A GlucoGlance White Paper
Integrating Wearable Sensor Fusion to Resolve Non-Insulin Glucose Drivers
Executive Summary
The primary challenge in automated insulin management is the “Grey Area”, glucose fluctuations that do not correlate with Insulin On Board (IOB) or Carbohydrate On Board (COB). This paper proposes a methodology for identifying and quantifying these “ghost” drivers by establishing a mathematical Variance (Δ) field. By fusing glucose, insulin, carbohydrate, therapy, activity, cardiovascular, thermal, and available physiological data with a user-defined metabolic blueprint, GlucoGlance can identify changing metabolic conditions, quantify their effect on insulin sensitivity, and project developing glucose states.
1. The Problem: The Static Insulin Model
Standard insulin therapy relies on static parameters (CF, CR, Basal). However, exercise and physiological stress introduce variables that alter insulin sensitivity dynamically. Without real-time metabolic data, these shifts are often misidentified as insulin-delivery failures or carb-count errors, leading to “stacking” and hypoglycemia.
2. The Proposed Solution: The Variance Field (Δ)
Our algorithm moves from a “Detection” model to a “Deviation” model. By establishing a 24-hour Basal Blueprint (Theoretical Floor), the system calculates an Expected Glucose State.
The Fundamental Equation:
- Any non-zero Variance must be attributed to a physical cause.
- Negative Variance + High BPM/Cadence: Identified as Metabolic Draw (Exercise-induced insulin sensitivity).
- Positive Variance + Low BPM: Identified as Absorption Resistance (Potential site failure or occlusion).
3. Data Acquisition & Methodology
The system utilizes a “Silent Data Stream” to minimize user burden:Metabolic Tachometer: Real-time HR polling via Health Services API (Android/WearOS) to measure absolute heart rate frequency.Relative Intensity: Comparison of current BPM to Resting Heart Rate (RHR) from Health Connect, establishing an intensity coefficient without requiring age/weight data.Physical Verification: Step Cadence (Steps per Minute) verifies if heart rate spikes are mechanical (exercise) or psychological (stress/caffeine).
3.1 Multi-Factor Metabolic Modulation
The original Variance Field used heart rate and physical activity as the primary means of identifying non-insulin glucose drivers. The implemented GlucoGlance model expands this approach by evaluating three independent classes of metabolic influence:
Exertion Sensitivity (Alpha): Represents the influence of physical activity and exertion on insulin sensitivity. Heart rate, resting heart rate, steps, cadence, and exercise activity can provide evidence of increased metabolic demand.
Exertion Multiplier
Mactivity = 1 + (0.5 × Alpha) × ((HR / RHR) × (Cadence / 100))
Where Alpha is the user-selected Exertion Sensitivity. A setting of 1.0 represents Standard sensitivity, 0 disables the influence, and 2.0 doubles the configured influence.
Stress Sensitivity (Beta): Represents the influence of physiological stress on glucose behavior and insulin sensitivity. Available wearable measurements, including heart rate, HRV, SpO₂, respiratory information, and EDA where supported, provide additional context for distinguishing physiological stress from physical exertion.
Stress Multiplier
Mstress = 1 − (0.3 × Beta) × EDA
The stress multiplier is applied only when the user is at rest. When cadence indicates physical activity, the stress multiplier remains neutral at 1.0. A safety floor prevents the multiplier from falling below 0.1.
Thermal Sensitivity (Gamma): Represents the influence of temperature-related physiological conditions. Skin temperature and available ambient temperature provide additional context for identifying thermal changes that may accompany illness, inflammation, environmental exposure, or other metabolic disturbances.
Thermal Multiplier
Mtemp = 1 + (0.02 × Gamma) × (Skin Temperature − 33°C)
The model uses 33°C as its baseline skin-temperature reference. Temperatures above or below that baseline modify the projected effect according to the user’s Thermal Sensitivity setting.
Alpha, Beta, and Gamma are independently adjustable within GlucoGlance. This allows the model to reflect differences between users rather than assuming that exertion, stress, and thermal conditions affect insulin sensitivity equally in every individual.
4. Reverse-Carb Attribution
To resolve the “Food Ghost” variable without dual-entry, the algorithm employs a Reverse-Carb Assumption. When a significant IOB increase is detected without a corresponding correction-need, the system back-calculates the dose into Estimated Carbs using the user’s established Carb Ratio (CR).
4.1 Forward Projection and Predictive Classification
Reverse attribution helps explain a glucose event after sufficient evidence exists to identify its probable driver. GlucoGlance extends this concept by using the same metabolic information to project glucose forward.
The core predictive calculation is:
ΔBGpred(t) = [(COBrate(t) × (ISF / ICR)) − (IOBactive(t) × ISF × Mactivity × Mstress × Mcycle)] × Mtemp
The resulting rate is then used to project glucose forward:
BGprojected(t+T) = BGactual(t) + (ΔBGpred(t) × T)
Where:
COBrate = current carbohydrate absorption rate
IOBactive = estimated current insulin activity rate
ISF = Insulin Sensitivity Factor
ICR = Insulin to Carbohydrate Ratio
Mactivity = exertion multiplier
Mstress = physiological stress multiplier
Mcycle = hormonal cycle multiplier
Mtemp = thermal multiplier
T = prediction horizon in minutes
The current implementation assumes that the calculated rate at the time of prediction remains constant across the prediction window. This provides a practical forward projection without requiring continuous mathematical integration of changing rates.
