Abstract
Goal
We sought to develop an instrument to screen for insulin resistance in non-diabetic patients with recent ischemic stroke or transient ischemic attack (TIA).
Materials and methods
Subjects were 7262 nondiabetic patients aged greater than or equal to 40 years with ischemic strokes or TIA within the past 6 months. Features were analyzed in bivariate analysis for association with insulin resistance, measured by the homeostasis model assessment of insulin resistance (HOMA-IR). Features significantly associated with HOMA-IR (P < .05) were entered into multivariable analysis. The magnitudes of regression coefficients from the multivariable model were used to assign point values for 2 diagnostic scoring instruments: a basic instrument that did not incorporate laboratory test values and an enhanced instrument that did. The performance of the instruments was tested using receiver operating characteristic (ROC) analysis.
Findings
In the basic model, 5 features were retained in the multivariable regression analysis: male gender, abdominal obesity, body mass index (BMI), elevated waist-to-hip ratio, and systolic blood pressure. In the enhanced model, 4 features were retained in the multivariable regression analysis: BMI, abdominal obesity, fasting glucose greater than or equal to 100 mg/dL, and triglyceride/high-density lipoprotein ratio. In the basic model, the area under the ROC curve (aROC) was .73 in the validation cohort. In the enhanced model, the aROC was .78 in the validation cohort.
Conclusions
Our 2 scoring systems performed well in identifying stroke patients with insulin resistance, but they are probably not sufficiently accurate for high-stake clinical decisions. We suggest strategies for improving the accuracy of future instruments.
Keywords: Stroke, ischemic stroke, transient ischemic attack, insulin resistance, obesity
Background
Insulin resistance describes a condition in which decreased tissue sensitivity to the metabolic effects of the hormone insulin results in hyperglycemia and compensatory hyperinsulinemia.1 Insulin resistance, along with insulin deficiency, is one of the fundamental physiologic defects in type 2 diabetes mellitus and has been independently associated with increased risk of development of diabetes, nonalcoholic fatty liver disease, cardiovascular disease, and ischemic stroke.1–5
Despite the known public health impact of insulin resistance, identification of nondiabetic patients with the condition remains challenging. Many patients have normal fasting glucose and so require unconventional testing for diagnosis of occult impairment. The gold standard is the hyperinsulinemic–euglycemic clamp. However, this method is expensive, time consuming, and impractical for use in routine clinical practice. Simpler indices for measuring insulin resistance, including the homeostasis model assessment of insulin resistance (HOMA-IR), the quantitative insulin sensitivity check index, and the McAuley index, rely on measures of serum fasting insulin, which are not standardized across laboratory platforms.6–9 Development of a clinical instrument that identifies patients with higher risk for insulin resistance could help clinicians identify a subset of patients who might benefit from more complex testing.
The value of screening and testing for insulin resistance depends on the availability of effective follow-up treatment. Weight loss, dietary modification, and drugs can each improve insulin sensitivity.10–13 Only drugs, however, have been tested for efficacy in preventing clinical outcomes. The PROactive trial demonstrated that treatment with pioglitazone reduced major adverse cardiovascular events in patients with type 2 diabetes.14,15 In addition, the Insulin Resistance Intervention after Stroke (IRIS) study trial recently demonstrated that pioglitazone was effective in preventing stroke and myocardial infarction among non-diabetic patients with a recent ischemic stroke or transient ischemic attack.16
In this study, using data obtained during the IRIS trial’s enrollment phase, we sought to develop a simple and reliable instrument based on routine clinical assessment and laboratory tests to identify patients at high risk for insulin resistance.
Research Methods
Study Population and Data Collection
The study population comprised subjects screened for eligibility to participate in the IRIS trial. IRIS was a multinational randomized trial that tested the effectiveness of pioglitazone, compared with placebo, for prevention of stroke and myocardial infarction among nondiabetic patients with a recent ischemic stroke or transient ischemic attack (TIA).16 To be eligible for randomization in IRIS, patients were required to have insulin resistance, defined as a homeostasis model assessment-insulin resistance (HOMA-IR) value (calculated as [fasting insulin, µU/mL × fasting glucose, mmol/L]/22.5) greater than 3.0. At the screening blood test, the subject’s age, gender, and modified Rankin Scale grade were recorded. For the current study, we included only IRIS subjects screened after November 2005 when data collection was broadened to include race, ethnicity, blood pressure, and body habitus measures. All blood tests were processed at a central laboratory (Esoterix, Inc., Austin, TX, or an affiliate laboratory) and included measurements of fasting insulin, glucose, total cholesterol, high-density lipoprotein (HDL), lowdensity lipoprotein, and triglycerides. If HOMA-IR exceeded 3.0, further testing for C-reactive protein and hemoglobin A1c (HgbA1c) was performed on the screening sample. Eligible subjects who were randomized into the IRIS trial underwent a detailed baseline interview and cognitive examination.
