HOMA-IR Calculator (Insulin Resistance)

Enter a fasting glucose and a fasting insulin from the same blood draw, in whichever units your report uses. You get HOMA-IR worked step by step, HOMA-B and QUICKI from the same two numbers, and a straight account of why the published cutoffs disagree with each other.

Was this calculator helpful?

Data reviewed: August 2026. Figures here come from published sources and change over time. How we verify

What HOMA-IR actually measures

Most people arrive here holding a lab report and a question that the report did not answer. Somebody ordered a fasting insulin alongside the usual fasting glucose, the numbers came back, and now there is a formula on the internet promising to turn the pair into a verdict. Before the arithmetic, one thing worth saying plainly: two numbers drawn from one arm on one morning cannot tell you who you are or what happens next. They are a snapshot of one moment of one system. This page will do the arithmetic carefully and show you every step of it, and then it will be honest about how much the answer is and is not worth.

HOMA stands for homeostatic model assessment. It came out of Oxford in 1985, when Matthews and colleagues built a mathematical model of the feedback loop between the liver and the pancreas: your liver releases glucose, rising glucose prompts the pancreas to release insulin, and insulin tells the liver to stop. In a person whose tissues respond well to insulin, it takes very little insulin to hold that balance. In a person whose tissues respond poorly, it takes a lot. So if you know the fasting glucose and the fasting insulin that a body has settled at, you can work backwards to an estimate of how much insulin that body needed to get there.

That is the whole idea. HOMA-IR is not a measurement of insulin resistance. It is a model's estimate of it, calculated from two numbers, and the model's authors have spent decades reminding people of the difference.

The formula

HOMA-IR = (fasting glucose in mg/dL × fasting insulin in uU/mL) ÷ 405

HOMA-IR = (fasting glucose in mmol/L × fasting insulin in uU/mL) ÷ 22.5

Insulin is in microunits per millilitre (uU/mL, which your lab may print as mIU/L) in both versions. Only the glucose unit changes, and with it the divisor.

The two constants are the same constant. 22.5 is the product of the fasting values the model treats as its reference point: 5 uU/mL of insulin multiplied by 4.5 mmol/L of glucose. Dividing by that product is what makes a normal weight healthy adult come out at 1. Multiply 22.5 by 18 to move it from mmol/L into mg/dL and you get 405. Use whichever divisor matches the unit your glucose was reported in, and you get the same answer.

Almost. Here is a small thing that no other calculator seems to mention. The 18 buried inside those constants is a rounded number. Glucose has a molecular weight of 180.156 grams per mole, so one mmol/L is really 18.0156 mg/dL, not 18. Run the same physical sample through both published forms and they disagree by exactly 0.087 percent, every time, for every sample. Neither form is wrong; the rounding is simply baked in. We mention it because it is the kind of thing that makes a careful person think they have made a mistake, and because 0.087 percent turns out to be an almost comically small worry compared to the one further down this page.

HOMA-B = (20 × fasting insulin in uU/mL) ÷ (fasting glucose in mmol/L − 3.5)

QUICKI = 1 ÷ (log insulin in uU/mL + log glucose in mg/dL)

Two more indices come free from the same two lab values, so this page calculates them too. HOMA-B is the other half of the 1985 paper: an estimate of beta cell function expressed as a percentage, scaled so the model's reference adult sits at 100. Read alongside HOMA-IR it describes a different thing, not how resistant the tissues look but how hard the pancreas appears to be working to hold the line. In mg/dL the same formula is (360 × insulin) ÷ (glucose − 63), and 360 and 63 are just 20 and 3.5 multiplied by that same 18.

QUICKI is the quantitative insulin sensitivity check index, published by Katz and colleagues in 2000. It is the identical pair of numbers on a logarithmic scale (base 10, and it always wants insulin in uU/mL with glucose in mg/dL whatever units your report used), and it runs the opposite way: higher means more insulin sensitive. Its authors reported a correlation of 0.78 against the glucose clamp, which is the reference method insulin sensitivity is actually measured with, and the log transform makes it better behaved at the extremes than HOMA-IR is.

Worked example

Fasting glucose 95 mg/dL, fasting insulin 12 uU/mL, both from the same morning draw:

HOMA-IR = (95 × 12) ÷ 405 = 1140 ÷ 405 = 2.81

The cross-check in the other unit system: 95 mg/dL is 5.2732 mmol/L, and 5.2732 × 12 ÷ 22.5 = 2.8124. The 0.087 percent gap described above, exactly as promised.

HOMA-B = (360 × 12) ÷ (95 − 63) = 4320 ÷ 32 = 135.0 percent of the model's reference beta cell output.

