Short-Term Repeatability of Insulin Resistance Indexes in Older Adults: The Atherosclerosis Risk in Communities Study

Abstract

Context

The homeostatic model assessment of insulin resistance (HOMA-IR) and triglyceride (TG)/high-density lipoprotein cholesterol (HDL-C) ratio (TG/HDL-C) are insulin resistance indexes routinely used in clinical and population-based studies; however, their short-term repeatability is not well characterized.

Objective

To quantify the short-term repeatability of insulin resistance indexes and their analytes, consisting of fasting glucose and insulin for HOMA-IR and TG and HDL-C for TG/HDL-C.

Design

Prospective cohort study.

Participants

A total of 102 adults 68 to 88 years old without diabetes attended an initial examination and repeated examination (mean, 46 days; range, 28 to 102 days). Blood samples were collected, processed, shipped, and assayed following a standardized protocol.

Main Outcome Measures

Repeatability was quantified using the intraclass correlation coefficient (ICC) and within-person coefficient of variation (CV). Minimum detectable change (MDC₉₅) and minimum detectable difference with 95% confidence (MDD₉₅) were quantified.

Results

For HOMA-IR, insulin, and fasting glucose, the ICCs were 0.70, 0.68, and 0.70, respectively; their respective within-person CVs were 30.4%, 28.8%, and 5.6%. For TG/HDL-C, TG, and HDL-C, the ICCs were 0.80, 0.68, and 0.91, respectively; their respective within-person CVs were 23.0%, 20.6%, and 8.2%. The MDC₉₅ was 2.3 for HOMA-IR and 1.4 for TG/HDL-C. The MDD₉₅ for a sample of n = 100 was 0.8 for HOMA-IR and 0.6 for TG/HDL-C.

Conclusions

Short-term repeatability was fair to good for HOMA-IR and excellent for TG/HDL-C according to suggested benchmarks, reflecting the short-term variability of their analytes. These measurement properties can inform the use of these indexes in clinical and population-based studies.

We quantified the short-term repeatability of HOMA-IR and TG/HDL-C. We found short-term repeatability was fair to good for HOMA-IR and excellent for TG/HDL-C according to suggested benchmarks.

Insulin resistance is a metabolic imbalance defined by reduced response to insulin action in target tissues (1); it almost always precedes the development of metabolic syndrome and type 2 diabetes (2). A standard reference method to measure insulin resistance is the euglycemic-hyperinsulinemic clamp (3, 4). However, its implementation is difficult in large, population-based studies because of its invasive, time-consuming, and costly protocol. As a result, insulin resistance indexes are routinely used as insulin resistance proxy indicators in clinical and population-based studies.

Insulin resistance indexes are based on common analytes that reflect different aspects of insulin resistance. The homeostatic model assessment of insulin resistance (HOMA-IR) reflects dysregulation of fasting glucose and insulin; it is strongly correlated with direct measures of insulin-mediated glucose uptake (r = −0.82) (5). The triglyceride (TG)/high-density lipoprotein cholesterol (HDL-C) ratio (TG/HDL-C) reflects the dyslipidemia seen in insulin resistance; it is strongly correlated with direct measures of insulin-mediated glucose uptake (r = 0.60) and is more correlated with these measures than either TG (r = 0.57) or HDL-C is individually (r = −0.40) (6). Both insulin resistance indexes are minimally invasive and commonly used in research. However, although insulin assays are not well standardized, plasma TG and HDL-C assays are well standardized and widely used in research as well as in clinical settings (7–12).

The short-term repeatability of insulin resistance indexes has not been well characterized, which hinders their interpretation in exposure-outcome associations (13, 14). We thus quantified and compared the short-term repeatability of commonly used insulin resistance indexes and their constituent analytes—consisting of fasting glucose and insulin for HOMA-IR and TG and HDL-C for TG/HDL-C—in a community-based population of older adults.

Materials and Methods

Study population

The Atherosclerosis Risk in Communities (ARIC) Study is an ongoing prospective study of 15,792 adults 45 to 64 years old from four US communities (Washington County, Maryland; Forsyth County, North Carolina; Jackson, Mississippi; and the northwestern suburbs of Minneapolis, Minnesota) (15). Complete examinations were conducted in 1987 to 1989 (visit 1), 1990 to 1992 (visit 2), 1993 to 1995 (visit 3), 1996 to 1998 (visit 4), and 2011 to 2013 (visit 5). Before all examinations, participants were asked to fast for 8 hours, to bring in medications taken within the past 2 weeks, and to refrain from smoking and vigorous physical activity at least 1 hour before the examination. Institutional review boards at each site reviewed the study, and written informed consent was obtained from all participants.

