory at the University of Minnesota, Minneapolis, MN, USA in 2005–2006. Glucose was measured using a hexokinase method. As different instruments were used in 2003–2004 (Roche/Hitachi 911) and 2005–2006 (Roche Cobas Mira), the glucose values from 2005–2006 were corrected to 2003–2004 values using linear regression, as suggested by NHANES. Insulin was measured using a RIA kit (Pharmacia Diagnostics AB, Uppsala, Sweden) in 2001–2002, a two-site immunoenzymometric assay (Tosoh Corporation, Toyama, Japan) in 2003–2004 and an ELISA immunoassay (Merocodia, Uppsala, Sweden) in 2005–2006. Insulin values from 2001–2002 and 2005–2006 were adapted to 2003–2004 values using linear regression( 18 ). Insulin resistance was estimated from glucose and insulin using the homeostatic model assessment for insulin resistance (HOMA-IR), calculated with the following equation: [fasting serum insulin (μU/ml) × fasting plasma glucose (mg/dl)]/405( Reference Matthews, Hosker and Rudenski 24 ).
Anthropometric measurements
For the assessment of height and weight during the physical examination, participants were dressed in underwear, disposable paper gowns and foam slippers. A digital scale was used to measure weight to the nearest 100 g, a fixed stadiometer to measure height to the nearest millimetre. BMI was calculated as weight in kilograms divided by the square of height in metres. WC was measured at the iliac crest to the nearest millimetre, using a steel tape( 18 ).
Covariable assessment
The covariables were chosen based upon prior studies on the association between 25(OH)D and serum lipids( Reference Jorde, Figenschau and Hutchinson 5 , Reference Li, Yin and Yao 14 , Reference Giovannucci 25 – Reference Jungert, Roth and Neuhauser-Berthold 33 ). Information on age, sex, self-identified ethnicity, level of education, physical activity, smoking behaviour, alcohol consumption and intake of prescribed cholesterol-lowering medication was obtained from the interview. Smoking behaviour was grouped into three categories. ‘Never’ applied to participants who reported never having smoked 100 cigarettes during their lifetime; ‘former’ applied to participants who reported having smoked at least 100 cigarettes during their lifetime but currently did not smoke, whereas ‘current’ applied to those who reported smoking either every day or some days or at least one cigarette per day. To categorize alcohol consumption, the reported number of alcoholic beverages consumed per week was calculated and grouped into: ‘none’ for women and men who reported no consumption of alcoholic beverages at all, ‘moderate’ for women and men who reported consuming at least one but not more than seven or fourteen alcoholic beverages per week, respectively, and ‘heavy’ for women and men who reported consuming more than seven or fourteen alcoholic beverages per week, respectively. For physical activity, all leisure-time activities in the past 30 d which were performed for at least 10 min were recorded. An activity which caused light sweating or slight to moderate increases in breathing or heart rate was considered to be moderate, an activity which caused heavy sweating or large increases in breathing or heart rate was considered to be vigorous. If participants performed both moderate and vigorous activities, the amount of vigorous activities defined the group allocation. The estimated glomerular filtration rate was calculated from serum creatinine using the CKD-EPI equation( Reference Levey, Stevens and Schmid 34 ) and chronic kidney disease was defined as a glomerular filtration rate <60 ml/min per 1·73 m2 ( 35 ).