QUICKI = 1 ÷ (log 12 + log 95) = 1 ÷ (1.0792 + 1.9777) = 1 ÷ 3.0569 = 0.3271.

Now the part that matters. That HOMA-IR of 2.81 sits above two of the five published thresholds in the table below and beneath the other three. Applied in the United States it would usually be called elevated; measured against the 90th percentile of a Spanish population it would not be; against a study of Americans of Mexican descent it would be comfortably inside the ordinary range. The number is fixed. The reading of it is not.

Getting the two lab values right

Three things go wrong before the arithmetic ever starts, and all three are worth more attention than the formula.

Both values must come from the same draw. This is the most common mistake by a distance. People pull a glucose from a routine panel taken in March and an insulin from a different test taken in July and multiply them together. HOMA models a single steady state, one moment in one feedback loop. Two numbers from two mornings describe two different states and their product describes neither.
It has to be a genuine fasting sample. Usually 8 to 12 hours, water only. Insulin responds to food far faster and far more dramatically than glucose does, so a coffee with milk on the way to the lab can move the insulin figure enough to change the answer materially while leaving the glucose looking innocent.
Check the unit printed on the report, not the unit you assume. Glucose comes in mg/dL in the United States and mmol/L nearly everywhere else. Insulin comes in uU/mL (which is the same thing as mIU/L) or in pmol/L. This page catches the impossible mix-ups for you, and it echoes both lab values back in both unit systems in the first step of the working, precisely so you can check them against your printout. It cannot catch a mix-up where both readings happen to be plausible numbers, and no calculator can.

On that last point, one conversion is worth knowing about. This page converts insulin using 1 uU/mL = 6.00 pmol/L, which is what follows from the WHO insulin standard. An enormous number of lab reports, textbooks, unit tables and online calculators use 6.945 instead, a factor that traces back to an international standard retired in the 1980s. Choosing one over the other changes the reported insulin, and therefore HOMA-IR, by about 14 percent. When you enter insulin in pmol/L this calculator shows you both answers rather than quietly picking one.

What the published reference ranges actually say

Here is the section most HOMA-IR pages skip, and it is the reason this one exists. There is no agreed cutoff. Not a contested cutoff, not a cutoff with regional variations around a settled centre: no agreed cutoff at all. These are real thresholds from real peer reviewed studies, and they disagree with each other by more than a factor of two and a half.

Threshold	Where it comes from
1.4 to 2.5	Various studies in Asian populations, where thresholds run consistently lower
2.05	Spain, EPIRCE study (n = 2,459), anchored to metabolic syndrome components
2.0 to 3.0	The band most commonly applied in United States clinical settings
2.5	The value commonly applied in analyses of NHANES data
2.86	Brazil, upper reference limit from 21,684 laboratory records on one assay
3.46	Spain, the 90th percentile of the same EPIRCE population as the 2.05 above
3.80	United States, a study of Americans of Mexican descent

Look at the two Spanish rows. Same country, same study, same 2,459 people. One threshold comes from taking the top tenth of the distribution and calling it resistant; the other comes from asking which value best flags the metabolic syndrome. Both are defensible. They are 69 percent apart. That single comparison tells you more about HOMA-IR cutoffs than any table of ranges ever will: the number you compare yourself against depends entirely on what question the researchers were asking and who they were asking it of.

Cutoffs also shift with age, sex, body mass, ethnicity, and pubertal stage, which is why a threshold derived in adults should never be pointed at a fourteen year old. The EPIRCE authors went further and reported that in non-diabetic women over 70, HOMA-IR did not usefully classify anybody at all; the confidence interval for its discriminating ability included pure chance.

One more piece of folklore worth defusing. You will often read that a HOMA-IR under 1.0 is optimal. That figure is not a finding. It is the model's calibration point: Matthews and colleagues set the scale so a normal weight healthy adult under 35 lands at 1, using reference fasting values of 5 uU/mL and 4.5 mmol/L. Saying that 1.0 is optimal is a statement about the arithmetic of the model, not a threshold derived from anybody's outcomes.

The assay problem, which is bigger than all of the above

If you take one thing from this page, take this one, because it is the thing that most changes what you should do with your number.

Insulin immunoassays are not standardized between laboratories. Glucose measurement is standardized and reliable and boring, which is exactly what you want from a lab test. Insulin is not. Different manufacturers' assays use different antibodies that respond differently to insulin, to proinsulin, and to its split products, and they have never been harmonized. When an American Diabetes Association workgroup compared assays against a common set of samples, the among-assay coefficients of variation ran from 12 percent to 66 percent, with a median of 24 percent. Giving every laboratory the same reference preparation did not fix it.