Statistical Methods
For this study, screened IRIS participants, both those found to be insulin resistant and non-insulin resistant, were divided randomly into a development set (60%) and a validation set (40%).17 Candidate predictive features were selected for examination in the development set based on associations with insulin resistance described in prior research (i.e., age, sex, race, Hispanic ethnicity, functional status, body mass index [BMI], central obesity, waist-to-hip ratio, blood pressure, fasting glucose, triglyceride/HDL ratio).18–21 We defined BMI using commonly accepted criteria for normal, overweight, and obese.22 Abdominal obesity was defined as greater than 88 cm for women and greater than 102 cm for men. Elevated waist-to-hip ratio was defined as greater than .9 in men and greater than .85 in women. Elevated systolic blood pressure (SBP) was defined as greater than or equal to 130 mmHg.23
Each candidate feature was analyzed individually in the development set for association with HOMA-IR defined as a continuous variable. Because HOMA-IR values were not distributed normally in the study population, a Spearman rank correlation coefficient was used to measure the strength of associations in bivariate analyses. Multiple linear regression analysis was then performed to examine the relationship between features that were significantly associated with HOMA-IR (P < .05) in bivariate analysis. A stepwise algorithm was used with P value for inclusion and retention of .05. Independent features selected in the final regression model were then used to create 2 diagnostic scoring instruments based on the relative magnitude of the regression coefficients: (1) an instrument that incorporates routine clinical variables not requiring blood samples (basic instrument); and (2) an instrument that incorporates routine clinical variables and common laboratory measurements (enhanced instrument).
Defining the presence of insulin resistance as HOMA-IR greater than 3.0, the diagnostic instruments were then tested in the development and validation sets for sensitivity, specificity, and diagnostic accuracy. In addition, we measured the performance of the instruments in both sets using receiver operating characteristic (ROC) analyses.
For randomized IRIS participants, additional patient features were analyzed for their association with HOMA-IR using the same methods as above. Our goal was to identify candidate variables for inclusion in a future improved instrument.
Results
Patients
Among the 7599 patients screened for the IRIS trial, 7262 were screened on or after November 1, 2005, and are included in the present study. The study population was split into a development cohort of 4357 patients and a validation cohort of 2905 patients. No significant differences in demographic or clinical data were observed between the development and validation cohorts (Table 1).
Table 1.
Characteristics of development and validation cohorts (n = 7262)
| Cohort | ||
|---|---|---|
|
|
||
| Feature | Development N = 4357 |
Validation N =2905 |
| Demographic features | ||
| Age (years, mean) | 64 ± 11 | 64 ± 11 |
| Male (%) | 64% | 65% |
| Black race (%) | 11% | 11% |
| Hispanic ethnicity (%) | 3% | 4% |
| Physical examination | ||
| BMI (kg/m2, mean) | 28 ± 5 | 28 ± 6 |
| Abdominal obesity | 52% | 51% |
| SBP (mmHg, mean) | 136 ± 19 | 137 ± 19 |
| DBP (mmHg, mean) | 81 ± 11 | 81 ± 11 |
| Modified Rankin Scale score (mean) | .9 ± 1.0 | .9 ± 1.0 |
| Laboratory data | ||
| HOMA-IR (mean) | 4.5 ± 4.1 | 4.4 ± 3.3 |
| LDL (mmol/L, mean) | 88 ± 32 | 89 ± 32 |
| HDL (mmol/L, mean) | 49 ± 13 | 50 ± 15 |
| TG (mg/dL, mean) | 128 ± 67 | 129 ± 69 |
| TG/HDL (mean) | 3.0 ± 2.1 | 3.0 ± 2.4 |
| Fasting glucose (mg/dL) (mean) | 97 ± 13 | 97 ± 16 |
| Fasting insulin (µU/mL) (mean) | 18 ± 13 | 18 ± 11 |
| Randomized into IRIS (no.) | 2245 | 1465 |
Abbreviations: BMI, body mass index; DBP, diastolic blood pressure; HDL, high-density lipoprotein; HOMA-IR, homeostasis model assessment of insulin resistance; LDL, low-density lipoprotein; SBP, systolic blood pressure; TG, triglyceride.
