Homeostastic Model Assessment and Insulin Sensitivity/Resistance

  1. Jerry Radziuk
  1. Department of Medicine, University of Ottawa, Ottawa Hospital, Ottawa, Ontario, Canada
  1. Corresponding author: Jerry Radziuk, jradziuk{at}uottawa.ca.


Homeostasis is the manifestation of a large network of integrated metabolic reactions and signaling pathways designed to maintain life and function in a stable manner. Insulin is an integral part of this regulatory network and affects virtually all its parts. The actions of insulin vary in time (1) as well as in the various tissues it affects as demonstrated, for example, by selective receptor deletion studies (2). Moreover, insulin individually affects a large number of reactions within these tissues. Lack of insulin, or of its action(s), results in diabetes. This has historically been viewed from the perspective of glucose metabolism, although, as pointed out by McGarry (3), a similar spectrum of deficiencies would have been identified first if lipids had been the primary focus of investigation. Insulin action as an entity is therefore, necessarily, ill-defined. It is embedded within the network of reactions and signals where it plays critical regulatory roles.

In some sense, measuring insulin action must be equally tenuous. When, where, and how it is measured is a matter of physiology, but also of history. Clinically then, resistance to insulin action was first defined on the backdrop of glucose metabolism (4). Specifically, diabetes was divided into insulin-sensitive and -insensitive types. Insulin insensitivity or resistance rose into more prominence when it was associated with other risk factors for the development of type 2 diabetes, such as obesity, dyslipidemia, hypertension, and hypercoagulopathy, grouped under the term hyperinsulinemic/insulin resistance syndrome (X) or currently, the metabolic syndrome. Collectively, this group of pathologies increased the risk of developing diabetes and was associated also with an increased risk of cardiovascular disease (5,6), hence the impetus to quantify insulin resistance.

Many methods consequently have been developed and new ones continue to arise. Given the diverse nature of insulin actions, the heterogeneity of the methods is not surprising. In this issue, the careful article by Ader et al. (7) serves to alert us once again to the when/where/how of the assessment of insulin sensitivity. It compares the change in insulin sensitivity in dogs after a high-fat diet, using the hyperinsulinemic-euglycemic clamp (8), an insulin-supplemented intravenous glucose tolerance test (IVGTT) with minimal model analysis (9), and the homeostatic model assessment (HOMA) method for insulin sensitivity/resistance analysis based on fasting glucose and insulin samples (10). It found in pooled studies that HOMA-insulin resistance (IR) did not correlate with the IVGTT- and clamp-based methods either before or after the fast. Neither were the diet-induced changes in the insulin sensitivity index (SI) found by the other two methods reflected in the HOMA calculation.

In many ways, these results echo the literature on such comparisons. HOMA-SI/IR have been widely applied, often in large-scale population studies, and frequently compared with other methods of SI determination. Largely, they correlate with more detailed indices of SI (clamp, IVGTT) (1114) and in some longitudinal studies (15,16), but less so in others (17), and not as well in some populations such as African American men (18) and Asian-Indian men that include a preponderance of lean diabetic patients (19). In animal models, mice (20) and rats (21) do not do so well with HOMA, but rhesus monkeys do better (22). Clearly, there is some diversity. This occurs in the comparison of any methods. For HOMA, systematic errors could be expected if the model description does not correspond to the population (or species) being studied.


The approaches used in Ader et al. (7) span the methods that have been developed to assess SI—from glucose clamping with the administration of exogenous insulin to the consideration of fasting glucose and insulin, interpreted from a more comprehensive model of glucose metabolism. The insulin-supplemented IVGTT is a hybrid where both endogenous insulin is stimulated with a glucose load and additional insulin given, interpreted through a purpose-driven model, which is therefore minimal. As has been pointed out, all these methods have a common core in that they are built around an insulin effect compartment (23) (i.e., the action of insulin), whether it be on peripheral tissues or the liver is reflected in a compartment or pool that is remote from the circulation (24,25). Transfer from circulating insulin to an effective insulin compartment creates a delay in the action of secreted or administered insulin.