Since HOMA-IR moves almost proportionally with insulin, that variation lands straight in your result. The same tube of your blood, split and sent to two laboratories using two different platforms, can come back as two HOMA-IR figures that a threshold table would read completely differently. Nothing about you would have changed.

Three practical consequences follow, and they are more useful than any cutoff:

A HOMA-IR from one lab cannot be compared to a threshold derived at another. Every number in the table above was derived on a particular assay in a particular population. Transferring it to your report is an approximation whose error you cannot see.
If you are tracking your own number over time, stay with one laboratory. Trended within a single assay, HOMA-IR is far more informative than it is as a single absolute value, because the assay bias cancels out. A change that coincides with a change of lab or a change of platform is not evidence of anything.
Do not read small differences as signal. On top of the assay problem, fasting insulin varies biologically from one morning to the next, because insulin is secreted in pulses. A repeatability study within the Atherosclerosis Risk in Communities cohort measured the within-person coefficient of variation over a short interval at 30.4 percent for HOMA-IR and 28.8 percent for insulin, against 5.6 percent for glucose. Glucose is the steady one. A move from 2.4 to 2.7 is well inside the noise.

The researchers who published QUICKI reached the same conclusion from the other direction and put it bluntly: a normal range needs to be established for each laboratory. That advice is more than twenty years old and it is still the right advice.

Where HOMA-IR is not valid at all

The model rests on two assumptions, and when either one fails the number it produces is not a poor estimate. It is meaningless.

The insulin has to be yours. HOMA works backwards from a feedback loop in which your pancreas is responding to your glucose. Injected insulin is not part of that loop, so for anyone using exogenous insulin the calculation has nothing to work backwards from.
The pancreas has to be able to secrete. In type 1 diabetes, and in long-standing type 2 diabetes where beta cell function has substantially failed, a low fasting insulin reflects a pancreas that cannot make insulin rather than a body that does not need much. The model cannot tell those two situations apart, and it will read the first one as excellent insulin sensitivity, which is precisely backwards.
It has to be a fasting steady state. Not a random sample, not a post-meal sample, not a sample drawn during acute illness, and not one taken from someone on a glucose infusion. The model is a description of the resting balance and there has to be a resting balance to describe.

Beyond those hard stops there are softer ones. HOMA-IR is a fasting index, so it mostly reflects hepatic insulin resistance and says relatively little about how your muscles handle a meal. It is at its best across groups of people, which is what it was designed for, and least reliable applied to one individual, which is what everyone uses it for. It also correlates only moderately with the glucose clamp in some populations and barely at all in others: in 110 non-diabetic Jamaican adults, HOMA-IR showed no significant correlation with clamp-measured or minimal-model insulin sensitivity, while still tracking body fat perfectly well. A tool that follows adiposity but not the thing it claims to measure is a tool to hold loosely.

None of that makes it useless. It makes it a screening and research index that can point at a question, and one that cannot answer it.

Insulin resistance in plain terms, and what generally moves it

Insulin resistance means that a given amount of insulin has less effect than it used to. The pancreas compensates by making more, which is why fasting insulin tends to rise long before fasting glucose does, and why an insulin measurement can be informative when glucose alone still looks fine. That compensation can go on for years. HOMA-IR is one attempt to put a number on where that quiet process has got to.

On what changes it, the evidence supports the unglamorous answers and does not support the exciting ones. Body weight, physical activity, sleep, and overall diet quality are the levers with real evidence behind them. The clearest demonstration is the Diabetes Prevention Program, which randomized 3,234 people at high risk and gave the lifestyle group two goals: lose at least 7 percent of body weight, and get at least 150 minutes of moderate activity a week. Over an average of 2.8 years that reduced progression to type 2 diabetes by 58 percent, which was better than the medication arm in the same trial managed.

That is deliberately as far as this page goes. We are not going to hand you a protocol, a supplement, an eating window, or a promise, because we do not know your history, your medications, your other results, or anything else about you. What the evidence supports in general is not the same as what is right for one person, and the distance between those two things is exactly where a clinician earns their keep.

When to talk to a clinician

Take the number to whoever ordered the test, and take the lab report with it rather than the calculated figure alone, since the assay matters as much as the arithmetic. It is worth a conversation if the value sits above the range your own laboratory reports, if it has moved meaningfully across repeat tests at the same lab, or if it arrives alongside other things: a rising fasting glucose or HbA1c, a waist measurement that has been climbing, high triglycerides with low HDL, a family history of type 2 diabetes, polycystic ovary syndrome, or a fatty liver found on a scan somebody ordered for another reason. HOMA-IR earns its place as one line in that picture and nowhere near the whole of it.