A repeatability study was conducted to include an initial examination and a repeated examination scheduled 4 to 8 weeks later following the same standardized protocol (16). Participants were recruited in the order of arrival at the initial examination (visit 5; 2011 to 2013). The study sample included 200 participants. Our analytic sample included 102 participants after exclusions for (1) diabetes, defined by fasting glucose level ≥126 mg/dL (7 mmol/L), nonfasting glucose level ≥200 mg/dL (11.1 mmol/L), or use of diabetes medications (n = 65); (2) missing diabetes status (n = 6); (3) fasting for <8 hours at either examination (n = 9); (4) missing measurements at either examination (n = 11); and (5) measurements >3 standard deviations (SDs) from the mean at either examination (n = 7).

Blood collection, processing, and assays

The standardized protocol used for specimen collection, processing, shipping, and assay was the same at all examinations. Blood samples were collected by trained staff following a standard venipuncture protocol, processed within 90 minutes, and shipped weekly to the Atherosclerosis Clinical Research Laboratory at the Baylor College of Medicine. The central laboratory was masked; bar code ID labels on blood specimens for the regular examinations and repeated examination were indistinguishable (17). The interassay coefficient of variation (CV) for insulin was 10.6% [mean, 12.9 μU/mL (89.6 pmol/L)] and for fasting glucose was 3.1% [mean, 112.9 mg/dL (6.3 mmol/L)]. The interassay coefficient of variation for TG was 4.9% [mean, 125.2 mg/dL (1.4 mmol/L)] and for HDL-C was 4.2% [mean, 51.7 mg/dL (1.3 mmol/L)] (16, 17).

Insulin resistance indexes and their constituent analytes

We calculated HOMA-IR, equal to the product of insulin in μU/mL and fasting glucose in mg/dL divided by 405 (9). Insulin and fasting glucose were assayed using a radioimmunoassay method (Cobas e411; Roche) and an enzymatic method (Olympus AU480; Beckman/Olympus), respectively (18). We calculated TG/HDL-C, equal to the ratio of TG in mg/dL to HDL-C in mg/dL (10). TG and HDL-C were assayed using an enzymatic method (Olympus AU480; Beckman/Olympus) and a direct method (Olympus AU480; Beckman/Olympus), respectively (18).

Statistical analysis

We calculated summary statistics to describe measurements at both examinations. We also generated Bland-Altman plots to compare measurements at both examinations.

We used linear mixed-effects models to partition total variance (${\sigma }_{total}^2$) into between-person variance (${\sigma }_{bp}^2$) and within-person variance (${\sigma }_{wp}^2$). We quantified repeatability using the intraclass correlation coefficient (ICC), given by

$$ICC={\sigma }_{bp}^2/{\sigma }_{total}^2={\sigma }_{bp}^2/({\sigma }_{bp}^2+{\sigma }_{wp}^2),$$

and the within-person CV, given by

$$CV=\sqrt{{\sigma }_{wp}^2}/\mu \times 100,$$

where $\mu$ = mean of measurements at both examinations. Suggested benchmarks for ICC were defined as follows: poor (0 to 0.39), fair to good (0.40 to 0.75), and excellent (0.76 to 1) (19).

We calculated the minimum detectable change with 95% confidence (MDC₉₅) for a one-sample study design to describe true change beyond measurement error, equal to:

$$MD{C}_{95}=SEM\times \sqrt{2}\times z_{\alpha /2},\text{where}SEM=\sqrt{{\sigma }_{wp}^2}\text{and}{z}_{\alpha /2}=1.96$$

SEM being the standard error of measurement (20, 21). We calculated the minimum detectable difference with 95% confidence (MDD₉₅) for a two-sample study design to describe true difference beyond measurement error, equal to:

$$MD{D}_{95}=\sqrt{(2\times {\sigma }_{total}^2)/n}\times \left(t_{1-\alpha ,df}+t_{1-\beta ,df}\right),\text{where}\alpha =0.025,\beta =0.05,\text{and}df=2n-2.$$

where df represents degrees of freedom. We also calculated the $MD{C}_{95}$ and $MD{D}_{95}$ as a percentage of the grand mean.

Additional analyses were conducted. We quantified the repeatability of the insulin resistance indexes and their analytes after log transformation. We quantified the repeatability of the insulin resistance indexes and their analytes including measurements >3 SDs from the mean at each examination. We quantified the repeatability of the lipid-based insulin resistance index and its constituent analytes after excluding adults with statin use 4 weeks before the initial examination.

