Validation of Steady-State Insulin Sensitivity Indices in Chronic Kidney Disease

Eligible subjects were aged 18–75 years and 1) had an estimated glomerular filtration rate (eGFR) <75 ml/min per 1.73 m² by the four-variable Modification of Diet in Renal Disease estimating equation using a single serum creatinine value standardized to the Cleveland Clinic Foundation laboratory or 2) received maintenance hemodialysis for end-stage renal disease. Exclusion criteria included pregnancy, known diabetes or fasting blood glucose >125 mg/dl at screening, corticosteroid use within the past 6 months, or a terminal diagnosis. Subjects were recruited by advertisement and by investigator-initiated query at primary care clinics, nephrology clinics, and maintenance hemodialysis facilities at or affiliated with the hospital of the University of Pennsylvania or the Philadelphia Veterans Administration Medical Center. Written informed consent was obtained from each subject in compliance with institutional review boards at these institutions.

The study procedure took place during a single admission to the General Clinical Research Center at the hospital of the University of Pennsylvania. After an overnight fast, subjects underwent a 3-h FSIVGTT, as previously described (25), with the dose of insulin administered modified to be 0.01, 0.02, or 0.03 units/kg (based on eGFR <30, 30–<60, or >60 ml/min per 1.73m², respectively).

Demographic variables and body weight and height were collected. Plasma immunoreactive insulin was measured in duplicate by double-antibody radioimmunoassay (Linco Research, St. Louis, MO). Plasma glucose was measured by automated glucose meter (YSI, Yellow Springs, OH) at bedside during the FSIVGTT. SI-FSIVGTT was derived using MINMOD Millennium (version 6.02) software (26).

Letting I₀ and G₀ equal the mean of four fasting plasma insulin and glucose values measured at time −15, −10, −5, and −1 min (immediately before the FSIVGTT), the following SI-SS indices were derived: HOMA of insulin resistance (IR-HOMA) = [I₀(μU/ml) × G₀(mmol/l)]/22.5, log(IR-HOMA), SI-HOMA = 1/(IR-HOMA), and QUICKI (quantitative insulin sensitivity check index) = 1/[log (I₀) + log (G₀)]. I₀, log(I₀), and 1/I₀ also were considered. SI-SS values based only on single insulin and glucose measurements, obtained at the −1 min time point, also were examined. Glucose values were converted to millimoles per liter (mmol/l) for SI-SS calculations.

Standard descriptive statistics were used. The assumption of normal distribution across the independent variables was examined. Treating SI-FSIVGTT as the dependent variable, univariable linear regression models for each SI-SS index were fit. Pearson correlation coefficients (r) were tested using one-sample t tests against a null hypothesis that the population correlation coefficient (ρ) = 0. Interactions between these correlations and potential covariables including eGFR, age, race (black versus nonblack), sex, and BMI were explored as discussed in results. In supplemental analyses, Spearman's rank correlations (r_s) also were obtained. A two-tailed P value <0.05 was the criterion for statistical significance. Analyses were performed using STATA 9.0.

Table 1 provides demographic and clinical characteristics of the 27 study subjects categorized by eGFR level. Median age was 47 years (range 21–73), median BMI was 30.0 kg/m² (18.2–73.3), 67% (18 of 27) of subjects were male, and 63% (17 of 27) were African American. Four subjects had end-stage renal disease requiring maintenance hemodialysis. Among the others, eGFR ranged from 16 to 70 ml/min per 1.73m² (overall median = 48).

I₀, G₀, IR-HOMA, and SI-FSIVGTT also are summarized in Table 1. I₀ and G₀ are reported as the within-subject mean of measurements at four time points over 15 min. The median (range) of these insulin and glucose values were 12.8 μU/ml (2.8–48.7) and 77 mg/dl (57–129), respectively. Between-subject variability, expressed as coefficient of variation (CV) ([SD divided by the mean] × 100), was much larger for insulin (71%) than for glucose (19%). There was moderate within-subject variability in fasting insulin levels, consistent with assay characteristics and known fluctuations in plasma insulin levels over short periods of time (27). The median (range) intrasubject CV for fasting insulin levels was 10% (4–40), and three of the 27 subjects had a CV >20% (22, 24, and 40%). By contrast, the median (range) intrasubject CV for fasting glucose levels was 3% (0.5–9).

