Development and Validation of a Method to Estimate Insulin Sensitivity in Patients With and Without Type 1 Diabetes

Context:

People with type 1 diabetes (T1D) have markedly reduced insulin sensitivity (IS) compared to their nondiabetic counterparts, and reduced IS is linked to higher cardiovascular risk.

Objective:

This study aimed to develop and validate an improved method for estimating IS in people with T1D.

Design:

Prospective cohort.

Setting:

Adults (36 with T1D, 41 nondiabetic) were recruited from the Coronary Artery Calcification in Type 1 Diabetes (CACTI) study for measurement of IS by hyperinsulinemic-euglycemic clamp to develop a clinically useful IS prediction equation (eIS) for T1D and nondiabetic individuals. These equations were then compared with previously published equations from the SEARCH and Pittsburgh Epidemiology of Diabetes Complications studies for the ability to predict measured IS in test sets of adults and adolescents from independent clamp studies.

Intervention:

None.

Main Outcome Measure:

Comparison of clamp-measured IS to estimated IS.

Results:

The best-fit prediction model (eIS) differed by diabetes status and included waist circumference, triglycerides, adiponectin, and diastolic blood pressure in all CACTI adults and insulin dose in adults with T1D (adjusted R² = 0.64) or fasting glucose and hemoglobin A1c (HbA1c) in nondiabetic adults (adjusted R² = 0.63). The eIS highly correlated with clamp-measured IS in all of the non-CACTI comparison populations (r = 0.83, P = .0002 in T1D adults; r = 0.71, P = .01 in nondiabetic adults; r = 0.44, P = .008 in T1D adolescents; r = 0.44, P = .006 in nondiabetic adolescents).

Conclusions:

eIS performed better than previous equations for estimating IS in individuals with and without T1D. These equations could simplify point-of-care assessment of IS to identify patients who could benefit from targeted intervention.

Insulin resistance is associated with dyslipidemia (1), hypertension, and hyperglycemia and is a recognized risk factor for cardiovascular disease (CVD) (2–5) and diabetic nephropathy (6–8). In individuals with type 1 diabetes (T1D), insulin resistance is associated with coronary artery calcification (6) and hard cardiovascular events (9), demonstrating that it is an important factor in CVD development. Furthermore, insulin resistance is an important outcome in research on insulin-sensitizing agents (10, 11). Although insulin resistance is a known feature of type 2 diabetes, it is less recognized in patients with T1D, who have fewer classical manifestations of insulin resistance such as obesity and dyslipidemia. However, clear evidence, both historic and recent, exists showing that insulin sensitivity (IS) is reduced by approximately half in both adults and adolescents with T1D (3, 12–14).

The euglycemic-hyperinsulinemic clamp is the “gold standard” for measurement of IS (15), but it is expensive, time-consuming, technically cumbersome, and, therefore, impractical for large epidemiological studies and for clinical use. In nondiabetic individuals and those with non-insulin-dependent diabetes, methods have been developed to estimate IS based on fasting levels of glucose and insulin, including the homeostasis model of assessment for insulin resistance (HOMA-IR) (16) and the quantitative IS check index (Quicki) equation (17). However, estimations of IS among patients with insulin-dependent diabetes using these equations can be problematic because fasting concentrations of glucose and insulin reflect exogenous treatment rather than underlying insulin and glucose metabolism and therefore can yield unreliable estimates of IS based on changes in insulin regimen or diet. Moreover, insulin-deficient patients with T1D are unable to produce insulin in response to glucose challenges in oral glucose tolerance tests and iv glucose tolerance tests, or the insulin-modified frequently sampled iv glucose test and minimal model (18). Thus, there is a need for a reliable, validated, and clinically accessible method for estimating IS in patients with T1D.

Several equations for estimating IS in patients with T1D have been developed. One equation was created by the landmark Pittsburgh Epidemiology of Diabetes Complications (EDC) study (13). The EDC equation estimates glucose disposal rate (Pittsburgh eGDR) and was developed in adults in the 1990s, before many of the contemporary improvements in diabetes treatment. Also, the Pittsburgh eGDR uses the older HbA1 method for estimating chronic glycemia, rather than hemoglobin A1c (HbA1c), the current standard measure (13). A similar equation was developed in the SEARCH for Diabetes in Youth study (19), with the goal of creating a single equation that was applicable to adolescents with T1D, type 2 diabetes, and those without diabetes (SEARCH IS score). However, in light of the apparent differences in the presentation of insulin resistance in type 1 and type 2 diabetes, combining participants with type 1 and type 2 diabetes into a single equation may also be less than ideal for studies focused on T1D alone.