Results

Mean age was 76 years, 68% of participants were female (n = 69), and 24% were African American (n = 24). Mean waist circumference was 99 cm. Mean time between measurements was 46 days (range, 28 to 102 days). See Table 1 for the distribution of the measurements at each examination.

Table 1.

Descriptive Statistics of Insulin Resistance Indexes and Their Constituent Analytes in a Short-Term Repeatability Study of Adults Aged 68‒88 Years (n = 102), the ARIC Study (2011‒2013)

	Measurement 1	Measurement 2	Difference	Absolute Difference
	Median (P5, P95)	Median (P5, P95)	Median (P5, P95)	Median (P5, P95)
HOMA-IR	2.2 (0.9, 5.6)	2.3 (1.0, 5.2)	0.1 (−1.6, 1.5)	0.5 (0.1, 2.5)
Insulin, μU/mL	9.3 (4.1, 21.3)	9.2 (4.2, 19.9)	0.4 (−6.4, 5.8)	2.2 (0.2, 9.9)
Fasting glucose, mg/dL	100.5 (87.0, 118.0)	101.0 (86.0, 119.0)	−0.5 (−13.0, 12.0)	6.0 (0.0, 15.0)
TG/HDL-C	1.9 (0.9, 4.6)	1.9 (0.9, 4.2)	0.1 (−1.0, 1.1)	0.4 (0.1, 1.5)
TG, mg/dL	103.0 (61.0, 183.0)	103.5 (57.0, 180.0)	4.0 (−53.0, 37.0)	15.5 (1.0, 75.0)
HDL-C, mg/dL	52.0 (34.0, 85.0)	54.0 (35.0, 82.0)	1.0 (−11.0, 9.7)	4.2 (1.0, 11.0)

Conversion factors: insulin (1 pmol/L = 0.14 μU/mL); fasting glucose (1 mmol/L = 18 mg/dL); TG (1 mmol/L = 88.57 mg/dL); HDL-C (1 mmol/L = 38.67 mg/dL).

Abbreviations: P5, 5th percentile; P95, 95th percentile.

For HOMA-IR, insulin, and fasting glucose, the ICCs were 0.70, 0.68, and 0.70, respectively; between-person variance contributed a majority to total variance. For TG/HDL-C, TG, and HDL-C, the ICCs were 0.80, 0.68, and 0.91, respectively; between-person variance contributed a majority to total variance. Repeatability was higher for TG/HDL-C than for HOMA-IR (ICC, 0.80 vs 0.70) (Table 2).

Table 2.

Variance Components of Insulin Resistance Indexes and Their Constituent Analytes in a Short-Term Repeatability Study of Adults Aged 68‒88 Years (n = 102), the ARIC Study (2011‒2013)

	Between-Person SD (% Total)	Within-Person SD (% Total)	Grand Mean	ICC		CV	MDC95 (% Mean)
				Exclude Outliers	Include Outliers
HOMA-IR	1.2 (70.0)	0.8 (30.0)	2.7	0.70	0.77	30.4	2.3 (84.3)
Insulin, µU/mL	4.5 (68.4)	3.0 (31.6)	10.5	0.68	0.77	28.8	8.4 (79.7)
Fasting glucose, mg/dL	8.6 (69.5)	5.7 (30.5)	101.1	0.70	0.56	5.6	15.7 (15.5)
TG/HDL-C	1.0 (80.2)	0.5 (19.8)	2.2	0.80	0.81	23.0	1.4 (63.6)
TG, mg/dL	33.1 (68.5)	22.4 (31.5)	109.0	0.68	0.76	20.6	62.2 (57.0)
HDL-C, mg/dL	14.8 (91.5)	4.5 (8.5)	55.3	0.91	0.92	8.2	12.5 (22.7)

Samples excluding and including measurements >3 SDs from the mean were n = 102 and n = 109, respectively. Conversion factors: insulin (1 pmol/L = 0.14 μU/mL); fasting glucose (1 mmol/L = 18 mg/dL); TG (1 mmol/L = 88.57 mg/dL); HDL-C (1 mmol/L = 38.67 mg/dL).

Within-person CVs for HOMA-IR, insulin, and fasting glucose were 30.4%, 28.8%, and 5.6%, respectively; dispersion was greater for insulin. Within-person CVs for TG/HDL-C, TG, and HDL-C were 23.0%, 20.6%, and 8.2%, respectively; dispersion was greater for TG. Dispersion for TG/HDL-C was lower than dispersion for HOMA-IR (CV, 23.0% vs 30.4%) (Table 2).

