HOMA-IR normal range: why there is no single number

Is there a normal range for HOMA-IR?

HOMA-IR has no single normal range that applies to everyone. The homeostasis model assessment published by Matthews and colleagues in Diabetologia in 1985 fixes one reference point: normal-weight normal subjects under 35 years are assigned an insulin resistance of 1 and a beta-cell function of 100%. HOMA-IR is calculated as fasting insulin (µU/mL) × fasting glucose (mmol/L) ÷ 22.5, so a value of 2 means the model estimates twice that reference degree of insulin resistance. The paper does not give an upper limit of normal or a cut-off that defines insulin resistance. Those have come from later studies, each in its own population. The only study below that set a true reference interval, in 2,153 healthy Japanese adults, found 0.4 to 2.4. Other studies used the 75th or 90th percentile of a healthy group, the top fifth of a population, or the value that best predicted metabolic syndrome, and got answers from 1.85 to 3.80.

What published HOMA-IR studies found

Published HOMA-IR cut-offs differ because each study answered a slightly different question in a different group of people. Yamada and colleagues took 2,153 healthy Japanese adults aged 20 to 79, with fasting glucose under 100 mg/dL and BMI under 25, and set the reference interval at 0.4 to 2.4, suggesting 2.5 or more as a reasonable marker of insulin resistance in Japanese people. Ascaso's group in Valencia found a 75th percentile of 2.6 in 65 adults with normal glucose metabolism. The Bruneck Study in Italy used the lower limit of the top fifth, 2.77, in 225 non-obese adults without metabolic disorders. The Brazilian BRAMS study found a 90th percentile above 2.7 in 297 healthy adults, and above 2.3 best identified metabolic syndrome. The Spanish EPIRCE study of 2,459 adults found 3.46 by the 90th percentile but 2.05 by metabolic syndrome, and Qu's study of 1,854 Mexican American adults in Texas found 3.80.

Why HOMA-IR cut-offs differ between studies

HOMA-IR cut-offs differ for four main reasons: the method, the population, the insulin assay and the version of the model. The method matters most. In the EPIRCE study the same 2,459 Spanish adults gave a threshold of 3.46 by the 90th percentile and 2.05 when it was set by metabolic syndrome, and the authors found that age changed the accuracy of HOMA-IR in non-diabetic women but not in men. The population matters because insulin resistance is common: Qu and colleagues found that 39.1% of their Mexican American sample had a HOMA-IR above 3.80, and noted that their result differed sharply from the commonly used clinical cut-off of 2.60. The insulin assay matters because results are not standardised: Manley's group found that 11 commercial insulin assays gave values differing by a factor of 2 on the same samples, and HOMA-IR rises in proportion to insulin.

The insulin unit adds more variation. Laboratories that report insulin in pmol/L convert it to µU/mL with a factor of 6 or 6.945; Knopp and colleagues argue that 6.00 is correct and that 6.945 under-reports insulin by about 15%. The model version matters too. The Oxford group's later computer model, HOMA2, gives different values from the 1985 formulas: in BRAMS, the 90th-percentile cut-off was 2.7 for HOMA1-IR and 1.8 for HOMA2-IR. A cut-off for one version cannot be applied to the other.

How the fasting sample changes a HOMA-IR result

HOMA-IR is defined on fasting, steady-state values, so the conditions of the blood sample change the result. The 1985 model reads the basal glucose and insulin at which a person's glucose and insulin feedback loop settles in the fasting state. Matthews and colleagues advised measuring fasting insulin over a 15-minute period in a rested person, because insulin is secreted in pulses and stress affects it, and wrote that sampling three times, 5 minutes apart, improves precision. They added that a single fasting sample taken in an outpatient clinic is unlikely to be a reliable guide to a person's insulin resistance or beta-cell function. They noted that overnight basal glucose is slightly lower than in a fasting sample taken after the person has got up, and that exercise or proinsulin measured as insulin could also affect the reading. The studies that set cut-offs used fasting samples too; Ascaso's group, for example, drew blood after a 12-hour overnight fast.

A sample taken after eating is not a fasting value, and HOMA-IR should not be calculated from it. The HOMA-IR calculator assumes both values come from the same fasting sample.

Limits of HOMA-IR as an index

HOMA-IR is a good group measure and an imprecise individual one. In the 1985 paper it correlated with insulin resistance measured by the euglycaemic clamp (Rs = 0.88) in 12 normal subjects and 11 people with type 2 diabetes, and Bonora and colleagues later found a correlation of −0.820 with clamp glucose disposal in 115 people with a range of glucose tolerance, concluding that HOMA can be used reliably in large or epidemiological studies with only a fasting sample. But repeat estimates in the same person varied with a coefficient of variation of 31% for insulin resistance and 32% for beta-cell function, and the authors wrote that this low precision limits its use. Insulin assays add error of their own: Miller and colleagues found that 7 of 10 commercial methods differed from a mass spectrometry reference by more than 15.5% in 36% to 100% of samples.

The 1985 model was built and tested in people without diabetes and in people with type 2 diabetes treated by diet alone, not in people injecting insulin. Wallace, Levy and Matthews, noting in 2004 that HOMA had been reported in more than 500 publications, gave cohort and epidemiological studies as appropriate uses and measuring beta-cell function in isolation as an inappropriate one. HOMA-%B, which the calculator shows beside HOMA-IR, should be read together with it.

Worked example: one HOMA-IR result, several cut-offs

This worked example shows why a HOMA-IR result has to be read against a study of a similar population; the numbers are illustrative and hand-checked. An adult has a fasting glucose of 90 mg/dL (5.0 mmol/L) and a fasting insulin of 10 µU/mL (60 pmol/L, using the factor of 6). HOMA-IR is 10 × 5.0 ÷ 22.5 = 2.22, the same as 10 × 90 ÷ 405. HOMA-%B is 20 × 10 ÷ (5.0 − 3.5) = 133%. Against the 1985 reference of 1.0, the model estimates a little over twice the insulin resistance of young, normal-weight adults. Against the published cut-offs, the answer depends on the study. The value lies inside the Japanese reference interval of 0.4 to 2.4, below the 2.6 of Ascaso's Spanish study, the 2.77 of Bruneck and the 3.46 and 3.80 thresholds, but above the 2.05 metabolic syndrome threshold from EPIRCE.

If the same insulin had been reported as 60 pmol/L and converted with 6.945 instead of 6, it would become 8.6 µU/mL and HOMA-IR 1.92, below every insulin resistance cut-off in this table. That gap is the size of the unit problem on its own. You can try other values in the HOMA-IR calculator.