How to test for insulin resistance

No routine clinical test directly diagnoses insulin resistance. Research methods can measure how sensitively the body’s tissues respond to insulin, but they require hours of monitored intravenous infusions and are not used in everyday care (NIDDK, DeFronzo et al. 1979).

Instead, clinicians order standardized glucose and A1C tests, which diagnose prediabetes and diabetes rather than insulin resistance itself (ADA Standards of Care 2026). Other measures, such as fasting insulin, Homeostatic Model Assessment for Insulin Resistance (HOMA-IR), the Triglyceride-glucose (TyG) index, and the triglyceride-to-HDL (TG/HDL-C) ratio, are used to estimate insulin resistance. The interpretation of these measures depends on several factors: type of laboratory assay used, the formula applied, the units of measurement used in the calculations, the referenced population cutoff, and what the cutoff was compared against.

This article discusses what each measure establishes and what it cannot. For more background on the condition itself, take a look at our guide on insulin resistance.

Is there a test for insulin resistance?

Yes, but only in research settings. Insulin resistance can be measured directly using the euglycemic insulin clamp. This is a research reference method, in which insulin is infused at a fixed rate while glucose is infused at a variable rate to hold blood glucose steady. The amount of glucose needed to maintain that steady state reflects how much glucose the tissues take up in response to insulin (DeFronzo et al. 1979).

DeFronzo, Tobin, and Andres described two distinct clamp techniques in 1979. The hyperglycemic clamp raises and holds blood glucose above normal to quantify beta-cell insulin secretion. The euglycemic insulin clamp holds glucose at normal levels during insulin infusion to quantify tissue insulin sensitivity (DeFronzo et al. 1979). Only the second method measures what most people mean by insulin resistance. Sources that refer to a single “glucose clamp” blur two methods answering different questions.

Note that a primary care office can’t order a clamp. It requires intravenous access, repeated blood sampling, and hours of staff time, so insulin resistance is inferred, not measured (NIDDK).

What clinicians order instead

In routine clinical care, clinicians diagnose prediabetes and diabetes using standardized blood tests to measure glucose levels and A1C (ADA Standards of Care 2026). These criteria exist because they are reproducible across laboratories and tied to studied outcomes.

Criteria per the American Diabetes Association Standards of Care (ADA 2026).

Even these standardized numbers contain a gray zone. The ADA’s lower bound for impaired fasting glucose is 100 mg/dL (5.6 mmol/L) (ADA 2026), while the WHO/IDF sets the lower bound at 110 mg/dL (6.1 mmol/L) (WHO/IDF 2006). A fasting glucose of 105 mg/dL, for example, would meet the ADA definition of prediabetes while falling below the WHO threshold. There are no biological changes on either end; risk runs along a continuum, and two expert bodies drew the boundary in different places.

For A1C, the WHO uses A1C ≥6.5% as a diagnostic cutoff for diabetes when strict testing quality controls are in place; however, it did not formally recommend how to interpret values below 6.5% (WHO 2011). So a reader comparing sources may find an intermediate A1C range attributed to one body and not another.

If you are trying to figure out how these conditions relate to one another, see insulin resistance vs. prediabetes vs. type 2 diabetes.

Fasting insulin

A fasting insulin test measures the concentration of insulin circulating in the blood after fasting. Insulin-resistant tissues require more insulin to keep glucose normal, so insulin may rise before glucose does. Unfortunately, this measurement is not standardized across laboratories.

A 2025 assessment of serum insulin measurement found that insulin immunoassays remain incompletely harmonized, meaning the same blood sample can return meaningfully different values depending on which platform runs it (Rohlfing et al. 2025). This is not a new problem; a standardization workgroup flagged it as a barrier to clinical use of insulin-based sensitivity measures years earlier (Staten et al. 2010), and even now, it has not been fully resolved.