Predictive Excess Insulin
GlucoGlance evaluates glucose 45 minutes into the future.
The trigger condition is:
BGprojected(t+45) < BGlow
When the projected glucose falls below the user’s configured low threshold, GlucoGlance calculates the carbohydrate amount required to move the projected glucose back toward the user’s target:
Carbsneeded = (BGtarget − BGprojected) × (ICR / ISF)
The resulting notification can therefore provide the projected glucose and a calculated carbohydrate amount rather than waiting for glucose to actually cross the low threshold.
Predictive Excess Carbs
GlucoGlance also evaluates glucose 60 minutes into the future.
The trigger condition is:
BGprojected(t+60) > BGhigh
When the projected glucose exceeds the user’s configured high threshold, GlucoGlance calculates the insulin offset required to move the projected glucose toward the user’s target:
Bolusrecommended = (BGprojected − BGtarget) / ISF
This allows GlucoGlance to identify a developing high-glucose condition before the high threshold has actually been reached.
5. Clinical Application: Site Efficiency Scoring
By utilizing the Pod-State Transition (Disconnected > Active) as a trigger, the system maps the Variance to specific infusion sites. Over time, this identifies physiological “Site Fatigue,” allowing clinicians to recommend more efficient rotation patterns based on quantitative absorption data rather than anecdotal reporting.
6. Hormonal Cycle Modeling and Rapid Cycle Shift
Insulin sensitivity can change throughout the menstrual cycle. These hormonal changes can produce periods of increased insulin resistance followed by a return toward greater insulin sensitivity.
GlucoGlance includes optional menstrual cycle tracking to provide this additional metabolic context. The user can enable Rapid Cycle Shift, record when a period begins, and provide an average cycle length. This establishes a cycle-relative reference that can be incorporated into Predictive Modeling.
The purpose of cycle modeling is not simply to identify the current phase of the menstrual cycle. Its primary purpose is to recognize a change in insulin sensitivity that may alter the expected effect of active insulin.
During a period of increased insulin resistance, insulin may appear to have less effect than expected. Additional insulin may therefore remain active when the metabolic state begins to change.
If resistance subsequently decreases, that existing IOB may have a greater glucose-lowering effect than it did during the more resistant state. Rapid Cycle Shift is intended to identify this transition and provide advance warning when the changing metabolic state may increase the risk of a future low.
Within the predictive equation, hormonal influence is represented by the Mcycle multiplier:
ΔBGpred(t) = [(COBrate(t) × (ISF / ICR)) − (IOBactive(t) × ISF × Mactivity × Mstress × Mcycle)] × Mtemp
The predictive calculation architecture therefore provides a specific position for hormonal-cycle influence alongside the exertion and stress multipliers.
In the implemented model, Mcycle is calculated from the user’s cycle-tracking settings rather than remaining neutral. On the day a period begins, Mcycle applies a configurable sensitivity boost, defaulting to 1.10, adjustable by the user. During the second half of the cycle, Mcycle applies a configurable resistance factor, defaulting to 0.80, also adjustable. Outside those windows, Mcycle remains neutral at 1.0.
These default multipliers are starting points, not clinical constants, since insulin sensitivity variation across the cycle differs meaningfully between individuals. Rapid Cycle Shift and the cycle-tracking interface therefore represent both the framework and the active calculation incorporating hormonal changes into Predictive Modeling.
7. From Variance Detection to Predictive Intervention
The original Predictive Metabolic Variance Model was designed to identify and explain glucose behavior that could not be accounted for by insulin and carbohydrate activity alone.
The implemented GlucoGlance model extends this concept from retrospective identification to forward prediction.
Traditional glucose alerts are primarily threshold based. They identify when glucose has already crossed a defined value. Predictive Modeling instead evaluates whether the current combination of glucose movement, active insulin, active carbohydrates, therapy parameters, and metabolic conditions indicates that a significant glucose condition is developing.
This allows GlucoGlance to identify potential excess insulin or excess carbohydrate conditions while the current glucose value may still be within the user’s configured range.
The model therefore operates across three stages:
Observe: Collect glucose, insulin, carbohydrate, therapy, and available physiological data.
Model: Evaluate how exertion, physiological stress, thermal conditions, and hormonal changes may alter the expected glucose and insulin response.
Predict: Project the developing glucose state and provide a notification when the projection meets the configured predictive conditions.
The objective is not to replace CGM alerts, insulin pump algorithms, or clinical judgment. It is to provide additional metabolic context by combining information that traditionally exists in separate systems.
Conclusion
By treating glucose management as a sensor-fusion problem rather than a simple insulin-tracking task, Predictive Metabolic Variance Modeling addresses the “Grey Area” of glycemic control: glucose behavior that cannot be adequately explained by insulin and carbohydrate activity alone.
The model began by using glucose, IOB, COB, basal requirements, heart rate, and physical activity to identify deviations from an expected metabolic state. GlucoGlance now extends that foundation through independently adjustable exertion, stress, and thermal sensitivity factors, hormonal cycle modeling, and forward glucose projection.
The result is a transition from detection to prediction.
Rather than only explaining why glucose moved after an event occurred, GlucoGlance can evaluate whether current conditions indicate developing excess insulin, excess carbohydrates, or a significant change in insulin sensitivity.
The objective remains the same as when the Variance Field was first proposed: make previously unexplained glucose behavior measurable. The implemented model takes the next step by using that measurement to identify what may happen next.