Introduction

Numerous epidemiologic studies have demonstrated a direct correlation between increasing body mass index (BMI) and elevated total cholesterol (TC), low-density lipoprotein cholesterol (LDL), and triglycerides (TG) and an inverse relationship with high density lipoprotein cholesterol (HDL) [1–3]. This association between BMI and lipoprotein levels, particularly LDL, has been suggested to be a contributing factor in the higher rates of cardiovascular events associated with obesity. However, the underrepresentation of individuals with obesity and especially morbid obesity in these studies limits conclusions with regard to expected lipid profiles in this subgroup.

Recently, observational studies of obese patients have confirmed a correlation between BMI and TG or HDL, but not with LDL levels [4]. These findings have raised the question of a possible “obesity paradox” where LDL levels may plateau or even fall with extreme levels of BMI [5, 6]. Therefore, we sought to study the association between BMI and lipoprotein levels in a population with a wide range of BMI subjects including some with morbid obesity.

Methods

Study Population

A retrospective case-control study of patients seen at the Cleveland Clinic Florida was performed. The study was approved by the Institutional Review Board. Consecutive obese patients (cases) were obtained from the obesity surgery database between August 31, 2000, and April 4, 2002. Obesity was defined as a BMI ≥ 30 kg/m2. Consecutive nonobese patients (controls) were obtained from a database of primary care physicians between May 1, 2004, and November 17, 2004.

Data Collection

Patient demographics and medical data were collected from an electronic medical record including the following: age, gender, height, weight, BMI, waist size, menopausal status, current tobacco use, history of diabetes, hypertension, coronary artery disease (CAD), myocardial infarction, and use of hormone replacement therapy, insulin, statin, or other cholesterol-lowering medication. Laboratory tests included a standard fasting lipid profile with TC, LDL, HDL, and TG.

Statistical Analysis

Descriptive analysis for categorical variables with frequency, percentage, and comparisons using chi-square tests were performed. Continuous variables were described as a mean ± standard deviation (SD) and compared by using t tests.

Primary outcomes were TC, LDL, HDL, and TG as continuous variables. For TG, a log-transformed outcome was created because this variable did not have a normal distribution. Pearson coefficients evaluated the correlation between continuous BMI with each of the above-described outcomes. Multivariable linear analysis was performed to evaluate the relation between BMI and each of the lipid measurements while adjusting for potential confounders (age, gender, diabetes, hypertension, current smoking, history of CAD, and statin use).

To further adjust for the potential selection bias when choosing cases and controls, we calculated the probability or propensity score to be a case or a control per patient. The closer the propensity score is to 1, the higher the probability to be a case. The propensity scores were derived from a logistic regression model that included case or control as the outcome and several independent variables (BMI, age, gender, current smoking, history of diabetes, hypertension, CAD, and statin use). We included the propensity scores in all multivariable linear regression analyses. We finally explored the association between BMI and lipids in men and women and in patients with and without use of statins. A p < 0.05 was considered statistically significant. Analyses were performed using SPSS 11.0 and S-Plus 6.1 software packages.

Results

Six hundred thirty-seven patients (female, 67%) were studied. Tables 1 and 2 show the distribution of patient characteristics and cholesterol profiles by BMI levels. The case group is younger with a higher prevalence of female gender and diabetes, whereas the control group has a higher prevalence of CAD and statin use. For the overall population, there was no association between BMI and TC (Pearson correlation coefficient = 0.002, p = 0.96) or LDL (correlation coefficient = 0.078, p = 0.06). There was an association between BMI and HDL (correlation coefficient = -0.267, p < 0.001) and log of TG (correlation coefficient = 0.163, p = 0.006).

