Abstract
Background
Increased triglycerides (TG) levels and decreased high-density lipoprotein cholesterol (HDL-C) levels are risk factors for type 2 diabetes (T2D). However, previous studies reached inconsistent results between ratio of TG-to-HDL-C (TG/HDL-C) and T2D risk, and it was unclear whether the association was modified by gender, body mass index, or fasting status.
Methods
Blood biomarkers, including TG and HDL-C levels, were assayed among 571 T2D cases and 571 controls in a case-control study nested within Singapore Chinese Health Study. Participants were free of diagnosed diabetes, cardiovascular disease, and cancer at blood collections (1999-2004). Incident self-reported T2D cases were identified at follow-up II interview (2006-2010). Controls were 1:1 matched for age, sex, dialect group and date of blood collection. Multivariable logistic regression models were used to compute the odds ratio (OR) and 95% confidence interval (CI) between lipid profile and T2D risk with adjustment for diabetes risk factors.
Results
The OR (95% CI) of T2D was 1.70 (1.39-2.09) per 1-SD increment in TG, and 1.72 (1.37-2.17) per 1-SD increment in TG/HDL-C ratio. The relations were stronger among female than male (P for interaction = 0.028 and 0.017, respectively), and stronger among lean participants (<23 kg/m2) than their overweight/obese counterparts (P for interaction = 0.051 and 0.046, respectively). TG and TG/HDL-C ratio improved T2D prediction with similar magnitude.
Conclusions
TG and TG/HDL-C ratio are independent risk factors for incident T2D, and they confer greater risk in women and in lean participants. TG/HDL-C ratio is not a better predictor of diabetes than TG alone.
Keywords: HDL cholesterol, nested case-control study, triglycerides, type 2 diabetes mellitus
Introduction
Dyslipidemia, especially increased triglycerides (TG) levels together with decreased high-density-lipoprotein cholesterol (HDL-C) levels, is strongly correlated with insulin resistance and type 2 diabetes (T2D).1 Mechanistic studies suggested that high TG and low HDL-C levels may be causal factors for T2D,2 and this inference was supported by various epidemiological and genetic studies. Prospective human studies have found that high TG and low HDL-C levels are independent risk factors for T2D, and TG and/or HDL-C values have been included in models to predict T2D risk.3, 4 Moreover, a recent Mendelian randomization study showed that genetic predisposition to high TG levels or low HDL-C levels were causally related to elevated T2D risk.5
Although recent studies demonstrated that the TG-to-HDL-C (TG/HDL-C) ratio could be better predictor for insulin resistance and cardiovascular disease than TG or HDL-C alone,6, 7 the association between TG/HDL-C ratio and T2D is not entirely clear yet. So far only a few prospective studies examined this issue,8-13 and the extent to which TG/HDL-C ratio is independently related to T2D risk has been variable. Some studies found positive association,8-12 while one study showed null association.13 Moreover, few studies have explored whether the association is modified by gender,8, 10 and fasting status.10 One study reported a slightly stronger association between TG/HDL-C ratio in women compared to men,8 although formal interaction test was not performed. Another study found association between TG/HDL-C ratio and T2D risk in both men and women, regardless of fasting status.10 Additionally, previous studies have found ethnic differences in ability of TG/HDL-C ratio to identify insulin resistance.6, 14 Since Asians develop T2D at lower body mass index (BMI) compared to western populations,15 lipids may play a more important role in T2D development in relatively lean Asian populations. However, to our best knowledge, no study has explored whether the association between TG/HDL-C ratio and T2D risk is modified by BMI. Moreover, it is unknown whether TG/HDL-C ratio adds substantial predictive information for T2D. Two studies reported the predictive ability of TG/HDL-C ratio together with other risk factors for T2D, without presenting discrimination improvement by TG/HDL-C ratio alone.8 Another study reported TG/HDL-C ratio improved T2D prediction, however, the sample size was small (74 T2D cases).9
To address these issues, we conducted a nested case-control study within the Singapore Chinese Health Study (SCHS) to examine the relationship between TG/HDL-C ratio and risk of incident T2D as well as possible interaction with gender, obesity and fasting status. Moreover, we evaluated the predictive utility of TG/HDL-C ratio, TG alone, HDL-C alone, on top of other diabetes risk factors.
Methods
Study population
The design of the SCHS has been described previously.16 Briefly, the SCHS was established between 1993 and 1998, and recruited 63,257 Chinese adults aged 45-74 years. At recruitment, an in-person interview was conducted using a structured questionnaire to collect health-related information. Follow-up I was conducted via telephone between 1999 and 2004 to update selected lifestyle habit and medical history. A total of 52,322 participants were re-contacted successfully, and 32,535 consenting participants donated their bio-specimens. Follow-up II was conducted from 2006 to 2010, and 39,528 participants were re-contacted successfully. Among the 32,535 participants with available bio-specimens, 25,477 (78.3%) were contacted in the follow-up II. The study protocol was approved by the Institutional Review Boards at the National University of Singapore.
Ascertainment of diabetes and other covariates
History of physician-diagnosed diabetes was asked at baseline and both follow-ups by the question: “Have you been told by a doctor that you have diabetes?” If the answer was “yes”, participants were also asked for the age at which they were first diagnosed. The robustness and accuracy (98.9%) of the self-reported diabetes data was validated in a separate study analyzing 1651 cohort participants (949 cases by medical records and 702 via phone interview).17
Body weight and height were self-reported at baseline and both follow-ups. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. Alcohol consumption was coded as never, weekly and daily drinking and smoking status as never, former, and current smoking based on the follow-up I questionnaires. Weekly moderate-to-vigorous activity levels (<0.5, 0.5-3.9, and ≥4.0 hours/week) and education level (no, primary school, secondary school and above) were obtained from the baseline questionnaires.