And if the number came back higher than you hoped, it is worth remembering what it is: an estimate, from a model, from two numbers, on one morning, on one machine that another laboratory's machine might well disagree with. It is a reason to ask a question. It is not a verdict, and it was never designed to be one.

If you are working through metabolic numbers more broadly, the BMI calculator and TDEE calculator get the same step by step treatment, the calorie calculator walks the Mifflin-St Jeor equation the same way, and the MAP calculator takes the same approach to a clinical number: show the working, state the reference ranges honestly, and grade nobody.

Sources

Where the numbers on this page come from. We go to the body that publishes the figure, not to another calculator. Figures on this page were checked against these sources in August 2026. See how we verify.

Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man, Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC. Diabetologia 1985;28:412-419. The original paper. Source of the HOMA-IR and HOMA-B approximating equations, the 22.5 and 20 and 3.5 constants, and the calibration that makes a healthy young adult score 1.
Use and Abuse of HOMA Modeling, Wallace TM, Levy JC, Matthews DR. Diabetes Care 2004;27(6):1487-1495. The model's own authors on where HOMA is appropriate and where it is not, written nineteen years after the original.
Insulin Assay Standardization: Leading to Measures of Insulin Sensitivity and Secretion for Practical Clinical Care, Staten MA et al, for the Insulin Standardization Workgroup. Diabetes Care 2010;33(1):205-206. The source of the assay figures on this page: among-assay coefficients of variation from 12 to 66 percent with a median of 24, and the finding that a shared reference preparation did not fix it.
Quantitative insulin sensitivity check index: a simple, accurate method for assessing insulin sensitivity in humans, Katz A, Nambi SS, Mather K, Baron AD, Follmann DA, Sullivan G, Quon MJ. J Clin Endocrinol Metab 2000;85(7):2402-2410. The QUICKI formula, with insulin in uU/mL and glucose in mg/dL, and its correlation of 0.78 with clamp-measured insulin sensitivity.
Detection of insulin resistance by simple quantitative insulin sensitivity check index QUICKI for epidemiological assessment and prevention, Hrebicek J, Janout V, Malincikova J, Horakova D, Cizek L. J Clin Endocrinol Metab 2002;87(1):144-147. Mean QUICKI of 0.366 in healthy adults with 0.357 the lower bound of the 95 percent range, and the paper's own conclusion that a normal range has to be established for each laboratory.
Insulin Units and Conversion Factors: A Story of Truth, Boots, and Faster Half-Truths, Knopp JL, Holder-Pearson L, Chase JG. Journal of Diabetes Science and Technology 2019;13(3):597-600. Why 1 uU/mL of insulin equals 6.00 pmol/L and not the widely printed 6.945, which traces back to a standard retired decades ago.
Insulin resistance (HOMA-IR) cut-off values and the metabolic syndrome in a general adult population: effect of gender and age: EPIRCE cross-sectional study, Gayoso-Diz P et al. BMC Endocrine Disorders 2013;13:47. One population, two defensible thresholds: 3.46 at the 90th percentile and 2.05 anchored to metabolic syndrome components, plus the finding that HOMA-IR did not usefully classify non-diabetic women over 70.
Proposal for fasting insulin and HOMA-IR reference intervals based on an extensive Brazilian laboratory database, Schrank Y et al. Archives of Endocrinology and Metabolism 2024. An upper reference limit of 2.86 derived from 21,684 records on one assay, with the authors noting how much published reference intervals differ from one another.
Short-Term Repeatability of Insulin Resistance Indexes in Older Adults: The Atherosclerosis Risk in Communities Study, Journal of Clinical Endocrinology and Metabolism 2018. Within-person coefficients of variation over a short interval: 30.4 percent for HOMA-IR and 28.8 percent for insulin, against 5.6 percent for glucose. The source of this page's warning not to read small changes as signal.
Limitations of fasting indices in the measurement of insulin sensitivity in Afro-Caribbean adults, Thompson DS et al. BMC Research Notes 2014;7:98. In 110 non-diabetic Jamaican adults, HOMA-IR showed no significant correlation with clamp-measured or minimal-model insulin sensitivity, though it did track adiposity.
Reduction in the Incidence of Type 2 Diabetes with Lifestyle Intervention or Metformin, Diabetes Prevention Program Research Group. New England Journal of Medicine 2002;346:393-403. The trial behind the lifestyle section: 3,234 people, a target of 7 percent weight loss and 150 minutes of activity a week, and a 58 percent reduction in progression to type 2 diabetes.