Statistical analysis
To account for the complex survey design, all analyses were adjusted for sampling probability, stratum and cluster effects using the SAS survey procedures SURVEYMEANS and SURVEYREG (SAS statistical software package version 9.3). A combined 6-year fasting weight for the three cycles was used( Reference Johnson, Paulose-Ram and Ogden 17 ). For proper variance estimation, the analytic sample was examined using the DOMAIN statement( Reference Johnson, Paulose-Ram and Ogden 17 ). The associations between 25(OH)D and the serum lipids were examined with multiple linear regression models, with the continuous lipids used as dependent variables. Normal distribution of the lipid variables was checked with visual inspection of the histograms. All lipids, except TAG, were considered normally distributed and after log transformation (loge) TAG attained a normal distribution as well. As the relationships of the lipids with both 25(OH)D and WC did not prove to be consistently linear, 25(OH)D and WC were categorized. Four 25(OH)D categories were used (<15 ng/ml, 15–<20 ng/ml, 20–<30 ng/ml, ≥30 ng/ml), based on cut-off points used by the US Institute of Medicine (16 and 20 ng/ml( 36 )) and by the US Endocrine Society (20 and 30 ng/ml( Reference Holick, Binkley and Bischoff-Ferrari 37 )). For WC, three categories were used. Normal waist was defined as WC <80 cm in women or <94 cm in men, abdominal overweight as WC of 80–<88 cm in women or 94–<102 cm in men, and abdominal obesity as WC ≥88 cm in women or ≥102 cm in men.
At first, the associations between 25(OH)D and the serum lipids were assessed in models adjusted for WC (main effect models). Subsequently, effect modification by WC was examined in two ways. First, cross-product interaction terms between 25(OH)D and WC categories were added to the main effect models (interaction models). Additionally, the associations between 25(OH)D and the serum lipids were assessed in models stratified by WC category (stratified models). All models were adjusted for age, sex, ethnicity, season of examination, physical activity, alcohol consumption, smoking status, level of education, kidney disease and intake of prescribed cholesterol-lowering medication. Further, to account for the change in the HDL-C measurement method, the HDL-C and LDL-C:HDL-C models were additionally adjusted for survey cycle. Additionally, all models were further adjusted for HOMA-IR. Linear trends across HOMA-IR quartiles were tested and, accordingly, the HDL-C, LDL-C:HDL-C and TAG models were adjusted for HOMA-IR, whereas the total cholesterol and LDL-C models were adjusted for HOMA-IR as well as HOMA-IR squared. In a sensitivity analysis, BMI was used instead of WC to operationalize obesity. Like for WC, three BMI categories were used, with normal weight being defined as BMI<25·0 kg/m2, overweight as BMI=25·0–<30·0 kg/m2 and obesity as BMI≥30·0 kg/m2. Sex- and ethnicity-specific differences in the interaction between 25(OH)D and WC with regard to the lipids were examined by adding three-way interaction terms (25(OH)D category×WC category×sex/ethnicity, together with the three respective cross-product terms) to the interaction models. A two-sided significance level of 0·05 was set, except for the interaction terms. Estimates of interaction effects have larger variances than estimates of additive effects, and thus the power of a statistical test to detect an interaction is lower( Reference Selvin 38 ). To compensate for this, a significance level of 0·1 was chosen for the interaction effects, which is considered more conventional for testing interactions( Reference Ziegler, Pritchett and Wang 39 ).
Results
Characteristics of the study population
The characteristics of the study population in total and according to 25(OH)D category are shown in Table 1. Serum 25(OH)D levels ranged from 3·6 to 79·3 ng/ml, with the mean level being 25·0 ng/ml. Participants with 25(OH)D levels <15 ng/ml were more likely to be female, non-white and inactive and more likely to have been examined between November and April. They were also more likely to be obese or abdominally obese. In participants with lower 25(OH)D levels, mean BMI, WC and HOMA-IR were considerably higher and LDL-C:HDL-C was marginally higher than in participants with higher 25(OH)D levels. No clear trend across the 25(OH)D categories was visible for the other serum lipids.
Characteristics of the study population in total and according to serum 25-hydroxyvitamin D (25(OH)D) category; non-pregnant, fasting adults aged ≥20 years, US National Health and Nutrition Examination Survey waves 2001–2006
AA, associates degree; HOMA-IR, homeostatic model assessment insulin resistance; WC, waist circumference; LDL-C, LDL cholesterol HDL-C, HDL cholesterol.
% indicate column percentages.
* Unweighted N.
† Prescribed cholesterol-lowering medication.
‡ Normal waist, WC <80 cm in women or <94 cm in men; abdominal overweight. WC of 80–<88 cm in women or 94–<102 cm in men; abdominal obesity, WC ≥88 cm in women or ≥102 cm in men.