Case and control selection
A nested case-control study of 571 cases and 571 matched controls within SCHS was established for the current analysis. All cases and controls were free of physician-diagnosed diabetes, cardiovascular disease and cancer at baseline interview as well as blood collection (1999-2004). Cases were those who reported to be diagnosed with incident diabetes during follow-up II visits (2006-2010). Controls were selected from the remaining participants who were free of T2D at follow-up II, and were matched for age (±3 years), date (±6 months) of blood collection, sex, and dialect group with the cases on a 1:1 ratio. In addition, the selected controls were screened for the presence of undiagnosed T2D at the time of blood donation by measuring haemoglobin A1c (HbA1C) level. All matched controls with HbA1C ≥6.0% were ineligible for the study and excluded, and a replacement control with the same matching criteria was randomly chosen among the remaining eligible controls.
Laboratory Procedures
20-mL peripheral blood was obtained from each consenting participant. The tubes were put on ice during transport from the participants’ homes to the laboratory immediately after the blood collection. All specimens were separated into various components (plasma, serum, red blood cells, and buffy coat). All specimens were subsequently stored in −80°C freezers for long-term storage. Frozen plasma aliquots from cases and controls were selected for simultaneous analysis at the same batch at the National University Hospital Reference Laboratory.
Plasma concentrations of total cholesterol, TG, HDL-C and high-sensitivity C-reactive protein (CRP) were measured via colorimetric method on a chemistry analyzer (AU5800 Analyzer, Beckman Coulter, Brea, California). Adiponectin levels were measured by ELISA (Bio-Rad Laboratories, Hercules, California). LDL-C levels were calculated using the Friedewald formula: LDL cholesterol = total cholesterol – HDL cholesterol – triglycerides/2.2; where TG did not exceed 4.5 mmol/L. Thus, 44 participants had missing values for LDL cholesterol because of high TG levels.
Statistical Analysis
Based on the distribution among control participants, lipid measures (HDL-C, LDL-C, TG, and TG/HDL-C ratio) were divided into tertiles, and the lowest tertile served as the reference group. We used conditional logistic regression to model the association between lipid measures and T2D risk with adjustment for age at blood collection (continuous), smoking status (never, past, and current smoker), alcohol intake (never, weekly or daily), weekly moderate-to-vigorous activity levels (<0.5, 0.5-3.9, and ≥4.0 hours/week), education level (no, primary school, secondary or above), history of hypertension (yes, no), fasting status (yes, no), BMI (continuous), and plasma concentrations of CRP and adiponectin in tertiles. We also calculated the odds ratio of T2D per standard deviation (SD) increment in lipid measures with the same adjustment abovementioned. We tested potential interactions between lipid measures and gender, obesity, and fasting status. We used unconditional logistic regression models with additional adjustment for matching factors on gender (male, female) and dialect group (Cantonese, Hokkien) when performing stratified analysis on obesity and fasting status.
To assess the predictive utility of each lipid measure, we established a parsimonious logistic regression model including education level, weekly activity, history of hypertension, and BMI using a forward selection procedure (P <0.05). The improvement in discrimination was examined by comparing area under receiver-operating characteristic curve (AUC) between the parsimonious model and the model plus one lipid measure using the method from DeLong.18 Moreover, due to the limitation of AUC such as its insensitivity to model improvement,19 we also evaluated the category-free net reclassification improvement (NRI) and integrated discrimination improvement (IDI) statistics.20 NRI is calculated from a sum of differences between higher predicted risk for events and lower predicted risk for non-events,20 while IDI is based on differences in integrated average sensitivities and ‘one minus specificities’ between models with and without the new biomarker.20 Furthermore, the goodness-of-fit of all models were assessed by Akaike information criteria (AIC), where a lower AIC value indicates better model fit. Additionally, we tested the predictive utility of these lipid measures in different gender, and BMI groups. Analyses were performed with Stata software, version 11.0 (Stata Corp, College Station, Texas).
Results
Among T2D cases, the mean age of diagnosis (SD) was 63.2 (6.4) years and the mean duration (SD) between blood donation and diagnosis of T2D was 4.0 (1.7) years. The baseline characteristics of cases and controls are shown in Table 1. Cases had high-risk profiles except for the matching factors. They had higher BMI, and were more likely to have history of hypertension. No significant differences were found for education level, cigarette smoking, alcohol consumption, physical activity levels and fasting status. Cases had higher baseline levels of CRP, TG, and TG/HDL-C ratio, and lower adiponectin and HDL-C levels. LDL-C levels were not diff0erent between cases and controls.
Table 1. Baseline characteristics and plasma biomarkers of diabetes cases and matched controls.*.
| Cases (n =571) | Controls (n =571) | P-value | |
|---|---|---|---|
| Age (years) at blood taken | 59.6 ± 6.1 | 59.7 ± 6.2 | - |
| Gender (Female) | 335 (58.7) | 335 (58.7) | - |
| Dialect (%) | - | ||
| Cantonese | 287 (50.3) | 287 (50.3) | |
| Hokkien | 284 (49.7) | 284 (49.7) | |
| Body mass index (kg/m2) | 24.8 ± 3.6 | 22.8 ± 3.3 | <0.001 |
| Level of education (%) | 0.26 | ||
| No formal education | 104 (18.2) | 99 (17.3) | |
| Primary school | 255 (44.7) | 233 (40.8) | |
| Secondary and above | 212 (37.1) | 239 (41.9) | |
| History of Hypertension (%) | 265 (46.4) | 148 (25.9) | <0.001 |
| Cigarette smoking (%) | 0.15 | ||
| Never smokers | 410 (71.8) | 425 (74.4) | |
| Former smoker | 63 (11.0) | 71 (12.4) | |
| Current smokers | 98 (17.2) | 75 (13.1) | |
| Weekly moderate activity (%) | 0.11 | ||
| <0.5 hours/week | 456 (79.9) | 454 (79.5) | |
| 0.5-3.9 hours/week | 82 (14.4) | 68 (11.9) | |
| ≥4 hours/week | 33 (5.8) | 49 (8.6) | |
| Alcohol Intake (%) | 0.81 | ||
| Abstainers | 498 (87.2) | 497 (87.0) | |
| Weekly drinkers | 55 (9.6) | 59 (10.3) | |
| Daily drinkers | 18 (3.2) | 15 (2.6) | |
| Fasting status (yes) | 178 (31.2) | 156 (27.3) | 0.15 |
| TC, mmol/L | 5.3 ± 1.0 | 5.2 ± 0.9 | 0.05 |
| HDL-C, mmol/L | 1.1 ± 0.2 | 1.2 ± 0.3 | <0.001 |
| LDL-C, mmol/L | 3.2 ± 0.9 | 3.2 ± 0.8 | 0.77 |
| TG, mmol/L | 2.2 (1.5-3.0) | 1.5 (1.1-2.2) | <0.001 |
| TG/HDL-C ratio | 2.0 (1.3-3.1) | 1.3 (0.8-2.1) | <0.001 |
| Adiponectin, μg/mL | 7.0 ± 2.7 | 9.1 ± 3.8 | <0.001 |
| C-reactive protein, mg/L | 1.8 (1.0-3.5) | 1.2 (0.6-2.3) | <0.001 |
Abbreviations: TC, total cholesterol; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; TG, triglycerides; TG/HDL-C ratio, triglyceride-to-high density lipoprotein cholesterol ratio.