For someone with a result, a fasting insulin value cannot be reliably compared against a cutoff published online unless that cutoff came from the same assay. There is also no universal threshold for fasting insulin. Published values vary substantially, so it is difficult to interpret a single number detached from its assay.

A similar issue exists with HOMA-IR, which has fasting insulin as one of its two inputs. In addition, the Endocrine Society’s framework for identifying patients at metabolic risk centers on blood pressure, waist circumference, fasting lipids, and glycemia as core measures. Fasting insulin and HOMA-IR are not part of that core set (Rosenzweig et al. 2019).

HOMA-IR and what its numbers mean

Homeostasis model assessment of insulin resistance (HOMA-IR) is not a test you can draw; rather, it is a calculation based on two components: fasting insulin and fasting glucose. The original HOMA1 formulas are unit-bound:

• HOMA1-IR = fasting insulin (µU/mL) × fasting glucose (mmol/L) / 22.5
• HOMA1-IR = fasting insulin (µU/mL) × fasting glucose (mg/dL) / 405

These two forms are equivalent when used correctly. Errors occur when the wrong divisor is paired with the glucose units, which produces a number that looks plausible but is roughly eighteen-fold off. If you are checking a value, confirm which glucose units the calculator expected. A calculator is available here: HOMA-IR calculator.

It is difficult to determine which HOMA-IR value is “normal” because there are at least five separate sources of variation:

• Insulin assays are not fully harmonized, so insulin measurement varies by laboratory (Rohlfing et al. 2025).
• The model version matters. HOMA1 and HOMA2, the updated computer model, are not interchangeable, so their outputs and cutoffs should not be swapped (Wallace et al. 2004).
• Thresholds are derived empirically in different populations, so a cutoff reflects the cohort, assay, and definition of insulin resistance used to generate it (Tahapary et al. 2022).
• The original validation was imprecise at the individual level.
• HOMA-IR varies within the same person over time, independent of any change in health.

To check how well HOMA-IR reflects insulin resistance, Matthews and colleagues compared it with more direct laboratory measurements (Matthews et al. 1985). They found that HOMA-IR generally tracked well with these more precise tests. However, the relationship was much stronger when comparing groups of people than when estimating insulin resistance in a single person. HOMA-IR is useful for studying patterns across populations but less precise as a standalone measure for an individual.

Biological variation adds another layer. In the European Biological Variation Study cohort, within-person biological variation for HOMA-IR was approximately 26.7% (Carobene et al. 2025). In plain terms, repeated measurements in the same stable person can differ noticeably without any change.

The effect of the chosen index is illustrated within a single large cohort. Among 7,875 fasting participants in the Qatar Biobank, the proportion classified as insulin resistant ranged from approximately 51% to 65% depending on which index was applied (Aliyu et al. 2025). That range describes index disagreement in that cohort; it is not a global or U.S. prevalence figure and does not establish cutoffs for other populations.

Taken together, there is no universally accepted normal HOMA-IR, and this article does not identify one. A HOMA-IR result is best read as one piece of context alongside standardized glucose and A1C testing, not as a diagnosis in itself.

Lipid-based estimates: TyG and the TG/HDL-C ratio

Two indices estimate insulin resistance from lipid and glucose values that most people already have from a routine panel. Neither measures insulin.

The TyG index

The triglyceride-glucose index (TyG index) combines fasting triglycerides and fasting glucose into a single number. The index was tested for sensitivity and specificity against the euglycemic clamp in a study with 99 participants (Guerrero-Romero et al. 2010). The study showed that the TyG index had high sensitivity (96.5%) and specificity (85.0%), making it a potentially useful tool to identify people with decreased insulin sensitivity.

