Ascertaining Kraft dynamic insulin response patterns following a 3-h 100 g oral glucose tolerance test seems to be the most reliable method for diagnosing hyperinsulinaemia. However, this test may be too resource-intensive for standard clinical use.
This study aims to see if Kraft patterns can be accurately predicted using fewer blood samples with sensitivity and specificity analyses.
St Joseph Hospital, Chicago, Illinois, United States and Human Potential Centre, Auckland University of technology, Auckland, New Zealand.
We analysed the results of 4185 men and women with a normal glucose tolerance, who had a 100 g oral glucose tolerance test with Kraft pattern analysis. Participants were dichotomised into normal–low insulin tolerance (Kraft I or V patterns) or hyperinsulinaemia (Kraft IIA–IV patterns). Sensitivity and specificity analysis was applied to available variables (including age, body mass index, fasting insulin or glucose) both individually and in combination.
Out of a maximal combined sensitivity and specificity score of 2.0, 2-h insulin level > 45 µU/mL attained the highest score (1.80). Two-hour insulin also attained the highest sensitivity (> 30 µU/mL, 0.98) and the highest specificity (> 50 µU/mL, 0.99) scores. Combining the 2-h insulin with other variables reduced the sensitivity and/or specificity. Dynamic measures had a better combined sensitivity and specificity compared to fasting or anthropological measures.
People with a 2-h plasma insulin level < 30 µU/mL are unlikely to be hyperinsulinaemic. Given that first-line treatment is lifestyle modification, we recommend that a 2-h plasma insulin level > 30 µU/mL following a 100 g oral glucose tolerance test be used to identify the hyperinsulinaemic individual.
Hyperinsulinaemia contributes to metabolic disease via inflammatory pathways, by increasing cellular growth and proliferation via IGF‑1, and being proatherosclerotic via decreased nitric oxide production, impaired fibrinolysis and increasing triglyceride production.
Diagnosing hyperinsulinaemia is challenging as most studies are based on the poorly defined concept of insulin resistance. Although the World Health Organization (WHO) has defined insulin resistance as ‘under hyperinsulinaemic-euglycaemic conditions, glucose uptake below lowest quartile for background population under investigation’,
The hyperglycaemic–euglycaemic clamp test is considered to be the ‘gold-standard’ test for insulin resistance.
There are a number of studies that use fasting insulin to indicate insulin resistance and/or hyperinsulinaemia. However, there are concerns that using a single fasting insulin level may not be sufficiently accurate because of the oscillatory nature of pancreatic insulin release.
The question remains as to whether insulin resistance tests should be used to determine hyperinsulinaemia. Although the two conditions are fundamentally intertwined, they are intrinsically different conditions.
Under normal physiological conditions, a person can only become hyperinsulinaemic when two conditions are met. The first is when a person has a degree of insulin resistance that may occur acutely or chronically. Acute insulin resistance can occur under a number of conditions, for example when glucose needs to be preferentially shunted to the brain or other body systems that can only rely on glucose for fuel (e.g. red blood cells). This may include fasting, hypoglycaemia or high cortisol levels. Acute insulin resistance may also occur with acute hyperglycaemia, when the GLUT4 transporters are downregulated.
Therefore, to effectively understand hyperinsulinaemia, a new method for diagnosis and monitoring needs to be developed. The most promising research has been based around insulin response patterns, formed during an oral glucose tolerance test. In 1975, Kraft demonstrated five distinct insulin response patterns arising during a 3-h 100 g oral glucose tolerance test.
Kraft pattern algorithm.
Kraft pattern | Description |
---|---|
Pattern I (normal insulin) | Fasting insulin ≤ 30 µU/mL 30-min or 1-h peak 2-h + 3-h sum < 60 µU/mL |
Pattern IIA (borderline) | Fasting insulin ≤ 50 µU/mL 30-min or 1-h peak 2-h + 3-h sum ≥ 60, < 100 µU/mL Fasting insulin 31–50 µU/mL 30-min or 1-h peak 2-h + 3-h sum < 60 µU/mL |
Pattern IIB (hyperinsulinaemia) | Fasting insulin ≤ 50 µU/mL 30-min or 1-h peak 2-h + 3-h sum ≥ 100 µU/mL |
Pattern III (hyperinsulinaemia) | Fasting insulin ≤ 50 µU/mL Delayed peak (2 or 3 h) |
Pattern IV (hyperinsulinaemia) | Fasting insulin > 50 µU/mL |
Pattern V (hypoinsulinaemia) | All values ≤ 30 µU/mL |
Hayashi and colleagues used different insulin response patterns. They measured plasma insulin at baseline and then at 30, 60 and 120 min during a 2-h, 75 g oral glucose tolerance test. By determining the timing of the insulin peak/s, as assessed by the responses, they showed an increased risk of developing type 2 diabetes in people who had an insulin response that peaked at 2 h compared to those who had an insulin peak at 30 or 60 min.