§ Normal weight, BMI <25·0 kg/m2; overweight, BMI=25·0–<30·0 kg/m2; obesity, BMI≥30·0 kg/m2.
Associations between serum 25-hydroxyvitamin D and serum lipids
The results of the main effect models in the total sample are shown in the online supplementary material, Supplemental Table 1. Lower 25(OH)D levels were significantly associated with lower HDL-C and higher TAG levels, while no significant association was found between 25(OH)D and total cholesterol, LDL-C or LDL-C:HDL-C. The adjusted effect estimates for the 25(OH)D categories from the interaction models, which indicate the mean differences with regard to the reference category of ≥30 ng/ml, are shown in Fig. 1 (β coefficients for total cholesterol, LDL-C, HDL-C and LDL-C:HDL-C; geometric mean ratios for TAG). Compared with participants having 25(OH)D levels ≥30 ng/ml, abdominally obese participants with 25(OH)D levels <15 ng/ml had a 0·15 mmol/l lower HDL-C level, a 0·28 units higher LDL-C:HDL-C and a 12 % higher TAG level (all significant), whereas no significant association was found in abdominally overweight or normal-waist participants. By contrast, lower 25(OH)D levels were associated with lower levels of total cholesterol and LDL-C, but this association was significant only in abdominally overweight participants with 25(OH)D levels between 15 and 20 ng/ml. The interaction between 25(OH)D and WC, however, was significant only for LDL-C:HDL-C (P for interaction=0·02), while no significant difference between the WC categories was found for total cholesterol (P for interaction=0·23), LDL-C (P for interaction=0·19), HDL-C (P for interaction=0·18) and TAG (P for interaction=0·28). The results of the stratified models (Supplemental Table 1) are in accordance with the results from the interaction models.
Interaction between waist circumference (—●—, abdominal obesity; – – ■ – –, abdominal overweight; · · · ▲ · · ·, normal waist) and serum 25-hydroxyvitamin D (25(OH)D) levels in their association with serum lipids among non-pregnant, fasting adults (n 4342) aged ≥20 years, US National Health and Nutrition Examination Survey waves 2001–2006. Adjusted β coefficients, with their 95 % confidence intervals represented by vertical bars, for the 25(OH)D categories from the interaction models on (a) total cholesterol (mmol/l), (b) HDL cholesterol (HDL-C; mmol/l), (c) LDL cholesterol (LDL-C; mmol/l) and (d) LDL-C:HDL-C. (e) Adjusted geometric mean ratios (GMR), with their 95 % confidence intervals represented by vertical bars, for the 25(OH)D categories from the interaction model on TAG (mmol/l). Reference category: 25(OH)D ≥30 ng/ml; all models adjusted for age, sex, ethnicity, season of examination, physical activity, alcohol consumption, smoking status, level of education, kidney disease and intake of prescribed cholesterol-lowering medication. HDL-C and LDL-C:HDL-C models additionally adjusted for survey cycle
Influence of adjustment for insulin resistance
Further adjustment of the models for HOMA-IR, as shown in the online supplementary material, Supplemental Table 2 (stratified models), as well as in Fig. 2 (interaction models: total cholesterol, LDL-C, HDL-C, LDL-C:HDL-C ratio and TAG), had little influence on the association of 25(OH)D with total cholesterol, LDL-C, HDL-C or LDL-C:HDL-C. For TAG, further adjustment for HOMA-IR resulted in changes in the adjusted geometric mean ratio, in particular in abdominally obese participants, where the association between lower 25(OH)D levels and lower TAG levels was attenuated and no longer significant. Consequently, after adjustment for HOMA-IR, the strength of the already non-significant interaction for TAG was further weakened (P for interaction=0·32).
