http://www.kidney-international.org

clinical investigation

& 2014 International Society of Nephrology

see commentary on page 243

Validation of insulin sensitivity surrogate indices and prediction of clinical outcomes in individuals with and without impaired renal function Ting Jia1, Xiaoyan Huang1,2, Abdul R. Qureshi1, Hong Xu1,3, Johan A¨rnlo¨v4,5, Bengt Lindholm1, Tommy Cederholm6, Peter Stenvinkel1, Ulf Rise´rus6 and Juan J. Carrero1,7 1

Divisions of Renal Medicine and Baxter Novum, Department of Clinical Science, Intervention, and Technology, Karolinska Institutet, Stockholm, Sweden; 2Division of Nephrology, Peking University Shenzhen Hospital, Peking University, Shenzhen, China; 3Division of Renal Medicine, Peking Union Medical College Hospital, Beijing, China; 4Department of Public Health and Caring Sciences, Section of Geriatrics, Uppsala University, Uppsala, Sweden; 5School of Health and Social Studies, Dalarna University, Falun, Sweden; 6Department of Public Health and Caring Sciences, Clinical Nutrition and Metabolism, Uppsala University, Uppsala, Sweden and 7Center for Molecular Medicine, Karolinska Institutet, Stockholm, Sweden

As chronic kidney disease (CKD) progresses with abnormalities in glucose and insulin metabolism, commonly used insulin sensitivity indices (ISIs) may not be applicable in individuals with CKD. Here we sought to validate surrogate ISIs against the glucose disposal rate by the gold-standard hyperinsulinemic euglycemic glucose clamp (HEGC) technique in 1074 elderly men of similar age (70 years) of whom 495 had and 579 did not have CKD (estimated glomerular filtration rate (eGFR) under 60 ml/min per 1.73 m2 (median eGFR of 46 ml/min per 1.73 m2)). All ISIs provided satisfactory (weighted j over 0.6) estimates of the glucose disposal rate in patients with CKD. ISIs derived from oral glucose tolerance tests (OGTTs) agreed better with HEGC than those from fasting samples (higher predictive accuracy). Regardless of CKD strata, all ISIs allowed satisfactory clinical discrimination between the presence and absence of insulin resistance (glucose disposal rate under 4 mg/kg/min). We also assessed the ability of both HEGC and ISIs to predict all-cause and cardiovascular mortality during a 10-year follow-up. Neither HEGC nor ISIs independently predicted mortality. Adjustment for renal function did not materially change these associations. Thus, ISIs can be applied in individuals with moderately impaired renal function for diagnostic purposes. For research matters, OGTT-derived ISIs may be preferred. Our data do not support the hypothesis of kidney function mediating insulin sensitivity (IS)–associated outcomes nor a role for IS as a predictor of mortality. Kidney International (2014) 86, 383–391; doi:10.1038/ki.2014.1; published online 29 January 2014 KEYWORDS: chronic kidney disease; clinical outcome; hyperinsulinemic euglycemic glucose clamp; insulin resistance; insulin sensitivity index; mortality Correspondence: Juan J. Carrero, Divisions of Renal Medicine and Baxter Novum, Karolinska University Hospital at Huddinge K56, Karolinska Institutet, Stockholm SE-14186, Sweden. E-mail: [email protected] Received 3 April 2013; revised 10 December 2013; accepted 12 December 2013; published online 29 January 2014 Kidney International (2014) 86, 383–391

Insulin resistance (IR) is commonly present in patients with chronic kidney disease (CKD) and presumably involved in the pathogenesis and cardiometabolic complications of this patient population.1 Several techniques are used to evaluate insulin sensitivity (IS) in humans, with the hyperinsulinemic euglycemic glucose clamp (HEGC) technique being considered the gold-standard method.2 Because of its technical requirements and complexity, HEGC cannot be applied in clinical practice and in epidemiological studies. Instead, IS indices (ISIs) obtained from glucose and insulin measured at fasting state or during an oral glucose tolerance test (OGTT) have been developed. These indices have often been used in epidemiological studies and at times in the clinical setting to discriminate IR individuals and/or search for associations between IS and clinical outcomes in the community,3 as well as in specific disease domains such as CVD,4,5 diabetes,6 or CKD7. CKD patients display several metabolic and hormonal alterations that influence many aspects of insulin and glucose metabolism, and as CKD progresses these abnormalities aggravate.8 At present, it is not known whether, and to what extent, the validity and accuracy of the commonly used ISIs originally developed in healthy populations are compromised in the context of renal dysfunction. A number of ISIs are currently used on the basis of insulin/glucose fasting samples or OGTT tests, some of which have been applied in studies including CKD patients in the absence of methodological validation.1 Whereas Hung et al.9 validated some of these ISIs against HEGC in 12 presumably anuric hemodialysis patients, confirmation of the applicability of ISIs in populations with moderate–severe kidney dysfunction is lacking. We have recently reported that in elderly (70 years old) men with CKD stages 3 and 4, IS as assessed by HEGC was not associated with mortality,10 opposing in part to previous reports using IS indices,7,11,12 and altogether evidencing the need of formal validation of these surrogates. Against this 383

clinical investigation

T Jia et al.: Validation of ISIs and prediction of clinical outcomes

Table 1 | General characteristics of study subjects and stratification by the presence of chronic kidney disease (CKD, estimated glomerular filtration rateo60 ml/min per 1.73 m2) All subjects n ¼ 1074

Non-CKD n ¼ 579

CKD n ¼ 495

61.4(44.8,78.7) 26.2±3.4 29 73 10 9.6(5.9,15.9) 1.4±0.7 5.5±0.9 42(21, 85)

68.5(61.4, 83.5) 26.0±3.3 24 70 10 9.1 (5.9, 15.6) 1.4±0.6 5.6±1.0 40 (20, 82)

46.3(38.9, 58.3) 26.3±3.5 34 78 10 9.6 (5.9, 16.5) 1.5±0.8 5.5±0.8 44 (21, 93)