Missing data:
Development: race (14); ethnicity (14); BMI (20); abdominal obesity (54); SBP (12); DBP (12); modified Rankin Scale (4); LDL (45); HDL (17); TG (16); TG/HDL ratio (17).
Validation: race (18); ethnicity (20); BMI (13); abdominal obesity (54); SBP (7); DBP (7); modified Rankin Scale (1); LDL (29); HDL (11); TG (10); TG/HDL ratio (11).
Development of the Basic and Enhanced Instruments
The results of the analyses of the development cohort are shown in Table 2 (bivariate analyses) and Table 3 (multivariate analyses).
Table 2.
Bivariate analysis
| Feature | No. | Median HOMA-IR |
R* | P value |
|---|---|---|---|---|
| Age | ||||
| <65 | 2339 | 3.9 | ||
| ≥65 | 2018 | 3.5 | .08 | <.0001 |
| Male gender | ||||
| No | 1568 | 3.6 | ||
| Yes | 2789 | 3.7 | .04 | .01 |
| Race | ||||
| White | 3625 | 3.6 | ||
| Black | 501 | 3.8 | ||
| Other | 169 | 3.9 | ||
| Uncertain | 48 | 3.6 | .02 | .07 |
| Hispanic ethnicity | ||||
| No | 4192 | 3.7 | ||
| Yes | 151 | 4.1 | .04 | .02 |
| Modified Rankin Scale grade | 4353 | .03 | .06 | |
| Body mass index (kg/m2) | ||||
| <25 | 1149 | 2.7 | ||
| 25–29 | 1788 | 3.6 | ||
| 30–34 | 940 | 4.6 | ||
| 35+ | 459 | 5.4 | .41 | <.0001 |
| Abdominal obesity† | ||||
| Absent | 2078 | 3.0 | ||
| Present | 2225 | 4.1 | .33 | <.0001 |
| Waist-to-hip ratio obesity‡ | ||||
| Absent | 2966 | 3.5 | ||
| Present | 1325 | 4.1 | .13 | <.0001 |
| Systolic BP (mmHg) | ||||
| <130 | 1683 | 3.5 | ||
| 130+ | 2661 | 3.8 | .07 | <.0001 |
| Diastolic BP (mmHg) | ||||
| <85 | 2825 | 3.5 | ||
| 85+ | 1519 | 4.0 | .09 | <.0001 |
| Glucose (mg/dL) | ||||
| <100 | 2886 | 3.2 | ||
| 100+ | 1470 | 4.9 | .38 | <.0001 |
| TG/HDL | ||||
| <1.6 | 1029 | 2.8 | ||
| 1.6–2.3 | 1067 | 3.4 | ||
| 2.4–3.5 | 1105 | 3.9 | ||
| 3.6+ | 1138 | 4.7 | .35 | <.0001 |
Abbreviations: BP, blood pressure; HDL, high-density lipoprotein; HOMA-IR, homeostasis model assessment of insulin resistance; TG, triglyceride.
Spearman rank correlation coefficient.
Greater than 88 cm for women and greater than 102 cm for men.
Greater than .9 in men and greater than .85 in women.
Table 3.
Multivariate analyses
| Basic model | Enhanced model | |||
|---|---|---|---|---|
|
|
|
|||
| Feature | Coefficient | Points | Coefficient | Points |
| Male gender | .72 | 2 | ||
| Body mass index (kg/m2) | ||||
| 25–29 | .77 | 2 | .49 | 1 |
| 30–35 | 1.62 | 4 | 1.14 | 2 |
| 35+ | 2.47 | 8 | 1.92 | 4 |
| Abdominal obesity* | .57 | 2 | .54 | 1 |
| Waist-to-hip ratio obesity† | .65 | 2 | ||
| Systolic BP ≥130 mmHg | .32 | 1 | ||
| Glucose ≥100 mg/dL | 1.99 | 4 | ||
| Triglycerides/HDL | ||||
| 1.6–2.3 | .35 | 1 | ||
| 2.4–3.5 | .83 | 2 | ||
| 3.6+ | 1.71 | 4 | ||
| Total possible points | 0–15 | 0–13 | ||
Abbreviations: BP, blood pressure; HDL, high-density lipoprotein.
Greater than 88 cm for women and greater than 102 cm for men.