Cases and controls are matched on age at blood taken (±3 years), gender, dialect, and date of blood collection (±6 months). Data are expressed as mean ± standard deviation for continuous variables (normally distributed) and median (interquartile range) for continuous variables (skewed distributed), and n (percentage) for categorical variables. P values based on the McNemar's Chi-square test for categorical variables, paired Student's t-test for normally-distributed continuous variables and Wilcoxon signed-rank test for skewed continuous variables.
Among the healthy control participants, TG/HDL-C ratio was inversely correlated with levels of HDL-C, LDL-C, and plasma adiponectin levels (Pearson's coefficient r = -0.76, -0.17, and -0.49, respectively; all P <0.001), and positively correlated with TG level, BMI and plasma CRP levels (Pearson's coefficient r =0.95, 0.19 and 0.14, respectively; all P <0.01; Supplementary table 1).
The associations between lipid measures and risk of incident T2D were presented in Table 2. Higher levels of TG and TG/HDL-C ratio were positively associated with risk of T2D, while increased HDL-C levels were inversely associated with T2D risk. However, LDL-C was not significantly associated with T2D risk. In the final model, T2D risk was increased by 70% per 1-SD increment in TG (odds ratio [OR] 1.70; 95% confidence interval [CI], 1.39-2.09), and 72% per 1-SD increment of TG/HDL-C ratio (OR 1.72; 95% CI, 1.37-2.17), while decreased by 32% per 1-SD increment in HDL-C (OR 0.68; 95% CI, 0.56-0.82).
Table 2. Odds ratios (95% confidence intervals) of type 2 diabetes associated with different levels of plasma lipid measures.
| Variables | Tertiles of lipid measures | P for trend* | R2 | Per 1 SD increment | ||
|---|---|---|---|---|---|---|
|
| ||||||
| T1 | T2 | T3 | ||||
| HDL-C | ||||||
| Cases/controls | 294/191 | 195/190 | 82/190 | |||
| Model 1† | 1.00 | 0.67 (0.49-0.90) | 0.27 (0.18-0.39) | <0.001 | 10% | 0.52 (0.44-0.62) |
| Model 2‡ | 1.00 | 0.71 (0.51-0.98) | 0.30 (0.20-0.44) | <0.001 | 14% | 0.55 (0.46-0.66) |
| Model 3§ | 1.00 | 0.82 (0.58-1.15) | 0.47 (0.31-0.73) | 0.001 | 18% | 0.68 (0.56-0.82) |
| LDL-C | ||||||
| Cases/controls | 187/189 | 182/187 | 167/186 | |||
| Model 1† | 1.00 | 1.06 (0.76-1.47) | 0.89 (0.60-1.30) | 0.62 | 6% | 1.03 (0.90-1.18) |
| Model 2‡ | 1.00 | 1.07 (0.75-1.52) | 0.87 (0.61-1.24) | 0.43 | 11% | 1.04 (0.91-1.20) |
| Model 3§ | 1.00 | 1.03 (0.70-1.52) | 0.88 (0.60-1.29) | 0.51 | 17% | 1.08 (0.93-1.26) |
| TG | ||||||
| Cases/controls | 81/194 | 158/187 | 332/190 | |||
| Model 1† | 1.00 | 2.23 (1.52-3.28) | 5.29 (3.56-7.88) | <0.001 | 13% | 2.09 (1.71-2.55) |
| Model 2‡ | 1.00 | 2.15 (1.42-3.24) | 5.21 (3.42-7.94) | <0.001 | 17% | 2.01 (1.65-2.46) |
| Model 3§ | 1.00 | 1.61 (1.04-2.50) | 3.59 (2.30-5.61) | <0.001 | 20% | 1.70 (1.39-2.09) |
| TG/HDL-C ratio | ||||||
| Cases/controls | 69/191 | 171/190 | 331/190 | |||
| Model 1† | 1.00 | 2.51 (1.70-3.72) | 5.35 (3.57-8.01) | <0.001 | 13% | 2.22 (1.79-2.77) |
| Model 2‡ | 1.00 | 2.42 (1.59-3.66) | 5.14 (3.36-7.89) | <0.001 | 17% | 2.14 (1.71-2.68) |
| Model 3§ | 1.00 | 1.87 (1.21-2.89) | 3.31 (2.09-5.23) | <0.001 | 19% | 1.72 (1.37-2.17) |
Abbreviations: R2, the coefficient of determination; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; TG, triglycerides; TG/HDL-C ratio, the ratio of triglyceride to high-density lipoprotein cholesterol.
Linear trend was tested using the median level of each tertile of blood lipid profile.
Model 1: adjusted for age at blood taken (continuous), smoking (never, past, current smoker), alcohol intake (never, weekly or daily), weekly activity (<0.5, 0.5-3 hours/week, and ≥4 hours/week), education level (no, primary school, secondary or above), history of hypertension (yes, no), and fasting status (yes, no);
Model 2: Model 1 plus adjusted for body mass index (continuous);
Model 3: Model 2 plus adjusted for plasma concentrations of C-reactive protein (mmol/L) and adiponectin (μg/mL) (both in tertiles).