The two-way distributions of SI-FSIVGTT and six of the SI-SS indices are provided graphically in Fig. 1, and Pearson correlation coefficients are listed in Table 2. All correlations were statistically significant. For indices obtained using the mean of four fasting insulin and glucose values over 15 min, Pearson correlation coefficients (|r|) ranged from 0.51 to 0.87 (P < 0.01 for each). For indices using single fasting insulin and glucose values, |r| ranged from 0.51 to 0.72 (P < 0.01 for each). Indices using log or reciprocal transformations to limit the impact of outlying data [log(IR-HOMA), SI-HOMA, quantitative insulin sensitivity check index, log(I₀), and 1/I₀] were more highly correlated with SI-FSIVGTT than indices using untransformed data (IR-HOMA and I₀). In general, indices based on I₀ alone were at least as highly correlated with SI-FSIVGTT as indices derived from both I₀ and G₀ data. By both the four and one time point approaches, 1/I₀ had the highest (or nearly highest) correlation with SI-FSIVGTT (r = 0.87 and 0.71, respectively).

Use of one time point rather than four to derive I₀ and G₀ substantially attenuated the correlations with SI-FSIVGTT for indices using reciprocal transformations (SI-HOMA, quantitative insulin sensitivity check index, and 1/I₀) but had more modest impact on the correlations with SI-FSIVGTT for the other indices. This finding was principally explained by one subject whose fasting plasma insulin levels were highly variable (CV 40% for the four fasting insulin values). After exclusion of this subject, Pearson correlation coefficients with SI-FSIVGTT for all SI-SS indices derived from single time point values differed by <2% from the corresponding correlation coefficients based on four time point values. Because SI-FSIVGTT was not normally distributed, Spearman's rank correlations (r_s) also were obtained. |r_s| = ≥0.77 (P < 0.0001) for all indices assessed by both the four and one time point approaches.

Findings among subjects stratified according to eGFR level were comparable with those combining subjects across eGFR levels. For indices based on the mean of four fasting time points, the log- or reciprocal-transformed indices were most highly correlated with SI-FSIVGTT at both higher and lower eGFR levels, whereas for indices based on a single fasting time points the log-transformed indices were most highly correlated with SI-FSIVGTT. By the four–time point approach, Pearson correlation coefficients (|r|) with SI-FSIVGTT among 13 subjects with eGFR <48 ml/min per 1.73m² (the median eGFR value) were 0.47 (P = 0.11) for IR-HOMA, 0.63 (P = 0.02) for log IR-HOMA, 0.49 (P = 0.09) for I₀, and ranged from 0.69 to 0.86 (all P < 0.01) for the other four SI-SS indices. Among 14 subjects with eGFR ≥48 ml/min per 1.73m², the Pearson correlation coefficient (|r|) was 0.63 (P = 0.02) for IR-HOMA and 0.85–0.89 (all P < 0.01) for the other six indices. In keeping with these findings, tests for whether the association of each SI-SS with SI-FSIVGTT changed according to eGFR level < or ≥48 ml/min per 1.73m² were not significant (P values 0.32–0.97). In additional analyses, the Pearson correlation coefficients (|r|) were 0.73–0.96 for eGFR 60–70 ml/min per 1.73m² (CKD stage 2; n = 4), 0.53 (IR-HOMA; P = 0.02) to 0.89 (1/I₀; P < 0.0001) for eGFR 15 to <60 ml/min per 1.73m² (CKD stages 3–4; n = 19), and 0.31–0.45 for eGFR <15 ml/min per 1.73m² (CKD stage 5; n = 4). The precision of the estimates and the significance levels for CKD stages 2 and 5 were limited due to the small sample size in these categories.

Each SI-SS index retained statistically significant association with SI-FSIVGTT in multivariable linear regression models (P < 0.01 for each SI-SS index, in separate models). In the same models, age, BMI, and eGFR level were not significantly associated with SI-FSIVGTT, whereas black race was associated with greater insulin resistance by SI-FSIVGTT than nonblack race (all P ≤ 0.05). In a series of exploratory analyses, correlations between SI-FSIVGTT and each SI-SS index also did not vary significantly when stratified by age, race, sex, or BMI.

Significant inverse correlations were found between G₀ and FSIVGTT-derived parameters of pancreatic β-cell function including both the disposition index (AIR^gluc × SI) and the acute insulin response to glucose (AIR^gluc) as follows (Fig. 2): disposition index (r = −0.59, P = 0.001), AIR^gluc (r = −0.49, P = 0.009). Conversely, no correlation was seen between G₀ and SI-FSIVGTT (r = 0.02, P = 0.93).