Many online TyG cutoffs are not accurate enough to be usable. Two different formula conventions circulate: Ln[(TG × glucose)/2], where the division happens inside the logarithm, and Ln(TG × glucose)/2, where the logarithm is divided by two. The original authors have clarified the intended calculation (Simental-Mendía & Guerrero-Romero 2020), and the persistent confusion has been flagged as needing correction (Jialal 2025). The two conventions generate entirely different numbers. For example, a TyG value near 4.7 and one near 8.7 may both be “correct” under different arithmetic. A TyG cutoff is uninterpretable unless you know which formula and which units produced it, which is why there is no universal threshold.

The TG/HDL-C ratio and why one cutoff does not fit everyone

The threshold is the problem with this ratio. The TG/HDL-C ratio divides fasting triglycerides by HDL cholesterol. While it does not measure insulin action, it identifies a lipid pattern associated with insulin resistance in some people but not in others.

Three studies illustrate why a single cutoff cannot be defined:

• Sumner and colleagues reported that fasting triglycerides and the TG/HDL-C ratio performed poorly as markers of insulin resistance among African American adults aged 20 to 50 (Sumner et al. 2005). These findings apply only to this specific population and cannot be generalized.
• Kim-Dorner and colleagues reached a similar conclusion in 149 adults aged 18 to 45. Triglycerides and the TG/HDL-C ratio did a reasonable job of flagging insulin resistance in white participants but a poor job in African American participants (Kim-Dorner et al. 2010).
• Li and colleagues analyzed 2,652 nondiabetic U.S. adults, defining hyperinsulinemia by fasting serum insulin rather than by clamp. Discrimination was broadly similar across non-Hispanic white, non-Hispanic Black, and Mexican American adults, yet the authors proposed different cutoffs: 3.0 for non-Hispanic white and Mexican American adults, and 2.0 for non-Hispanic Black adults (Li et al. 2008).

These are patterns seen across groups, not rules for any one person. Units matter too. Because triglycerides and HDL cholesterol are converted between mg/dL and mmol/L by different factors, the ratio changes depending on which units are used. A ratio of 3.0 calculated in mg/dL is about 1.3 in mmol/L. If you see a cutoff, check which units it’s based on.

Tests sold directly to consumers

Direct-to-consumer metabolic panels typically include fasting glucose, fasting insulin, a lipid panel, A1C, and calculated indices such as HOMA-IR and TyG. Because the measures are the same, the panels inherit the same limitations, such as assay variability and formula ambiguity. Those limitations don’t disappear when the measures are packaged together and returned with a color-coded reference range.

Questions to ask regarding these panels include which assay produced the insulin value, which formula version generated the index, which units the cutoff assumes, and which population the reference range was derived in. Those details determine whether a flagged result means anything comparable to a result from a different vendor.

One index found on some panels is LP-IR, the lipoprotein insulin resistance index. It is a composite score derived from lipoprotein particle measurements made by nuclear magnetic resonance spectroscopy (Shalaurova et al. 2014). Like TyG and TG/HDL-C, it does not measure insulin concentration or tissue insulin action. Cutoffs for LP-IR are laboratory-specific; it is not an official diagnostic criterion found in clinical guidelines.

Continuous glucose monitoring without diabetes

A continuous glucose monitor (CGM) measures glucose in interstitial fluid, the fluid between cells. It neither measures insulin concentration nor tissue insulin sensitivity. Glucose variability and insulin resistance are related concepts, but not the same measurement. A spiky glucose trace does not measure insulin action.

For diagnosis, the ADA considers the evidence insufficient to support CGM for screening or diagnosing prediabetes or diabetes (ADA 2026). A narrative review of CGM use in people not living with diabetes described the available evidence as limited and heterogeneous, with no standardized benchmarks for interpreting the resulting metrics in this group (Oganesova et al. 2024).

Without defined normal ranges for people without diabetes, a CGM app’s assessment of a post-meal rise reflects the manufacturer’s chosen thresholds rather than a validated diagnostic standard.