Frequently asked questions

What is a normal HOMA-IR?

There is no single agreed answer, and any page that gives you one is overstating what the evidence supports. Published thresholds for insulin resistance run from about 1.4 in some Asian populations to about 3.8 in a study of Americans of Mexican descent, with 2.0 to 3.0 the range most often used in United States practice. The model itself is calibrated so a healthy young adult scores 1, but that is a normalization choice by its authors rather than a measured cutoff.

What HOMA-IR indicates insulin resistance?

It depends whose reference population you use. The value most commonly applied in United States analyses is 2.5. A large Spanish study derived 3.46 from the 90th percentile of its own population and 2.05 from the same people when the threshold was anchored to metabolic syndrome instead. A Brazilian laboratory database of 21,684 records gave 2.86. The honest summary is a range of roughly 1.4 to 3.8, and where you fall inside it is a conversation for your clinician.

Do I need to fast for a HOMA-IR test?

Yes, and both values must come from the same fasting draw. HOMA is a model of the fasting steady state, the balance between what your liver releases and what your pancreas secretes when nothing has been eaten. A glucose taken fasting and an insulin taken after lunch do not describe a steady state, and the arithmetic on that pair means nothing. Most labs ask for 8 to 12 hours with water only.

Can I compare HOMA-IR results from different labs?

Not reliably. Insulin immunoassays are not standardized between manufacturers, and an ADA workgroup found among-assay coefficients of variation from 12 to 66 percent with a median of 24. The same blood can produce meaningfully different insulin values on different platforms, and HOMA-IR moves with insulin almost proportionally. If you are tracking your own number over time, stay with one laboratory, and treat a change that coincides with a change of lab as unproven.

Why are there two HOMA-IR formulas, one dividing by 405 and one by 22.5?

They are the same constant in two glucose unit systems. 22.5 is 5 uU/mL of insulin multiplied by 4.5 mmol/L of glucose, the fasting values the model treats as its reference, and 405 is that same 22.5 multiplied by 18 to move it into mg/dL. Use the divisor that matches the unit your glucose was reported in. As a curiosity, the true molar conversion for glucose is 18.0156 rather than 18, so the two forms disagree by 0.087 percent, which is nothing next to assay variation.

Is HOMA-IR accurate if I take insulin?

No. HOMA assumes the insulin in your blood is insulin your own pancreas made in response to your own glucose, and it works backwards from that feedback loop. Injected insulin breaks the assumption completely, so the number produced is not interpretable. The same applies to type 1 diabetes and to long-standing type 2 diabetes where beta cell function has substantially failed, because a low insulin there reflects a pancreas that cannot secrete rather than a body that does not need to.

What does HOMA-B tell me that HOMA-IR does not?

HOMA-B estimates beta cell function, scaled so the model's reference adult sits at 100 percent, and it comes free from the same two lab values. Read together they describe two different things: how much insulin it is taking to hold your fasting glucose, and how hard the pancreas is working to supply it. HOMA-B is undefined when fasting glucose is at or below 3.5 mmol/L (63 mg/dL), because the formula divides by glucose minus that offset.

What is the difference between HOMA-IR and QUICKI?

The same two numbers, differently arranged. QUICKI is 1 divided by the sum of the base 10 logarithms of fasting insulin in uU/mL and fasting glucose in mg/dL, and it runs the opposite direction: higher means more insulin sensitive. Katz and colleagues reported a correlation of 0.78 with clamp-measured insulin sensitivity when they published it in 2000. Because it is a log scale it is better behaved at the extremes, which is the main reason researchers reach for it.

How is fasting insulin reported, and does the unit matter?

United States labs generally report uU/mL, also written mIU/L, and much of the rest of the world reports pmol/L. This page converts using 1 uU/mL = 6.00 pmol/L, which follows from the WHO insulin standard. Many reports and calculators use 6.945 instead, a factor traceable to a standard retired in the 1980s, and choosing it instead changes the answer by about 14 percent. Always check the unit printed on your report before typing anything.

My HOMA-IR is high. What actually moves it?

The evidence broadly supports the unglamorous things: body weight, physical activity, sleep, and overall diet quality. The Diabetes Prevention Program, which randomized 3,234 people, used a target of 7 percent weight loss and 150 minutes of activity a week and cut progression to type 2 diabetes by 58 percent. That is the size of effect ordinary changes can have. What none of it means is that you should act on one calculated number alone, and this page is not the right place to plan anything.