Every constant this tool uses, where it comes from, and whether it carries guideline weight or is a convention. Nothing here is a black box — if a number cannot be audited, it should not be trusted.
Guideline traceable to a published guideline Label from product prescribing information Empiric conventional rule of thumb, no validation trial
Everything downstream depends on this single number, so the tool offers three routes to it and shows which was used.
Insulin actually given — basal plus nutritional plus correction — over up to three days, weighted toward the most recent day because it best reflects current physiology.
The weights and multipliers are conventions chosen to damp single-day noise while staying responsive. They are not derived from trial data.
ADA Standards of Care in Diabetes—2026, §16, gives 0.3–0.6 units/kg/day as the usual starting total daily dose, calculated from body weight and insulin sensitivity — which is the range the tool offers. The individual modifiers below are conventions, not guideline figures; §16 notes that people with renal insufficiency should receive lower doses but publishes no formula, so the eGFR adjustments here are the tool's own and are labelled empiric.
| Factor | Adjustment | Rationale |
|---|---|---|
| Base (selectable) | 0.3 – 0.6 U/kg | conservative through markedly insulin-resistant |
| Dialysis | − 0.15 | reduced insulin clearance, glucose falls during and after sessions |
| eGFR < 30 | − 0.15 | prolonged insulin half-life |
| eGFR 30 – 44 | − 0.10 | requirements fall as renal function declines |
| Age ≥ 70 | − 0.05 | blunted counter-regulation, higher hypoglycaemia consequence |
| BMI ≥ 40 | + 0.15 | insulin resistance |
| BMI 30 – 39.9 | + 0.10 | insulin resistance |
| Glucocorticoid ≥ 40 mg pred-eq | + 0.20 | dose-dependent insulin resistance |
| Glucocorticoid ≥ 20 mg pred-eq | + 0.15 | |
| Glucocorticoid < 20 mg pred-eq | + 0.10 | |
| Hepatic failure | − 0.10 | impaired gluconeogenesis and glycogen reserve |
| Frailty / limited life expectancy | − 0.05 | deliberate de-intensification |
| Impaired hypoglycaemia awareness | − 0.05 | unheralded events |
| Type 1 floor | 0.30 U/kg | basal requirement is obligatory |
| Overall clamp | 0.15 – 1.0 | bounds against entry error |
Total body weight is used, with resistance handled through the U/kg factor rather than an adjusted-weight formula.
The interval used must be one where both the rate and the glucose were stable — a rate still being titrated does not represent a requirement.
Hypoglycaemia below 70 mg/dL within 24 hours applies a fixed × 0.8 to whichever TDD was derived, and raises the suggested target. This is deliberately blunt: the presence of a hypoglycaemic event is treated as more informative than the precision of the preceding calculation.
These are the "1800 rule" and "1500 rule". Empiric They originate in outpatient type 1 pump populations and were never validated for inpatient use, for type 2 diabetes, or for the acutely ill. Values from 1500 to 2200 all appear in practice.
The practical consequence is that the correction factor is a starting estimate for titration, not a measured property of the patient. It is most reliable when the TDD it came from was itself real — insulin actually administered to a patient whose status has not changed.
In mmol/L the equivalent constants are K ÷ 18, so 1800 becomes 100 and 1500 becomes 83.3. The tool holds all internal arithmetic in mg/dL and converts for display.
Rather than a fixed 50 mg/dL band, each band is one correction factor wide, so each step up the scale is exactly one dose increment.
Correction begins above the target rather than at it, so a patient sitting at goal is not dosed. The scale stops at the cap — default 10 U, and separately flagged if the top dose exceeds 0.15 U/kg — because a requirement that large is a reason to examine the patient, not to give more insulin from a chart.
Any glucose can be checked directly against the formula:
This is the component most missing from conventional scales, and the mechanism behind a substantial share of inpatient hypoglycaemia. A scale that reads only the current glucose cannot see the dose given two hours ago.