This study demonstrates, in subjects with CKD, a strong and significant correlation between insulin sensitivity estimates derived from fasting glucose and insulin data and the MINMOD-derived parameter of SI-FSIVGTT. Validity of these indices has been demonstrated in the general population (14,15) and in some (28,29), but not all (23), select populations. A single study has validated the use of SI-SS indices in renal transplant recipients (30), but the relevance of that study to the nontransplant CKD population is limited by important factors. These include the impact of immune-suppressive agents on glucose metabolism (31) and the excellent graft function of its subjects, which impairs its relevance to the setting of advanced renal insufficiency. In contrast, 23 of 27 subjects in the current study had K/DOQI stages 3–5 (glomerular filtration rate <60 ml/min per 1.73m²) CKD (32). An additional study did correlate IR-HOMA with SI-clamp in a population that contained a minority of subjects with elevated serum creatinine (17 of 113 = 15%) but did not report correlation specifically in this group with renal insufficiency (33). Therefore, the current study provides the first specific validation of SI-SS indices across a wide spectrum of CKD in nontransplant subjects.

The sample size of this study was relatively small, and replication of these results at other centers will strengthen these conclusions. However, the overall correlations demonstrated here between SI-SS indices and SI-FSIVGTT are statistically significant and similar in strength to those seen in multiple studies across diverse populations (16,28–30). Our results specifically support the utility of the SI-SS indices in stage 3 CKD, a critical group because of both its large size (3) and the substantial associated risk of CVD (4). This suggests that CKD does not disrupt the correlation between hepatic insulin sensitivity, which the SI-SS indices principally reflect, and insulin sensitivity at peripheral tissues including skeletal muscle and adipose, which contribute substantially to insulin sensitivity estimated with dynamic methods such as the FSIVGTT (16,34). The small number of subjects with stage 5 CKD precludes any conclusion specifically pertaining to this group.

Our data indicate that valid insulin sensitivity estimates can be effectively derived from single fasting insulin and glucose measurements in nondiabetic CKD subjects. This is in agreement with data from a diabetic population (15) and greatly enhances the feasibility of using SI-SS indices in large CKD studies. Further, and as seen in other studies (29,30,35), parameters derived from both insulin and glucose values were not superior to either raw fasting insulin values or transformations of fasting insulin (logI₀, 1/I₀). Fasting glucose concentration may contribute relatively little information regarding insulin sensitivity in nondiabetic populations, as it is maintained in a relatively restricted range, whereas fasting insulin concentrations vary widely according to insulin sensitivity (35). This was reflected in the current study by the much larger between-subject CV for insulin than for glucose. Finally, SI-SS indices using logarithmic or inverse transformations were uniformly more highly correlated with SI-FSIVGTT than untransformed SI-SS indices (IR-HOMA and I₀), suggesting that data transformation should be utilized. Overall, our findings indicate that measurement of a single fasting insulin value will permit effective insulin sensitivity estimation in nondiabetic CKD subjects.

As fasting blood glucose did correlate significantly with FSIVGTT-derived parameters of β-cell function but did not correlate with SI-FSIVGTT, it is likely that β-cell function plays a much more important role than insulin sensitivity in determining fasting blood glucose concentrations in this nondiabetic CKD population. This inverse correlation between fasting blood glucose and β-cell function has been observed in subjects with normal glucose tolerance and impaired fasting glucose (36), as well as in pancreatic islet transplant recipients (37). This is consistent with the postulate that insulin secretion will compensate for insulin resistance in the setting of CKD if β-cell function is adequate.

Insulin sensitivity estimation has been used in large-scale studies for multiple purposes. These include exploration of the association between insulin resistance and CVD (38–42), as well as the investigation of other potential emerging CVD risk factors (17) and the physiologic impact of genetic variations within metabolically active genes (18,43). Therefore, the use of these indices will significantly expand the hypotheses that may be addressed in Chronic Renal Insufficiency Cohort and other CKD population-based studies, including those exploring the relationship between insulin resistance and both CVD and the progression of renal insufficiency in this select population.

Study conduct and manuscript preparation were supported by the following National Institutes of Health Grants: M01-RR00040 (University of Pennsylvania General Clinical Research Center), the K12 (RR017625-03) Career Development Award (to M.C.), the K23 (DK-066327) Career Development Award (to B.R.), and UO1-DK060984.