Greater than .9 in men and greater than .85 in women.
In the basic model not requiring data from blood samples, 8 features were found to be significantly associated with HOMA-IR with P ≤ .05 (Table 2). Of these 8 features, 5 were found to be independent predictors in the multivariable regression analysis (Table 3) and allocated points in the model: male gender (2 points); BMI (25–29 kg/m2 [2 points]; 30–34 kg/m2 [4 points]; ≥35 kg/m2 [8 points]); abdominal obesity (2 points); elevated waist-to-hip ratio (2 points); and SBP (1 point). Points for features in the basic model yield a total score between 0 and 15 (Table 3).
In the enhanced model incorporating data from blood samples, 2 additional features were found to be significantly associated with HOMA-IR with P ≤ .05 (Table 3). Of these 10 features, 4 were retained in the multivariable regression analysis (Table 3): BMI (25–29 kg/m2 [1 point]; 30–34 kg/m2 [2 points]; ≥35 kg/m2 [4 points]); abdominal obesity (1 point); fasting glucose greater than or equal to 100 mg/dL (4 points); and triglyceride/HDL ratio (1.6–2.3 [1 point]; 2.4–3.5 [2 points]; ≥3.6 [4 points]). Points for features in the enhanced model yield a total score between 0 and 13 (Table 3).
Performance in the Development and Validation Cohorts
Using a HOMA-IR value of greater than 3.0 to define the presence of insulin resistance, the sensitivity, specificity, positive predictive value, and negative predictive value for each point value for both models are displayed in Tables 4 and 5. In the basic instrument, an optimal score of 6 had a sensitivity of 63% and a specificity of 69%, with a positive predictive value of 78%. In the enhanced instrument, an optimal score of 5 had a sensitivity of 68% and a specificity of 73%, with a positive predictive value of 81%. In the basic instrument, the area under the ROC curve (aROC) was .71 in the development cohort and .73 in the validation cohort. In the enhanced instrument, the aROC was .77 in the development cohort and .78 in the validation cohort.
Table 4.
Performance of the basic instrument
| Development cohort (n = 4280) | |||||
|
| |||||
| Score cutoff (greater than or equal to) | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | No. of patients with score less than cutoff |
|
| |||||
| 0 | 100 | 0 | 63 | — | 0 |
| 1 | 99 | 5 | 65 | 79 | 107 |
| 2 | 97 | 11 | 65 | 70 | 243 |
| 3 | 92 | 24 | 68 | 64 | 581 |
| 4 | 84 | 41 | 71 | 60 | 1072 |
| 5 | 76 | 53 | 74 | 56 | 1477 |
| 6 | 63 | 69 | 78 | 52 | 2091 |
| 7 | 54 | 75 | 79 | 48 | 2435 |
| 8 | 38 | 86 | 83 | 45 | 3024 |
| 9 | 31 | 90 | 84 | 43 | 3288 |
| 10 | 18 | 95 | 86 | 40 | 3724 |
| 11 | 16 | 96 | 87 | 40 | 3793 |
| 12 | 13 | 97 | 88 | 39 | 3891 |
| 13 | 9 | 98 | 87 | 38 | 3996 |
| 14 | 2 | 99 | 84 | 37 | 4212 |
| 15 | 1 | 99 | 80 | 37 | 4240 |
|
| |||||
| Validation cohort (n = 2844) | |||||
|
| |||||
| Score cutoff (greater than or equal to) | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | No. of patients with score less than cutoff |
|
| |||||
| 0 | 100 | 0 | 64 | — | 0 |
| 1 | 99 | 3 | 64 | 72 | 47 |
| 2 | 98 | 9 | 66 | 72 | 127 |
| 3 | 93 | 22 | 68 | 65 | 349 |
| 4 | 86 | 44 | 73 | 64 | 713 |
| 5 | 78 | 55 | 75 | 58 | 970 |
| 6 | 63 | 71 | 79 | 52 | 1399 |
| 7 | 55 | 76 | 80 | 49 | 1600 |
| 8 | 40 | 87 | 85 | 45 | 1976 |
| 9 | 32 | 91 | 86 | 43 | 2165 |
| 10 | 18 | 96 | 89 | 40 | 2483 |
| 11 | 16 | 96 | 89 | 39 | 2515 |
| 12 | 13 | 98 | 92 | 39 | 2584 |
| 13 | 10 | 99 | 93 | 38 | 2656 |
| 14 | 3 | 100 | 93 | 37 | 2783 |
| 15 | 2 | 100 | 92 | 36 | 2807 |
Abbreviations: NPV, negative predictive value; PPV, positive predictive value.
