Anthropometric and Biochemical Correlations of Insulin Resistance in a Middle-Aged Maltese Caucasian Population

Background Insulin resistance (IR) is associated with increased cardiovascular disease risk, and with increased all-cause, cardiovascular, and cancer mortality. A number of surrogate markers are used in clinical practice to diagnose IR. The aim of this study was to investigate the discriminatory power of a number of routinely available anthropometric and biochemical variables in predicting IR and to determine their optimal cutoffs. Methods We performed a cross-sectional study in a cohort of middle-aged individuals. We used receiver operator characteristics (ROC) analyses in order to determine the discriminatory power of parameters of interest in detecting IR, which was defined as homeostatic model assessment-insulin resistance ≥2.5. Results Both the lipid accumulation product (LAP) and visceral adiposity index (VAI) exhibited good discriminatory power to detect IR in both males and females. The optimal cutoffs were 42.5 and 1.44, respectively, in males and 36.2 and 1.41, respectively, in females. Serum triglycerides (TG) and waist circumference (WC) similarly demonstrated good discriminatory power in detecting IR in both sexes. The optimal cutoffs for serum TG and WC were 1.35 mmol/L and 96.5 cm, respectively, in men and 1.33 mmol/L and 82 cm, respectively, in women. On the other hand, systolic and diastolic blood pressure, liver transaminases, high-density lipoprotein cholesterol, serum uric acid, ferritin, waist-hip ratio, “A” body shape, thigh circumference, and weight-adjusted thigh circumference all had poor discriminatory power. Conclusions Our data show that LAP, VAI, TG, and WC all have good discriminatory power in detecting IR in both men and women. The optimal cutoffs for TG and WC were lower than those currently recommended in both sexes. Replication studies are required in different subpopulations and different ethnicities in order to be able to update the current cut points to ones which reflect the contemporary population as well as to evaluate their longitudinal relationship with longer-term cardiometabolic outcomes.


Introduction
Hyperinsulinaemia and insulin resistance (IR) are associated with increased cardiovascular disease risk [1][2][3], as well as with increased cardiovascular, cancer, and all-cause mortality [4,5].Te dysfunction associated with insulin resistance is largely restricted to the phosphatidylinositol 3kinase pathway rather than the mitogen-activated protein kinase (MAPK) pathway.Te former mediates the anabolic efects of insulin, whilst the latter mediates the mitogenic and proinfammatory efects of [6].Te hyperinsulinaemia associated with IR therefore fuels increased MAPK pathway activity.Te resultant enhanced mitogenecity probably mediates the increased cancer risk associated with IR.Te chronic subclinical proinfammatory state drives endothelial dysfunction [7], which in turn predisposes to atherosclerosis and to hypertension.Te dyslipidaemia typically associated with IR, namely, decreased and dysfunctional high-density lipoprotein (HDL), increased very low-density lipoprotein, and the generation of small dense and oxidized low-density lipoprotein (LDL), also contributes to the increased cardiovascular risk.Furthermore, oxidized LDL may also increase the risk of certain cancers [8].
However, routine quantifcation of IR is not readily available in clinical practice.Te euglycemic insulin clamp is the gold standard for measuring IR, whereby subjects are given continuous insulin infusion with plasma glucose levels being maintained constant by varying the rate of glucose infusion.Te glucose infusion rate is therefore a measure of insulin sensitivity [9].Whilst being a valuable research tool, this is clearly impractical for use in a clinical setting.A number of surrogate markers of IR have therefore been devised.Tese include the Homeostatic Model Assessment of Insulin Resistance (HOMA-IR) [10] and Quantitative Insulin Sensitivity Check Index (QUICKI) [11].However, both require the measurement of fasting serum insulin which is often not available in routine clinical care.
A number of anthropometric and biochemical parameters are therefore used as surrogate indices of IR.However, data comparing the discriminatory power of these parameters are lacking.Furthermore, the cutofs for each parameter is uncertain, with diferent bodies using diferent cutofs.It is also important to note that these cutofs were developed around 20 years ago.Many secular changes might have contributed to change in the optimal cut-ofs of the various parameters to predict IR.Tese include changes in dietary habits [12,13], an increase in adiposity [14][15][16], changes in body fat distribution patterns [16,17], and a decrease in muscle mass [18,19].Additionally, anthropometric-based indices of IR require crosspopulation replication and validation to account for regional diferences in body composition and obesity prevalence which are partly determined by population genetic structure.
Te aim of the present study was to determine the discriminatory power of the various anthropometric and biochemical parameters in predicting IR and to determine the optimal cutofs using receiver operating characteristic (ROC) analysis in a contemporary population.IR was defned as HOMA-IR ≥ 2.5.We used this cutof since it has been shown to predict increased mortality in large population-based studies [20,21].In view of sex diferences in the relationship of anthropometric and biochemical parameters with IR [22], we investigated males and females separately.