5.3±2.0 5.2±2.4 107.1±20.0

5.4±2.0 5.4±2.5 104.0±18.2

5.1±2.0 4.9±2.3 110. 7±21.4

OGTT test Glucose AUC, h*mmol/l Insulin AUC, h*pmol/l AUCins/glu, pmol/mmol

59(18, 107) 937(443, 2013) 18(6, 48)

59 (18, 106) 909 (405, 1965) 17 (6, 47)

59 (18, 107) 995 (482, 2331) 20 (7, 50)

Insulin resistance surrogate indices HOMA-IR HOMA-AD FIRI QUICKI McAuley’s index Matsuda index Stumvoll index

1.7(0.8, 3.8) 170(60, 574) 1.5(0.7, 3.4) 0.15±0.01 7.7±1.9 10.3(4.8, 21.1) 9.0±1.5

1.7 (0.8, 3.6) 166 (60, 569) 1.5 (0.7, 3.2) 0.15±0.02 7.9±1.9 10.4 (5.0, 22.4) 9.1±1.52

1.8 (0.8, 4.0) 175 (60, 585) 1.6 (0.8, 3.6) 0.15±0.02 7.5±1.8 10.1 (4.2, 19.3) 8.9±1.6

Parameters 2

eGFR, ml/min per 1.73 m BMI, kg/m2 Cardiovascular disease, % Hypertension, % Diabetes mellitus, % Adiponectin, ng/ml Triglycerides, mmol/l Fasting glucose, mmol/l Fasting insulin, pmol/l

Hyperinsulinemic euglycemic glucose clamps Glucose disposal M, mg/kg/min M/I Insulin60 min, mm/l

Abbreviations: AUC, area under the curve; AUCins/glu, ratio of the total areas under the curve for insulin and glucose; BMI, body mass index; eGFR, estimated glomerular filtration rate; FIRI, fasting insulin resistance index; HOMA-AD, homeostasis model assessment corrected by adiponectin; HOMA-IR, homeostasis model assessment of insulin resistance; Insulin60 min, insulin concentration at 60 min in clamp test; OGTT, oral glucose tolerance test; M/I, glucose disposal (M) per unit of plasma insulin (I) during the last 60 min of the hyperinsulinemic euglycemic glucose clamp; QUICKI, quantitative insulin sensitivity check index. Values are mean±s.d., median (10th to 90th percentile), or percentage.

background, our goal was to validate and compare the reliability and accuracy of commonly used ISIs against HEGC in a post hoc analysis of a large cohort of elderly men with (n ¼ 495) and without (n ¼ 579, serving as the control group) manifest CKD. As HEGC measures the glucose disposal rate in response to a standardized insulin challenge, it is not subjected to bias by renal function.1 In a second step, we tested (i) the diagnostic performance of surrogate ISIs to discriminate from a state of IR as assessed by HEGC prediction, and (ii) their ability to predict hard outcomes, either all-cause or cardiovascular-related deaths during a long follow-up period. RESULTS Validation of agreement

The demographics and clinical characteristics of studied subjects together with their estimated ISIs are described in Table 1. Individuals with CKD presented impaired IS by all methods considered. As the kidneys clear around 50% of peripheral insulin, the higher insulin concentration at steady state of insulin infusion observed in CKD versus non-CKD individuals (insulin concentration after 60 min 110.7±21.4 vs. 104.0±18.2 mm/l, Po0.001) is expected. This motivates using the M/I value instead of the M value for validation of 384

agreement analysis (see Materials and Methods). ISIs derived from fasting samples tended to have lower correlation coefficients (rho) in CKD subjects than in non-CKD subjects, although no statistically significant differences were observed. ISIs derived from OGTT samples had similar correlation coefficients in both groups (Table 2). To evaluate the agreement between ISIs and HEGC, weighted k was calculated (Figure 1). The agreements of OGTT-derived indices (Matsuda and Stumvoll indices) with HEGC were in general higher than those of fasting-derived indices. In any case, the agreement observed for the vast majority of ISIs was considered satisfactory for clinical use (weighted k40.6),13 irrespective of the presence of CKD. Results in addition show that the agreement for each of the indices studied was somewhat lower in CKD individuals as compared with nonCKD individuals, with the exception of Stumvoll index. We observed similar results after excluding diabetic individuals (data not shown). The predictive accuracy of ISIs versus HEGC-assessed M/I value was evaluated with random calibration models and the analysis of leave-one-out cross-validation type root meansquared error of prediction (CVPE) residuals. As shown in Table 3, the highest predictive accuracy (thus lowest CVPE values) was found for the OGTT-derived ISIs Matsuda and Kidney International (2014) 86, 383–391

clinical investigation

T Jia et al.: Validation of ISIs and prediction of clinical outcomes

Stumvoll, followed by the fasting-derived ISIs QUICKI and McAuley. Similar results were obtained in stratified analysis by the presence of CKD. Predictive accuracy as assessed by the less robust residual of root mean-squared error of prediction (RMSE) showed similar results (data not shown). For simplicity of data presentation, comparative analysis on prediction of outcomes focused on Matsuda, Stumvoll, QUICKI, and McAuley indices because of their predictive accuracy, and, in addition, on HOMA-IR, because of its common use in the scientific literature. Figure 2 plots HEGC-measured IS versus the predicted HEGC IS from the regression models of QUICKI and Matsuda. They both

Table 2 | Spearman’s correlation coefficients (rho) for associations between insulin sensitivity indices and M/I, as estimated by hyperinsulinemic euglycemic glucose clamp in subjects with and without manifest chronic kidney disease (CKD) Non-CKD n ¼ 579

CKD n ¼ 495

Wolfe’s test

rho

rho

Z (P)

