Insulinology derives the correction factor from the patient's own total daily dose, then builds a scale where every band is one correction factor wide. It shows its arithmetic, subtracts insulin still on board, converts between insulin products and regimens, and declines to produce a scale where a scale is the wrong instrument.
A pre-printed sliding scale gives the same correction to a 45 kg woman on 12 units a day and a 140 kg man on 180. One of them is being under-treated and the other is being set up for hypoglycaemia. Correction insulin is the one component of a regimen that is trivially individualised — the arithmetic takes a few seconds — and it is routinely the component that is not.
| Effective TDD | 39.2 U/day |
| Correction factor | 46 mg/dL / U |
| 151 – 200 | 1 U |
| 201 – 250 | 2 U |
| 251 – 300 | 3 U |
| 301 – 350 | 4 U |
From insulin actually administered over the previous days, weighted toward the most recent. From weight for the insulin-naïve, adjusted for renal function, age, BMI, steroids and hepatic failure. Or extrapolated from a stable intravenous infusion rate.
The tool shows the full arithmetic from total daily dose through to the band width, and names which constants are guideline-derived and which are empiric rules of thumb. A clinician who cannot audit a number should not trust it.
Nutrition state, glucocorticoid dose in prednisone-equivalents, dialysis, SGLT2 inhibitors, sulfonylureas, concentrated insulin, impaired hypoglycaemia awareness and frailty each change what the tool says — rather than sitting in a static block of small print.
Basal and prandial product switches with the dose adjustments that actually apply, plus regimen changes — infusion to basal-bolus, injections to pump, premix to basal-bolus. Cross-class substitutions are refused outright rather than calculated.
A calculator that produces a confident number for every input is more dangerous than one that stops. Insulinology suppresses its output entirely in four situations, and explains what to use instead.
Needs intravenous insulin with fluid and electrolyte replacement. A subcutaneous scale will not close an anion gap — and on an SGLT2 inhibitor, the glucose can be near normal while the patient is acidotic.
Subcutaneous absorption is erratic with poor peripheral perfusion, so a subcutaneous dose has an unpredictable effect. This is infusion territory.
Targets are far tighter and requirements shift across gestation. Standard correction scales are not validated in pregnancy.
Paediatric dosing uses different constants and carries a higher hypoglycaemia risk. The tool was built for adults and should not be extrapolated.
The ADA Standards of Care and the Endocrine Society guideline on inpatient hyperglycaemia both discourage sliding-scale insulin as sole therapy: it treats hyperglycaemia reactively, after the fact, and produces more glycaemic variability than scheduled basal–bolus dosing.
What this tool produces is the correction component. It is displayed alongside the scheduled basal and nutritional doses it is meant to accompany, deliberately, so that the output cannot be read as a standalone regimen.
Insulinology is an educational calculator. It is not a medical device, is not FDA-cleared, and has undergone no clinical validation. It performs no patient-specific verification, holds no patient data, and sends nothing anywhere — everything runs in your browser and nothing is stored or logged.
The constants it uses are conventional rules of thumb, not validated equations. The 1800 and 1500 rules were derived from outpatient type 1 pump populations. The weight-based starting figures, the 50/50 basal–nutritional split, the glucocorticoid estimate and the infusion conversion fraction are all starting estimates that require titration. Insulin-on-board uses a linear decay approximation of what is really a curvilinear process. Every output requires independent verification against the patient, local protocol, and your own judgement before any insulin is ordered.
The methods page sets out every constant, where it comes from, and what the tool does not model.