Our previous research suggests that Kraft patterns should be preferred to the Hayashi patterns as Kraft patterns demonstrated less variation.
It is also plausible that other clinical features influence, or are influenced by, hyperinsulinaemia. For example, Hayashi and colleagues demonstrated that different glucose response patterns were produced depending on the patient’s insulin response curve.
Sensitivity and specificity analyses are statistical binary classification measures used to assess the proportions of correctly diagnosed people suspected of having a clinical diagnosis. Sensitivity measures the proportion of correctly identified people with the clinical condition (sick), while specificity measures the proportion of correctly identified people without the clinical condition (healthy) as according to the methods of Altman and Bland.
A total of 15 000 patients and healthy volunteers were referred for an oral glucose tolerance test at St Joseph Hospital, Chicago, IL, USA between 1972 and 1992. Data collected included plasma glucose, plasma insulin, age, gender, height and weight.
Subjects fasted overnight for 10–16 h. A fasting venous blood sample was taken; 100 g of glucose (Glucola, Miles/Ames, Elkhardt, IN, USA) was ingested and venous samples at 30 min, 60 min and each subsequent hour for between 3 and 5 h. The blood specimens were measured for glucose and insulin. Originally, the ferricyanide method (Autoanalyzer, Technicon Corporation, Tarrytown, NJ, USA) was used to analyse glucose, but this was later changed to plasma glucose oxidase method (Autoanalyzer, Technicon Corporation; Vitros, Johnson and Johnson Clinical Diagnostics, Inc., Rochester, NY, USA). According to the methods of Passey and colleagues, glucose samples analysed with the ferricyanide method were adjusted downward by 10 mg/dL to account for the systematic error.
Plasma insulin was determined from the samples stored at –70 °C by a commercial double-antibody solid phase radioimmunoassay, (Pharmacia insulin RIA 100, Pharmacia Diagnostics AB, Uppsala, Sweden). The Phadebas Insulin Test had duplicate procedure precision of 1 standard deviation = ±5 µU in measurements up to 150 µU.
Exclusion criteria included a body mass index (BMI) ≤ 17.9kg/m2 because of the potential confounder of concurrent illness. Women aged between 20 and 45 years were excluded because of the potential confounder of pregnancy.
From this data set, we included 2161 men aged older than 20 years and 2024 women aged older than 45 years, who had a normal glucose tolerance as defined by WHO criteria (1999) and also had age, height and weight recorded – a total of 4185 participants (
Participant characteristics.
Characteristics | Total |
---|---|
4185 | |
Female | 2024 (48) |
Age (years) | |
Male | 44.9 (15.2) |
Female | 59.1 (9.4) |
BMI (kg/m2) | 25.9 (4.7) |
Plasma insulin (µU/mL) | |
0 min | 13 (13) |
30 min | 87 (56) |
60 min | 105 (73) |
120 min | 77 (62) |
180 min | 40 (41) |
Plasma glucose (mg/dL) | |
0 min | 86 (10) |
30 min | 152 (32) |
60 min | 146 (43) |
120 min | 101 (22) |
180 min | 82 (25) |
Frequency data are reported as
This study uses current clinical practices and sensitivity and specificity calculations to logically derive whether Kraft’s patterns can be simplified. Area under the curve calculations were performed using the trapezoidal rule. Statistical analysis was performed using Microsoft Excel 2010 or IBM SPSS Statistics 22. Sensitivity and specificity calculations were performed as according to the methods of Altman and Bland.
The variables to be tested individually and in combination within the sensitivity and specificity calculations included BMI, age, HOMA2 %B, HOMA2 %S, HOMA2 IR, oral glucose insulin sensitivity (OGIS) and plasma glucose or insulin levels from each time point (0 min, 30 min, 1 h, 2 h and 3 h). HOMA2 variables and OGIS were calculated using their respective calculators.
Sensitivity and specificity calculations can only be performed with a dichotomised test outcome. Therefore, as depicted in
Sensitivity and specificity calculation table.
This study was granted ethical approval by Health and Disability Ethics Committee (New Zealand) on 30 October 2013. Approval reference: 13/CEN/166. AUTEC reference: 13/337.
As depicted in
Sensitivity and specificity calculations (further data on file).