Prediction of clinical outcomes

A state of IR was defined as M value o4 mg/kg/min,14 and receiver operating characteristic (ROC) analysis was performed to test the diagnostic performance of selected ISIs as test variables. Results are shown in Table 4, which also reports the sensitivity and specificity for each index. All four surrogate ISIs showed similar and optimal diagnostic performance, with ROC values above 0.80 in both CKD and non-CKD subjects. Areas under the curve (AUCs) from OGTT-derived ISIs tended to be higher. Results were confirmed when re-defining IR as below the 25% of M distribution (o3.85 mg/kg/min) in our population (data not Table 3 | Predictive accuracy (CVPE) calculated from random calibration analysis of insulin sensitivity surrogate indices against M/I All subjects n ¼ 1074

IS indices derived from fasting samples HOMA-IR  0.71 HOMA-AD  0.70 FIRI  0.71 QUICKI 0.71 McAuley’s index 0.64

 0.67  0.65  0.67 0.67 0.64

 1.25  1.50  1.25 1.25 0

IS indices derived from OGTT  0.56 AUCins Matsuda index 0.77 Stumvoll index 0.71

 0.57 0.74 0.73

 1.38(0.17) 1.14 (0.25)  0.68 (0.50)

(0.21) (0.13) (0.21) (0.21) (1)

Abbreviations: AUCins, area under the curve for insulin; CKD, chronic kidney disease; FIRI, fasting insulin resistance index; HOMA-IR, homeostasis model assessment of insulin resistance; HOMA-AD, homeostasis model assessment corrected by adiponectin; IS, insulin sensitivity; OGTT, oral glucose tolerate test; QUICKI, quantitative insulin sensitivity check index. Comparisons of correlation coefficients are performed by R to Z transformed Wolfe’s test. P-value was tested in the two-tailed test.

1.0

show a good degree of correlation in individuals with and without CKD.

Non-CKD n ¼ 579

CKD n ¼ 495

IS indices derived from fasting samples QUICKI 1.47 McAuley’s index 1.52 HOMA-IR 1.66 HOMA-AD 1.67 FIRI 1.67

1.52 1.62 1.74 1.69 1.74

1.39 1.41 1.57 1.64 1.58

IS indices derived from OGTT Matsuda index Stumvoll index AUCins

1.48 1.42 1.72

1.31 1.32 1.48

1.40 1.38 1.61

Abbreviations: AUCins, area under the curve for insulin; CKD, chronic kidney disease; FIRI, fasting insulin resistance index; HOMA-IR, homeostasis model assessment of insulin resistance; HOMA-AD, homeostasis model assessment corrected by adiponectin; IS, insulin sensitivity; OGTT, oral glucose tolerate test; QUICKI, quantitative insulin sensitivity check index. CVPE is leave-one-out cross-validation root mean-squared error of prediction as calculated from calibration analysis against HEGC (see Methods).

Indices derived from fasting samples

Indices derived from OGTT

0.9 0.8 Weighted 

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 HOMA-IR HOMA-AD

FIRI

QUICKI McAuley’s index

AUCins

Matsuda Stuvmoll index index

Figure 1 | Reliability coefficients between insulin sensitivity indices and M/I as estimated by hyperinsulinemic euglycemic glucose clamps in subjects without (black bars) and with (white bars) manifest chronic kidney disease. Data are shown as 44 weighted k and 95% confidence intervals. AUCins, area under the curve for insulin; FIRI, fasting insulin resistance index; HOMA-IR, homeostasis model assessment of insulin resistance; HOMA-AD, homeostasis model assessment corrected by adiponectin; OGTT, oral glucose tolerance test; QUICKI, quantitative insulin sensitivity check index. Kidney International (2014) 86, 383–391

385

clinical investigation

T Jia et al.: Validation of ISIs and prediction of clinical outcomes

shown). We then compared the AUCs of the two indices showing the best accuracy against HEGC, that is, QUICKI and Matsuda. As shown in Figure 3, Matsuda index

performed significantly better than QUICKI in CKD subjects, and it showed a similar performance in non-CKD subjects.

15

0.75

10 Sensitivity

M/I predicted by QUICKI

1.00

5

0.50

0.25 non-CKD, r 2=0.38 CKD, r 2=0.37

0

χ =1.73, P =0.19

0.00 0.00

0

5

10

Matsuda, ROC=0.89 QUICKI, ROC=0.88 Reference

2

0.25

15

M/I

0.50 1-Specificity Non-CKD

0.75

1.00

0.75 15 Sensitivity

M/I predicted by Matsuda index

1.00 20

10

0.50

0.25

5

0

5

10 M/I

15

2 χ =12.4, P =0.0004

0.00

non-CKD, r 2=0.42 CKD, r 2=0.47

0

Matsuda, ROC=0.91 QUICKI, ROC=0.86 Reference

0.00

20

0.25

0.50 0.75 1-Specificity

1.00

CKD

Figure 2 | Comparison between measured M/I and predicted M/I from QUICKI and Matsuda indices in non-CKD (N ¼ 579) subjects and CKD subjects (N ¼ 495). Predicted M/I was calculated using the leave-one-out cross-validation analysis of the calibration models, as described in Materials and Methods. The solid line indicates ideal predictive accuracy. Square of correlation coefficients (r2) is shown in each panel. Circle symbols represent non-CKD participants and square symbols represent CKD participants. CKD, chronic kidney disease; QUICKI, quantitative insulin sensitivity check index.

Figure 3 | Comparison of receiver operating characteristic (ROC) curves for QUICKI and Matsuda indices to predict a state of insulin resistance (Mo4 mg/kg/min) in CKD and non-CKD subjects. Data are shown as sensitivity/1-specificity curves. The black curve represents the Matsuda index and the gray curve represents QUICKI. The comparison of the two AUC was tested by binomial exact confidence interval. AUC, areas under curves; CKD, chronic kidney disease; QUICKI, quantitative insulin sensitivity check index; ROC, receiver operating characteristic.

