Calculate Insulin Resistance

HOMA-IR – the Homeostasis Model Assessment of Insulin Resistance – estimates how hard your body has to work to keep blood sugar in check. It was derived in 1985 by Matthews and colleagues in Oxford ^[1], and it has become the most widely used way to gauge insulin resistance from a routine blood test, because it needs only two fasting values: glucose and insulin.

The idea is simple. In a healthy person, a modest amount of insulin keeps fasting glucose normal. When cells stop responding well to insulin (insulin resistance), the pancreas compensates by secreting more – so fasting insulin creeps up, often years before fasting glucose does. HOMA-IR captures that imbalance as a single number, calibrated so a metabolically normal reference person scores 1.0 ^[1].

It is a screening surrogate, not a gold-standard measurement. The reference method – the hyperinsulinaemic-euglycaemic clamp – is impractical outside research, and HOMA-IR agrees with it only moderately at the individual level ^[2]. Think of HOMA-IR as a useful orientation, not a verdict.

There are two algebraically identical forms of the formula – they differ only by the glucose unit:

In plain words: the more insulin it takes to keep your blood sugar normal, the more insulin-resistant you are.

Conventional (glucose in mg/dL): HOMA-IR = fasting insulin (µU/mL) × fasting glucose (mg/dL) ÷ 405.
SI (glucose in mmol/L): HOMA-IR = fasting insulin (mU/L) × fasting glucose (mmol/L) ÷ 22.5.

The two denominators are linked by the glucose conversion factor: 405 = 22.5 × 18. Insulin in µU/mL, µIU/mL and mU/L are all numerically identical (1:1), so no conversion is needed there. If your lab reports insulin in pmol/L, this calculator divides by 6.0 to get µU/mL. The insulin-standardisation literature (Knopp and colleagues 2019 ^[9]; the ADA SI table) argues 6.0 is the more defensible factor than the widely-copied 6.945, which would understate your insulin (and so your HOMA-IR) by about 14%. The choice is assumption-dependent and the field is not fully standardised, and it only matters when a lab reports in pmol/L.

The same inputs also yield two companion numbers. HOMA-%B, steady-state beta-cell function (normal ≈ 100%), is part of the original 1985 model ^[1]. The sensitivity figure we label HOMA-%S is simply 100 ÷ HOMA-IR, so a normal subject reads 100%; strictly, the named "%S" output belongs to the later HOMA2 computer model, but for HOMA1 this reciprocal carries the same meaning.

Here is the most important thing most calculators won't tell you: there is no single universal HOMA-IR cutoff. The threshold for "insulin resistance" depends on the population studied, the insulin assay used, and which percentile you call abnormal – published cutoffs range from roughly 1.4 to 3.9 ^[7][8].

The traffic-light bands below are a screening orientation for a general European/German adult audience, anchored to German population data – the Gutenberg Health Study reported a healthy median of 1.09 and a 97.5th-percentile upper reference limit of 2.35 ^[5] – with the upper band consistent with the classic European Bruneck cutoff of 2.77 ^[6]. They apply to HOMA1-IR (this calculator) only. Some European cohorts that anchor the cutoff to metabolic-syndrome outcomes land lower, around 1.85 to 2.0 (these usually take the top quartile of fasting insulin or HOMA-IR rather than a fixed value), so treat 2.35 as the conservative end of the grey zone.

Because insulin immunoassays are not standardised, the same blood sample can give HOMA-IR values that differ almost two-fold between labs ^[10][11]. Always read your number against your own laboratory's reference range rather than treating any fixed value as absolute.

HOMA-IR is genuinely useful, but it is easy to over-read. Keep these caveats in mind:

The formula above is HOMA1: the original 1985 linear approximation. It is fast, transparent and matches the most familiar published cutoffs – which is exactly what this calculator gives you.

The original authors later showed that the "correct" evaluation uses a non-linear computer model, HOMA2 ^[3], refined further as iHOMA2 ^[4]. HOMA2 is more accurate for modern insulin assays and across the full clinical range, and it also accounts for things the linear model can't. The trade-off: it can't be reduced to a simple equation – it is solved iteratively by the official Oxford Diabetes Trials Unit calculator.

One rule matters above all: never cross-apply cutoffs. HOMA2-IR runs systematically lower than HOMA1-IR – a HOMA1 threshold near 2.7 corresponds to roughly HOMA2-IR 1.8 ^[8]. If your glucose or insulin is at the extremes, this calculator points you to the Oxford HOMA2 tool, because that is where the linear approximation is least trustworthy ^[2].

HOMA-IR is the familiar baseline, but it is not the only fasting model, and the most advanced one is not actually Oxford's HOMA2. HOMA2 is a closed, licensed program whose constants were never fully published, so nobody can reproduce it faithfully. That is why this tool links out to the official Oxford calculator for the rare extreme cases, rather than shipping a look-alike that would quietly be wrong.

Instead, the advanced panel above computes SPINA-Carb, an open, mechanistic model (Dietrich et al. 2022, 2024) that reconstructs the structural constants of the glucose-insulin control loop from the same two fasting values ^[12][13]. It returns insulin sensitivity (SPINA-GR), beta-cell capacity (SPINA-GBeta), and a fasting disposition index (SPINA-DI) that mimics what normally needs an oral glucose tolerance test. In multi-cohort studies SPINA-DI out-discriminated HOMA-IR, QUICKI and even the dynamic (OGTT) disposition index for diagnosing diabetes ^[13].

Two insulin-free surrogates round it out for people who only have a basic panel: the triglyceride-glucose index (TyG), the most widely studied insulin-free marker ^[14], and METS-IR, which adds HDL and BMI and is also validated against the euglycemic clamp ^[15]. All of these are open published formulas with no license burden, computed natively here, which is how we can be a richer reference than a single linked-out tool.