Test variable | Sensitivity | Specificity | Sum SS |
---|---|---|---|
2-h insulin > 30 µU/mL | 0.98 | 0.62 | 1.60 |
OGIS < 600 mL/min/m2 | 0.95 | 0.34 | 1.30 |
2-h insulin – fasting insulin > 30 µU/mL | 0.90 | 0.83 | 1.73 |
2-h glucose > 80 mg/dL | 0.90 | 0.38 | 1.28 |
HOMA2 %B > 20 | 0.87 | 0.40 | 1.27 |
1-h insulin > 50 µU/mL | 0.86 | 0.49 | 1.36 |
2-h insulin > 45 µU/mL | 0.85 | 0.95 | 1.80 |
Age > 35 years | 0.85 | 0.24 | 1.09 |
2-h insulin – fasting insulin > 35 µU/mL | 0.84 | 0.92 | 1.76 |
2-h glucose – fasting glucose > 0 mg/dL | 0.83 | 0.47 | 1.31 |
Fasting insulin > 5 µU/mL | 0.83 | 0.46 | 1.29 |
1-h insulin > 60 µU/mL | 0.80 | 0.61 | 1.40 |
2-h insulin > 50 µU/mL | 0.79 | 0.99 | 1.78 |
3-h insulin > 20 µU/mL | 0.79 | 0.85 | 1.64 |
2-h insulin > 45 µU/mL and 2-h glucose > 80 mg/dL | 0.78 | 0.96 | 1.74 |
OGIS < 500 mL/min/m2 | 0.70 | 0.84 | 1.54 |
2-h insulin > 45 and 2-h glucose > 90 | 0.69 | 0.97 | 1.67 |
2-h glucose – fasting glucose > 10 mg/dL | 0.68 | 0.67 | 1.35 |
2-h insulin – fasting insulin > 50 µU/mL | 0.65 | 1.00 | 1.64 |
2-h glucose > 100 mg/dL | 0.63 | 0.73 | 1.35 |
Age > 50 years | 0.61 | 0.52 | 1.13 |
3-h insulin > 30 µU/mL | 0.60 | 0.99 | 1.58 |
Fasting glucose > 85 mg/dL | 0.56 | 0.46 | 1.02 |
2-h insulin > 45 µU/mL and 2-h glucose > 100 mg/dL | 0.55 | 0.98 | 1.54 |
BMI > 25 kg/m2 | 0.55 | 0.61 | 1.16 |
BMI > 25 kg/m2, 2-h insulin > 30 µU/mL | 0.54 | 0.83 | 1.37 |
Fasting insulin > 10 µU/mL | 0.54 | 0.79 | 1.32 |
HOMA2 IR > 0.2 | 0.52 | 0.81 | 1.32 |
2-h glucose – fasting glucose > 20 mg/dL | 0.50 | 0.81 | 1.31 |
Age > 35 years and BMI > 25 kg/m2 | 0.48 | 0.70 | 1.17 |
BMI > 30 kg/m2 | 0.16 | 0.91 | 1.07 |
2-h insulin > 20 µU/mL | 0.12 | 0.99 | 1.11 |
Fasting glucose > 80 mg/dL and fasting insulin > 20 µU/mL | 0.09 | 0.99 | 1.08 |
SS, total sum of sensitivity and specificity.
The highest overall score was 2-h insulin > 45 µU/mL (1.80) and the highest specificity was 2-h insulin > 50 µU/mL (0.99). The 2-h insulin alone achieved high scores for sensitivity, but this score dropped if applied in combination with another variable such as glucose. For example, 2-h glucose > 80 mg/dL achieved scores of 0.9, 0.38 and 1.28 for sensitivity, specificity and the total sum, respectively, and 2-h insulin > 45 µU/mL achieved scores of 0.85, 0.95 and 1.8 for sensitivity, specificity and the total sum respectively. However, the combination of 2-h glucose > 80 mg/dL and 2-h insulin > 45 µU/mL only attained a score of 0.78 for sensitivity, 0.96 for specificity and a combined result of 1.74. Although this is still a very good score, the sensitivity is lower than using 2-h insulin in isolation.
Oral glucose insulin sensitivity < 600 mL/min/m2 attained the highest score (1.30) of the measures for insulin resistance with a very high sensitivity score (0.95). HOMA2 variables did not score highly overall: HOMA2 %B > 20 scored 1.27, while HOMA2 IR > 0.2 scored 1.32.
A receiver operating characteristic (ROC) curve confirmed these sensitivity and specificity calculations (
Receiver operating characteristic (ROC) curve.
Area under the curve calculations for receiver operating characteristic analysis.