Table 4 | Receiver operating characteristic (ROC) analysis of selected insulin sensitivity surrogate indices comparing their performance to discriminate a state of insulin resistance (Mothan 4 mg/kg/min) All subjects n ¼ 1074

Non-CKD n ¼ 579

CKD n ¼ 495

AUC (95% CI)

AUC (95% CI)

Sensitivity, %

1-Specificity, %

AUC (95% CI)

Sensitivity, %

1-Specificity, %

IS indices derived from fasting samples HOMA-IR 0.87 (0.85–0.90) QUICKI 0.85 (0.83–0.88) McAuley’s index 0.82 (0.79–0.85)

0.88 (0.85–0.91) 0.88 (0.85–0.91) 0.82 (0.78–0.86)

0.79 0.79 0.75

0.17 0.17 0.25

0.86 (0.82–0.90) 0.86 (0.82–0.91) 0.83 (0.79–0.87)

0.79 0.79 0.71

0.18 0.18 0.18

IS indices derived from OGTT Matsuda index 0.90 (0.88–0.92) Stumvoll index 0.90 (0.88–0.92)

0.89 (0.86–0.92) 0.89 (0.86–0.92)

0.88 0.85

0.21 0.21

0.91 (0.87–0.94) 0.90 (0.97–0.93)

0.86 0.88

0.16 0.22

Abbreviations: AUC, areas under the curve; CI, confidence interval; CKD, chronic kidney disease; HOMA-IR, homeostasis model assessment of insulin resistance; IS, insulin sensitivity; OGTT, oral glucose tolerate test; QUICKI, quantitative insulin sensitivity check index.

386

Kidney International (2014) 86, 383–391

clinical investigation

T Jia et al.: Validation of ISIs and prediction of clinical outcomes

Table 5A | Receiver operating characteristic analysis and Cox proportional hazard models of insulin sensitivity (IS) surrogate indices regarding their predictive performance for mortality (by any cause) during follow-up HR for mortality of 1-s.d. increase Area under the curve

Crude

Model 1

Model 2

Hyperinsulinemic euglycemic glucose clamp Glucose M value M/I

0.57 (0.53–0.61) 0.56 (0.52–0.60)

0.79 (0.70–0.89) 0.81 (0.72–0.92)

0.87 (0.74–1.03) 0.93 (0.79–1.19)

0.88 (0.75–1.04) 0.95 (0.81–1.12)

IS indices derived from fasting samples HOMA-IR QUICKI McAuley’s Index

0.53 (0.49–0.57) 0.53 (0.49–0.57) 0.53 (0.49–0.57)

1.10 (1.00–1.20) 0.90 (0.80–1.02) 1.11(1.00–1.23)

1.04 (0.92–1.18) 0.96 (0.82–1.12) 1.05 (0.92–1.20)

1.03 (0.92–1.17) 0.97 (0.83–1.13) 1.04 (0.91–1.19)

IS indices derived from OGTT Matsuda index Stumvoll index

0.55 (0.51–0.59) 0.52 (0.48–0.56)

1.10 (1.00–1.20) 0.82 (0.74–0.92)

1.08 (0.96–1.20) 0.93 (0.76–1.12)

1.06 (0.95–1.19) 0.92 (0.76–1.11)

Abbreviations: CI, confidence interval; HOMA-IR, homeostasis model assessment of insulin resistance; HR, hazards ratio; M/I, glucose disposal (M) per unit of plasma insulin (I) during the last 60 min of the hyperinsulinemic euglycemic glucose clamp; OGTT, oral glucose tolerate test; QUICKI, quantitative insulin sensitivity check index. Data are presented as area under the curve (95% CI) and hazard ratios (95% CI). Model 1 is adjusted for the following established confounders of the association between IS and mortality: history of CVD, smoking, lipid medication, physical activity and waist circumference. Model 2 is adjusted as in Model 1 plus eGFR (per s.d. increase).

During a median follow-up of 10 (range 0.8–12.4) years, 288 participants died (incidence rate 2.68/100 person-years at risk), and of those 120 participants died from cardiovascular causes (incidence rate 1.12/100 person-years at risk). ROC analyses tested the prediction capacity, and Cox analyses explored the hazards of 1-s.d. increase on prediction of allcause mortality (Table 5A) or CVD-specific mortality (Table 5B) for selected indices and the gold standard. HEGC-assessed M and M/I predicted both all-cause mortality and CVD deaths, with AUCs ranging from 0.56 to 0.58, which denotes a relatively poor prediction capacity. The same was true for the studied indices, all of which presented similar or lower AUCs than the gold standard. The Cox proportional hazard model showed that only HEGC IS was associated with all-cause mortality in crude models, but the association was lost after adjustment for confounders. HEGC-assessed IS and QUICKI, Matsuda, and Stumvoll indices were associated with the prediction of cardiovascularspecific deaths in crude analysis, but the statistical significance was lost after adjustment for confounders. In both cases, further correction for renal function did not materially change the magnitude of the associations observed. DISCUSSION

Although measurement of the glucose-disposal rate by HEGC provides high precision in the sophisticated research setting, there is a need to examine and validate more practical methods. In line with previous community studies,15–21 the correlation coefficients in the non-CKD individuals of our study were in the range 0.50–0.75, making it a valid reference to compare with. Generally, ISIs in our study were satisfactory surrogates of IS in individuals with CKD as compared with the gold standard. Furthermore, the predictive accuracy of ISIs in CKD was similar to our non-CKD control population. This complements and expands the ISI validation recently reported by Hung et al.9 in 12 hemodialysis patients. Indices derived Kidney International (2014) 86, 383–391

