Lipid indices as simple and clinically useful surrogate markers for insulin resistance in the U.S. population

This study aimed to compare the accuracy of novel lipid indices, including the visceral adiposity index (VAI), lipid accumulation product (LAP), triglycerides and glucose (TyG) index, TyG-body mass index (TyG-BMI), and TyG-waist circumference (TyG-WC), in identifying insulin resistance and establish valid cutoff values. This cross-sectional study used the data of 11,378 adults, derived from the United States National Health and Nutrition Examination Survey (1999–2016). Insulin resistance was defined as a homeostasis model assessment-insulin resistance value above the 75th percentile for each sex and race/ethnicities. The area under the curves (AUCs) were as follows: VAI, 0.735; LAP, 0.796; TyG index, 0.723; TyG-BMI, 0.823, and; TyG-WC, 0.822. The AUCs for TyG-BMI and TyG-WC were significantly higher than those for VAI, LAP, and TyG index (vs. TyG-BMI, p < 0.001; vs. TyG-WC, p < 0.001). The cutoff values were as follows: VAI: men 1.65, women 1.65; LAP: men 42.5, women 42.5; TyG index: men 4.665, women 4.575; TyG-BMI: men 135.5, women 135.5; and TyG-WC: men 461.5, women 440.5. Given that lipid indices can be easily calculated with routine laboratory tests, these values may be useful markers for insulin resistance risk assessments in clinical settings.

www.nature.com/scientificreports/ These parameters were proposed as a useful surrogate measure of insulin resistance [25][26][27][28] . In addition, several studies have evaluated modified indices that combine TyG index and obesity indices such as body mass index (BMI) and waist circumference (WC) [29][30][31] . However, limited evidence is available regarding the discriminatory accuracy and cutoff values of these novel lipid indices for detecting insulin resistance. Therefore, this study aimed to compare the accuracy of novel lipid indices in identifying insulin resistance using a representative sample of the US population and establish valid cutoff values for IR.

Results
The study included 11,378 adults (men 5478, women 7900; mean age, 40 years) from the National Health and Nutrition Examination Survey (NHANES) 1999-2016 (Fig. 1). Participants' demographic and clinical characteristics were compared based on the presence or absence of IR, and the results are shown in Table 1. Age, BMI,  Table 1. Baseline characteristics of the participants in NHANES 1999-2016. Data are presented as mean ± standard deviation or number (%). BMI body mass index, WC waist circumference, BP blood pressure, HbA1c hemoglobin A1c, HDL high-density lipoprotein; HOMA-IR homeostasis model assessment-insulin resistance, VAI visceral adiposity index, LAP lipid accumulation product, TyG index triglycerides and glucose index. *Insulin resistance was defined as a HOMA-IR value above the 75th percentile for each sex and race/ ethnicity. † Other race included non-Hispanic Asian and multi-racial Americans.  Table 2.
The receiver operating characteristic (ROC) curve for IR is presented in Fig. 2. The AUC was 0.723 for TyG index and 0.735 for VAI (Table 3). The AUC of LAP (0.796) was significantly higher than that of TyG index (p < 0.001). However, the AUCs of TyG-BMI (0.823) and TyG-WC (0.822) were significantly higher than that of LAP (vs. TyG-BMI, p < 0.001; vs. TyG-WC, p < 0.001). Subgroup analysis according to sex and race/ethnicities showed that TyG-BMI and TyG-WC had the highest AUC in every subgroup. Further analysis using 1:1 propensity score matching (PSM) data with age, sex, and race/ethnicities showed similar results ( Table 3 Table 2. Distribution of indirect parameters for insulin resistance according to race/ethnicity and sex. IQR interquartile range, HOMA-IR homeostasis model assessment-insulin resistance, VAI visceral adiposity index, LAP lipid accumulation product, TyG index triglycerides and glucose index, BMI body mass index, WC waist circumference. *Other race included non-Hispanic Asian and multi-racial Americans.  www.nature.com/scientificreports/ cutoff values with their corresponding sensitivity, specificity, and odds ratio (OR) of insulin resistance according to sex and race/ethnicities are summarized in Table 4.