Methods
Tis was a cross-sectional study consisting of 521 middleaged, (41 ± 10 years) non-institutionalized individuals of Maltese Caucasian descent.Te sample population was identifed after a letter of invitation (recruitment letter) was sent either electronically (via email) or via post to individuals who ft the eligibility criteria for this survey.Initially, invites were sent to employees who worked at the Mater Dei Teaching Hospital in Malta, and subsequently, a convenience type of sampling was carried out whereby the recruited individual (index person) was allowed to invite other colleagues/friends/family members as new participants in the study through word of mouth or by passing on the recruitment letter via email or post.Tis method of recruitment was similar to that used by Buscemi et al. in the ABCD study [23].Te exclusion criteria were active malignancy or terminal illness, type 1 diabetes, pregnancy, genetic or endocrine causes of overweight or underweight (apart from controlled thyroid disorders), and inability to give voluntary informed consent.Figure 1 shows the fowchart of participant recruitment.A dedicated questionnaire was used to capture baseline demographic data relating to age, sex, past medical and surgical history, and a detailed drug history including use of antihypertensives, and hypolipemic agents.
Anthropometric measurements were taken with the participants dressed in light clothing and without shoes, using validated equipment which was calibrated in accordance with WHO recommendations [24].Body weight was measured to the nearest 0.1 kg, whilst height and all circumferences to the nearest 0.1 cm.Height was measured using a calibrated stadiometer.Body mass index (BMI) was calculated as the weight (in kg) divided by the square of the height (in meters).Waist circumference (WC) was measured over the abdomen halfway between the bottom of the rib cage and superior iliac crest; the hip circumference (HC) was measured over the widest diameter around the buttocks with participants standing with their feet together and after full expiration.Te neck circumference (NC) was measured at level of the mid-cervical spine [25].Te mid-upper arm circumference (MUAC) was measured at the midpoint of the distance between the acromion and olecranon process, with the elbow fexed at a 90 °and the arm held parallel to the side of the body.Te thigh circumference (TC) was   Journal of Nutrition and Metabolism measured at the level of the gluteal fold with the thigh muscles fully relaxed.All circumferences were taken with the subjects standing upright, with shoulders and thighs relaxed, facing the investigator [26].Waist-to-height ratio (WHtR), waist-to-hip ratio (WHR), waist-to-thigh ratio, and arm-toheight ratio were calculated as WC (cm)/height (cm), WC (cm)/HC (cm), WC (cm)/TC (cm), and MUAC/height, respectively.Te conicity index (CI) was calculated using the formula CI = waist/(0.109× √weight (kg)/height (m) [27], the body adiposity index (BAI) was calculated using the formula BAI = (HC/height 3/2 ) − 18 [28], abdominal volume index (AVI) was measured according to the formula AVI = (2 cm (waist) 2 + 0.7 cm (waist-hip) 2 )/1000 [29], whilst A-type body shape index (ABSI) was measured using the formula ABSI = WC/BMI 2/3 × height ½ [30].Blood pressure was measured after 5 minute rest in the seated position using a clinically validated digital sphygmomanometer with an appropriately sized cuf for each participant and using the average of the second and third readings for analyses, in accordance with the European Society of Hypertension Guidelines [31].
Fasting plasma glucose (FPG), HbA 1c , lipid, and other biochemical parameters including renal profle, liver function tests, and thyroid function tests were assayed.All investigations were performed at the Biochemistry Laboratory of Mater Dei Hospital, Malta using automated and regularly calibrated analysers.FPG and lipid profle assays were performed using COBAS INTEGRA ® Analysers (Roche diagnostics, Basel, Switzerland) machines.Fluoridecontaining tubes were used for the collection of samples for the estimation of FPG so as to reduce glycolysis.Te FPG assay was based on hexokinase and glucose oxidase enzyme reactions.Blood for the lipid profle assessment was collected in a serum clot activator tube.Te lipid variables measured were total cholesterol, HDL-cholesterol (HDL-C), LDL-cholesterol (LDL-C), and triglycerides (TG).Haemoglobin A 1c was measured using the Bio-Rad variant II HbA 1c program (California, USA), which utilises the principle of high pressure liquid chromatography.Insulin was quantifed by sandwich ELISA (Diagnostic Automation, USA) using a Mithras ® microplate reader for absorbance determination as per the manufacturer's instructions.Samples were assayed in duplicate using 50 μL of serum HOMA-IR was calculated using the formula: fasting serum insulin (μU/ml) × fasting plasma glucose (mmol/L)/22.5 [10].
All participants gave their written informed consent stating willingness to participate in this study as well as to undergo physical examination and biochemical testing.Te statistical signifcance of diferences in proportions was assessed using the two proportions z test.Spearman's rankorder coefcient was used to explore the strength and direction of association between quantitative variables.
Receiver operating characteristic (ROC) analysis was used to compute the area under curve (AUC) to assess the performance of anthropometric and biochemical parameters, and indices derived thereof, in discriminating subjects with insulin resistance (defned by the categorical cutof HOMA-IR ≥ 2.5).
Te highest Youden index (sensitivity + specifcity − 1) was used to determine optimal cutof points.ROC analysis was performed using the easyROC R application [36], and cutof values were determined using the OptimalCutpoints R package [37].All analyses were performed using IBM SPSS version 26 and R v.3.4.2.A p value of <0.05 was considered signifcant.

