SURVIVE AND THRIVE, SMARTGUIDE
SmartGuide’s three prediction layers, a teaching resource
A one-page resource for clinicians and educators onboarding people onto the Accu-Chek SmartGuide. Each of the three prediction layers does a different job. Teach them in this order, and the wearer goes home with a usable mental model rather than three blinking icons. Designed to be printed at A4 and slipped into the SmartGuide group session pack.
The frame to teach with
None of these layers prevents an event. Each gives an earlier window to act. The three models are independent (Herrero 2024, J Diabetes Sci Technol): a 30-minute hypo classifier, a 2-hour glucose forecast curve, and a 9pm-to-2am overnight risk score. They share the CGM data; they do not share each other’s logic. Wearers who understand the time horizons act on each appropriately. Wearers who do not, end up either over-trusting the 2-hour worm or under-trusting the overnight RAG light.
Layer 1, the active safety net
Low Glucose Predict, 30-minute hypo classifier
What it does
Classifies the next 30 minutes as high-risk or not for hypoglycaemia. When high-risk, an alarm fires.
What it does not do
It does not predict the size or depth of the low; it does not catch every event (sensitivity is not 100%). It is reactive, not preventive.
When to teach it
First, before either of the other two layers. This is the active safety net the wearer will rely on while learning the system.
What the wearer sees
An alarm on the SmartGuide app, not the Predict app. The Predict app holds the longer-horizon layers.
What to do when it fires
Finger prick to confirm if uncertain, treat per usual hypo plan, plan the next 30 minutes (driving, meeting, sleep). Re-check after the response window.
Teaching pitfall
Wearers who silenced “predictive low” alerts on previous CGMs because they fired too often. The SmartGuide LGP fires meaningfully less; ask them to keep it on for two weeks before deciding.
Layer 2, action and awareness
Glucose Predict, 2-hour forecast curve
What it does
Projects a glucose trace 2 hours forward, with a 50% confidence band drawn around it. Refreshes every 5 minutes.
What it does not do
It does not see meals or insulin doses the wearer has not entered. It does not account for alcohol, exercise, or illness. A wide band means low confidence.
When to teach it
After the LGP. Frame it as “the curve you plan around for the next hour, the curve you act on for the next forty-five minutes”. The split is the teach.
What the wearer sees
The Predict app shows the trace as a coloured worm with a shaded band. Narrow band, high confidence. Wide band, low confidence.
What to do when it lights up
0 to 45 min: act on the curve direction (eat ahead of a low, defer a correction if a fall is shown). 45 min to 2 h: plan around it (meeting timing, drive, exam, run).
Teaching pitfall
Selling the 2-hour mark as reliable. It is not. Wearers who treat the 2-hour endpoint as actionable will be disappointed quickly, and the disappointment is the form factor of a sensor falling out of routine wear.
Layer 3, the night plan
Night Low Predict, overnight risk score
What it does
Issues a Red, Amber, or Green prediction of overnight low risk between 9pm and 2am, refreshable every 20 minutes. Personalises over ~28 days of input data.
What it does not do
It is a population-level confidence band, not a personal probability. It misses approximately 45% of overnight hypoglycaemia events at the published threshold; LGP remains the active safety net during the night.
When to teach it
Third. The wearer should already trust the LGP alarm and the GP curve. The night plan is what unlocks confidence to sleep.
What the wearer sees
A Red, Amber, or Green indicator before bed. Re-requestable every 20 minutes until 2am.
What to do when it lights up
Red, first half: fast carbohydrates before bed are the option most aligned with the risk (discuss with care team). Red, second half: lower model confidence, conversation about protein, slow-acting carbohydrates, or a temporary basal reduction (if on a pump). Amber: re-check in 20 minutes, watch for trend.
Teaching pitfall
Treating both halves of the night identically. The ROC AUC asymmetry (0.902 vs 0.730) means a red light at 1am and a red light at 4am are not the same signal. Land the asymmetry at onboarding.
Teaching order at a glance
Day 0, the group onboarding session. Layer 1 first, framed as the active safety net. Layer 2 second, with the 45-minute action / awareness split landed concretely. Layer 3 third, with the night-half asymmetry named.
Day 1, the calibration day. Two paired finger-prick calibrations at approximately 12 and 14 hours after insertion. The conversation about finger pricking is part of the teach, not an inconvenience to apologise for. After day one, no further finger pricks are needed.
Days 2 to 28, the personalisation period. Wearers should not draw conclusions from the first few nights. Night Low Predict needs the input data to reach its full personalised accuracy. The first month is the learning month for the system, not just for the wearer.
Day 14 onward, the end-of-wear drop. Accuracy holds reasonably (Mader 2024, 85.9% within plus or minus 20/20 on days 13 to 14 versus 92.8% on day 2). Schedule the next sensor before the current one ends.
About this resource
Produced alongside Episode 40 of the GNL Podcast with Amy Jolley (Lead Educator, Diabetes Technology Network UK; Highly Specialised Dietitian, Salford NHS Foundation Trust). Evidence audited against the GNL Grace wiki (Herrero 2024, Mader 2024). Independent of Roche Diagnostics. This content is for educational exploration only. It describes average responses and general principles. It is not medical advice and cannot replace individual clinical guidance from your diabetes care team.
SmartGuide device guide · Episode 40 show notes · All GNL resources