We further performed stratified analysis to explore whether the observed associations were modified by gender (Table 3), BMI (Table 4), or fasting status (Supplementary table 2). The associations of TG and TG/DHL-C ratio with T2D risk were stronger among women than among men (P for interaction = 0.028 and 0.017, respectively), and stronger among lean participants (<23 kg/m2) than among overweight/obese participants (≥23 kg/m2) (P for interaction = 0.051 and 0.046, respectively). The associations were consistent across all other subgroups and interaction tests were not statistically significant.
Table 3. Odds ratios (95% confidence intervals) of type 2 diabetes associated with tertile levels of plasma lipid measures in men and women separately.*.
| T1 | T2 | T3 | P for trend† | Per 1 SD increment | |
|---|---|---|---|---|---|
| HDL-C | |||||
| Men | |||||
| Cases/controls | 116/79 | 84/79 | 36/79 | ||
| OR (95% CI) | 1.00 | 0.89 (0.54-1.47) | 0.56 (0.29-1.05) | 0.09 | 0.71 (0.53-0.94) |
| Women | |||||
| Cases/controls | 188/112 | 101/112 | 46/111 | ||
| OR (95% CI) | 1.00 | 0.73 (0.44-1.19) | 0.48 (0.26-0.88) | 0.02 | 0.69 (0.53-0.90) |
| LDL-C | |||||
| Men | |||||
| Cases/controls | 78/77 | 71/77 | 66/76 | ||
| OR (95% CI) | 1.00 | 1.19 (0.65-2.15) | 0.89 (0.46-2.15) | 0.82 | 1.06 (0.83-1.36) |
| Women | |||||
| Cases/controls | 111/112 | 107/111 | 103/109 | ||
| OR (95% CI) | 1.00 | 0.92 (0.53-1.61) | 0.81 (0.49-1.34) | 0.44 | 1.03 (0.84-1.27) |
| TG‡ | |||||
| Men | |||||
| Cases/controls | 46/79 | 69/79 | 121/78 | ||
| OR (95% CI) | 1.00 | 1.09 (0.60-1.98) | 2.25 (1.16-4.36) | 0.007 | 1.61 (1.16-2.24) |
| Women | |||||
| Cases/controls | 42/115 | 86/111 | 207/109 | ||
| OR (95% CI) | 1.00 | 1.82 (0.98-3.39) | 5.03 (2.58-9.81) | <0.001 | 1.85 (1.40-2.44) |
| TG/HDL-C ratio§ | |||||
| Men | |||||
| Cases/controls | 46/79 | 69/79 | 121/78 | ||
| OR (95% CI) | 1.00 | 1.01 (0.53-1.90) | 1.98 (1.04-3.79) | 0.01 | 1.61 (1.12-2.34) |
| Women | |||||
| Cases/controls | 35/112 | 80/112 | 220/111 | ||
| OR (95% CI) | 1.00 | 1.68 (0.87-3.26) | 4.98 (2.48-10.0) | <0.001 | 1.80 (1.34-2.39) |
Abbreviations: OR, odds ratio; CI, confidence interval; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; TG, triglycerides; TG/HDL-C ratio, the ratio of triglyceride to high-density lipoprotein cholesterol.
Multivariate model adjusted for age at blood taken (continuous), smoking (never, past, current smoker), alcohol intake (never, weekly or daily), physical activity (<0.5, 0.5-3.9, and ≥4 hours/week), education level (no, primary school, secondary or above), history of hypertension (yes, no), fasting status (yes, no), BMI (continuous), plasma concentrations of C-reactive protein (mmol/L) and adiponectin (μg/mL) (both in tertiles).
Linear trend was tested using the median level of each tertile of blood lipid profile.
P for interaction between TG and sex is 0.028.
P for interaction between TG/HDL and sex is 0.017.
Table 4. Odds ratios (95% confidence intervals) of type 2 diabetes associated with tertile levels of plasma lipid measures stratified by level of body mass index (BMI).*.
| T1 | T2 | T3 | P for trend† | Per 1 SD increment | |
|---|---|---|---|---|---|
| HDL-C | |||||
| BMI <23 kg/m2 | |||||
| Cases/controls | 95/106 | 69/106 | 22/106 | ||
| OR (95% CI) | 1.00 | 0.94 (0.59-1.50) | 0.40 (0.21-0.76) | 0.01 | 0.65 (0.50-0.85) |
| BMI ≥23 kg/m2 | |||||
| Cases/controls | 189/85 | 122/84 | 74/84 | ||
| OR (95% CI) | 1.00 | 0.69 (0.45-1.06) | 0.55 (0.45-1.06) | 0.02 | 0.75 (0.61-0.93) |
| LDL-C | |||||
| BMI <23 kg/m2 | |||||
| Cases/controls | 58/105 | 61/106 | 53/103 | ||
| OR (95% CI) | 1.00 | 1.02 (0.62-1.66) | 0.90 (0.55-1.49) | 0.58 | 1.10 (0.89-1.36) |
| BMI ≥23 kg/m2 | |||||
| Cases/controls | 120/83 | 130/84 | 114/81 | ||
| OR (95% CI) | 1.00 | 1.09 (0.71-1.68) | 1.05 (0.67-1.63) | 0.24 | 1.04 (0.87-1.25) |
| TG‡ | |||||
| BMI <23 kg/m2 | |||||
| Cases/controls | 18/107 | 54/105 | 114/106 | ||
| OR (95% CI) | 1.00 | 2.56 (1.33-4.93) | 5.06 (2.65-9.68) | <0.001 | 1.88 (1.45-2.44) |
| BMI ≥23 kg/m2 | |||||
| Cases/controls | 75/85 | 101/84 | 209/84 | ||
| OR (95% CI) | 1.00 | 1.24 (0.76-2.00) | 2.14 (1.33-3.43) | 0.001 | 1.33 (1.06-1.67) |
| TG/HDL-C ratio§ | |||||
| BMI <23 kg/m2 | |||||
| Cases/controls | 19/106 | 50/106 | 117/106 | ||
| OR (95% CI) | 1.00 | 2.30 (1.21-4.40) | 4.85 (2.51-9.35) | <0.001 | 1.88 (1.42-2.49) |
| BMI ≥23 kg/m2 | |||||
| Cases/controls | 61/85 | 125/84 | 199/84 | ||
| OR (95% CI) | 1.00 | 1.69 (1.03-2.75) | 2.38 (1.44-3.95) | 0.002 | 1.30 (1.01-1.67) |
Abbreviations: BMI, body mass index; OR, odds ratio; CI, confidence interval; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; TG, triglycerides; TG/HDL-C ratio, the ratio of triglyceride to high-density lipoprotein cholesterol.