Results
Five hundred and twenty-one subjects participated in the study (331 females and 190 males).Te median (interquartile range) age was 41 (6.0) years.Table 1 shows subject characteristics stratifed by sex and HOMA-IR ≥ 2.5.As expected, subjects with HOMA-IR ≥ 2.5 had higher BMI, WC, FPG, and TG but a lower HDL-C.Tere was also a higher proportion of use of antihypertensive medication in both males and females with HOMA-IR ≥ 2.5 and of lipidlowering pharmacotherapy in females with HOMA-IR ≥ 2.5. Figure 2 shows a correlation matrix of HOMA-IR with quantitative anthropometric and biochemical indices.As expected, there were signifcant positive correlations between HOMA-IR and anthropometric or biochemical indices of adiposity.
In males, the LAP had the best discriminatory power to detect IR (area under the curve (AUC) � 0.79) (Table 2 and Figure 3).Te highest Youden index for LAP corresponds to a value of 42.5, with a sensitivity of 86% and a specifcity of 63%.Te VAI, TG/HDL-C ratio, and TG also had good discriminatory power (AUC of 0.780.79 and 0.75, respectively) (Table 2 and Figure 3).A value of VAI of 1.44 had 86% sensitivity and 65.8% specifcity, whilst a triglyceride level of 1.33 mmol/L had a sensitivity of 76.2% and a specifcity of 63.7%.
In females, VAI, LAP, and the TG/HDL-C ratio had equivalent discriminatory power to detect IR (AUC of 0.82 for VAI and TG: HDL-C ratio and 0.81 for LAP) (Table 2 and Figure 4).A value of LAP of 36.2 had a sensitivity of 75.5% and a specifcity of 80.4% to detect IR, a value of VAI of 1.41    had 79.6% sensitivity and 77.8% specifcity, whilst a TG/ HDL-C ratio of 0.78 had a sensitivity of 77.6% and a specifcity of 76.9%.TG also had good discriminatory power (AUC � 0.78), with a value of triglyceride level of 1.35 mmol/ L having a sensitivity of 65.3% and a specifcity of 85.9%.