Table 5.
Performance of the enhanced instrument
| Development cohort (n = 4278) | |||||
|
| |||||
| Score cutoff (greater than or equal to) | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | No. of patients with score less than cutoff |
|
| |||||
| 0 | 100 | 0 | 64 | — | 0 |
| 1 | 97 | 15 | 67 | 77 | 312 |
| 2 | 93 | 32 | 70 | 72 | 688 |
| 3 | 85 | 50 | 75 | 66 | 1184 |
| 4 | 77 | 61 | 78 | 60 | 1579 |
| 5 | 68 | 73 | 81 | 56 | 2019 |
| 6 | 55 | 83 | 85 | 52 | 2509 |
| 7 | 44 | 91 | 89 | 48 | 2921 |
| 8 | 31 | 95 | 92 | 44 | 3364 |
| 9 | 24 | 97 | 93 | 42 | 3591 |
| 10 | 14 | 99 | 95 | 40 | 3888 |
| 11 | 9 | 99 | 97 | 39 | 4016 |
| 13 | 3 | 100 | 97 | 37 | 4190 |
|
| |||||
| Validation cohort (n = 2837) | |||||
|
| |||||
| Score cutoff (greater than or equal to) | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | No. of patients with score less than cutoff |
|
| |||||
| 0 | 100 | 0 | 64 | — | 0 |
| 1 | 98 | 16 | 68 | 81 | 204 |
| 2 | 93 | 35 | 72 | 75 | 480 |
| 3 | 86 | 53 | 76 | 68 | 799 |
| 4 | 78 | 64 | 80 | 62 | 1066 |
| 5 | 68 | 74 | 82 | 56 | 1337 |
| 6 | 55 | 84 | 86 | 51 | 1664 |
| 7 | 44 | 90 | 88 | 47 | 1941 |
| 8 | 29 | 95 | 92 | 43 | 2255 |
| 9 | 24 | 97 | 94 | 42 | 2370 |
| 10 | 14 | 99 | 96 | 39 | 2569 |
| 11 | 9 | 100 | 98 | 38 | 2672 |
| 12 | 2 | 100 | 100 | 36 | 2797 |
| 13 | 2 | 100 | 100 | 36 | 2798 |
Abbreviations: NPV, negative predictive value; PPV, positive predictive value.
Additional Variables in the Randomized Cohort
Four additional variables were found to be significantly associated with HOMA-IR among randomized IRIS participants: (1) hypertension history; (2) current cigarette smoking; (3) increasing C-reactive protein; and (4) increasing HgbA1c. However, because these associations were found in the randomized cohort, which comprises only subjects with HOMA-IR greater than 3.0, these variables could not be included in the final prediction instrument.
Discussion
Our goal was to develop a simple screening instrument for insulin resistance that was adequately sensitive and specific for use in selecting patients with a recent ischemic stroke or TIA for more advanced testing. An optimal instrument would identify almost all patients with insulin resistance with adequate specificity to avoid unnecessary advanced testing in persons without insulin resistance. Unfortunately, neither our basic nor enhanced instrument demonstrated performance characteristics required for clinical care. At an optimal score of 5, the better of the two instruments (i.e., the enhanced model) had a sensitivity of 68% and a specificity of 73%. Lower scores achieved better sensitivity, as expected, but were associated with specificity values that are probably unacceptable for a clinical test of this purpose and considering the complexity of more definitive testing.
Several other groups have developed clinical instruments for the diagnosis of insulin resistance based on routine clinical data and demographics, although not in patients with established cerebrovascular disease. Forst et al. proposed the IRIS II score (unrelated to the IRIS trial upon which this study is based) to diagnose insulin resistance, defined by the authors as HOMA-IR greater than 2, in patients with diabetes mellitus.24 The score assigns point values to categorical levels of BMI, fasting blood glucose, fasting triglycerides, and fasting HDL. The authors found a significant correlation between their scoring system and HOMA-IR (r = .42, P < .0001) with a positive predictive value of .95, a specificity value of .95, and a sensitivity value of .34 at a score above 70. However, unlike our prediction instrument, which was developed in nondiabetic patients, the IRIS II score was developed to predict insulin resistance in patients with known diabetes mellitus who have a very high pretest probability of having the condition.