HOMA-IR estimates insulin resistance, or its inverse, SI, simply from basal measurements of plasma insulin and glucose. Optimally, SI is determined from a nomogram (or computer program [10,14,26,27]), but an approximation is often used:Formulawhere I and G are fasting insulin and glucose levels, measured at three different times, to minimize pulsatile effects.

This approximation is derived from the steady-state solution of organ-based equations, describing the glucose-insulin feedback loops. A version of these has been abstracted here (10,26):FormulaFormulaFormulaFormulaFormulaThe insulin kinetic model with the effect compartment identified as receptor-bound insulin is:FormulaFormulaFormulaFormulaFormulaFormulaWithout defining all the parameters, G is circulating glucose, Ge and Geh are effective glucose peripheral and hepatic glucose concentrations (i.e., G multiplied by the sensitivity to glucose, SG and SGL, respectively). IP, IH, IE are peripheral, hepatic, and peripheral receptor (effect compartment) insulin levels; ISR is the insulin secretion rate; and IDR is the peripheral delivery rate. R and RL are peripheral and liver receptor concentrations, and P5 and L5 are peripheral and liver insulin sensitivities (assumed the same for a single estimate of SI). The parameters are identified in previous experiments and are assigned. The insulin sensitivity is estimated from either Eq. 1 or Eqs. 2−12.

Hyperinsulinemic-Euglycemic Clamp

The most straightforward method of estimating a response to insulin is to administer (infuse) insulin and measure the response. In order to circumvent counterregulatory reactions and the effects of varying glucose levels (e.g., Eqs. 3 and 5), glucose is infused at variable rates to maintain basal (isoglycemic) or normal (euglycemic) levels (8). Using an insulin infusion, a steady-state of both insulin and glucose infusion is achieved. Either the glucose infusion rate (GIR) or the ratio of the GIR to insulin (Iss, or incremental insulin above basal) is then used as a measure of SI. (A correction can be made for suppression of insulin secretion using C-peptide levels before and during the clamp.) Thus for insulin sensitivity:FormulaIf SI at different levels of glucose is to be compared, this equation can be divided by the steady-state clamp glucose level, Gss, as GIR/Gss is the metabolic clearance of glucose, the entity that is more directly regulated by insulin. The equation is simple, although the technique is technically demanding. Because this approach isolates the changes due to insulin by maintaining a constant glycemia, it has been the definitive methodology in assessing SI.

IVGTT With Minimal Modeling

Instead of requiring a steady-state of glucose and insulin, this method models the response of glucose clearance to insulin in a systemic effect compartment, identifying all the parameters, including SI (9).

Further Observations on the Methods

Peripheral and Liver Dose-Responses to Insulin

HOMA-based SI is at basal fasting levels of insulin and glucose, whereas all methods based on challenging the system with exogenous insulin will necessarily deal with a response at higher insulin concentrations. Figure 1A shows the dose-responses to insulin for endogenous glucose production or output (EGP) and for glucose uptake (GU) in humans, measured separately using tracer methods (28,29). It can be seen that basal insulin levels define an operating point in midrange of the EGP dose-response and at an early phase of the increase in GU, whereas with clamping insulin concentrations (e.g., Ader et al. [7]), EGP is suppressed and the insulin sensitivities correspond to the insulin sensitivity of GU. A different balance of sensitivities is being measured by HOMA, a basal method, and by those methods that perturb the system.

Figure 1

A: Dose-response to insulin of glucose production and utilization (uptake). Adapted from ref. 28 with point in hypoinsulinemic range from ref. 29. Normal fasting operating point is indicated by the arrow. B: Plot of Eq. 14 using data from diabetic subjects in ref. 27. Intercept of the regression line, with GN = 3.5, corresponds to SI/SIN and is 0.2, which is lower than the average SI (HOMA) of 0.4.

Model Structures

More detailed models can include organ effects but all parameters cannot be identified in a single study so that only relevant parameters (SI, β-cell function) can be varied to accommodate the data. Changes in parameters could however have different effects because the model structure determines the definition of insulin sensitivity, for example, when it involves a nonlinear, saturable GU.