Multivariate model adjusted for age at blood taken (continuous), gender (male, female), dialect group (Cantonese, Hokkien), smoking (never, past, current smoker), alcohol intake (never, weekly or daily), physical activity (<0.5, 0.5-3.9, and ≥4 hours/week), education level (no, primary school, secondary or above), history of hypertension (yes, no), fasting status (yes, no), plasma concentrations of C-reactive protein (mmol/L) and adiponectin (μg/mL) (both in tertiles).
Linear trend was tested using the median level of each tertile of blood lipid profile.
P for interaction between TG and BMI (<23 kg/m2, ≥ 23.0 kg/m2) is 0.051.
P for interaction between TG/HDL and BMI (<23 kg/m2, ≥ 23.0 kg/m2) is 0.046.
The summary statistics for the predictive performance of lipid measures was presented in Table 5. Compared with the AUC of the base model, inclusion of HDL-C, TG or TG/HDL-C ratio significantly improved AUCs (P <0.05) with similar magnitude. Furthermore, the positive values of NRIs and IDIs also suggested that inclusion of lipid measures improved T2D risk prediction significantly (all P <0.001). Although TG had slightly higher NRI and IDI than TG/HDL-C ratio and HDL-C, overall the three lipid measures had similar improvement in T2D prediction. Moreover, the NRI table stratified for T2D cases and controls was presented in Supplementary table 3. Specifically, the inclusion of HDL-C resulted in 24.3% of T2D cases correctly assigned to a higher predicted T2D risk and 6.5% of controls correctly assigned to a lower predicted T2D risk. When TG and TG/HDL-C ratio was added to the model separately, 0.9% and -5.4% of cases were reclassified as higher T2D risk, while 40.4% and 42.9% of controls were reclassified as lower T2D risk, respectively. When stratified by gender, the improvement of three lipid measures in T2D prediction was only found in women (Supplementary table 4). However, when stratified by BMI level, the performance of lipid measures was consistent across all subgroups (Supplementary table 5).
Table 5. Summary statistics to evaluate the improved classification of type 2 diabetes by plasma lipid measures.
| Variable | Multivariate model* | |||
|---|---|---|---|---|
|
| ||||
| Discrimination (AUC [95% CI]) | Calibration (AIC) | NRI (%) | IDI (%) | |
| Base model | 0.69 (0.66-0.72) | 670 | ||
| HDL-C | 0.72 (0.69-0.74)† | 622 | 30.8 | 3.64 |
| TG | 0.73 (0.70-0.75)† | 615 | 41.3 | 4.31 |
| TG/HDL-C ratio | 0.72 (0.70-0.75) † | 618 | 37.5 | 3.99 |
Abbreviations: AUC, area under ROC curve; CI, confidence interval; AIC, Akaike information criterion; NRI, net reclassification improvement; IDI, integrated discrimination improvement; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; TG, triglycerides; TG/HDL-C ratio, the ratio of triglyceride to high-density lipoprotein cholesterol.
Base model included education level (no, primary school, secondary and above), physical activity (<0.5, 0.5-3.9, ≥4 hours/week), history of hypertension (yes, no), and body mass index (continuous).
Multivariate model adjusted for all the variables included in the base model.
Compare with the base model, the increment in AUC value is statistically significant (P <0.05).
Discussion
In this case-control study nested within the Singapore Chinese Health Study, we found that TG and TG/HDL-C ratio were associated with increased risk of incident T2D, while HDL-C was related to a lower risk of T2D. Moreover, the positive associations with TG and TG/HDL-C ratio were significantly stronger among women and lean participants compared to their counterparts. Furthermore, TG, HDL-C, and TG/HDL-C ratio all significantly improved T2D risk prediction, but TG/HDL-C ratio was not a better predictor than TG alone.
Our finding of positive association between TG/HDL-C ratio and T2D risk was supported by previous studies in Chinese,9 Iranian,8, 12 Japanese,10 and Caucasian men,11 despite the heterogeneity in statistical methods (studies modeled TG/HDL-C ratio as continuous variable,8, 9 binary variable with a cutoff point,11, 12 or quartiles10). However, one prospective study showed null association.13 This study was conducted in a group of Iranian people (without diabetes) with family history of diabetes, and participants had much higher BMI and TG/HDL-C ratio compared to the current study (mean BMI: 29 vs. 23; TG/HDL-C ratio: 4.0 vs 1.3); therefore, TG/HDL-C ratio may not predict T2D risk in this specific high-risk population. In our study, we found similar strength of association with TG and TG/HDL-C ratio for diabetes risk, which was in accordance with previous studies.8, 9 Moreover, LDL-C was not associated with T2D risk in our study, which was in agreement with the findings from the Framingham Heart Study.21
The possible causal relation between high TG and low HDL-C levels and T2D risk was supported by several lines of evidence. Lipotoxicity, inflammation and endoplasmic reticulum stress are three widely-accepted mechanisms causing insulin resistance.22-24 Recent data indicate that high TG levels provoke lipotoxicity, cause overload of free fatty acids levels in the skeletal muscle and pancreas, and lead to insulin resistance, beta-cell dysfunction and apoptosis.25 Meanwhile, high TG levels could directly promote inflammation or endoplasmic reticulum stress.1 Additionally, low HDL-C levels may influence glucose homeostasis through reducing insulin secretion, insulin sensitivity and direct glucose uptake by muscle via the AMP-activated protein kinase.22 Results from prospective studies across ethnic groups also reported that high TG levels and low HDL-C levels were independent risk factors for T2D risk.3, 4 Moreover, clinical trials conducted in high-risk populations showed that drugs to decrease TG levels and increase HDL-C levels (e.g. bezafibrate) can lower the T2D risk.26 Furthermore, a recent Mendelian randomization study testing 25 loci (account for 10% of the genetic variation) further supported the causal relation between high TG and low HDL-C levels with T2D risk,5 although a previous Mendelian randomization study with fewer loci (10 loci accounting for 3-5% of the genetic variation) implied no causal relation.27 Taken together, these findings suggested that we may be able to prevent T2D by modulating TG and HDL-C levels.