from OGTT (such as Matsuda and Stumvoll) tested, however, better than those derived from fasting samples, followed by QUICKI. Because IR in CKD is primarily localized in skeletal muscle,8 impaired IS may not be fully reflected by ISIs from fasting samples, which are more likely to reflect hepatic IR. Instead, OGTT-derived indices, likely capturing peripheral IR,22 may be preferred in the context of CKD. Typical additional metabolic disorders inherent to CKD, such as gastrointestinal disorders and impaired absorption of nutrients, inflammation, sodium and fluid retention, and the decline of insulin clearance by the kidneys, may further motivate why OGTT-derived indices should be preferable.1 With regard to the OGTT indices studied, and opposing to a previous report in healthy volunteers,23 the area ‘under the curve for insulin’ did not show a good agreement with HEGC. Furthermore, our community-based study of elderly individuals with impaired renal function did not agree with a previous study in stable renal transplant recipients24 concerning the better agreement of McAuley’s index over other indices. Alterations in the metabolism of blood lipids after renal transplantation because of the use of immunosuppressant therapies are likely to be of importance when interpreting these divergences.25 Because of the role of kidneys in insulin elimination,26 we hypothesized that higher insulin levels in CKD may reflect impaired kidney function instead of, or in addition to, true IR. Although Kappa statistics seemed to indicate that fasting sample–derived ISIs relate somewhat worse with HEGC in CKD subjects than in those without CKD, our analyses of predictive method accuracy and diagnostic discrimination suggest that this may not be the case. In community-based studies, ISIs estimated from fasting27–29 or dynamic testing30 have been associated with the risk of CVD, and IR is considered to be closely related to important risk factors such as hypertension, dyslipidemia, and hyperglycemia.31 Community studies addressing the association between ISIs and (cardiovascular) deaths have, 387

clinical investigation

T Jia et al.: Validation of ISIs and prediction of clinical outcomes

Table 5B | Receiver operating characteristic analysis and Cox proportional hazard models of selected insulin sensitivity (IS) surrogate indices regarding their predictive performance for cardiovascular-related mortality during follow-up Area under the curve

Crude

1-s.d. HR for mortality Model 1

Model 2

Hyperinsulinemic euglycemic glucose clamp Glucose M value 0.58 (0.53–0.64) M/I 0.58 (0.53–0.64)

0.74 (0.61–0.89) 0.74 (0.61–0.90)

0.83 (0.65–1.07) 0.88 (0.69–1.13)

0.85 (0.66–1.09) 0.91 (0.71–1.16)

IS indices derived from fasting samples HOMA-IR 0.58 (0.52–0.63) QUICKI 0.58 (0.52–0.64) McAuley’s Index 0.58 (0.52–0.64)

1.19 (1.07–1.32) 0.74 (0.61–0.90) 1.21 (1.08–1.36)

1.14 (1.00–1.29) 0.81 (0.64–1.02) 1.15 (1.00–1.33)

1.13 (0.99–1.28) 0.82 (0.65–1.03) 1.14 (0.99–1.31)

IS indices derived from OGTT Matsuda Index Stumvoll Index

0.78 (0.62–0.96) 0.76 (0.64–0.90)

0.85 (0.66–1.10) 0.78 (0.59–1.03)

0.87 (0.68–1.12) 0.72 (0.58–1.01)

0.57 (0.51–0.62) 0.58 (0.52–0.64)

Abbreviations: CI, confidence interval; HOMA-IR, homeostasis model assessment of insulin resistance; HR, hazards ratio; M/I, glucose disposal (M) per unit of plasma insulin (I) during the last 60 min of the hyperinsulinemic euglycemic glucose clamp; OGTT, oral glucose tolerate test; QUICKI, quantitative insulin sensitivity check index. Data are presented as area under the curve (95% CI) and hazard ratios (95% CI). Model 1 is adjusted for established confounders in the association between IS and mortality: history of CVD, smoking, lipid medication, physical activity and waist circumference. Model 2 is adjusted for model 1 plus eGFR (per s.d. increase).

however, reported conflicting results.27,32–36 Our analysis shows in that, among an elderly population of Swedish men, IS by neither the gold-standard nor surrogate indices was associated with (cardiovascular) deaths during follow-up. When weak associations were observed, they disappeared after adjustment for established confounders. One possible explanation for these negative associations pertains to the old age of the included subjects, which is similar to the average age of participants in studies showing weak or no association with hard end points.34–36 Another possibility to explain this divergence in findings could be the influence of renal function in the association between IS and hard end points. Indeed, de Boer et al.34 recently showed that the association between OGTT-estimated IS and mortality was considerably attenuated after adjustment of cystatin-C eGFR in 3138 elderly individuals (465 years) participating in the Cardiovascular Health Study. The authors proposed that renal function could therefore be a confounder or a mediator in this relationship. Using a similar approach, our analysis cannot confirm this effect attenuation by renal function. Although both studies did include elderly community individuals, the fact that our survey was restricted to men aged 70 years may limit full comparison with de Boer et al.34 Furthermore, it is also possible that our participant selection renders a narrow eGFR range that may not reflect/capture the broader renal function spectrum. The use of gold-standard techniques, the large sample size, and the strict inclusion criteria of our cohort (same sex, age, geographical distribution, and ethnicity) facilitate testing of unbiased estimations of the agreement between methods and diagnostic usefulness, and these factors should be considered as strengths in our study. Nonetheless, there are some additional limitations that need to be taken into account when interpreting the results. First, our results may not necessarily be extrapolated to other age categories, or to women. Second, the investigated cohort comprised relatively 388

healthy individuals, attributed in part to the lower prevalence of cardiovascular risk factors in Nordic countries and to the nature of this population-based screening program, whereby individuals who are more concerned about their health and lifestyle are likely prone to participate. In this sense, there is, for instance, a narrow range of BMI among the included participants. Third, because of logistical reasons, blood sampling and OGTT were not performed on the same day as HEGC. Finally, the categorization into CKD/non-CKD is solely based on eGFR, and the vast majority of CKD individuals had an eGFR that classified them into stage 3. To conclude, our study demonstrates that ISIs developed in healthy populations can be correctly applied in subjects with moderately impaired renal function. In these individuals, OGTT-derived indices (Matsuda and Stumvoll) and QUICKI may be recommended in view of their higher reliability and accuracy. These indices can be used for diagnostic purposes (diagnosis of a state of IR). However, they showed a poor prediction performance for hard end points in this population of elderly individuals with prospective long-term follow-up. Consideration of renal function in these prognostic models did not materially change the magnitude of the associations observed.