Interaction between waist circumference (—●—, abdominal obesity; – – ■ – –, abdominal overweight; · · · ▲ · · ·, normal waist) and serum 25-hydroxyvitamin D (25(OH)D) levels in their association with serum lipids, additionally adjusted for homeostatic model assessment insulin resistance (HOMA-IR), among non-pregnant, fasting adults (n 4342) aged ≥20 years, US National Health and Nutrition Examination Survey waves 2001–2006. Adjusted β coefficients, with their 95 % confidence intervals represented by vertical bars, for the 25(OH)D categories from the interaction models on (a) total cholesterol (mmol/l), (b) HDL cholesterol (HDL-C; mmol/l), (c) LDL cholesterol (LDL-C; mmol/l) and (d) LDL-C:HDL-C. (e) Adjusted geometric mean ratios (GMR), with their 95 % confidence intervals represented by vertical bars, for the 25(OH)D categories from the interaction model on TAG (mmol/l). Reference category: 25(OH)D ≥30 ng/ml; all models adjusted for age, sex, ethnicity, season of examination, physical activity, alcohol consumption, smoking status, level of education, kidney disease and intake of prescribed cholesterol-lowering medication. HDL-C and LDL-C:HDL-C models additionally adjusted for HOMA-IR and survey cycle; total cholesterol and LDL-C models additionally adjusted for HOMA-IR and HOMA-IR2; TAG model additionally adjusted for HOMA-IR
Sensitivity analyses
No significant differences in the interaction between 25(OH)D category and WC category were found according to sex (all P values for all three-way interaction terms >0·1). However, a significant difference according to ethnicity was found for HDL-C, LDL-C:HDL-C and TAG. In order to examine these differences further, the respective interaction models were stratified by ethnicity. While the interactions between 25(OH)D category and WC category in non-Hispanic whites, the largest ethnic group in our study population, were similar to the interactions in the total sample, the analyses for the other ethnic groups were not sufficiently powered to draw a valid conclusion (data not shown).
Using BMI to operationalize obesity resulted in a different group allocation. Only 31 % of the participants were classified as obese, but 34 % as overweight and 35 % as having a normal weight. As compared with abdominally obese and abdominally overweight participants, respectively, the β coefficients and geometric mean ratios for HDL-C, LDL-C:HDL-C and TAG were smaller in the obese and larger in the overweight participants (see online supplementary material, Supplemental Fig. 2). These differences were particularly strong for TAG, where the strongest association was found in overweight participants and where a significant association was no longer found in abdominally obese participants. For total cholesterol and LDL-C, the β coefficients were bigger in the obese and smaller in the overweight participants as compared with the main analysis (Supplemental Fig. 2).
Discussion
In the present large, cross-sectional sample of adults aged ≥20 years, lower 25(OH)D levels were significantly associated with lower HDL-C levels, higher LDL-C:HDL-C and higher TAG levels in abdominally obese participants, but not in abdominally overweight or normal-waist participants. In contrast, lower 25(OH)D levels were associated with lower levels of total cholesterol and LDL-C in abdominally overweight and normal-waist participants. However, a significant difference in the associations between 25(OH)D and the lipids according to WC category was only found for LDL-C:HDL-C. Further adjustment for HOMA-IR attenuated the association between 25(OH)D and TAG in abdominally obese participants.