Therefore, an IS prediction equation (eIS) was developed from and validated in the Coronary Artery Calcification in Type 1 Diabetes study and was then applied to independent populations of youth and adults with clamp-measured IS to assess its performance. In addition, the performance of eIS was compared to the previously published SEARCH IS score and the Pittsburgh eGDR prediction equations.

Subjects and Methods

CACTI clamp study population

Adults were recruited from the Coronary Artery Calcification in Type 1 Diabetes (CACTI) cohort as previously reported in detail (1, 12, 20–22). Inclusion criteria for this study included ages of 27–61 years at the current visit and, in participants with T1D, insulin requirement within a year of diagnosis, current insulin therapy, T1D diagnosed before age 30 or a physician diagnosis of T1D, and long-standing diabetes (mean duration, 23 ± 8 y). We collected data from 87 adults (40 with T1D and 47 nondiabetic controls, frequency matched for age, gender, and weight) recruited between 2005 and 2008 who underwent a three-stage euglycemic-hyperinsulinemic clamp as previously described (12). Further inclusion criteria for the clamp study included body mass index (BMI) between 18 and 40 kg/m², HbA1c ≤9.5%, blood pressure (BP) <160/100 mm Hg, albumin excretion rate <200 μg/min, and triglycerides <400 mg/dL. For the development of the prediction equation, only study participants with an average final steady-state glucose concentration between 85 and 95 mg/dL were included, resulting in 77 study participants (36 with T1D and 41 nondiabetic). The Colorado Multiple Institutional Review Board approved the protocol, and written informed consent was obtained from all study participants.

CACTI euglycemic-hyperinsulinemic clamp visit

Participants were maintained on a standardized macronutrient composition diet (50% carbohydrate, 20% protein, 30% fat) provided by the Clinical Translational Research Center (CTRC) with total calories based on dual-energy x-ray absorptiometry fat free mass (FFM) for 3 days before their clamp. Participants were admitted to the inpatient CTRC unit before dinner on the evening before their clamp study and fasted overnight and through the clamp protocol. In individuals with T1D, sc insulin was withheld and replaced overnight with an insulin infusion as previously described (12). Blood samples for determination of baseline insulin and glucose concentrations were drawn over 30 minutes before initiation of the clamp protocol.

A three-stage euglycemic-hyperinsulinemic clamp was then initiated using the method described previously (12). Briefly, a primed continuous infusion of insulin was administered at 4 mU/m²/min for 1.5 hours, 8 mU/m²/min for 1.5 hours, and then 40 mU/m²/min for the final 1.5 hours. A variable amount of 20% dextrose was infused to maintain blood glucose at a target of 90 mg/dL. Glucose infusion rate (GIR) was determined from the mean steady-state iv dextrose infusion in the final 30 minutes of the clamp and is reported per kilogram of FFM (mg/kg/FFM/min). GIR was used as the gold-standard method for measured whole-body IS.

Independent adult and adolescent comparison populations

The adult comparison population consisted of 25 premenopausal women (T1D, n = 12; nondiabetic, n = 13) from the Women, Insulin and Sex Hormones (WISH) study who completed a three-stage (4, 8, and 40 mU/m²/min) hyperinsulinemic-euglycemic clamp using the same methods described for the CACTI cohort (12). The adolescent comparison group consisted of 75 adolescents (T1D, n = 36; nondiabetic, n = 39) from the RESistance to InSulin in type 1 ANd type 2 diabetes (RESISTANT) and the Effects of Metformin on CaRdiovascular Function in AdoLescents with type 1 Diabetes (EMERALD) studies who completed a three-stage (8, 16, and 80 mU/m²/min) hyperinsulinemic-euglycemic clamp (3).

Laboratory assays

Adiponectin was measured using the RIA methodology (Millipore).