Linear decay is a simplification. Real subcutaneous insulin action is curvilinear, peaking at one to two hours, so a linear model under-estimates residual effect early after a dose — the direction that matters, since that is when stacking occurs. The stacking alert exists to compensate: it fires whenever a correction is being considered within 75% of the duration of action, independently of the arithmetic.
Minimum spacing between corrections: 3–4 hours for rapid analogs, 5–6 hours for regular insulin.
Empiric Basal 50% of TDD, nutritional 50% divided across meals — the conventional starting split. The tool shows these alongside the correction scale deliberately, so that the scale cannot be read as a standalone regimen.
Carbohydrate ratio, when shown, uses the 500 rule: 500 ÷ TDD grams per unit.
Nutritional insulin is held automatically when the patient is NPO, on continuous feeds, or on parenteral nutrition — but basal is never held, and in type 1 the tool says so explicitly on every render.
Doses are converted to prednisone-equivalents Label using the standard potency ratios: hydrocortisone 20 mg = prednisone 5 mg = methylprednisolone 4 mg = dexamethasone 0.75 mg.
Where NPH matching is suggested, it uses the convention of 0.1 U/kg per 10 mg prednisone-equivalent, capped at 0.4 U/kg. Empiric Dexamethasone is handled separately because its 24-hour action does not produce the afternoon-weighted pattern that morning prednisone does.
| Situation | Target | Basis |
|---|---|---|
| Non-critically ill inpatients — what this tool is for | 100 – 180 mg/dL | Guideline ADA 2026 Rec 16.5b |
| Critically ill — hard stop, use intravenous insulin | 140 – 180 mg/dL | Guideline ADA 2026 Rec 16.5a |
| Threshold for starting or intensifying insulin | ≥ 180 mg/dL | Guideline ADA 2026 Rec 16.4 |
| Before, during and after surgery | 100 – 180 mg/dL | Guideline ADA 2026 Rec 16.15 — new in 2026 |
| Elective surgery, preceding 3 months | A1C < 8% | Guideline ADA 2026 Rec 16.14 — new in 2026 |
| Frailty, limited life expectancy, impaired awareness, advanced CKD, recent hypoglycaemia, hepatic failure | toward the upper end | Empiric de-intensification within the range |
The non-critically ill target moved from 140–180 to 100–180 mg/dL, widening the acceptable range downward. The tool defaults to correcting toward 140 — the middle of that range rather than its floor, so a correction dose does not aim the patient at 100 and overshoot. Enter a different target if your institution's protocol says otherwise.
Perioperative glucose (16.15) and a preoperative A1C goal (16.14) are both new recommendations in 2026, and both now fire as warnings when the perioperative box is ticked.
The tool suggests a target from the entered risk factors and flags when the chosen target is lower than suggested, but it does not override the user's entry.
Four conditions suppress the scale entirely. This is the core safety design: a calculator that answers every question confidently is more dangerous than one that stops.
| Condition | Why a subcutaneous scale is the wrong instrument |
|---|---|
| DKA or HHS | Requires intravenous insulin with fluid and electrolyte replacement. A subcutaneous scale will not close an anion gap. On SGLT2 inhibitors the glucose may be near-normal while the patient is acidotic. |
| Shock, vasopressors, severe oedema | Subcutaneous absorption is erratic with poor peripheral perfusion, making the effect of a given dose unpredictable. |
| Pregnancy or breastfeeding | Targets are substantially tighter and requirements change across gestation. Standard scales are not validated in pregnancy. |
| Age under 18 | Paediatric dosing uses different constants with higher hypoglycaemia risk. The tool was built for adults. |
The insulin converter deliberately remains available during a hard stop, since an infusion-to-subcutaneous conversion is precisely what a DKA transition requires.
