Predictive Modeling in GlucoGlance
GlucoGlance Predictive Modeling looks beyond your current glucose value.
Instead of only asking “What is my glucose now?”, Predictive Modeling looks at glucose movement, active insulin, carbohydrates, therapy settings, and available physiological information to help evaluate where your glucose may be heading and whether the conditions affecting your glucose are changing.
The original Predictive Metabolic Variance Model was designed around a simple observation: insulin does not necessarily have the same effect on the body throughout every hour of every day. Physical activity, physiological stress, temperature-related changes, and other metabolic conditions can change how the body responds.
When wearable health data is available, GlucoGlance can use information such as heart rate, HRV, SpO₂, activity and exercise, and skin temperature as additional inputs for evaluating those changing conditions.
Predictive Modeling brings those concepts together with GlucoGlance’s existing glucose, insulin, carbohydrate, and therapy data to provide information before a developing condition becomes obvious from the glucose value alone.
Three Predictive Notifications
GlucoGlance currently provides three types of Predictive Notifications. Each can be enabled or disabled independently.
Predictive Excess Insulin
Predictive Excess Insulin looks ahead for a situation where the amount of active insulin may become greater than the amount needed for the projected glucose condition.
Rather than waiting until glucose has already reached a low value, GlucoGlance can provide:
Where glucose is projected to go
Approximately when it may get there
A calculated carbohydrate amount to offset the projected excess insulin
For example:
Glucose projected at 62 mg/dL in 35 min. Consume 14g carbs.
The important distinction is that this is predictive. A current glucose reading does not have to be low for GlucoGlance to recognize that the combination of glucose movement and active insulin may be developing into a future low.
Predictive Excess Carbs
Predictive Excess Carbs looks in the opposite direction.
GlucoGlance evaluates whether the current glucose trajectory indicates that glucose may continue rising beyond the desired range.
When that condition is identified, the notification can provide:
The projected glucose peak
Approximately when that peak is expected
The calculated insulin offset associated with the projection
For example:
Glucose projected to peak at 215 mg/dL in 50 min. Calculated offset: 1.2 U.
Again, the objective is not simply to report that glucose is already high. The model is attempting to identify the developing condition while glucose is still moving toward it.
Rapid Cycle Shift
Rapid Cycle Shift addresses a different problem.
Sometimes the assumptions that were reasonable earlier are no longer reasonable now.
A period of increased insulin resistance can be followed by a relatively rapid return toward greater insulin sensitivity. When that happens, insulin already on board can become more significant than it appeared to be when it was originally delivered.
Rapid Cycle Shift is intended to recognize that change.
A notification can therefore identify a sensitivity shift and place the existing IOB into that new context.
For example:
Sensitivity Shift Active: Insulin now 1.1x more potent. Consider 3g carbs to prevent low.
The sensitivity multiplier and resulting carb amount are calculated from your own configured settings and current insulin on board, not a fixed universal number.
This is fundamentally different from a conventional glucose threshold alert. The concern is not simply the glucose value. It is that the metabolic conditions surrounding the glucose and active insulin have changed.
Advanced Model Settings
Not every person responds equally to the physiological factors used by the model.
For that reason, GlucoGlance provides three sensitivity controls under Predictive Model Settings > Advanced.
Exertion Sensitivity (Alpha)
Alpha controls the model’s sensitivity to physical exertion and activity.
Exercise and activity can substantially alter glucose behavior and insulin response. The Alpha setting allows the user to adjust how strongly exertion-related information influences Predictive Modeling.
Stress Sensitivity (Beta)
Beta controls sensitivity to physiological stress.
Available wearable information such as heart rate, HRV, and SpO₂ can provide additional context about physiological conditions occurring alongside glucose changes.
The Beta setting controls how strongly this information influences the model.
Thermal Sensitivity (Gamma)
Gamma controls sensitivity to temperature-related physiological changes.
Skin-temperature information can provide another piece of metabolic context and may help identify conditions associated with illness, inflammation, or other thermal changes.
Unlike Alpha and Beta, Thermal Sensitivity can also be turned Off when the user does not want thermal information included. The implemented Advanced settings provide independent controls for Alpha, Beta, and Gamma.
Cycle Modeling
GlucoGlance also includes an optional Cycle component within Predictive Model Settings.
Hormonal changes across the menstrual cycle can be associated with changing insulin sensitivity. Rather than treating insulin sensitivity as permanently fixed, cycle information provides another source of context for identifying periods of increased resistance and subsequent sensitivity changes.
The Cycle settings allow the user to enable Rapid Cycle Shift, record when a period begins, maintain the last recorded period, provide an average cycle length, and adjust two sensitivity values: how much stronger insulin acts at period onset, and how much weaker it acts during the luteal phase. Both default to typical starting points and can be tuned to match individual patterns.
Once sufficient cycle information exists, GlucoGlance can use that information as another component of its predictive model.
This is particularly important to Rapid Cycle Shift because the clinically interesting event is not simply that resistance exists. It is the transition from one metabolic state toward another while active insulin may still be present.
Prediction Is Different From a Threshold Alert
Traditional glucose alerts are generally reactive:
Glucose crossed a value. Alert the user.
Predictive Modeling is intended to answer a different question:
Given what is happening now, what condition appears to be developing?
That distinction is why GlucoGlance combines multiple pieces of information rather than relying on the current glucose number alone.
A person can therefore be within their normal glucose range while the model identifies a developing excess-insulin condition, excess-carbohydrate condition, or sensitivity shift.
The Model Is Personalized
Predictive Modeling uses the user’s own GlucoGlance configuration and available data.
The Alpha, Beta, and Gamma controls allow the physiological components of the model to be adjusted rather than assuming that exertion, stress, and temperature have identical effects on everyone.
Likewise, Predictive Notifications are based on the glucose, insulin, carbohydrate, and therapy information available to GlucoGlance.
As a result, the model is not simply applying a generic glucose curve to every user.
Your Data Remains Local
The wearable health information used by Predictive Metabolic Variance Modeling is processed and stored locally on the device.
GlucoGlance does not upload this physiological information to a cloud server or share it with third parties as part of Predictive Modeling.
From Modeling to Actionable Information
The purpose of Predictive Modeling is not to produce another collection of numbers.
It is to take information GlucoGlance already has, combine it with changing metabolic and physiological conditions when available, and turn that information into something understandable:
Where does glucose appear to be heading?
When might it get there?
Is there excess insulin or excess carbohydrate contributing to that projection?
Has insulin sensitivity changed enough that the existing IOB needs to be viewed differently?
Those questions are what connect the Predictive Metabolic Variance Model to the three Predictive Notifications now available in GlucoGlance.
Want the Technical Details?
For the mathematical concepts and original technical foundation behind this feature, see the Predictive Metabolic Variance Modeling White Paper.