We are aware of 1 instrument, which, like ours, uses exclusively clinical information, not including fasting insulin level, to identify nondiabetic patients with insulin resistance. Stern et al. used regression tree analysis to create 3 rule-based models to identify diabetic and nondiabetic patients with insulin resistance, defined as less than 28 µmol/min*kg lean body mass by euglycemic insulin clamp.25 Stern’s first model included HOMA score as a component variable and therefore was not comparable to ours. Stern’s second model used nonlaboratory clinical variables and, therefore, was comparable in construction to our basic model. In this model, a patient is identified as insulin resistant if BMI is greater than 28.7 kg/m2 or greater than 27 kg/m2 and there is family history of diabetes. This rule had a sensitivity of 78.7% and a specificity of 76%. Stern’s third model included clinical variables but also incorporated lipid values other than fasting insulin, similar to our enhanced model. In this model, a patient was identified as insulin resistant based on the following: (1) BMI is greater than 28.7 kg/m2; (2) BMI is greater than 27.0 kg/m2 and family history of diabetes is positive; or (3) family history of diabetes is negative, but triglycerides are greater than 2.44 mmol/L. This rule had a sensitivity of 81.3% and a specificity of 76.3%.
Although our instrument selected similar features as Stern’s models, we have several theories about why our instrument did not perform as well. First, Stern’s models defined insulin resistance by the euglycemic insulin clamp, the gold standard for the measurement of insulin resistance. Our instrument, on the other hand, used HOMA-IR as the definition of insulin resistance, which may not be an accurate enough measure of insulin resistance. Although the results of the HOMA-IR correlate with the results obtained from the euglycemic clamp and other advanced tests, correlations are not perfect and range from .53 to .66.26,27 In addition, there is no standardized reference range for HOMA-IR. Our criterion of HOMA-IR greater than 3.0 was based on limited epidemiologic data.6,24 Furthermore, Stern’s second and third models include family history of diabetes, probably a highly predictive feature for insulin resistance, which likely contributes to the models’ performance. Information about family history was not available in our database. Lastly, comparison between the performance of our instruments and Stern’s is difficult given the different populations in which the instruments were developed. Stern’s study population comprised 2321 subjects from 3 major sources: (1) the European Group for the Study of Insulin Resistance consisting of Caucasian subjects from multiple European countries with normal fasting glucose and normal glucose tolerance test; (2) the Pima Indian Study consisting of Pima Indian adults aged 18–35 years; and (3) studies performed in San Antonio, TX, in a mostly Mexican-American population.28–30 The demographic information of Stern’s population is not described in detail, but includes heterogeneous groups of subjects that included many adults with insulin resistance, particularly from the native American and Latino populations, known to be at significantly increased risk for this condition, and, ultimately, for type 2 diabetes.
We recognize several other limitations to our study. First, our predictive instrument was validated using a split sample technique.17 However, by selecting development and validation cohorts from the same population, we were better able to match the 2 groups and maximized the extent to which the demographic characteristics were similar for the development and validation groups. Other limitations include the restriction of the study population to Western geographic regions.
Future efforts toward producing an improved scoring system for risk of insulin resistance should include testing the variables found in our randomized cohort to be associated with the magnitude of HOMA-IR: history of hypertension, cigarette smoking, C-reactive protein, and HgbA1c. Of note, neither C-reactive protein nor HgbA1c was evaluated by Stern. Stern’s research would suggest that family history should also be considered.
In conclusion, we have developed 2 instruments that incorporate routine clinical variables to predict risk for insulin resistance in men and women aged 40 years and older with a recent ischemic stroke or TIA. However, we were not able to demonstrate improved performance over a simpler instrument that used a more accurate gold standard in a diverse population. We recommend that future studies developing instruments to predict insulin resistance use the euglycemic–hyperinsulinemic clamp as the measure of insulin resistance. We also recommend that future studies incorporate family history of diabetes and hypertension, cigarette smoking, C-reactive protein, and HgbA1c into their analyses. With high enough sensitivity and specificity, these prediction instruments could help clinicians identify patients at high risk for insulin resistance in order to prevent important downstream complications including diabetes mellitus, coronary disease, and ischemic stroke.
Acknowledgments
Grant support/funding: The U.S. National Institute of Neurological Disorders and Stroke (NINDS) 1 U01 NS044876.