Statistical analyses

Development of the eIS

Baseline characteristics were compared by sex and diabetes status using t tests for continuous variables and a χ² test for categorical variables. Test of normality was conducted with the Kolmogorov-Smirnov and Shapiro-Wilks tests. Variables that were positively skewed, not normally distributed, were natural log-transformed for analysis. An eIS was first modeled in a block randomized training set of 26 adults with T1D and 32 nondiabetic adults from the CACTI study, and the findings were then applied to and confirmed in a validation set of 10 adults with T1D and nine nondiabetic adults from the CACTI study, based on an a priori power calculation. There were no significant differences between the test and validation sets for age, HbA1c, waist circumference, triglycerides, diabetes duration, or insulin dose.

GIR was natural log-transformed for analysis. Easily measured clinical parameters, known to influence IS, that were considered for inclusion in the CACTI model included age, diabetes duration (in T1D individuals), BMI, waist circumference, waist:hip ratio (WHR), systolic and diastolic BP, hypertension (defined as BP ≥ 140/90 mm Hg or antihypertensive treatment), total cholesterol, low-density lipoprotein (LDL)-cholesterol, high-density lipoprotein (HDL)-cholesterol, triglycerides, adiponectin, family history of diabetes, fasting free fatty acids, HbA1c, fasting glucose (in nondiabetics), and insulin dose (in T1D participants). Most CACTI study participants were non-Hispanic white (95% of T1D and 83% of nondiabetic individuals). There was no difference in measured IS by race in the cohort (P = .74 in T1D, P = .97 in nondiabetics), so race was not further considered in the modeling. Scatterplots were used to assess whether the independent predictors were linearly associated with GIR, and there was no evidence of a nonlinear relationship.

Next, a rigorous process was used to assess the models developed using the clinical parameters above. More specifically, multivariable linear regression models were fit separately for the test set of study participants with T1D and those without diabetes to select the 20 models with the highest adjusted R². The “best-fit” models for each group (T1D and nondiabetic) were selected from among the models that maximized the adjusted R² based on the models best predicting measured IS in the validation set. For variables that were highly collinear (ie, waist circumference and BMI), the variable with the strongest univariate association was used. Interactions by both sex and diabetes were also considered for all variables, and the selected model resulted in different intercepts and factors by diabetes status. The eIS equation developed in the training set was tested in the validation set, and the 95% limits of agreement of IS difference by average IS in the validation set were calculated as described by Bland and Altman (23). The correlation of the eIS and the clamp-measured IS was examined using Spearman correlation coefficients. An additional step was then performed excluding fasting measures and variables not commonly measured in clinical practice (adiponectin).

Evaluating the performance of eIS in independent cohorts

Once the eIS was developed and validated within the CACTI cohort, it was then tested in independent adult and adolescent populations in whom hyperinsulinemic-euglycemic clamp studies were performed. The performance of the three prediction equations (eIS, SEARCH IS score, and Pittsburgh eGDR) was compared to the GIR from the independent adult and adolescent studies. Clamp-measured IS was compared to estimated IS using Spearman correlation coefficients. Statistical analyses were performed using SAS software, version 9.3, of the SAS System for Windows. P values <.05 were considered statistically significant.

Results

Characteristics of CACTI study participants used to develop the eIS are shown in Table 1. Similar to previously reported findings, CACTI adults with and without T1D did not differ in terms of age, BMI, waist circumference, WHR, BP, or prevalence of hypertension (12). Measured IS was significantly lower in both men and women with T1D compared to men and women without diabetes. Spearman coefficients for correlations of IS with the clinical factors used to develop the eIS are shown in Appendix Table 1 by diabetes status.

The best-fit model (eIS) for the CACTI adult participants with T1D (Table 2) included waist circumference, daily insulin dose per kilogram body weight, adiponectin, triglycerides, and diastolic BP (DBP). The formula to calculate eIS is: exp (4.06154 − 0.01317 * waist [cm] − 1.09615 * insulin dose [daily units per kg] + 0.02027 * adiponectin [μg/mL] − 0.27168 * triglycerides [mmol/L (−0.00307 for mg/dL)] − 0.00733 * DBP [mm Hg]). Because patients are not always able to fast for clinical visits, we next removed fasting measures from the best model to fit an additional nonfasting model (eIS-nf) (T1D model 2), which included waist circumference and daily insulin dose. Similarly, because adiponectin is not routinely measured, we then fit an additional model excluding adiponectin (eIS-exA) (T1D model 3), which included waist circumference, daily insulin dose per kilogram body weight, triglycerides, and DBP. Adjusted R² for models 1, 2, and 3 were 0.67, 0.61, and 0.67, respectively.