In this community-based sample of older adults, the MDC₉₅ for HOMA-IR was 2.3, for insulin was 8.4 µU/mL (58.3 pmol/L), and for fasting glucose was 15.7 mg/dL (0.9 mmol/L), representing respective percentage changes between two time points of 84.3%, 79.7%, and 15.5%. The MDC₉₅ for TG/HDL-C was 1.4, for TG was 62.2 mg/dL (0.7 mmol/L), and for HDL-C was 12.5 mg/dL (0.3 mmol/L), representing respective percentage changes between two time points of 63.6%, 57.0%, and 22.7%. The detectable percentage change was lower for TG/HDL-C than for HOMA-IR (percentage change, 63.6% vs 84.3%) (Table 2).

For a study of N = 100 in each of two samples, a mean difference <0.8 for HOMA-IR, <2.8 µU/mL for insulin (19.4 pmol/L), and <5.3 mg/dL (0.3 mg/dL) for fasting glucose was attributable to measurement error; the corresponding percentage differences between two samples were 28.4%, 26.2%, and 5.2% (Fig. 1). A mean difference <0.6 for TG/HDL-C, <20.5 mg/dL (0.2 mmol/L) for TG, and <7.9 mg/dL (0.2 mmol/L) for HDL-C was attributable to measurement error; the corresponding percentage differences between two samples were 26.4%, 18.8%, and 14.4% (Fig. 2). The detectable percentage difference was somewhat lower for TG/HDL-C than for HOMA-IR (percentage difference, 26.4% vs 28.4%).

Figure 1.

Minimum detectable difference (MDD₉₅) of (A) insulin resistance index HOMA-IR and constituent analytes, (B) insulin, and (C) fasting glucose in a short-term repeatability study of adults 68 to 88 years old (n = 102) in the ARIC Study (2011 to 2013). Sample sizes per group were 5, 10, 50, 100, 500, and 1000. Grand mean for HOMA-IR was 2.7, for insulin was 10.5 μU/mL, and for fasting glucose was 101.1 mg/dL. Conversion factors: insulin (1 pmol/L = 0.14 μU/mL); fasting glucose (1 mmol/L = 18 mg/dL).

Figure 2.

Minimum detectable difference (MDD₉₅) of (A) insulin resistance index TG/HDL-C and constituent analytes, (B) TG, and (C) HDL-C in a short-term repeatability study of adults 68 to 88 years old (n = 102) in the ARIC Study (2011 to 2013). Sample sizes per group were 5, 10, 50, 100, 500, and 1000. Grand mean for TG/HDL-C was 2.2, for TG was 109.0 mg/dL, and for HDL-C was 55.3 mg/dL. Conversion factors: TG (1 mmol/L = 88.57 mg/dL); HDL-C (1 mmol/L = 38.67 mg/dL).

We quantified the repeatability of the insulin resistance indexes and their analytes after log transformation, motivated by skew in their distribution, as well as higher differences at higher mean levels of measurements (Figs. 3 and 4; Supplemental Figs. 1 and 2). For log-transformed HOMA-IR, insulin, and fasting glucose, the ICCs were 0.74, 0.73, and 0.70, respectively. For log-transformed TG/HDL-C, TG, and HDL-C, the ICCs were 0.82, 0.74, and 0.92, respectively (Supplemental Figs. 3 and 4; Supplemental Table 1). Because the ICCs were similar before and after log transformation, our main results were presented in the original units.

Figure 3.

Bland-Altman plots of (A) insulin resistance index HOMA-IR and constituent analytes, (B) insulin, and (C) fasting glucose in a short-term repeatability study of adults 68 to 88 years old (n = 102) in the ARIC Study (2011 to 2013). Mean difference (solid line) and limits of agreement (dashed lines) are displayed in the plot region; their respective values are displayed in the right-hand corner. Conversion factors: insulin (1 pmol/L = 0.14 μU/mL); fasting glucose (1 mmol/L = 18 mg/dL).

Figure 4.

Bland-Altman plots of (A) insulin resistance index TG/HDL-C and constituent analytes, (B) TG, and (C) HDL-C in a short-term repeatability study of adults 68 to 88 years old (n = 102) in the ARIC Study (2011 to 2013). Mean difference (solid line) and limits of agreement (dashed lines) are displayed in the plot region; their values are displayed in the right-hand corner. Conversion factors: TG (1 mmol/L = 88.57 mg/dL); HDL-C (1 mmol/L = 38.67 mg/dL).