Possible Adjustments

If the definition of SI relative to normal (SI/SIN) is restricted to the insulin-dependent part of glucose metabolic clearance, then (23):FormulaIndeed, data from Hosker et al. (27) suggests that such a relationship exists (Fig. 1B). It further suggests that HOMA-SI works best when glucose is near-normal and that adjustments may be made. Should such adjustments be useful, the estimates of SI/SIN can be achieved on a population basis, which is the recommended venue for HOMA use.

Interestingly, also, when morning fasting glucose levels are elevated as in type 2 diabetes, this appears to correspond to a high EGP, which declines during the day to near-normal rates (30). Glycemia decreases correspondingly. Remaining elevation in glucose appears largely due to decreases in GU. As the insulin levels do not increase as EGP falls, this is not an insulin-dependent effect. Although this may be clinically difficult to accomplish, one wonders whether fasting measurements of SI later in the afternoon might correspond more closely to clamp measurements.

When glycemia is close to normal, the simplified measure of HOMA-IR corresponds very closely to insulin concentrations. This would be expected from the definition and a very close correlation has been demonstrated in humans (31). The data in Ader et al. (7) also seem to be in support of this observation as insulin concentrations increased by 30% when SI (clamp) fell by a third.

The distinction between impaired fasting glucose (IFG) and impaired glucose tolerance (IGT) in the pathogenesis of diabetes may well correspond to different loci for insulin resistance. It has been shown that IFG appears to be due to an increase in EGP, whereas IGT arises from a decreased response of GU to insulin (32). These altered balances in peripheral and hepatic sensitivities could change HOMA calculations as they would be most evident under basal conditions (33).

β-Cell Function

As suggested by Ader et al. (7), β-cell function may be important in determining the predictive value of HOMA-SI/IR. If the insulin secretion is insufficient to compensate for decreases in SI, it would be expected that glucose concentrations increase—concomitantly increasing HOMA-IR. This could mean, however, that changes in stimulated insulin secretion do not necessarily reflect those under basal conditions.

As has been pointed out by Ader et al., it is very important to consider the linkages between changes in SI and those in insulin secretion in the progression to diabetes. In this context, the reciprocal relationship between the two has been demonstrated using HOMA2-based calculations of SI and β-cell function (34), although autocorrelation based on formulae could contribute.


Populations are critical. The HOMA method infers the degree of insulin resistance from the basal levels of endogenous insulin and glucose. It is based on a series of equations that describe the glucose-insulin feedback loops. It comprises a large number of parameters that cannot be identified in a single study. It depends on population estimates of these parameters to provide baseline system behavior. Deviations from this baseline are able to provide the estimates of SI/IR required. If populations are different (for example, diabetes in the lean is different from that in the obese [19]), then the parameter set may be different. If that for another population is used, less robust estimates of SI/IR will result. It also should be pointed out that when glucose is lowered therapeutically by stratagems that do not rely on increasing SI or glucose sensitivity of the β-cell (e.g., sodium-glucose cotransporter 2 inhibitors [26]) parameter sets of more comprehensive models may need to change. Methods that rely on the administration of exogenous insulin and measure the response do not depend on such a parameter set to describe endogenous relationships.

For large population studies and clinical tracking, the simplicity of the HOMA approach is very attractive. Ader et al. (7) have called attention to its potential limitations. Sensitivity of the user to the population being studied, the circumstances of the measurements, and possible improvements should yield better estimates of the sensitivity to insulin. It will be interesting to see how careful adjustment of the fixed model parameters of the HOMA equations (such as those listed here), appropriate to the population or treatment being used that are now possible with the new interactive version (26) of HOMA (iHOMA2), will improve HOMA-based estimates of insulin sensitivity.

Article Information

Duality of Interest. No potential conflicts of interest relevant to this article were reported.


  • See accompanying article, p. 1914.

Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. See http://creativecommons.org/licenses/by-nc-nd/3.0/ for details.


| Table of Contents

This Article

  1. doi: 10.2337/db14-0116 Diabetes vol. 63 no. 6 1850-1854
  1. Free via Open Access: OA

Google Scholar