What has held up, what hasn’t, and what remains disputed

What has held up. The euglycemic insulin clamp remains the research reference method for quantifying tissue insulin sensitivity, distinct from the hyperglycemic clamp used for beta-cell secretion (DeFronzo et al. 1979). Standardized glucose and A1C criteria remain the gold standard for diagnosing diabetes and prediabetes (ADA 2026). Non-standardization of the insulin assay is a documented and current issue (Rohlfing et al. 2025). And surrogate indices demonstrably depend on the formula convention and the population in which their thresholds were derived (Tahapary et al. 2022, Jialal 2025).

What has not held up, or is not yet supported. Whether a blood test can reliably catch insulin resistance early has not been validated. A good correlation with a reference method does not create a standardized threshold for diagnosing an individual. Correlation describes how two measures move together across a group, while diagnosis requires a reproducible cutoff on a standardized assay.

What remains disputed. Whether and how surrogate indices should move from epidemiologic and research use into clinical practice remains open. Researchers continue to debate this issue (Tahapary et al. 2022). Knowing that a question is unresolved is useful; it explains why two clinicians can look at the same HOMA-IR value and weigh it differently.

Frequently asked questions

What is a normal HOMA-IR?

There is no single universally accepted normal cutoff, and a cutoff shown without its assay, formula, units, and source population leaves out the conditions it depends on. Four things affect the number independently of your health: insulin assays are not fully harmonized, so the same sample yields different insulin values at different laboratories; within-person biological variation for HOMA-IR was approximately 26.7% in the European Biological Variation Study cohort; HOMA1 and HOMA2 are different models whose outputs and cutoffs are not interchangeable; and thresholds are derived in specific populations using specific definitions of insulin resistance.

Can a blood test tell you if you have insulin resistance?

No routine blood test diagnoses insulin resistance the way standardized glucose and A1C criteria diagnose prediabetes and diabetes. Research methods can measure insulin sensitivity directly, and surrogate blood-based estimates exist, but they are inferences, not actual diagnostic tests. A surrogate result outside a reference range flags a pattern worth discussing but does not establish a diagnosis on its own.

What is the difference between a fasting insulin test and HOMA-IR?

Fasting insulin is measured in a blood sample. HOMA-IR is calculated from two values: fasting insulin and fasting glucose. Because insulin is an input to the calculation, HOMA-IR carries forward every limitation of the insulin assay, including the incomplete harmonization across platforms. HOMA-IR also depends on which model version was used, since HOMA1 and HOMA2 are not interchangeable.

Do you need to fast before insulin resistance testing?

It depends on the measure. Fasting insulin and fasting glucose are, by definition, fasting measurements, and HOMA-IR is calculated using both. The TyG index and the TG/HDL-C ratio are typically calculated from fasting samples in the studies that evaluated them. A1C does not require fasting. Fasting durations are determined by the ordering laboratory or study protocol, not by the index itself.

Conclusion

Insulin resistance can be measured directly, but only with research methods such as the euglycemic insulin clamp. Routine clinical care instead diagnoses prediabetes and diabetes using standardized glucose and A1C criteria. Fasting insulin, HOMA-IR, TyG, and TG/HDL-C are surrogate estimates that are informative in populations and research. However, these estimates depend on the assay, formula, units, model version, and population cutoff. They cannot establish an individual diagnosis on their own. For further reading, see our guide on insulin resistance.