In our main analyses, we excluded measurements >3 SDs from the mean at each examination (n = 7). To gauge their influence, we quantified the repeatability of the insulin resistance indexes and their constituent analytes including all observations. The replicate measurements for the outliers across the two examinations were not consistent, as indicated by the differences in measurements between the two examinations (Supplemental Figs. 5 and 6). Including outliers, the ICCs for HOMA-IR, insulin, and fasting glucose were 0.77, 0.77, and 0.56, respectively. Including outliers, the ICCs for TG/HDL-C, TG, and HDL-C were 0.81, 0.76, and 0.92, respectively. Thus, consistent with our main results, repeatability was somewhat higher for TG/HDL-C than for HOMA-IR (ICC, 0.81 vs 0.77) (Table 2).

We quantified the repeatability of TG/HDL-C, TG, and HDL-C after excluding statin use. Mean TG/HDL-C, TG, and HDL-C levels did not differ by statin use (Supplemental Table 2). The ICCs in adults without statin use for TG/HDL-C, TG, and HDL-C were 0.79, 0.63, and 0.91, respectively. Thus, short-term repeatability of the latter in adults without statin use (n = 52) did not differ from our main finding in all adults (n = 102) (Supplemental Table 3).

Discussion

In a community-based population of older adults, the short-term repeatability was fair to good for HOMA-IR and excellent for TG/HDL-C according to suggested benchmarks (19). The short-term repeatability of the indexes was consistent with the short-term repeatability of their respective analytes (mean, 46 days; range, 28 to 102 days).

Given its formulation, we interpreted our findings in the context of our study population (22). We recognize that participants who attend a study may differ in general from participants who do not attend a study. We endeavored to ensure internal validity by recruiting participants at random, by implementing a standardized protocol, and by using exclusion criteria that could affect measurement accuracy. As such, our findings represent the intraindividual variability of the insulin resistance indexes among free-living older adults without diabetes.

Short-term repeatability for insulin, fasting glucose, TG, and HDL-C has been reported in prior studies. In adults 45 to 64 years old examined 1 to 2 weeks apart (n = 40), the ICC for insulin was 0.81, for fasting glucose was 0.84, for TG was 0.85, and for HDL-C was 0.94 (22, 23). Under similarly standardized conditions, we observed lower ICCs, which may have been due to the longer time elapsed between repeated examinations. In adults 20 years and older (n = 2426), the within-person CV for insulin was 25.2% [mean, 9.56 µU/mL (66.36 pmol/L)], for fasting glucose was 8.3% [mean, 92.52 mg/dL (5.14 mmol/L)], for TG was 28.8% [mean, 132.61 mg/dL (1.49 mmol/L)], and for HDL-C was 12.4% [mean, 50.66 mg/dL (1.31 mmol/L)] (24). We observed consistently higher within-person CVs for insulin and TG among older adults.

Our results inform the use of insulin resistance indexes in clinical and population-based research. Adding new information, we characterized the short-term repeatability of a glucose and insulin–based index, as well as the short-term repeatability of an alternative TG and HDL-C‒based index that circumvents well-known issues with insulin assays and standardization. We calculated the MDC₉₅ and MDD₉₅ for each index, which can be used to inform future one-sample study designs (e.g., longitudinal studies with repeated measurements) and two-sample study designs (e.g., clinical studies with an intervention), respectively.

As a limitation to consider, although participants agreed to fast following a late snack on the night before the examination, fasting time (≥8 hours before venipuncture) was self-reported. In an attempt to address this issue, participants were excluded for measurements >3 SDs from the mean at either examination.

In summary, short-term repeatability was fair to good for HOMA-IR and excellent for TG/HDL-C according to suggested benchmarks, reflecting the short-term variability of their constituent analytes (mean, 46 days; range, 28 to 102 days). These measurement properties can inform the use of such metrics in clinical and population-based studies.

Glossary

Abbreviations:

ARIC

Atherosclerosis Risk in Communities

CV

coefficient of variation

df

degrees of freedom

HDL-C

high-density lipoprotein cholesterol

HOMA-IR

homeostatic model assessment of insulin resistance

ICC

intraclass correlation coefficient

MDC₉₅

minimum detectable change with 95% confidence

MDD₉₅

minimum detectable difference with 95% confidence

SD

standard deviation

SEM

standard error of measurement

TG

triglyceride

TG/HDL-C

triglyceride/high-density lipoprotein cholesterol ratio

${\sigma }_{bp}^2$

between-person variance

${\sigma }_{total}^2$

total variance

${\sigma }_{wp}^2$

within-person variance