MATERIALS AND METHODS Study sample The individuals were community-dwelling individuals from the third examination cycle of the Uppsala Longitudinal Study of Adult Men (ULSAM: www.pubcare.uu.se/ULSAM/), which includes all men born between 1920 and 1924 and residing in Uppsala County, Sweden. In this cycle, 1221 Caucasian men underwent clinical examinations at the age of 70–71 years.36,37 For the current analysis, subjects who were lacking HEGC test (n ¼ 61), or who were on oralhypoglycemic agents and insulin treatment (n ¼ 60) or lacked serum cystatin-C (n ¼ 26), were excluded. A total of 1074 subjects were finally included in this analysis, of which 108 (10%) subjects had Kidney International (2014) 86, 383–391

T Jia et al.: Validation of ISIs and prediction of clinical outcomes

diabetes mellitus but were not receiving diabetes mellitus-specific medication. The study was approved by the Ethics Committee of the Faculty of Medicine at Uppsala University, and informed consent was obtained from all subjects. Blood samples were drawn in the morning after overnight fasting. Plasma and serum determinations were performed at the Department of Clinical Chemistry, University Hospital, Uppsala, which is accredited according to the Swedish Board for Accreditation and Conformity Assessment (Swedac) standard ISO/IEC 17025. Glucose, insulin, and triglycerides were measured with standard methods.37 Adiponectin was measured by using time-resolved immunofluorometric assay (R&D, Abingdon, UK). Diabetes mellitus was defined as fasting plasma glucose X7.0 mmol/l, or 2-h postload glucose levels X11.1 mmol/l, or the use of oral hypoglycemic agents or insulin. Serum cystatin-C was measured by latex-enhanced reagent (Dade Behring, Deerfield, IL) with Behring BN ProSpec analyser (Dade Behring) and used to calculate eGFR. The total analytical imprecision of the method was 4.8% at 0.56 mg/l and 3.7% at 2.85 mg/l. GFR was calculated from cystatin C by using the formula Y ¼ 77.24CystC  1.2623, which has been shown to be closely correlated with iohexol clearance.38 Manifest CKD was defined as eGFRo60 ml/min per 1.73 m2 according to the KDIGO/ KDOQI classification,39 and causes of CKD were not ascertained. Previous CVD was established as a history of any CVD as recorded in the Swedish Hospital Discharge Registry (International Classification of Diseases (ICD)-8 codes 390–458 or ICD-9 codes 390–459). Hypertension was defined as systolic blood pressure X140 mm Hg, diastolic blood pressureX90 mm Hg, and/or use of antihypertensive medication. Classification of antihypertensive and lipid-lowering drugs was performed according to the, at that time, current list of pharmaceutical specialties available in Sweden (FASS 1992/1993). Waist circumference was measured at the level of the umbilicus. Smoking habits (yes/no) and physical activity were recorded by a standardized questionnaire detailed elsewhere.40 The Swedish national register of deaths was used to define total mortality and cardiovascular mortality (defined as death from ischemic heart disease, cerebrovascular disease, or other cardiovascular disease; ICD-9 codes 410–414, 430–438 or ICD-10 codes I20-I25, I60-I69/ G45). The register includes all Swedish citizens, which minimizes loss to follow-up. End of follow-up was December 2003, and the median follow-up duration is 10.1 years.

Hyperinsulinemic euglycemic glucose clamp and oral glucose tolerance tests The HEGC was performed as detailed by DeFronzo et al.2 with slight modifications described elsewhere.37 In brief, two intravenous infusion lines were placed, one into an antecubital vein and the other into a hand or wrist vein. After a 10-min priming infusion, insulin (Actrapid Human; Novo, Copenhagen, Denmark) infusion was held constant at a concentration of 660 pmol/l for 120 min. Plasma glucose was measured every 5 min to be clamped at the euglycemic level (5.1 mmol/l) by infusion of variable amounts of 20% dextrose solution. The total body glucose disposal rate (M value), which is the basic clamp-derived IS index, was the average value of the glucose infusion rate during the final 60 min of the 120min study (steady state). In addition, the M/I value was also calculated, which accounts for the insulin concentrations during the last 60 min of the clamp, and thus represents the amount of glucose metabolized (that is, taken up by the body) per unit of plasma insulin. The clamp measurement in this cohort is considered to be Kidney International (2014) 86, 383–391

clinical investigation

reliable, as replication tests showed a high intra-individual variation of 0.93 and low coefficient of variation of 0.12.41 In addition, the OGTT test was performed in conjunction with blood sampling. This test occurred on a separate day but within the same week as HEGC. All subjects completed a standard 75-g OGTT after a 10-h overnight fasting. Subjects ingested a solution containing 75 g of dextrose, and venous blood samples were drawn at time points 0, 30, 60, 90, and 120 min for measurement of plasma glucose and insulin. As published elsewhere,42 the intra-individual variation of the OGTT test following repeated measurements in a subsample of 17 individuals was 0.57 and the coefficient of variation was 0.50. Surrogate indices of IS Several commonly used ISIs were investigated, including those derived from fasting blood samples (homeostasis model assessment for insulin resistance;16 homeostasis model assessment corrected by adiponectin;15 fasting insulin resistance index; quantitative insulin sensitivity check index;17 and McAuley’s index18) and those derived from OGTT. The latter include the Matsuda21 and the Stumvoll20 indices, as well as the area under the curve for insulin and glucose concentrations during whole OGTT, estimated with the formula AUC ¼ 3ðM30  M0 þ 2ðM60  M0Þ þ 2ðM90  M0Þ þ M120  M0Þ;