References

1. National Institute of Diabetes and Digestive and Kidney Diseases. Insulin Resistance & Prediabetes. NIDDK health information.
2. DeFronzo RA, Tobin JD, Andres R. Glucose clamp technique: a method for quantifying insulin secretion and resistance. Am J Physiol. 1979;237(3):E214-E223. doi:10.1152/ajpendo.1979.237.3.E214
3. American Diabetes Association Professional Practice Committee. 2. Diagnosis and Classification of Diabetes: Standards of Care in Diabetes—2026. Diabetes Care. 2026;49(Supplement_1):S27-S49. doi:10.2337/dc26-S002
4. World Health Organization; International Diabetes Federation. Definition and diagnosis of diabetes mellitus and intermediate hyperglycaemia. Geneva: WHO; 2006.
5. World Health Organization. Use of Glycated Haemoglobin (HbA1c) in the Diagnosis of Diabetes Mellitus. Geneva: WHO; 2011.
6. Rohlfing C, Petroski G, Hatten-Beck M, et al. The current status of serum insulin measurements and the need for standardization. Clin Chem Lab Med. 2025;63(12):2442-2446. doi:10.1515/cclm-2025-0552
7. Staten MA, Stern MP, Miller WG, Steffes MW, Campbell SE. Insulin assay standardization. Diabetes Care. 2010;33(1):205-206. doi:10.2337/dc09-1206
8. Rosenzweig JL, Bakris GL, Berglund LF, et al. Primary Prevention of ASCVD and T2DM in Patients at Metabolic Risk. J Clin Endocrinol Metab. 2019;104(9):3939-3985. doi:10.1210/jc.2019-01338
9. Wallace TM, Levy JC, Matthews DR. Use and abuse of HOMA modeling. Diabetes Care. 2004;27(6):1487-1495. doi:10.2337/diacare.27.6.1487
10. Tahapary DL, Pratisthita LB, Fitri NA, et al. Challenges in the diagnosis of insulin resistance. Diabetes Metab Syndr. 2022;16(8):102581. doi:10.1016/j.dsx.2022.102581
11. Matthews DR, Hosker JP, Rudenski AS, et al. Homeostasis model assessment. Diabetologia. 1985;28(7):412-419. doi:10.1007/BF00280883
12. Carobene A, Kilpatrick E, Bartlett WA, et al. The biological variation of insulin resistance markers (EuBIVAS). Clin Chem Lab Med. 2025;63(1):110-117. doi:10.1515/cclm-2024-0672
13. Aliyu U, Toor SM, Abdalhakam I, et al. Evaluating indices of insulin resistance in a large biobank cohort. Front Endocrinol. 2025;16:1591677. doi:10.3389/fendo.2025.1591677
14. Guerrero-Romero F, Simental-Mendía LE, González-Ortiz M, et al. The product of triglycerides and glucose. J Clin Endocrinol Metab. 2010;95(7):3347-3351. doi:10.1210/jc.2010-0288
15. Simental-Mendía LE, Guerrero-Romero F. The correct formula for the triglycerides and glucose index. Eur J Pediatr. 2020;179(7):1171. doi:10.1007/s00431-020-03644-1
16. Jialal I. Confusion Concerning the Calculation of the Triglyceride-Glucose Index. Metab Syndr Relat Disord. 2025;23(3):135-136. doi:10.1089/met.2024.0193
17. Sumner AE, Finley KB, Genovese DJ, et al. Fasting triglyceride and the TG-HDL cholesterol ratio in African Americans. Arch Intern Med. 2005;165(12):1395-1400. doi:10.1001/archinte.165.12.1395
18. Kim-Dorner SJ, Deuster PA, Zeno SA, et al. Should triglycerides and the TG/HDL-C ratio be used as surrogates for insulin resistance? Metabolism. 2010;59(2):299-304. doi:10.1016/j.metabol.2009.07.027
19. Li C, Ford ES, Meng YX, Mokdad AH, Reaven GM. Does the association of the TG/HDL-C ratio with fasting insulin differ by race/ethnicity? Cardiovasc Diabetol. 2008;7:4. doi:10.1186/1475-2840-7-4
20. Shalaurova I, Connelly MA, Garvey WT, Otvos JD. Lipoprotein insulin resistance index. Metab Syndr Relat Disord. 2014;12(8):422-429. doi:10.1089/met.2014.0050
21. Oganesova Z, Pemberton J, Brown A. The use of continuous glucose monitors in people not living with diabetes. Diabet Med. 2024;41(9). doi:10.1111/dme.15369