In addition, our study showed significant gender differences in the relation of TG and TG/HDL-C ratio with T2D risk, and the associations were significantly higher in women than men. Similarly, the same finding was also reported in an Iranian population, although formal interaction test was not performed.8 In addition, the observed gender difference in the association was also supported by other studies of TG and T2D risk prediction in a Japanese population,10 and of TG/HDL-C ratio and insulin resistance in population of primarily European ancestry.28 However, further studies are warranted to confirm this finding in other populations. The underlying mechanism for the observed gender difference is unclear and we speculate that it may be due to sex hormones. Decline in estrogen levels after menopause leads to dysregulation of glucose and lipid metabolism,29 and thus may put women under greater risk of developing T2D. Similar pattern has been observed for the association between TG levels and increased cardiovascular disease risks where it is stronger in women than men,30 and the National Cholesterol Education Program has suggested more aggressive targets for HDL-C and TG in women to prevent cardiovascular disease compared to men.31 Therefore, the finding from the present study, along with previous studies,8, 10, 28 implied that the gender difference should be considered in T2D prevention as well.
Moreover, our study found slightly higher relations of TG and TG/HDL-C ratio with T2D risk in lean people than their heavier counterparts. Although the underlying mechanism is unclear yet, we thought that one possible explanation for the observed interaction with obesity might be due to the differences in the dominant pathogenesis pathway driving T2D development between lean and obese individuals. In lean people, dyslipidemia may play a more important causal role in T2D development, while in obese counterparts, other mechanism related to adiposity, such as low-grade inflammation, may play a more prominent role. Alternatively, in obese people, the increased T2D risk by increased TG may be attenuated by other mechanisms. For example, adiponectin was associated with significantly decreased risk of T2D in obese people, but its effect was less apparent in non-obese counterparts in Western and Asian populations.32, 33 Nevertheless, the sample size for the subgroup analysis was small and the results should be interpreted cautiously.
The finding also highlights that lean appearance does not necessarily equal to being healthy. Therefore, it is important for clinical practice to identify lean people with high TG levels as high risk population of developing T2D, and lean people may receive greater benefit from decreasing T2D risk by altering lipid levels compared to overweight/obese people. However, Asians tend to develop T2D at lower BMI compared to western populations,15 which may partly due to their propensity to store fat viscerally rather than subcutaneously.34 In addition, compared with non-Asian populations, Asians are more insulin resistant even in relatively lean individuals, and had lower beta cell function to overcome insulin resistance.35 Due to the biological differences between Asians and non-Asians, whether the observed interaction with BMI in the current study remains in non-Asian populations is unclear.
Our study found that TG, HDL-C, and TG/HDL-C ratio improved T2D risk prediction with similar magnitude. In lind with our study, Hadaegh et al8 also found same improvements in AUCs by TG, HDL-C (only in women) and TG/HDL-C ratio. Additionally, He et al9 also found similar AUC improvements between TG and TG/HDL-C, and the value was similar to the present study. Although recent studies suggested that TG/HDL-C ratio could be better predictor for insulin resistance and cardiovascular disease than TG or HDL-C alone,6, 7, 36 the superiority of TG/HDL-C ratio in disease prediction was not observed for T2D.
The strength of the current study was the prospective design and hence the presumed lack of recalled bias in exposure data (questionnaires, collected bio-specimen) prior to T2D diagnosis. However, there are some limitations as well. First, we measured TG and HDL-C only once at baseline, thus may not represent long-term lipid profile. However, these would lead to non-differential misclassification and may underestimate the association. Second, incident diabetes was obtained from self-reported information, although our validation study suggested that individuals who reported to have diabetes were more likely to be true diabetes cases, undiagnosed diabetes may still exist in the population. Third, more than 70% of blood samples were non-fasting, and therefore may influence the lipid levels. However, we performed stratified analysis among fasting and non-fasting group, and found similar associations between lipid measures and T2D risk in two groups, indicating non-fasting status did not influence the associations in the present study. Recent European clinical guideline also concluded that “fasting is not routinely required for determination of a lipid profile”. 37 Moreover, current study used matched case-control study design, although it is valid in studying associations, recent studies pointed out it may introduce bias when studying predictive utility of biomarkers. 38, 39
Conclusion
In conclusion, the current study demonstrated a strong, dose-dependent relation of TG and TG/HDL-C ratio with T2D risk, and the association was stronger in women and in lean participants compared to their counterparts. Additionally, both TG and TG/HDL-C ratio improved T2D risk prediction significantly, but TG/HDL-C ratio was not a better predictor than TG alone. Our findings suggest that women and lean people may receive greater benefit from decreasing T2D risk by altering lipid levels compared to men and overweight/obese people. Further research is needed to validate the findings and investigate the biochemical and genetic mechanisms for the associations and the heterogeneity observed between TG levels and the development of T2D.
Supplementary Material
Key points.
-
Significant findings of the study
TG and TG/HDL-C ratio was independently associated with increased risk of incident type 2 diabetes. Both associations were significantly stronger in women and lean participants compared to their counterparts.
TG and TG/HDL-C ratio improved T2D prediction with similar magnitude.
-
What this study adds
This study conducted formal interaction tests and showed that the association between TG/HDL-C ratio and T2D risk was stronger in women and in lean participants. Additionally, the predictive utility of TG/HDL-C ratio for T2D is no better than TG alone.