where M is the value of glucose or insulin at the specified time.19 Agreement and calibration model analysis of surrogate ISIs The correlation coefficient between the indices and HEGC was tested by Spearman’s rank test (rho) and compared between CKD and non-CKD groups with the two-tailed R to Z transformed Wolfe’s test.43 Because ISIs and HEGC provide different units of measure, classical Bland–Altman agreement analysis could not be plotted. To assert the coefficient of reliability among methods, 44 weighted k and its 95% confidence interval were performed.44 Because ISIs are measured with error from an experimental population, random calibration model analysis was performed for each IS index to assess the accuracy against the gold standard.45 It was assumed that the random error has Gaussian distribution with a mean of 0 and a constant variance. Although HEGC was also measured with error, it was assumed that the measurement error of HEGC was very small relative to the measurement error of simple IS surrogates and was neglected from our calibration models. We regressed the measured M/I for each subject on each IS indices in the entire cohort and after stratification for the presence of CKD. The fitted calibration model was used to generate values of predicted M/I by each ISI. For analysis of each surrogate index, two types of predicted residuals were considered and used to study predictive accuracy and random error: the root mean-squared error of prediction (RMSE) and leave-one-out cross-validation type root mean-squared error of prediction (CVPE). Smaller values of RMSE and CVPE indicate better prediction. RMSE is likely to underestimate prediction errors, whereas CVPE is more robust. As a sensitivity analysis, correlation coefficients and Cohen’s k were performed in 966 nondiabetes mellitus subjects (that is, excluding diabetic individuals not taking insulin). Comparative prediction of clinical outcomes by ISIs We tested and compared the discriminatory power of ISIs to predict clinical outcomes by ROC analyses and Cox proportional hazard models. The AUC was used as a measure of how well a continuous variable predicts the outcome of interests, and accuracy was assessed 389

clinical investigation

by computing sensitivity and specificity (false-positive rate; 1-sensitivity). Statistical comparison of AUCs from selected ROC curves was performed by the method of binomial exact confidence interval. The first clinical outcome considered was a state of IR. To allow comparisons with the study of Hung et al.,14 this was as defined by an M value less than 4 mg/kg/min.14 In addition, we also tested, as a sensitivity analysis, the prediction of IR defined as below the 25th percentile of M (3.85 mg/kg/min) in our study population. The second clinical outcome considered was death, from any cause or due to CVD-specific causes, which was recorded, with no loss to follow-up, after a median of 10 years. Cox proportional hazard models study the association with mortality of HEGC and ISIs. To allow comparison in outcome prediction between indices, hazard ratios (and 95% confidence intervals) are shown per one s.d. increase in the exposure. We considered a crude model and an adjusted model for established confounders in the association between IS and mortality (that is, history of CVD, smoking, use lipid medication, physical activity, and waist circumference). As a final step, we adjusted for eGFR to test the hypothesis that renal function may modify the association between IS and mortality in the community. All statistic analyses were performed by using the STATA 12.0 software (StataCorp LP., College Station, TX).

T Jia et al.: Validation of ISIs and prediction of clinical outcomes

10.

11.

12.

13. 14.

15.

16.

17.

18. 19.

DISCLOSURE

BL is affiliated with Baxter Healthcare. All the other authors declared no competing interests.

20.

21.

ACKNOWLEDGMENTS

Baxter Novum is the result of a grant from Baxter Healthcare Corporation to Karolinska Institutet. This work was supported by grants from the Swedish Research Council, Swedish Heart-Lung Foundation, Marianne and Marcus Wallenberg foundation, Excellence of Diabetes Research in Sweden (EXODIAB), Westman Foundation, the Centre for Gender Medicine, and the Strategic Research Programme in Diabetes at Karolinska Institutet. TJ is supported by grants from the China Scholarship Council.

24.

REFERENCES

25.

1.

2.

3.

4.

5.

6.

7.

8. 9.

390

Pham H, Utzschneider KM, de Boer IH. Measurement of insulin resistance in chronic kidney disease. Curr Opin Nephrol Hy 2011; 20: 640–646. DeFronzo RA, Tobin JD, Andres R. Glucose clamp technique: a method for quantifying insulin secretion and resistance. Am J Physiol 1979; 237: E214–E223. Hedblad B, Nilsson P, Engstrom G et al. Insulin resistance in non-diabetic subjects is associated with increased incidence of myocardial infarction and death. Diabetic Med 2002; 19: 470–475. Bonora E, Formentini G, Calcaterra F et al. HOMA-estimated insulin resistance is an independent predictor of cardiovascular disease in type 2 diabetic subjects—Prospective data from the Verona Diabetes Complicated Study. Diabetes Care 2002; 25: 1135–1141. Hanley AJG, Williams K, Stern MP et al. Homeostasis model assessment of insulin resistance in relation to the incidence of cardiovascular disease— The San Antonio Heart Study. Diabetes Care 2002; 25: 1177–1184. Katz A, Nambi SS, Mather K et al. Quantitative insulin sensitivity check index: A simple, accurate method for assessing insulin sensitivity in humans. J Clin Endocr Metab 2000; 85: 2402–2410. Chen J, Muntner P, Hamm LL et al. Insulin resistance and risk of chronic kidney disease in nondiabetic US adults. J Am Soc Nephrol 2003; 14: 469–477. DeFronzo RA, Alvestrand A, Smith D et al. Insulin resistance in uremia. J Clin Invest 1981; 67: 563–568. Hung AM, Sundell MB, Egbert P et al. A comparison of novel and commonly-used indices of insulin sensitivity in African American chronic hemodialysis patients. Clin J Am Soc Nephrol 2011; 6: 767–774.

22.

23.

26. 27.

28.

29.

30.

31.

32. 33.

34.