Discussion
In this study, we investigated the discriminatory accuracy of novel lipid indices for IR and confirmed that LAP showed significantly higher AUC than TyG index and VAI. There was a significant increase in AUC when BMI or WC was combined with TyG index, exhibiting an even higher discriminatory accuracy than that of LAP. Another important aspect of this study is that the cutoff value of each parameter for IR was presented using large-scale data, facilitating the clinical application of each parameter. Although the mechanism through which lipid indices cause IR remains unclear, numerous studies have reported that glucolipotoxicity is a key mechanism in the modulation of IR 32,33 . Ectopic lipid accumulation in the liver and skeletal muscle tissue activates pathways that are associated with IR, leading to a decrease in the glucose uptake in muscle tissue and glycogen synthesis in the liver [34][35][36][37] . Insulin resistance in muscle tissue due to ectopic lipid accumulation increases hepatic lipogenesis and leads to IR in the liver and hyperlipidemia [38][39][40] . In addition, macrophage infiltration into white adipose tissue increases lipolysis, which stimulates hepatic triglyceride synthesis, thereby, promoting hyperlipidemia 33 . Macrophage-induced lipolysis in white adipose tissue also leads to increased hepatic gluconeogenesis and results in hyperglycemia through increased fatty acid delivery to the liver, which results in increased glycerol conversion to glucose [41][42][43] .
VAI uses BMI, WC, and triglyceride and HDL cholesterol levels to evaluate IR and was proposed by Amato et al. in 2010 21 . First proposed by Kahn et al. using the NHANES data 22,23 , LAP is calculated using WC and fasting triglyceride levels. In a previous study conducted by Amato et al., VAI showed a significant inverse correlation with insulin sensitivity measured using a HIEG clamp, providing evidence that VAI can be a surrogate marker for IR 21 . In addition, VAI was reported to be associated with the glucose distribution rate evaluated through the HIEG clamp test in a study on patients with type 1 diabetes mellitus (DM) 26 , and it was shown to be inversely correlated with HIEG clamp tested insulin sensitivity in studies conducted on women with polycystic ovary syndrome (PCOS) in South Korea and China 27,44 . In the case of LAP, a small-scaled study with PCOS patients demonstrated a modest inverse correlation with insulin sensitivity measured through HIEG clamp test 27 . However, most of these studies using the HIEG clamp test had a small sample, and the clinical application of VAI and LAP requires an investigation of appropriate cutoff values through large scale population-based studies. In addition to studies measuring insulin resistance directly, there are numerous studies on the accuracy of VAI and LAP in assessing HOMA-IR defined insulin resistance. However, studies on cutoff values were mainly conducted on a small number of PCOS patients [45][46][47] . Thus, a study with a larger population with healthy adults is necessary. The current study is significant as it presents the cutoff values of VAI and LAP by sex, using a large and healthy population. Interestingly, the VAI cutoff value identified in this study is similar to the VAI cutoff value of 1.6-1.8 reported in the previous studies with PCOS patients 27,45,46 . However, in the case of LAP, previous studies have Table 3. Area under the curve for each parameter for insulin resistance according to race/ethnicity and sex. HOMA-IR homeostasis model assessment-insulin resistance, VAI visceral adiposity index, LAP lipid accumulation product, TyG index triglycerides and glucose index, BMI body mass index, WC waist circumference, AUC area under the curve, PSM propensity score matching with age, sex, and race/ethnicity. *Other race included non-Hispanic Asian and multi-racial Americans. www.nature.com/scientificreports/ reported diverse cutoff values ranging from 18.5 to 33.8, and this study shows a higher cutoff value than that reported in the preceding studies 27,45,46 . TyG index was proposed as a useful surrogate measure of IR by Guerrero-Romero et al. in 2008 24 . Despite the small scale of previous studies, TyG index displayed an inverse correlation with insulin sensitivity measured through HIEG clamp and MMAMG 25,28 . Furthermore, various epidemiological studies have reported that TyG index is associated with the incidences of cardiovascular disease (CVD) and DM, indicating that TyG index is able to predict diseases that result from IR [48][49][50] . However, there are a few studies on the estimation of the cutoff value of TyG index for IR. Guerrero-Romero et al. suggested that the best value of the TyG index for the diagnosis of IR was 4.68 using the HIEG clamp test with a small sample size 25 . In a study with Korean NHANES, the cutoff values for metabolic syndrome, which is a pathological condition associate with IR, were 4.76 in men and 4.71 in women 19 . In a prospective cohort study with Korean population, the cutoff value to predict DM was 4.69 49 . In our study, the cutoff value of TyG index was 4.66 in men and 4.57 in women, which is similar to those of previous studies.
Recently, there have been studies on indices that combine adiposity status with the TyG index [29][30][31] . Considering that adipose tissues secrete inflammatory cytokines, adipokines, and reactive oxygen species, contributing to a variety of metabolic problems 51-53 , compound indices with TyG index and obesity parameters such as BMI and WC might be better indicators of IR than TyG index alone. Several studies reported that the compound indices were significantly associated with metabolic abnormalities such as high blood pressure, nonalcoholic fatty liver disease, prediabetes, DM, and hyperuricemia 29-31,54-56 . In addition, recent studies have indicated that TyG-BMI or TyG-WC is more effective in the identification of IR than VAI, LAP, and TyG index 29 31 . However, such studies were conducted only in Asian populationss, and no studies have been conducted on other ethnic populations. Therefore, the current study is meaningful as it confirms using large-scale data that TyG-BMI or TyG-WC can be an effective surrogate marker for IR in the US population Table 4. Cutoff values for each parameter and their corresponding sensitivity, specificity, and odds ratios for insulin resistance. HOMA-IR homeostasis model assessment-insulin resistance, VAI visceral adiposity index, LAP lipid accumulation product, TyG index triglycerides and glucose index, BMI body mass index, WC waist circumference. *Adjusted for Age, race, smoking status, and blood pressure. † Other race included non-Hispanic Asian and multi-racial Americans. www.nature.com/scientificreports/ with various races/ethnicities. Nonetheless, to accurately assess the correlation of IR with TyG-BMI and TyG-WC, verification through the HIEG clamp test is required as has been performed for HOMA-IR and TyG index. The present study has several strengths. This study is the largest to evaluate the performance of the novel lipid indices to identify insulin resistance in the general US population. In addition, this study conducted various subgroup analyses of IR, with age, sex, and PSM data, to minimize the bias caused by heterogeneity due to demographic characteristics. Moreover, it is important to propose valid cutoff values for each lipid index so that they can be used as a reference in clinical settings for identifying groups at risk for IR. To the best of our knowledge, this is the first study that evaluated the performance and cutoff values of TyG-BMI and TyG-WC in a non-Asian population. However, considering that this is a cross-sectional study, further prospective studies are required to validate the relationship between each surrogate measure and cardiovascular risk factors.