Acknowledgments
We thank Siew-Hong Low of the National University of Singapore for supervising the fieldwork of the Singapore Chinese Health Study, and Renwei Wang for the maintenance of the cohort study database. We also thank the founding principal investigator of the Singapore Chinese Health Study, Mimi C. Yu. This study was supported by the National Medical Research Council, Singapore (NMRC/CIRG/1354/2013) and National Institutes of Health, USA (RO1 CA144034 and UM1 CA182876).
Footnotes
Disclosure: No conflicts of interest declared.
References
- 1.Li N, Fu J, Koonen DP, Kuivenhoven JA, Snieder H, Hofker MH. Are hypertriglyceridemia and low HDL causal factors in the development of insulin resistance? Atherosclerosis. 2014;233:130–8. doi: 10.1016/j.atherosclerosis.2013.12.013. [DOI] [PubMed] [Google Scholar]
- 2.von Eckardstein A, Sibler RA. Possible contributions of lipoproteins and cholesterol to the pathogenesis of diabetes mellitus type 2. Curr Opin Lipidol. 2011;22:26–32. doi: 10.1097/MOL.0b013e3283412279. [DOI] [PubMed] [Google Scholar]
- 3.Wilson PW, Meigs JB, Sullivan L, Fox CS, Nathan DM, D’Agostino RB., Sr Prediction of incident diabetes mellitus in middle-aged adults: the Framingham Offspring Study. Arch Intern Med. 2007;167:1068–74. doi: 10.1001/archinte.167.10.1068. [DOI] [PubMed] [Google Scholar]
- 4.Chien K, Cai T, Hsu H, et al. A prediction model for type 2 diabetes risk among Chinese people. Diabetologia. 2009;52:443–50. doi: 10.1007/s00125-008-1232-4. [DOI] [PubMed] [Google Scholar]
- 5.Qi Q, Liang L, Doria A, Hu FB, Qi L. Genetic predisposition to dyslipidemia and type 2 diabetes risk in two prospective cohorts. Diabetes. 2012;61:745–52. doi: 10.2337/db11-1254. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.McLaughlin T, Reaven G, Abbasi F, et al. Is there a simple way to identify insulin-resistant individuals at increased risk of cardiovascular disease? Am J Cardiol. 2005;96:399–404. doi: 10.1016/j.amjcard.2005.03.085. [DOI] [PubMed] [Google Scholar]
- 7.Kannel WB, Vasan RS, Keyes MJ, Sullivan LM, Robins SJ. Usefulness of the triglyceride-high-density lipoprotein versus the cholesterol-high-density lipoprotein ratio for predicting insulin resistance and cardiometabolic risk (from the Framingham Offspring Cohort) Am J Cardiol. 2008;101:497–501. doi: 10.1016/j.amjcard.2007.09.109. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Hadaegh F, Hatami M, Tohidi M, Sarbakhsh P, Saadat N, Azizi F. Lipid ratios and appropriate cut off values for prediction of diabetes: a cohort of Iranian men and women. Lipids Health Dis. 2010;9:85. doi: 10.1186/1476-511X-9-85. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.He S, Wang S, Chen X, et al. Higher ratio of triglyceride to high-density lipoprotein cholesterol may predispose to diabetes mellitus: 15-year prospective study in a general population. Metabolism. 2012;61:30–6. doi: 10.1016/j.metabol.2011.05.007. [DOI] [PubMed] [Google Scholar]
- 10.Fujihara K, Sugawara A, Heianza Y, et al. Utility of the triglyceride level for predicting incident diabetes mellitus according to the fasting status and body mass index category: the Ibaraki Prefectural Health Study. J Atheroscler Thromb. 2014;21:1152–69. doi: 10.5551/jat.22913. [DOI] [PubMed] [Google Scholar]
- 11.Vega GL, Barlow CE, Grundy SM, Leonard D, DeFina LF. Triglyceride-to-high-density-lipoprotein-cholesterol ratio is an index of heart disease mortality and of incidence of type 2 diabetes mellitus in men. J Investig Med. 2014;62:345–9. doi: 10.2310/JIM.0000000000000044. [DOI] [PubMed] [Google Scholar]
- 12.Sadeghi M, Talaei M, Parvaresh Rizi E, Dianatkhah M, Oveisgharan S, Sarrafzadegan N. Determinants of incident prediabetes and type 2 diabetes in a 7-year cohort in a developing country: The Isfahan Cohort Study. J Diabetes. 2015;7:633–41. doi: 10.1111/1753-0407.12236. [DOI] [PubMed] [Google Scholar]
- 13.Janghorbani M, Amini M. Utility of serum lipid ratios for predicting incident type 2 diabetes: the Isfahan Diabetes Prevention Study. Diabetes Metab Res Rev. 2015 doi: 10.1002/dmrr.2770. [DOI] [PubMed] [Google Scholar]
- 14.Sumner AE, Finley KB, Genovese DJ, Criqui MH, Boston RC. Fasting triglyceride and the triglyceride-HDL cholesterol ratio are not markers of insulin resistance in African Americans. Arch Intern Med. 2005;165:1395–400. doi: 10.1001/archinte.165.12.1395. [DOI] [PubMed] [Google Scholar]
- 15.Yoon KH, Lee JH, Kim JW, et al. Epidemic obesity and type 2 diabetes in Asia. Lancet. 2006;368:1681–8. doi: 10.1016/S0140-6736(06)69703-1. [DOI] [PubMed] [Google Scholar]
- 16.Hankin JH, Stram DO, Arakawa K, et al. Singapore Chinese Health Study: development, validation, and calibration of the quantitative food frequency questionnaire. Nutr Cancer. 2001;39:187–95. doi: 10.1207/S15327914nc392_5. [DOI] [PubMed] [Google Scholar]
- 17.Odegaard AO, Pereira MA, Koh WP, Arakawa K, Lee HP, Yu MC. Coffee, tea, and incident type 2 diabetes: the Singapore Chinese Health Study. Am J Clin Nutr. 2008;88:979–85. doi: 10.1093/ajcn/88.4.979. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44:837–45. [PubMed] [Google Scholar]