Table 2 similarly shows the best-fit model (eIS) among participants without diabetes (nondiabetic model 1), which included waist circumference, adiponectin, triglycerides, and DBP as well as both fasting glucose and HbA1c. The equation for this model was: exp (7.47237 − 0.01275 * waist [cm] − 0.24990 * HbA1c [%] − 0.35730 * fasting glucose [mmol/L (− 0.01983 for mg/dL)] + 0.01905 * adiponectin [μg/mL] − 0.28673 * triglycerides [mmol/L (−0.00324 for mg/dL)] − 0.00588 * DBP [mm Hg]). In addition, Table 2 shows the best-fit model not including fasting variables (eIS-nf) (nondiabetic model 2) and not including adiponectin (eIS-exA) (nondiabetic model 3). Adjusted R² for models 1, 2, and 3 were 0.68, 0.65, and 0.68, respectively. We also included a model with the HOMA-IR equation, based on fasting glucose and insulin (nondiabetic model 4) because HOMA-IR was more strongly correlated with measured IS than Quicki (HOMA-IR, r = −0.55, P < .0001; Quicki, r = 0.40, P < .05). A file to estimate IS using eIS is included (Supplemental Data).

The best-fit models developed in the training set were then validated in a separate, a priori, randomly selected group of CACTI clamp study participants. Estimated IS was calculated for each study participant in the validation group and was compared to measured IS from the clamp study. Correlation coefficients were significant for all models (Appendix Table 2). Interactions by sex were tested for the variables examined, and none were significant.

Characteristics of the independent adult and adolescent populations, used to compare the three IS prediction equations, are shown in Table 3. As expected, the individuals with T1D in both groups had higher HbA1c, fasting glucose, and insulin levels. In the adult population from the WISH Study, women with T1D were slightly older than the nondiabetic women, and they had a significantly higher BMI and waist circumference but lower fasting triglyceride levels. In the adolescents, there was no difference in BMI or waist circumference between the two groups, but WHR was greater in the nondiabetic girls compared to the diabetic girls. Among adolescent girls, total cholesterol and triglycerides were lower in the participants with T1D compared to those without diabetes.

As shown in Table 4, the eIS was strongly correlated with clamp-measured IS in all participants in the adult and adolescent cohorts. The SEARCH IS score was positively correlated with measured IS in the CACTI study adults and the adolescents but was not significantly correlated with measured IS in the women from the WISH cohort. Pittsburgh eGDR correlated with the measured IS only in the nondiabetic CACTI study adults. In the nondiabetic individuals, HOMA-IR was negatively correlated with measured IS in both the CACTI adult and adolescent cohorts, but not in the WISH adult comparison population. eIS had stronger correlation coefficients compared to all of the other IS estimating equations, with the exception of the nondiabetic adolescents.

When examining the best estimated IS model using the nonfasting clinical measures (model 2, Table 2), eIS-nf was highly positively correlated with clamp-measured IS in both the comparison adult (r = 0.77; P = .006) and adolescent (r = 0.40; P = .02) populations with T1D. Additionally, eIS-nf was positively correlated with measured IS, although not significantly, in the nondiabetic adults from the WISH study (r = 0.46; P = .14) and a similar, but significant, relationship was observed in the adolescents without diabetes (r = 0.45; P = .004) (Appendix Table 3). When examining the best eIS model excluding adiponectin (model 3, Table 2), eIS-exA was positively correlated with measured IS in both the T1D and nondiabetic adults from the WISH study (T1D, r = 0.79, P = .004; nondiabetic, r = 0.58, P = .04) and adolescents (T1D, r = 0.50, P = .002; nondiabetic, r = 0.44, P = .005) (Appendix Table 3).