| Conversion | Factor | Basis |
|---|---|---|
| Twice-daily basal → once-daily long-acting analog | × 0.8 | Label Lantus PI: 80% of the total NPH dose. Toujeo PI: 80% of a twice-daily NPH or detemir dose. |
| Off glargine U-300 → any U-100 basal | × 0.8 | Empiric not a label figure — see the note below |
| Once-daily basal → glargine U-300 | × 1.0 | Label Toujeo PI, but expect to need a higher dose over time |
| Any basal → degludec, adults | × 1.0 | Label Tresiba PI: same total daily unit dose (80% is the paediatric figure) |
| Between long-acting analogs, once daily | × 1.0 | Label starting point, then titrate |
| Daily basal ↔ weekly basal | × 7 / ÷ 7 | Empiric arithmetic only — loading-dose practice varies by indication and region, confirm against the current label |
| Between rapid-acting analogs | × 1.0 | Label |
| Regular ↔ rapid analog | × 1.0 | Label dose unchanged, but timing, duration and correction constant all change |
| Between premixes | × 1.0 | Empiric two-thirds morning, one-third evening |
| To or from U-500 | × 1.0 units | Empiric units convert one-for-one, volume does not — many reduce 10–20% at the switch |
| Injections → pump | × 0.8 | Empiric continuous delivery is more efficient |
| Pump → injections | × 1.0 | Empiric 0.8–0.9 if hypoglycaemia-prone |
| Infusion → subcutaneous | × 0.75 | Empiric of the extrapolated 24-hour IV requirement |
An earlier version of this page tagged it Label. That was wrong, and the correction is recorded in the changelog. The Toujeo prescribing information gives a figure for switching to Toujeo — unit-for-unit from a once-daily basal, 80% from a twice-daily NPH or detemir dose — but gives no number for switching off it. The EU Summary of Product Characteristics simply directs the prescriber to the label of the insulin being switched to.
The 20% reduction is a reasonable inference from the pharmacology — U-300 needs roughly 10–20% more units than a U-100 basal for equivalent effect — and it is what most endocrinologists teach. It is still an inference. The tool applies it and says so on screen.
Basal and prandial insulin are not interchangeable. Substituting one for the other at the same dose produces either DKA or severe hypoglycaemia, so the tool returns a refusal rather than a number. Converting on or off a premix changes the structure of the regimen rather than the dose, and is routed to the regimen tab instead.
It also performs no verification that the entered values are correct. A mistyped total daily dose produces a confidently wrong scale, which is why the tool flags implausible values — above 1.5 U/kg/day, below 0.2 U/kg/day, or a correction factor outside 15–100 mg/dL per unit — rather than silently accepting them.
Insulinology is an educational calculator. It is not a medical device, is not FDA-cleared, and has undergone no clinical validation. It does not diagnose, does not predict, and does not direct dosing autonomously — it performs transparent arithmetic on values a clinician enters, and every output requires independent verification before any insulin is ordered.
No data leaves the browser. The tool is a single static HTML file with no backend, no analytics, no third-party scripts, and no network requests. Nothing entered is transmitted, stored, or logged. Closing the tab discards everything.
Consult current editions — inpatient glycaemic guidance is revised regularly, and this page is not a substitute for the primary documents.
Verified against the 2026 edition on 4 August 2026. The Standards are revised annually and the section numbering shifts between editions — re-check the recommendation numbers above against the current release before relying on them.
The 1800 and 1500 rules are long-standing conventions from the insulin pump literature rather than guideline recommendations, and are labelled empiric throughout.
Clinical constants in this tool have already changed once. Anyone who used it before a change is entitled to know what moved, so every change to a clinical value is recorded here.
Current version: 1.0 — released 5 August 2026.
The tool records which edition it was verified against and the date the next is due. Once that date passes, the Guideline basis panel opens by itself and warns that the values need re-checking. Separately, a scheduled monthly job checks whether a new ADA edition has appeared and reports what changed.
Nothing updates a clinical value automatically. That is deliberate. A constant changes only when a clinician reads the new edition and edits the source — the automation exists to catch drift, not to act on it.
If a number here looks wrong, please say so — that is the most useful thing anyone can do with this tool. Include the inputs you entered, what the tool produced, and what you expected.
Corrections to clinical content are made against a primary source, recorded in the changelog above, and deployed. There is no tracking of who reported what, and no analytics on this site.