Test variable | Area under the curve |
---|---|
Insulin 120 min | 0.965 |
Sum insulin 0 min + insulin 120 min | 0.963 |
Difference insulin 120 min - insulin 0 min | 0.948 |
Insulin 180 min | 0.912 |
Insulin area under the curve 3 h | 0.897 |
Insulin 60 min | 0.771 |
Glucose 120 min | 0.740 |
Insulin 0 min | 0.737 |
HOMA2 %IR | 0.736 |
Glucose area under the curve 3 h | 0.720 |
HOMA2 %B | 0.713 |
Insulin 30 min | 0.706 |
Glucose 60 min | 0.673 |
Glucose 180 min | 0.656 |
Age | 0.596 |
Body mass index (BMI) | 0.590 |
Glucose 30 min | 0.550 |
Glucose 0 min | 0.518 |
HOMA2 %S | 0.264 |
OGIS | 0.150 |
HOMA, homeostasis model assessment; OGIS, oral glucose insulin sensitivity.
This study aimed to determine if there was a test that could simplify the diagnosis of hyperinsulinaemia in people with normal glucose tolerance, as defined by Kraft patterns IIa, IIb, III and IV. We were looking for a test with a high degree of sensitivity that required the least amount of resources, including time. We found that if the fasting insulin was < 30 µU/mL, then a 2-h plasma insulin level > 30 µU/mL following a 100 g, 2-h oral glucose tolerance test (OGTT) provided the highest degree of sensitivity in predicting a hyper insulinaemic pattern.
Although other variable combinations attained a higher combined sensitivity and specificity score, there are a number of reasons why we believe that the combination of a fasting insulin level combined with an OGTT and a 2-h plasma insulin cut-off of 30 µU/mL is the most useful test to recommend for both clinical and research practice.
If a person returned a fasting insulin level of > 30 µU/mL, then this alone should be considered diagnostic for hyperinsulinaemia, but lower levels cannot exclude the condition.
Using the lowest 2-h insulin level that maintained a reasonable sensitivity and specificity seemed the most appropriate clinical decision. Although a 2-h level > 45 µU/mL attained the highest summed score of 1.8, it had a lower sensitivity of 0.85 compared with 0.98 for a 30 µU/mL cut-off. A sensitivity score of 1.0 means that everybody who is tested for the disease, who truly has the disease, will be given a correct diagnosis. When sensitivity scores are decreased to 0.85, this means 15% of people who truly have the disease will be told, falsely, that they have a negative result. A lower specificity score increases the possibility of false negative results, or when people will be told that they have the disease, when they are, in fact, disease free.
The decision to err on the side of sensitivity or specificity also depends on the available management strategies. If the potential treatment is associated with significant risks relative to benefits, then the decision may be based on specificity. For hyperinsulinaemia, the potential first-line treatments include diet and physical activity.
One criticism of using insulin is that 2-h levels have a high degree of variability. Our previous study showed that the repeatability coefficient of 2-h plasma insulin following a 100 g OGTT was approximately 45 µU/mL (282 pmol/L).
Fasting insulin levels ≤ 30 µU/mL were not useful in determining hyperinsulinaemia. Levels at the lowest end of the current reference range had a high sensitivity, but low specificity. We partially agree with current recommendations that neither hyperinsulinaemia nor insulin resistance should be diagnosed on the basis of a fasting insulin test,
Diagnostic algorithm for hyperinsulinaemia.
We were disappointed that we could not recommend a single fasting test for hyperinsulinaemia: variables such as BMI, fasting glucose and fasting insulin, either alone or in combination did not attain sufficient sensitivity or specificity. Haemoglobin A1c (HbA1c) was not collected by Kraft, but is likely only useful for determining glucose status as plasma insulin elevation may be detected prior to changes in HbA1c.
A significant limitation to our study is the lack of long-term health outcomes because of the cross-sectional nature of the Kraft database. We cannot, at this stage, evaluate the effectiveness of this test in actually predicting the risk of future disease. Previous work has shown that elevated 2-h insulin levels are associated with increased risk of developing type 2 diabetes
Hyperinsulinaemia is conclusively linked with many metabolic diseases,
This article is based on a chapter of Dr Crofts’ PhD thesis ‘Understanding and diagnosing hyperinsulinaemia’ (Auckland University of Technology), which was supported by the National Heart Foundation (NZ) Ref.: 1522.
The authors declare that they have no financial or personal relationships that may have inappropriately influenced them in writing this article.
C.A.P.C. made substantial contributions to the study concept and design, data collection, analysis and interpretation; drafted the article; and performed data and statistical analysis. G.S. made contributions to the article concept and critical revisions. M.C.W. made contributions to the article concept, design, data analysis and interpretation, and critical revisions. C.Z. made contributions to the article concept and critical revisions. J.R.K. was responsible for data collection and made contributions to the article drafting and data analysis.
This research was supported by the National Heart Foundation (NZ) Ref.: 1522, but no other sponsorship.
Data sharing is not applicable to this article as no new data were created or analysed in this study.
The views and opinions expressed in this article are those of the authors and do not necessarily reflect the official policy or position of any affiliated agency of the authors.