Journal of Nutrition and Metabolism
Of the anthropometric parameters in females, the WC had the best discriminatory power (AUC of 0.76), followed closely by BMI (AUC 0.74) (Table 1).Te optimal cutof for WC to predict insulin resistance in females was 82 cm with a sensitivity of 85.7% and a specifcity of 53.3%.Te optimal cutof for BMI in females was to 31.9 kg/m 2 , with a sensitivity of 59% and a specifcity of 80%.
In males, BMI (AUC � 0.73) and waist circumference (AUC � 0.70) were the strongest anthropometric predictors of insulin resistance (Table 2).Te optimal cutof for WC to predict insulin resistance in males was 96.5 cm with a sensitivity of 72.1% and a specifcity of 60.3%, while the optimal cut-of for BMI in males was 29.1 kg/m 2 , with a sensitivity of 74.4% and a specifcity of 64.4%.Waist-hip ratio, BAI, AVI, FI, HDL-C, serum uric acid, liver transaminase, and weightadjusted thigh circumference all had poor discriminatory power, whereas ferritin levels, systolic and diastolic blood pressure and "A" body shape did not exceed signifcance thresholds in ROC analysis.

Discussion
Te lipid accumulation product (LAP), which incorporates both the WC and TG in its calculation, exhibited the highest discriminatory power in males and also performed very well     in females.We also found that both the WC and TG individually had good discriminatory power in both sexes.Te WC is a well-established marker of central adiposity, which in turn is strongly associated with IR.Although it also measures abdominal subcutaneous fat, which is believed to be less unhealthy than visceral fat, we found that WC is a strong predictor of IR in both sexes.It performed better than the BMI, which is consistent with previous data [38].WC has also been shown to predict incident type 2 diabetes [39,40] and cardiovascular disease independently of BMI [41,42].TG levels have been shown to exhibit a strong independent association with IR [43] and with type 2 diabetes [44,45].Insulin stimulates lipoprotein lipase activity; IR therefore results in reduced lipoprotein lipase activity [46][47][48], leading to increased triglyceride levels.Since circulating non-esterifed fatty acids (also known as free fatty acids) are a major determinant of hepatic triglyceride production and packaging into very low density lipoprotein [49][50][51], serum triglyceride levels may be marker of free fatty acid levels.Te latter are thought to be causally related to insulin resistance [52,53].Tey also inhibit lipoprotein lipase [54], resulting in a further increase in circulating triglyceride levels.
HDL-C had poor discriminatory power in both men and women, and TG/HDL-C ratio was not signifcantly better than TG on its own.HDL-C exhibits a higher heritability than other lipids [55].It also has much higher heritability when compared to IR [56], implying that environmental factors that afect IR have much less impact on HDL-C.Furthermore, many of the genetic polymorphisms that have been shown to afect HDL-C concentrations would not be expected to afect insulin resistance [57][58][59][60].It should also be noted that, although epidemiological data show that low HDL-C is associated with increased cardiovascular disease, most known genetic variants that afect HDL-C levels do not increase cardiovascular disease risk [57,61,62].Dysfunctional HDL may be more important in identifying insulin resistance [63][64][65], but this is not captured by measuring HDL-C levels.
We found optimal cutofs for WC in both males and females to be lower than those currently in use.In fact the optimal cut-ofs were 96.5 cm in males and 82 cm in females.Te National Cholesterol Education Program/Adult Treatment Panel III (NECP) recommends a cutof of 102 cm in males and 88 cm in females in its defnition of metabolic syndrome [66].Meig's et al. [67] and Hamer and Stamatakis [68] use the same cutofs in their defnition of metabolic health.Tese NCEP cutofs were developed over 20 years ago.Tere is evidence of secular changes in body fat distribution [16,17] and a decrease in muscle mass [18,19].Tis might have contributed to IR occurring at a lower WC than previously.Furthermore, a decline in serum testosterone levels has been reported in males [69,70].Since low androgen levels are associated with a greater increase in visceral fat area compared to subcutaneous fat area [71], this may also have contributed to IR occurring at a smaller WC.
We found that in our contemporary cohort, the optimal cutof for TG to predict insulin resistance was 1.35 mmol/L in males and 1.33 mmol/L in females, which is much lower than the 1.7 mmol/L recommended by NCEP-ATPIII [66] and many others.Tere are surprisingly little data to support the use of 1.7 mmol/L cutof.Indeed, there is evidence that cardiovascular disease risk starts to increase at much lower levels.For example, a triglyceride level >0.68 mmol/l was found to predict incident cardiovascular disease risk in Korean subjects [72].Te best cutof for nonfasting triglycerides to predict ischaemic heart disease in Japanese subjects has recently been reported to be 1.24 mmol/L [73].

