What the model does

Each bolus you enter contributes a fraction of its original dose as still-active insulin at the time of the calculation. That fraction is determined by how long ago the bolus was given, using a well-established decay curve for rapid-acting insulin analogues. Total IOB is the sum across all entered boluses.

Evidence basis

Rapid-acting analogues (Humalog, NovoRapid, Fiasp) follow a well-validated biexponential decay — peak action approximately 60–90 minutes, duration approximately 4 hours (shorter for Fiasp). This pharmacokinetic model underpins all commercial bolus calculators and AID algorithm IOB tracking.

What the model does

Raw IOB is normalised to body weight. The same number of units of active insulin represents a very different physiological burden in a smaller versus a larger person. All downstream calculations use the weight-normalised IOB figure, not the raw unit total.

Evidence basis

Weight-based dosing is standard pharmacological practice. Physiological insulin burden scales with distribution volume, which correlates strongly with body weight. This approach is consistent with T1D Exchange data and CGM exercise datasets from the diabetes research literature.

What the model does

The weight-normalised IOB figure is classified into one of four bands — Safe, Low-Moderate, Moderate-High, and Elevated — each mapped to a level of exercise-associated hypoglycaemia risk. Higher bands trigger progressively stronger guidance around carbohydrate intake and whether to proceed with the session.

The exact thresholds defining each band are part of the proprietary algorithm.

Evidence basis

Pre-exercise IOB is a primary predictor of exercise-associated hypoglycaemia — identified as a key variable in the T1DEXI real-world dataset (Bergford S et al. Diabetes Technol Ther 2023;25(9):602–611). No single clinical trial directly defines safe IOB/kg thresholds for exercise. Band thresholds were derived from analysis of the T1DEXI dataset and Pemberton JS service evaluation data (BWCFT, unpublished, 2019); exact values are proprietary GNL IP.

Moderate confidence because published studies use varied definitions of IOB and different activity protocols. The bands represent a reasonable clinical synthesis, not a single directly evidence-derived value.

What the model does

Activities are classified by their aerobic demand. Aerobic exercise (walking, cycling, swimming, dancing, cross-trainer) has the strongest and most predictable glucose-lowering effect. Resistance exercise may initially raise glucose. HIIT has a mixed response — an early rise followed by sustained lowering. The algorithm applies a directional multiplier to the drop estimate based on the selected activity type.

Evidence basis

Aerobic exercise increases glucose uptake in working muscle and enhances insulin sensitivity for hours post-exercise — well established in the literature. Resistance exercise drives an acute catecholamine and glucagon response. HIIT has complex biphasic glucose effects. Sources: Bally et al. 2016, Colberg et al. 2016, Yardley et al.

What the model does

A falling glucose trend at the start of exercise increases the estimated drop; a rising trend reduces it. The modifier is additive to the base drop estimate. CGM trend inputs correspond to standard arrow conventions: rapid fall, fall, stable, rise, rapid rise.

Evidence basis

CGM trend is a validated predictor of near-term glucose direction (Kovatchev 2006; Klonoff 2014). Combining current glucose value with trend direction is standard practice in AID algorithm design and was incorporated into exercise guidance with Riddell et al. 2017.

What the model does

The Explorer outputs an estimated glucose drop at 10, 20, and 30 minutes of moderate aerobic exercise. The base drop rate scales with IOB exposure band and is directionally adjusted by activity type and glucose trend. Output is expressed as a range — not a single number — to reflect inherent uncertainty.

Evidence basis

Aerobic exercise causes measurable glucose lowering within 10–15 minutes. In studies with moderate active insulin, drops of 1–3 mmol/L over 30 minutes are commonly reported (Riddell MC et al. Lancet Diabetes Endocrinol 2017; Tsalikian E et al. J Pediatr 2005;147:528–534; Riddell MC et al. Diabetes Care 2023;46(4):698–709). The model uses mid-range estimates; individual variation around these estimates is large.

What the model does

If starting glucose is below 5.0 mmol/L, the tool flags a safety warning based on published guideline thresholds. Above 14.0 mmol/L, the ISPAD/EASD consensus recommends a ketone check before exercise. The tool does not block the user — it displays the guideline context clearly without suppressing the rest of the output.

Evidence basis

The 5.0 mmol/L lower threshold before aerobic exercise is from Riddell et al. Lancet Diabetes Endocrinol 2017 (EASD/ISPAD consensus) — the most widely cited lower safety threshold in current clinical guidelines. The upper 14.0 mmol/L threshold for ketone check is from Adolfsson et al. ISPAD 2022 guidelines. Both are high-confidence, guideline-consistent values.

What the model does

Up to eight bolus doses can be entered, each with its own units and time since administration. Total IOB is the sum of residual insulin from each dose. Stacking can produce high cumulative IOB even when individual boluses appear modest — for example, a meal bolus from 90 minutes ago plus a correction from 30 minutes ago.

Evidence basis

Insulin stacking is a well-recognised cause of unexplained hypoglycaemia in T1D, particularly pre-exercise. Validated by ADA/ISPAD bolus calculator guidelines. The individual dose calculations are each pharmacokinetically sound; the summation introduces no additional modelling uncertainty.