Methodology - Insulin Resistance Calculator

About the metrics

HOMA-IR: (fasting glucose [mg/dL] × fasting insulin [uU/mL]) / 405 — a simple fasting-state index of insulin resistance.
WBISI (Matsuda Index): 10000 / √[(fasting G × fasting I) × (mean G × mean I)] over the standard 0/30/60/90/120-min OGTT window — a whole-body insulin sensitivity index.
Cobelli Oral Minimal Model SI: Fits a two-compartment glucose/remote-insulin ODE model to the full OGTT glucose curve by nonlinear least squares, using measured insulin as a forcing input. SG and the distribution volume V are fixed to published population constants (median reference values from Dalla Man, Caumo, Basu, Rizza, Toffolo & Cobelli, 2004, a two-tracer validation study) rather than freely estimated, since SG and SI cannot both be reliably identified from OGTT data alone. SI is computed by a more flexible 8-parameter piecewise-linear gut-absorption model (the primary estimator shown here), with the simpler single-parameter gamma-shaped model retained and computed alongside it as a fast cross-check. Reported alongside a glucose-fit RMSE so low-confidence fits (poor convergence, a fitted parameter at its physiological bound, high fit error, or an SI value above the typical physiological range) are flagged — a low or near-zero SI is treated as a valid result (severe insulin resistance), not an error. When the two models disagree substantially on the same subject, that disagreement is shown as a note rather than hidden, since it signals genuine uncertainty in the fit for that individual's data.

This tool is provided for research and educational purposes only. It is not a medical device and must not be used to guide clinical decisions.

Simplifications relative to the published method

This tool's Cobelli oral minimal model fit is a simplified implementation, not a certified equivalent of the original method. Specifically:

Gut glucose absorption Ra(t) is modeled with an 8-knot piecewise-linear curve (the primary estimator) rather than the full published absorption model; a simpler single-parameter gamma-shaped curve is computed alongside it as a fast cross-check.
SG and the distribution volume V are fixed to published population constants rather than individually estimated per subject, to avoid a known structural identifiability problem between SG and SI.
Even the published method's own tracer-validated accuracy is limited at the individual level (see "How accurate is the minimal-model SI, really?" below) - this is a property of the oral minimal model itself, not specific to this implementation.

Results should be interpreted with these simplifications in mind and validated against established clinical methods before any downstream use.

Limitations

How accurate is the minimal-model SI, really?: The 2004 tracer-validation study above found good population-level agreement between the oral minimal model and a two-tracer reference method (r = 0.86 across 88 subjects), but substantial individual-level deviation even under those ideal, tracer-measured conditions: the average absolute difference from the reference was ~29%, ranging from 0.4% to 160% across subjects. A discrepancy of similar magnitude against another reference method (HOMA-IR, WBISI, or a different lab's minimal-model fit) should be expected, and is not by itself evidence of a bug in this tool.
Does the 180-minute protocol length matter?: This tool's template stops at 180 minutes. In a separate study of 68 adolescent girls with obesity (Bartlette et al., 2021), minimal-model SI estimated from only 2-3 hours of OGTT data was lower than the estimate from a full 6-hour reference protocol, and a 4-hour truncated version came closest to that 6-hour reference. In plain terms: if glucose has not returned to baseline by the time sampling stops, the resulting SI may be underestimated, and this is more likely in people who are more insulin-resistant (since their glucose takes longer to come back down). That finding comes from adolescent girls with obesity specifically, so how far it applies to other ages, sexes, or body types is uncertain - we can't say by how much (if any) this specific tool is affected, only that the possibility is worth keeping in mind for subjects whose glucose is still elevated at the 180-minute mark.

References

This implementation is based on the following published methods:

Dalla Man C, Caumo A, Cobelli C. The oral glucose minimal model: estimation of insulin sensitivity from a meal test. IEEE Trans Biomed Eng. 2002;49(5):419-429.
Dalla Man C, Caumo A, Basu R, Rizza RA, Toffolo G, Cobelli C. Minimal model estimation of glucose absorption and insulin sensitivity from oral test: validation with a tracer method. Am J Physiol Endocrinol Metab. 2004;287(4):E637-E643. (source of the fixed SG, V, f, and p2-prior values used here)
Cobelli C, Dalla Man C, Toffolo G, Basu R, Vella A, Rizza R. The oral minimal model method. Diabetes. 2014;63(4):1203-1213.
Cobelli C, Dalla Man C. Minimal and maximal models to quantitate glucose metabolism. J Diabetes Sci Technol. 2022;16(5):1270-1298.
Matsuda M, DeFronzo RA. Insulin sensitivity indices obtained from oral glucose tolerance testing. Diabetes Care. 1999;22(9):1462-1470. (WBISI/Matsuda index)
Matthews DR, et al. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia. 1985;28(7):412-419. (HOMA-IR)
Bartlette K, et al. Validation of a shortened oral minimal model protocol in adolescent girls with obesity. Metabolism Open. 2021;9:100078. (protocol-duration sensitivity of minimal-model SI)