Xu H, Huang X, A¨rnlo¨v J et al. Clinical correlates of insulin sensitivity and its association with mortality among men with CKD stages 3 and 4. Clin J Am Soc Nephrol 2014; doi:10.2215/cjn.05230513. Shinohara K, Shoji T, Emoto M et al. Insulin resistance as an independent predictor of cardiovascular mortality in patients with end-stage renal disease. J Am Soc Nephrol 2002; 13: 1894–1900. Becker B, Kronenberg F, Kielstein JT et al. Renal insulin resistance syndrome, adiponectin and cardiovascular events in patients with kidney disease: the mild and moderate kidney disease study. J Am Soc Nephrol 2005; 16: 1091–1098. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977; 33: 159–174. Hung AM, Sundell MB, Egbert P et al. A comparison of novel and commonly-used indices of insulin sensitivity in African American chronic hemodialysis patients. Clin J Am Soc Nephrol 2011; 6: 767–774. Matsuhisa M, Yamasaki Y, Emoto M et al. A novel index of insulin resistance determined from the homeostasis model assessment index and adiponectin levels in Japanese subjects. Diabetes Res Clin Pract 2007; 77: 151–154. Matthews DR, Hosker JP, Rudenski AS et al. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 1985; 28: 412–419. Katz A, Nambi SS, Mather K et al. Quantitative insulin sensitivity check index: a simple, accurate method for assessing insulin sensitivity in humans. J Clin Endocrinol Metab 2000; 85: 2402–2410. McAuley KA, Williams SM, Mann JI et al. Diagnosing insulin resistance in the general population. Diabetes Care 2001; 24: 460–464. Purves RD. Optimum numerical integration methods for estimation of area-under-the-curve (AUC) and area-under-the-moment-curve (AUMC). J Pharmacokinet Biopharm 1992; 20: 211–226. Stumvoll M, Mitrakou A, Pimenta W et al. Use of the oral glucose tolerance test to assess insulin release and insulin sensitivity. Diabetes Care 2000; 23: 295–301. Matsuda M, DeFronzo RA. Insulin sensitivity indices obtained from oral glucose tolerance testing: comparison with the euglycemic insulin clamp. Diabetes Care 1999; 22: 1462–1470. Reaven GM, Brand RJ, Chen YD et al. Insulin resistance and insulin secretion are determinants of oral glucose tolerance in normal individuals. Diabetes 1993; 42: 1324–1332. Yeni-Komshian H, Carantoni M, Abbasi F et al. Relationship between several surrogate estimates of insulin resistance and quantification of insulin-mediated glucose disposal in 490 healthy nondiabetic volunteers. Diabetes Care 2000; 23: 171–175. Oterdoom LH, de Vries AP, van Son WJ et al. Validation of insulin resistance indexes in a stable renal transplant population. Diabetes Care 2005; 28: 2424–2429. Harris KP, Russell GI, Parvin SD et al. Alterations in lipid and carbohydrate metabolism attributable to cyclosporin A in renal transplant recipients. Br Med J (Clin Res Ed) 1986; 292: 16. Duckworth WC, Bennett RG, Hamel FG. Insulin degradation: progress and potential. Endocr Rev 1998; 19: 608–624. Pyorala K. Relationship of glucose tolerance and plasma insulin to the incidence of coronary heart disease: results from two population studies in Finland. Diabetes Care 1979; 2: 131–141. Tuomilehto J, Hu G, Qiao Q et al. Hyperinsulinemia as an independent risk factor for cardiovascular mortality in non-diabetic European men and women. Eur Heart J 2003; 24: 686–686. Despres JP, Lamarche B, Mauriege P et al. Hyperinsulinemia as an independent risk factor for ischemic heart disease. N Engl J Med 1996; 334: 952–957. Rutter MK, Meigs JB, Sullivan LM et al. Insulin resistance, the metabolic syndrome, and incident cardiovascular events in the Framingham Offspring Study. Diabetes 2005; 54: 3252–3257. Defronzo RA, Ferrannini E. Insulin resistance—a multifaceted syndrome responsible for NIDDM, obesity, hypertension, dyslipidemia, and atherosclerotic cardiovascular-disease. Diabetes Care 1991; 14: 173–194. Facchini FS, Hua N, Abbasi F et al. Insulin resistance as a predictor of age-related diseases. J Clin Endocr Metab 2001; 86: 3574–3578. Welborn TA, Wearne K. Coronary heart disease incidence and cardiovascular mortality in Busselton with reference to glucose and insulin concentrations. Diabetes Care 1979; 2: 154–160. de Boer IH, Katz R, Chonchol MB et al. Insulin resistance, Cystatin C, and mortality among older adults. Diabetes Care 2012; 35: 1355–1360. Kidney International (2014) 86, 383–391

clinical investigation

T Jia et al.: Validation of ISIs and prediction of clinical outcomes

35.

36.

37.

38.

39.

Niskanen L, Rauramaa R, Miettinen H et al. Carotid artery intima-media thickness in elderly patients with NIDDM and in nondiabetic subjects. Stroke 1996; 27: 1986–1992. Welin L, Eriksson H, Larsson B et al. Hyperinsulinaemia is not a major coronary risk factor in elderly men. The study of men born in 1913. Diabetologia 1992; 35: 766–770. Zethelius B, Lithell H, Hales CN et al. Insulin sensitivity, proinsulin and insulin as predictors of coronary heart disease. A population-based 10year, follow-up study in 70-year old men using the euglycaemic insulin clamp. Diabetologia 2005; 48: 862–867. Larsson A, Malm J, Grubb A et al. Calculation of glomerular filtration rate expressed in mL/min from plasma cystatin C values in mg/L. Scandinavian J Clin Lab Invest 2004; 64: 25–30. Levey AS, Coresh J, Balk E et al. National Kidney Foundation practice guidelines for chronic kidney disease: evaluation, classification, and stratification. Ann Int Med 2003; 139: 137–147.

Kidney International (2014) 86, 383–391

40.

41. 42.

43. 44. 45.

Byberg L, Zethelius B, McKeigue PM et al. Changes in physical activity are associated with changes in metabolic cardiovascular risk factors. Diabetologia 2001; 44: 2134–2139. L’Brown J. Efficacy of combined interferon and ribavirin for treatment of hepatitis C. Lancet 1998; 351: 78–79. Berglund L, Berne C, Svardsudd K et al. Early insulin response and insulin sensitivity are equally important as predictors of glucose tolerance after correction for measurement errors. Diabetes Res Clin Pract 2009; 86: 219–224. Wolfe DA. On testing equality of related correlation coefficients. Biometrika 1976; 63: 214–215. Cyr L, Francis K. Measures of clinical agreement for nominal and categorical data: the kappa coefficient. Comput Biol Med 1992; 22: 239–246. Chen H, Sullivan G, Quon MJ. Assessing the predictive accuracy of QUICKI as a surrogate index for insulin sensitivity using a calibration model. Diabetes 2005; 54: 1914–1925.

391