Discussion

In the present study, we developed and tested prediction equations for IS using easily measured clinical factors that explained 61–64% of the variance in measured insulin resistance. Our study is unique in that it included adults both with and without T1D in generating IS estimation equations for each population. Moreover, when applied to independent cohorts of adults and adolescents, eIS from the CACTI study showed better agreement with clamp-measured IS in a contemporary independent cohort than the previously published prediction equations from the SEARCH and EDC studies, possibly due to differences in factors affecting IS in adolescents who are obese or have type 2 diabetes and changes in the population of individuals with T1D over the past 20 years or more, respectively. Our data suggest that the CACTI prediction equations can be broadly applied to patients with and without T1D using common clinical measures. Using identical clamp methods, separate prediction equations were developed for individuals with and without T1D, allowing for differing strengths of association for parameters as well as the inclusion of factors specific to each group, such as daily insulin dose in people with T1D and fasting insulin and glucose in nondiabetic individuals.

Although elevated fasting glucose, insulin, and HbA1c all predicted insulin resistance in adults without diabetes, there was no association between glycemic control as measured by HbA1c and insulin resistance in participants with T1D in the CACTI study (12) or in the WISH cohort (unpublished) or any of our adolescent T1D studies (3). This is in contrast to 20th century studies that showed associations of HbA1 and insulin resistance at worse levels of glucose control than those of our study participants, potentially due to corresponding metabolic dysfunction and elevated counter-regulatory hormones in the setting of marked hyperglycemia (25, 26). Daily insulin dose was a strong predictor of the degree of insulin resistance in the CACTI, WISH, and adolescent study cohorts. In support of these findings, the pre-Diabetes Complications and Control Trial clamp studies showed improved IS and insulin dose without changing HbA1 (27).

The performance of all of the CACTI models (eIS, eIS-nf, and eIS-exA) was similar, and these results demonstrate that they are robust and capable of estimating IS using a variety of easily obtainable clinical measures. Furthermore, in comparison to the published SEARCH IS score and Pittsburgh eGDR equations, the eIS performed better in adults with and without T1D, adolescents with T1D, and similarly to the SEARCH IS score equation, in nondiabetic adolescents.

When the Pittsburgh eGDR equation was tested in the independent comparison groups, it did not correlate significantly with the clamp-measured IS among adults or adolescents with T1D, but it did correlate with measured IS in the nondiabetic CACTI study participants, most likely due to the stronger relationship between HbA1c and IS in these groups. Also in contrast to the results from the eGDR equation (13), HbA1c was not included in the eIS best-fit model among patients with T1D; instead, insulin dose and triglycerides were included. Furthermore, in CACTI, waist circumference and DBP were stronger predictors than WHR and hypertension, respectively. There are several potential explanations for these differences. The CACTI study population was 10 years older than the EDC study population, although age alone is not likely the main explanation because eIS also performed better in adolescents. CACTI participants on average, had relatively good glycemic control, with a mean HbA1c of 7.6% compared to 9.5% in the EDC study, although again our adolescents were not as well controlled (mean HbA1c, 8.5%; upper limit, 12%). However, it should be noted that both our model and the Pittsburgh eGDR model demonstrated that insulin resistance in T1D relates to abdominal adiposity and BP. Additionally, the participants in the EDC study were selected based on predetermined cutoffs of triglycerides, HDL-cholesterol, WHR, and HbA1 to obtain a balanced group of study participants in each tertile of predicted IS, assumptions that may have influenced the results. Our study participants, in contrast, were not screened on variables thought to influence IS. Notably, triglyceride levels were dramatically lower in our study, which better reflects the broad population of T1D patients seen today (28). Overall, our study was able to improve on the adjusted R² reported in the EDC study (R² = 0.571) in both the T1D (R² = 0.64) and nondiabetic (R² = 0.63) participants. More importantly, eIS more accurately estimated clamp-measured IS when compared to the Pittsburgh eGDR equation.

In adolescents with type 1 and type 2 diabetes combined, we recently reported in the SEARCH study that waist circumference, triglycerides, and HbA1c predict IS as measured by an 80 μU/m² euglycemic-hyperinsulinemic clamp (19). It is likely that HbA1c is a more important factor predicting IS in adolescents with type 2 diabetes, perhaps accounting for these different findings. Moreover, of the variables in the SEARCH model, HbA1c explained the least variance in measured IS. Despite the inclusion of HbA1c, the SEARCH IS score equation still correlated with measured IS in both the adolescents and CACTI adults with and without T1D, most likely due to the strong influence of waist circumference and triglycerides. However, the strength of association for the SEARCH IS score was significantly less than that for the eIS equation when tested in the independent adult comparison cohort and similar in magnitude in the adolescent comparison cohort.