Our results are mainly in line with previous cross-sectional studies on the association between serum 25(OH)D and serum lipids. Few studies reported no significant association with any lipid at all( Reference Saedisomeolia, Taheri and Djalali 31 , Reference Chen, Sha and Chen 32 ). By contrast, as in the total, unstratified sample in the present study, 25(OH)D levels in previous studies were found to be significantly positively associated with HDL-C( Reference Jorde, Figenschau and Hutchinson 5 , Reference Li, Yin and Yao 14 , Reference Giovannucci 25 , Reference Fraser, Williams and Lawlor 28 – Reference Park and Lee 30 , Reference Jungert, Roth and Neuhauser-Berthold 33 ) and significantly inversely associated with TAG( Reference Jorde, Figenschau and Hutchinson 5 , Reference Li, Yin and Yao 14 , Reference Giovannucci 25 , Reference Yin, Sun and Zhang 29 , Reference Park and Lee 30 ) in most studies. For total cholesterol and LDL-C, the findings from previous studies are conflicting. A significant positive association with 25(OH)D has been reported before( Reference Jorde, Figenschau and Hutchinson 5 ), but more often, as in the total sample of the present study, a positive but non-significant association was found( Reference Giovannucci 25 , Reference Fraser, Williams and Lawlor 28 ). Other studies reported a significantly inverse association of 25(OH)D with total cholesterol and LDL-C( Reference Karhapaa, Pihlajamaki and Porsti 27 , Reference Yin, Sun and Zhang 29 ) or found no clear relationship( Reference Melamed, Michos and Post 26 , Reference Park and Lee 30 , Reference Jungert, Roth and Neuhauser-Berthold 33 ). The association between 25(OH)D and LDL-C:HDL-C has been examined to a lesser extent, but results from previous studies report a significantly inverse association( Reference Jorde, Figenschau and Hutchinson 5 , Reference Jungert, Roth and Neuhauser-Berthold 33 ), which is in line with our results. Little is known about differences in the associations between 25(OH)D and lipids according to weight or WC status. Only Jorde et al. examined this association stratified by BMI group( Reference Jorde, Figenschau and Hutchinson 5 ). Contrary to our results, a positive association of 25(OH)D with total cholesterol and LDL-C was found in all BMI groups, although it was significant only in overweight and obese individuals. They also found positive associations with HDL-C and negative associations with TAG in all three BMI groups, but strongest in overweight individuals, which is in line with the results of our sensitivity analysis using BMI to operationalize obesity.
Our results suggest that an association between higher 25(OH)D levels and a favourable lipid profile in particular occurs in abdominally obese individuals. There are several mechanisms which could underlie such a relationship. Adiponectin, for instance, an adipokine whose levels are low in obese subjects, was found to have beneficial effects on lipid metabolism, such as on the HDL assembly in the liver, and low levels of adiponectin are associated with dyslipidaemia( Reference Jung and Choi 13 ). 25(OH)D levels were found to be positively associated with adiponectin levels, in particular in subjects with a high BMI, which was also found to be a significant effect modifier in the association between 25(OH)D and adiponectin( Reference Vaidya, Forman and Underwood 12 ). The underlying pathway may be an inhibitory effect of 1,25-dihydroxyvitamin D, the active vitamin D metabolite, on the adipose tissue renin–angiotensin system, which is over-activated during obesity( Reference Vaidya, Forman and Underwood 12 ). Another possible mechanism behind the association between 25(OH)D and lipids is LPL, an enzyme that catalyses the lipolysis of TAG and whose reduced expression and activity is a pathway for dyslipidaemia in obesity( Reference Klop, Elte and Cabezas 40 ). Low levels of LPL result in hypertriacylglycerolaemia, which in turn leads to decreased levels of HDL-C( Reference Jung and Choi 13 , Reference Klop, Elte and Cabezas 40 ). 25(OH)D was found to be significantly positively associated with LPL in a large Chinese cohort( Reference Huang, Li and Wang 11 ). In the same study, both 25(OH)D and LPL were found to be inversely associated with insulin resistance( Reference Huang, Li and Wang 11 ). Insulin is known to stimulate LPL activity and it is also an important regulator for the mobilization of NEFA from the adipose tissue( Reference Klop, Elte and Cabezas 40 ); an uncontrolled release of NEFA is one of the main mechanisms of dyslipidaemia in obesity( Reference Jung and Choi 13 ). In our study, further adjustment for HOMA-IR attenuated the association between 25(OH)D and TAG, which supports the notion of insulin resistance as an underling mechanism. However, this relationship was found only in abdominally obese individuals. Thus, it is possible that the inverse association of 25(OH)D with TAG and, in turn, the positive association with HDL-C and the inverse association with LDL-C:HDL-C, is detectable only during obesity, when insulin sensitivity as well as the LPL action are reduced. In fact, being abdominally obese was previously found to significantly modify the association between 25(OH)D and insulin resistance in data from NHANES 2001–2006( Reference Kabadi, Lee and Liu 41 ). Further, in a meta-analysis of randomized controlled trials, vitamin D supplementation led to an non-significant decrease of TAG and a non-significant increase of HDL-C in obese subjects, while the effect was opposite in normal-weight subjects( Reference Wang, Xia and Yang 8 ). It is thus possible that a significant effect of vitamin supplementation on TAG and HDL-C in obese individuals will be found in one of the large vitamin D trials that are currently being conducted( Reference Meyer, Holvik and Lips 42 ).