Conclusion
The present study supports the clinical relevance of novel lipid indices in identifying IR in the general US population. Considering that lipid indices can be easily calculated with routine laboratory tests, they can be useful markers of insulin resistance risk assessments in clinical settings. Moreover, the cutoff values presented in our study may be useful in interpreting the results of lipid indices for IR.

Methods
Study population. The NHANES is a cross-sectional study that uses a representative sample of the population living in the United States. The NHANES, administered by the Centers for Disease Control and Prevention every two years, consists of health and nutrition surveys, physical examinations, and laboratory tests. Of the 92,062 individuals who participated in the NHANES between 1999 and 2016, 11,378 adult participants were included in this study after excluding those who were aged < 20 years (n = 42,550), those with missing or incomplete anthropometric and fasting laboratory data (n = 28,694), those on medication for dyslipidemia and hypertension (n = 6598), those with DM, CVD, stroke, cancer, chronic liver disease, and estimated glomerular filtration rate < 60 ml/min/1.73 m 2 (n = 2842) (Fig. 1).
Anthropometric and laboratory measurements. Waist circumference was measured using a flexible tape between the uppermost lateral border of the right ilium and that of the left ilium. BMI was defined as the weight in kilograms divided by the height in meters squared (kg/m 2 ). Blood pressure was measured 3 times in the sitting position, with at least 5 min of rest in between each reading. The mean value of the three recorded blood pressure readings was used in this study. Fasting blood glucose and lipid levels were measured using the enzymatic method, and fasting insulin was measured using an immune-enzymometric assay. Detailed sample collection and processing instructions are described in the NHANES Laboratory Procedures Manual 57 .
Calculation of parameters for insulin resistance. Parameters for insulin resistance were calculated as follows [21][22][23][24][25]29 Statistical analysis. Data were presented as mean with standard deviation, or number with prevalence (%) of IR status. Between groups, the differences were determined using t-tests and a Pearson chi-square test. The values of each lipid index for IR were presented as median and interquartile range. To compare the relative diagnostic strength of each lipid index for insulin resistance, AUC was compared using the ROC curve; de Long's test was used to identify the surrogate measures that were significantly superior for insulin resistance. The cutoff value of each lipid index was determined as the value with the highest Youden index score. Considering the heterogeneity of demographic characteristics such as sex and race/ethnicities, subgroup analyses for IR were www.nature.com/scientificreports/ performed. Further analysis was performed by 1:1 PSM with age, sex, and race/ethnicities. This was performed using the "MatchIt" package with nearest-neighbor 1-to-1 matching 59 . Furthermore, OR of HOMA-IR defined IR was checked using the multivariate logistic regression models based on the estimated cutoff values. Statistical analysis was performed using IBM SPSS Statistics ver. 24.0 (IBM Co., Armonk, NY, USA) and R ver. 3.1.0 (R Foundation for Statistical Computing, Vienna, Austria; www.r-proje ct.org). The results were considered statistically significant if the p-value was less than 0.05. . All participants volunteered and provided written informed consent before enrolment. All participants' records were anonymized before being accessed by the authors. All methods were carried out in accordance with the principles contained in the Declaration of Helsinki.

Ethics statement. This study was approved by the institutional review board of Kangnam Sacred
Received: 27 August 2019; Accepted: 8 December 2020