Footnotes
Author contributions: Jin Xu: literature search, study design, figures, writing, data interpretation. Catherine Viscoli: literature search, study design, data collection, data analysis, data interpretation, figures, editing. Gary Ford: editing. Mark Gorman: editing. Walter Kernan: study conception, study design, data collection, data interpretation, writing, editing.
References
- 1.Einhorn D, Reaven GM, Cobin RH, et al. American College of Endocrinology position statement on the insulin resistance syndrome. Endocr Pract. 2003;9:237–252. [PubMed] [Google Scholar]
- 2.Hyvarinen M, Tuomilehto J, Mahonen M, et al. Hyperglycemia and incidence of ischemic and hemorrhagic stroke-comparison between fasting and 2-hour glucose criteria. Stroke. 2009;40:1633–1637. doi: 10.1161/STROKEAHA.108.539650. [DOI] [PubMed] [Google Scholar]
- 3.Kernan WN, Inzucchi SE, Viscoli CM, et al. Insulin resistance and risk for stroke. Neurology. 2002;59:809–815. doi: 10.1212/wnl.59.6.809. [DOI] [PubMed] [Google Scholar]
- 4.Lee M, Saver JL, Hong KS, et al. Effect of pre-diabetes on future risk of stroke: meta-analysis. BMJ. 2012;344:e3564. doi: 10.1136/bmj.e3564. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.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 Complications Study. Diabetes Care. 2002;25:1135–1141. doi: 10.2337/diacare.25.7.1135. [DOI] [PubMed] [Google Scholar]
- 6.Ascaso JF, Pardo S, Real JT, et al. Diagnosing insulin resistance by simple quantitative methods in subjects with normal glucose metabolism. Diabetes Care. 2003;26:3320–3325. doi: 10.2337/diacare.26.12.3320. [DOI] [PubMed] [Google Scholar]
- 7.Vaccaro O, Masulli M, Cuomo V, et al. Comparative evaluation of simple indices of insulin resistance. Metabolism. 2004;53:1522–1526. doi: 10.1016/j.metabol.2004.05.017. [DOI] [PubMed] [Google Scholar]
- 8.McAuley KA, Williams SM, Mann JI, et al. Diagnosing insulin resistance in the general population. Diabetes Care. 2001;24:460–464. doi: 10.2337/diacare.24.3.460. [DOI] [PubMed] [Google Scholar]
- 9.Otten J, Ahren B, Olsson T. Surrogate measures of insulin sensitivity vs the hyperinsulinaemic-euglycaemic clamp: a meta-analysis. Diabetologia. 2014;57:1781–1788. doi: 10.1007/s00125-014-3285-x. [DOI] [PubMed] [Google Scholar]
- 10.Esposito K, Maiorino MI, Ciotola M, et al. Effects of a Mediterranean-style diet on the need for antihyperglycemic drug therapy in patients with newly diagnosed type 2 diabetes: a randomized trial. Ann Intern Med. 2009;151:306–314. doi: 10.7326/0003-4819-151-5-200909010-00004. [DOI] [PubMed] [Google Scholar]
- 11.Knowler WC, Barrett-Connor E, Fowler SE, et al. Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin. N Engl J Med. 2002;346:393–403. doi: 10.1056/NEJMoa012512. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Petersen KF, Dufour S, Befroy D, et al. Reversal of nonalcoholic hepatic steatosis, hepatic insulin resistance, and hyperglycemia by moderate weight reduction in patients with type 2 diabetes. Diabetes. 2005;54:603–608. doi: 10.2337/diabetes.54.3.603. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Tuomilehto J, Lindstrom J, Eriksson JG, et al. Prevention of type 2 diabetes mellitus by changes in lifestyle among subjects with impaired glucose tolerance. N Engl J Med. 2001;344:1343–1350. doi: 10.1056/NEJM200105033441801. [DOI] [PubMed] [Google Scholar]
- 14.Dormandy JA, Charbonnel B, Eckland DJ, et al. Secondary prevention of macrovascular events in patients with type 2 diabetes in the PROactive Study (PROspective pioglitAzone Clinical Trial In macroVascular Events): a randomised controlled trial. Lancet. 2005;366:1279–1289. doi: 10.1016/S0140-6736(05)67528-9. [DOI] [PubMed] [Google Scholar]