Although we report the largest study to date to develop a prediction equation for IS in adults both with and without T1D and then compare it to other published equations in independent cohorts of adults and adolescents, there are several important limitations to our study. The current study focused on simple clinical measurements potentially related to insulin resistance, and other variables that were not assessed in the current study might better predict IS. Moreover, adiponectin currently does not have a standardized assay but remained in the best-fit model because it contributed to the prediction of IS independent of adiposity and significantly improved the model fit. Adiponectin is becoming more widely used, but we realize the limitation in the assay and thus derived models excluding adiponectin. Additionally, there may be certain populations in which eIS does not perform well. However, the main strength of this study was that eIS was developed for estimating IS using simple, clinically available measures for adults with and without T1D. Another strength is that the eIS also performed well in adolescents, despite the wider range in HbA1c, greater degree of insulin resistance, shorter duration of diabetes, and other adolescent-specific factors characteristic of this population. The sample size used to develop eIS was similar to that used to develop the previously published models (eIS, n = 36 T1D participants; Pittsburgh eGDR, n = 24; SEARCH IS score, n = 39).

Insulin resistance is increasingly being recognized as a risk factor for CVD and other complications of diabetes, but due to the difficulty of performing clamp studies, it is not practical to measure IS directly in large epidemiological studies. The application of an equation to estimate IS using easily measured clinical factors could therefore be used to examine further the relationship of IS with complications and the impact of interventions on IS in people with T1D. In addition, this equation could be used to identify those at highest risk of complications and allow individualization of intensified preventive measures, thus providing an immediate clinical application for a point-of-care assessment of IS.

Acknowledgments

The study was performed at the Barbara Davis Center for Childhood Diabetes in Denver, Colorado, and at the Clinical Translational Research Centers (CTRC) at the University of Colorado and Children's Hospital Colorado and was supported by the National Institutes of Health (NIH) Grant M01 RR000051 and Colorado Clinical and Translational Sciences Institute (CCTSI) Grant UL1 TR000154. Support was provided by the NIH National Heart, Lung and Blood Institute Grants R01 HL61753, R01 HL079611, and R01 HL11309; the American Diabetes Association Junior Faculty Award 1-10-JF-50 and 7-13-CD-10 (to J.K.S.-B.), American Diabetes Association Grant 7-11-CD-08, the Juvenile Diabetes Research Foundation Grant 11-2010-343 and 17-2013-313, NIH Building Interdisciplinary Research Careers in Women's Health (BIRCWH) Grant K12 5K12HD057022-04, the National Center for Research Resources Grant K23 RR020038-01, NIH Grant R56 DK088971 (to K.J.N.), Office of Research in Women's Health BIRCWH K12 Program (to I.E.S.), and Diabetes Endocrinology Research Center Clinical Investigation Core Grant P30 DK57516.

Author Contributions: J.K.S.-B. and L.M.D. researched data, analyzed data, and wrote the manuscript. D.M.M., P.B., and I.E.S. researched data, contributed to the discussion, and reviewed the manuscript. K.J.N. and B.C.B. designed the study, researched data, contributed to the discussion, and reviewed the manuscript. M.R. designed the study, contributed to the discussion, and reviewed the manuscript. J.K.S.-B. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Disclosure Summary: The authors have nothing to disclose. There are no potential conflicts of interest relevant to this article to report.

Abbreviations

 
• BMI

  body mass index

 
• BP

  blood pressure

 
• CVD

  cardiovascular disease

 
• DBP

  diastolic BP

 
• eGDR

  estimated glucose disposal rate

 
• eIS

  IS prediction equation

 
• eIS-exA

  eIS excluding adiponectin (model 3)

 
• eIS-nf

  eIS nonfasting (model 2)

 
• FFA

  fat free mass

 
• GIR

  glucose infusion rate

 
• HbA1c

  hemoglobin A1c

 
• HDL

  high-density lipoprotein

 
• HOMA-IR

  homeostasis model of assessment for insulin resistance

 
• IS

  insulin sensitivity

 
• LDL

  low-density lipoprotein

 
• Quicki

  quantitative IS check index

 
• T1D

  type 1 diabetes

 
• WHR

  waist:hip ratio.

References