The association of lower 25(OH)D levels with lower levels of total cholesterol and LDL-C, which was significant only in abdominally overweight participants with a 25(OH)D level between 15 ng/ml and <20 ng/ml, requires further investigation. To our knowledge, a similar relationship of 25(OH)D with total cholesterol and LDL-C has not been found in previous studies. In the meta-analysis mentioned above, vitamin D supplementation led to an increase in LDL-C, but this was, contrary to our results, significant only in obese subjects( Reference Wang, Xia and Yang 8 ). As our result may be a chance finding, more research is necessary before further conclusions can be drawn.
Our study has several limitations. Due to the cross-sectional design of NHANES, our results cannot give evidence on the causality or direction of the observed associations. Further, our results were dominated by non-Hispanic whites, due to the large proportion of this ethnic group in the study population. Although we tried to minimize confounding by adjusting our models for a variety of covariables, the possibility of residual confounding remains. Specifically, overall physical activity and smoking, which were used as covariables in the current analysis, might not be sufficient to capture a healthy lifestyle involving increased sun exposure due to more frequent outdoor physical activity as well as a diet rich in vitamin D. Such a lifestyle could be the common cause of both higher levels of 25(OH)D and a more favourable lipid profile. We also did not adjust our models for serum parathyroid hormone and calcium, which are both possible confounders in the relationship between vitamin D and CVD. However, in another study using data from NHANES 2001–2006, adjustment for both factors increased, rather than decreased, the strength of the association of 25(OH)D with TAG and HDL-C and only moderately decreased the strength of the already non-significant association with LDL-C( Reference Fraser, Williams and Lawlor 28 ). Finally, we did not correct our analyses for multiple testing, although we examined five different outcomes. As the evaluation of all lipids was planned and, as discussed above, we had a basis for expecting our results to be biologically plausible, we decided to not correct for multiple testing as suggested by Rothman( Reference Rothman 43 ). However, had we corrected our analyses for multiple testing, the association of 25(OH)D with HDL-C and LDL-C:HDL-C in abdominally obese participants as well as the interaction between 25(OH)D and WC on LDL-C:HDL-C would have still been significant.
Conclusion
Our results suggest that lower 25(OH)D levels are more strongly associated with an unfavourable lipid profile in individuals with abdominal obesity than in individuals with abdominal overweight or in normal-waist participants. Given that obese individuals generally have low 25(OH)D levels, vitamin D deficiency may be regarded as a currently unaccounted risk factor for dyslipidaemia in this population group. However, more research is needed to assess whether the interaction found in the present analysis is causal and if vitamin D supplementation is effective for the treatment of dyslipidaemia in obese individuals.
Acknowledgements
Acknowledgements: The authors thank the CDC’s NCHS, Hyattsville, MD, USA, for making the data available for analysis. Financial support: This work was supported by the Kompetenznetz Adipositas (Competence Network Obesity) funded by the German Federal Ministry of Education and Research (grant number FKZ: 01GI1121B). The Competence Network Obesity had no role in the design, analysis or writing of this article. Conflict of interest: None. Authorship: S.V. and R.S. conceived the present study. S.V. performed the statistical analysis and drafted the manuscript. R.S. substantially contributed to the conception and interpretation of the statistical analysis and critically revised the manuscript. J.B., A.P. and B.T. gave advice on designing the present study, substantially contributed to the conception and interpretation of the statistical analysis and critically revised the manuscript. All authors approved of the final version of the manuscript. Ethics of human subject participation: The US NHANES 2001–2006 were reviewed and approved by the NCHS Research Ethics Review Board. Written informed consent was obtained from all participants. The data used for the present study are from anonymized public-use data files available for download on the website of the NCHS.
Supplementary material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S1368980016001762