- 19.Pepe MS, Janes H, Longton G, Leisenring W, Newcomb P. Limitations of the odds ratio in gauging the performance of a diagnostic, prognostic, or screening marker. Am J Epidemiol. 2004;159:882–90. doi: 10.1093/aje/kwh101. [DOI] [PubMed] [Google Scholar]
- 20.Pencina MJ, D’Agostino RB, Sr, D’Agostino RB, Jr, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med. 2008;27:157–72. doi: 10.1002/sim.2929. discussion 207-12. [DOI] [PubMed] [Google Scholar]
- 21.Wilson PW, Kannel WB, Anderson KM. Lipids, glucose intolerance and vascular disease: the Framingham Study. Monogr Atheroscler. 1985;13:1–11. [PubMed] [Google Scholar]
- 22.Drew BG, Rye KA, Duffy SJ, Barter P, Kingwell BA. The emerging role of HDL in glucose metabolism. Nat Rev Endocrinol. 2012;8:237–45. doi: 10.1038/nrendo.2011.235. [DOI] [PubMed] [Google Scholar]
- 23.Kraegen EW, Cooney GJ, Turner N. Muscle insulin resistance: a case of fat overconsumption, not mitochondrial dysfunction. Proc Natl Acad Sci U S A. 2008;105:7627–8. doi: 10.1073/pnas.0803901105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Ikonen E. Cellular cholesterol trafficking and compartmentalization. Nat Rev Mol Cell Biol. 2008;9:125–38. doi: 10.1038/nrm2336. [DOI] [PubMed] [Google Scholar]
- 25.Unger RH, Zhou YT. Lipotoxicity of beta-cells in obesity and in other causes of fatty acid spillover. Diabetes. 2001;50(Suppl 1):S118–21. doi: 10.2337/diabetes.50.2007.s118. [DOI] [PubMed] [Google Scholar]
- 26.Tenenbaum A, Motro M, Fisman EZ, Schwammenthal E, Adler Y, Goldenberg I, et al. Peroxisome proliferator-activated receptor ligand bezafibrate for prevention of type 2 diabetes mellitus in patients with coronary artery disease. Circulation. 2004;109:2197–202. doi: 10.1161/01.CIR.0000126824.12785.B6. [DOI] [PubMed] [Google Scholar]
- 27.De Silva NM, Freathy RM, Palmer TM, et al. Mendelian randomization studies do not support a role for raised circulating triglyceride levels influencing type 2 diabetes, glucose levels, or insulin resistance. Diabetes. 2011;60:1008–18. doi: 10.2337/db10-1317. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Salazar MR, Carbajal HA, Espeche WG, et al. Relation among the plasma triglyceride/high-density lipoprotein cholesterol concentration ratio, insulin resistance, and associated cardio-metabolic risk factors in men and women. Am J Cardiol. 2012;109:1749–53. doi: 10.1016/j.amjcard.2012.02.016. [DOI] [PubMed] [Google Scholar]
- 29.Faulds MH, Zhao C, Dahlman-Wright K, Gustafsson JA. The diversity of sex steroid action: regulation of metabolism by estrogen signaling. J Endocrinol. 2012;212:3–12. doi: 10.1530/JOE-11-0044. [DOI] [PubMed] [Google Scholar]
- 30.Hokanson JE, Austin MA. Plasma triglyceride level is a risk factor for cardiovascular disease independent of high-density lipoprotein cholesterol level: a meta-analysis of population-based prospective studies. J Cardiovasc Risk. 1996;3:213–9. [PubMed] [Google Scholar]
- 31.National Cholesterol Education Program. Second Report of the Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel II) Circulation. 1994;89:1333–445. doi: 10.1161/01.cir.89.3.1333. [DOI] [PubMed] [Google Scholar]
- 32.Yamamoto S, Matsushita Y, Nakagawa T, Hayashi T, Noda M, Mizoue T. Circulating adiponectin levels and risk of type 2 diabetes in the Japanese. Nutr Diabetes. 2014;4:e130. doi: 10.1038/nutd.2014.27. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Wannamethee SG, Lowe GD, Rumley A, Cherry L, Whincup PH, Sattar N. Adipokines and risk of type 2 diabetes in older men. Diabetes Care. 2007;30:1200–5. doi: 10.2337/dc06-2416. [DOI] [PubMed] [Google Scholar]
- 34.Misra A, Khurana L. Obesity and the metabolic syndrome in developing countries. J Clin Endocrinol Metab. 2008;93:S9–30. doi: 10.1210/jc.2008-1595. [DOI] [PubMed] [Google Scholar]
- 35.Chan JC, Yeung R, Luk A. The Asian diabetes phenotypes: challenges and opportunities. Diabetes Res Clin Pract. 2014;105:135–9. doi: 10.1016/j.diabres.2014.05.011. [DOI] [PubMed] [Google Scholar]
- 36.Narayan KM. Type 2 Diabetes: Why We Are Winning the Battle but Losing the War? 2015 Kelly West Award Lecture Diabetes Care. 2016;39:653–63. doi: 10.2337/dc16-0205. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.da Luz PL, Favarato D, Faria-Neto JR, Jr, Lemos P, Chagas AC. High ratio of triglycerides to HDL-cholesterol predicts extensive coronary disease. Clinics (Sao Paulo) 2008;63:427–32. doi: 10.1590/S1807-59322008000400003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Pepe MS, Fan J, Seymour CW, Li C, Huang Y, Feng Z. Biases introduced by choosing controls to match risk factors of cases in biomarker research. Clin Chem. 2012;58:1242–51. doi: 10.1373/clinchem.2012.186007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Ganna A, Reilly M, de Faire U, Pedersen N, Magnusson P, Ingelsson E. Risk prediction measures for case-cohort and nested case-control designs: an application to cardiovascular disease. Am J Epidemiol. 2012;175:715–24. doi: 10.1093/aje/kwr374. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.