Table 1 Summary of categorical variables by BMI
Full size table
Table 2 Summary of continuous variables by BMI
Full size table

Multivariable linear analysis while adjusting for confounders and propensity scores confirmed a statistically significant association between BMI and both HDL and TG (Tables 3 and 4). HDL was also associated with age, gender, and use of statins, whereas TG was associated with history of hypertension, smoking, and use of statins. There was a 0.25 mg/dL decline in HDL per each unit of increase of BMI and a 1 mg/dL per unit increase in TG per unit increase in BMI. The association between BMI and HDL was significant in women only, after adjusting for covariates (p < 0.01; Fig. 1a). Similarly, the association between BMI and TG was significant in women only (p < 0.001; Fig. 1b). LDL levels were not associated with BMI even after adjustment for confounders and propensity scores (Table 5, Fig. 1c and Fig. 2).

Table 3 Multivariable linear model, outcome: HDL cholesterol (n = 597)
Full size table
Table 4 Multivariable linear model, outcome: log of TG (n = 598)
Full size table
Fig. 1
figure 1

a Distribution of HDL cholesterol across different categories of BMI, stratified by gender. b Distribution of the log of TG across different categories of BMI, stratified by gender. c Distribution of LDL across different categories of BMI, stratified by gender

Full size image
Fig. 2
figure 2

Correlation between LDL cholesterol and BMI, stratified by gender

Full size image
Table 5 Outcome LDL (n = 583)
Full size table

Discussion

In a broad range of BMI patients including the morbidly obese, we found a strong positive association between BMI and TG and a significant inverse association with HDL. However, there was a lack of association of BMI to LDL and TC. The lack of a correlation between BMI and LDL in a broad range of BMI patients is in contrast to findings derived from epidemiologic studies that did not include a high proportion of patients with high BMI.

The Framingham study and other large epidemiologic studies have reported a correlation between obesity and increased coronary heart disease-related events [7–9]. The proposed mechanisms for cardiac events in obese patients include increased atherosclerosis related to dyslipidemia, [10] insulin resistance, hypertension, and diabetes [11, 12].The effects of dyslipidemia on cardiac events is generally considered to be largely due to increased TC [13] and LDL cholesterol [14–20]. However, recent data have suggested that this paradigm of a direct correlation between BMI and LDL cholesterol may be oversimplified. In fact, an “obesity paradox” has been suggested with a weaker than expected correlation between LDL and BMI [21 – 26]. In the NHANES III, the cohort with the highest BMI of ≥ 35 kg/m2 had a lower prevalence of hyperlipidemia (TC >240 or self-reported hyperlipidemia) than the three cohorts with lower BMI. A study by Drapeau et al. found that the extremely (BMI >61 kg/m2) and severely (BMI >46 and <60 kg/m2) obese had a better LDL than the moderately obese (BMI >31-45 kg/m2), whereas the HDL mean level did not deteriorate linearly once the obesity BMI threshold (BMI 30 kg/m2) was passed [6].

Some investigators have suggested that LDL particle size but not its level may be a more important determinant of atherogenicity. Small experimental studies have found that changes in metabolic conditions such as obesity affect predominantly LDL particle size more so than LDL absolute levels [6, 27]. Rainwater et al. showed that a BMI >29.2 kg/m2 is inversely and significantly correlated to LDL particle size (r = -0.128). Particle size is affected by insulin resistance, which is largely due to obesity [28–30]. This change in particle size was shown to result in a greater atherogenic milieu independent of LDL level.

The limitations to our study should be noted. This is a retrospective study of modest size deriving obese patients selected for consideration for obesity surgery. The size of the study is significantly smaller than epidemiologic studies. A small correlation between BMI and LDL may have been evident in a larger sample size. Most variables that affect LDL levels were accounted for in our analysis, although the study was underpowered to determine if the association is present in subgroups of patients. Similar to most studies of patients undergoing obesity surgery, the patient population was predominantly young premenopausal females.

In summary, the common belief that increasing BMI results in a substantially higher TC and LDL may not extend to morbidly obese patients. It is clear that the effect of obesity on insulin resistance results in an increase in TG and a lowering of HDL; however, the effect on LDL level and particle size is not well established. Further study is warranted to evaluate this correlation between obesity and lipid particles.