- 15.Wilcox R, Kupfer S, Erdmann E. Effects of pioglitazone on major adverse cardiovascular events in high-risk patients with type 2 diabetes: results from PROspective pioglitAzone Clinical Trial In macro Vascular Events (PROactive 10) Am Heart J. 2008;155:712–717. doi: 10.1016/j.ahj.2007.11.029. [DOI] [PubMed] [Google Scholar]
- 16.Kernan WN, Viscoli CM, Furie KL, et al. Pioglitazone after Ischemic Stroke or Transient Ischemic Attack. N Engl J Med. 2016;374:1321–1331. doi: 10.1056/NEJMoa1506930. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Katz MH. Multivariable analysis: a practical guide for clinicians and public health researchers. 3. New York, NY: Cambridge University Press; 2011. [Google Scholar]
- 18.Farin HM, Abbasi F, Reaven GM. Body mass index and waist circumference both contribute to differences in insulin-mediated glucose disposal in nondiabetic adults. Am J Clin Nutr. 2006;83:47–51. doi: 10.1093/ajcn/83.1.47. [DOI] [PubMed] [Google Scholar]
- 19.Riserus U, Arnlov J, Berglund L. Long-term predictors of insulin resistance: role of lifestyle and metabolic factors in middle-aged men. Diabetes Care. 2007;30:2928–2933. doi: 10.2337/dc07-0360. [DOI] [PubMed] [Google Scholar]
- 20.Abbasi F, Brown BW, Jr, Lamendola C, et al. Relationship between obesity, insulin resistance, and coronary heart disease risk. J Am Coll Cardiol. 2002;40:937–943. doi: 10.1016/s0735-1097(02)02051-x. [DOI] [PubMed] [Google Scholar]
- 21.McLaughlin T, Abbasi F, Cheal K, et al. Use of metabolic markers to identify overweight individuals who are insulin resistant. Ann Intern Med. 2003;139:802–809. doi: 10.7326/0003-4819-139-10-200311180-00007. [DOI] [PubMed] [Google Scholar]
- 22.Jensen MD, Ryan DH, Apovian CM, et al. 2013 AHA/ACC/TOS guideline for the management of overweight and obesity in adults: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines and The Obesity Society. J Am Coll Cardiol. 2014;63:2985–3023. doi: 10.1016/j.jacc.2013.11.004. [DOI] [PubMed] [Google Scholar]
- 23.Grundy SM, Cleeman JI, Daniels SR, et al. Diagnosis and management of the metabolic syndrome: an American Heart Association/National Heart, Lung, and Blood Institute scientific statement: executive summary. Crit Pathw Cardiol. 2005;4:198–203. doi: 10.1097/00132577-200512000-00018. [DOI] [PubMed] [Google Scholar]
- 24.Forst T, Standl E, Hohberg C, et al. IRIS II study: the IRIS II score—assessment of a new clinical algorithm for the classification of insulin resistance in patients with type 2 diabetes. Diabet Med. 2004;21:1149–1153. doi: 10.1111/j.1464-5491.2004.01298.x. [DOI] [PubMed] [Google Scholar]
- 25.Stern SE, Williams K, Ferrannini E, et al. Identification of individuals with insulin resistance using routine clinical measurements. Diabetes. 2005;54:333–339. doi: 10.2337/diabetes.54.2.333. [DOI] [PubMed] [Google Scholar]
- 26.Bonora E, Targher G, Alberiche M, et al. Homeostasis model assessment closely mirrors the glucose clamp technique in the assessment of insulin sensitivity: studies in subjects with various degrees of glucose tolerance and insulin sensitivity. Diabetes Care. 2000;23:57–63. doi: 10.2337/diacare.23.1.57. [DOI] [PubMed] [Google Scholar]
- 27.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. doi: 10.1007/BF00280883. [DOI] [PubMed] [Google Scholar]
- 28.Gulli G, Ferrannini E, Stern M, et al. The metabolic profile of NIDDM is fully established in glucose-tolerant offspring of two Mexican-American NIDDM parents. Diabetes. 1992;41:1575–1586. doi: 10.2337/diab.41.12.1575. [DOI] [PubMed] [Google Scholar]
- 29.Ferrannini E, Vichi S, Beck-Nielsen H, et al. Insulin action and age. European Group for the Study of Insulin Resistance (EGIR) Diabetes. 1996;45:947–953. doi: 10.2337/diab.45.7.947. [DOI] [PubMed] [Google Scholar]
- 30.Lillioja S, Mott DM, Zawadzki JK, et al. In vivo insulin action is familial characteristic in nondiabetic Pima Indians. Diabetes. 1987;36:1329–1335. doi: 10.2337/diab.36.11.1329. [DOI] [PubMed] [Google Scholar]