Strengths and Limitations.
We studied a reasonablysized, well-characterized, representative cohort of middleaged individuals.Te decrease in muscle mass and function and the changes in fat distribution with ageing make it likely that the discriminatory power of the various parameters and their respective optimal cutofs are diferent in elderly individuals.We therefore believe that it is important to study diferent age groups separately.We used standard procedures in measuring anthropometric parameters, whilst biochemical parameters were analysed in a single laboratory with appropriate quality controls.
Our study was a cross-sectional one, and it is therefore not possible to make any conclusions on intermediate or long-term outcomes.Although the euglycaemic clamp is usually considered to be the gold standard measure of IR, there is a very good correlation with HOMA-IR [10,74].Furthermore, values of HOMA-IR ≥ 2.5 as we used in our study have been reported to predict with increased mortality [20,21].
We studied Maltese Caucasians since all other racial groups are underrepresented in our population.Our results therefore need to be replicated in other subpopulations.

Conclusions
Our data show that in a Maltese Caucasian middle-aged population, both the LAP and VAI exhibited good discriminatory power to detect IR (defned as HOMA-IR ≥ 2.5) in both sexes.Te optimal cutofs in males were 47.4 and 1.64, respectively, whilst in females the optimal cutofs were 36.1 and 1.42, respectively.TG and WC also had good discriminatory power in both sexes, but with lower cutofs than those currently recommended by NCEP-ATPIII.In fact, the optimal cutofs for TG were 1.35 mmol/ L in males and 1.33 mmol/L in females, whilst those for WC were 96.5 cm in males and 82 cm females.Our results therefore suggest that current cutofs need to be revised downwards in this population, and future longitudinal studies are required to investigate further their relationship with hard outcomes such as type 2 diabetes, cardiovascular disease, and mortality.

2
Ethical and data protection approvals were granted from the University of Malta Research Ethics Committee (Ref No: 06/ 2016) of the Faculty of Medicine and Surgery and the Information and Data Protection Commissioner respectively.2.1.Statistical Methods.Normality of continuous variables was assessed by the Shapiro-Wilk and Kolmogorov-Smirnov tests.All continuous parameters exhibited a skewed nonnormal distribution, and nonparametric statistics with medians and interquartile ranges are presented.

Table 1 :
Patient characteristics by gender and HOMA-IR ≥ 2.5

Figure 3 :
Figure 3: Receiver operator characteristic curves for lipid accumulation index, visceral adiposity index, and waist circumference to predict insulin resistance (HOMA-IR > 2.5) in males.

Figure 4 :
Figure 4: Receiver operator characteristic curves for lipid accumulation index, visceral adiposity index, and waist circumference to predict insulin resistance (HOMA-IR > 2.5) in females.

Table 2 :
Area under the curve of receiver operator characteristics curves for various anthropometric parameters for predicting HOMA-IR ≥ 2.5, stratifed by sex.