The CGM error model

The CGM error model is a key component of the ReplayBG framework, and it is designed to simulate the errors associated with these devices.

Model abstract class

The CGM error model is implemented as an abstract class CGM, which defines the interface for all CGM error models. This class can be extended by specific CGM error model implementations.

The constructor of the abstract class defines the following attributes:

• ts: The sample time of the cgm sensor in minutes (default: 5).
• max_lifetime: The maximum lifetime of the CGM sensor in minutes (default: 1440).

The abstract class defines the following methods:

• connect_new_cgm(connected_at: int=0): Connects a new CGM device, setting the time at which it is connected. Used internally to reset state across multiple replays.
• add_offset(to_add: float): Adjusts the model's internal offset. Rarely needs overriding.
• measure(ig: float, past_ig: list[float], t: float): Function that provides a CGM measure from the past interstitial glucose values. It takes as input the interstitial glucose values ig, all the past interstitial glucose values past_ig, and the relative time t in days since the startup of the sensor.

Default CGM error model

The default CGM error model available in ReplayBG is the Vettoretti19CGM model as published in Vettoretti et al., Sensors, 2019, which is a factory-calibrated device with a lifespan of 10 days and a sample time of 5 minutes.

Model equations are:

$\{\begin{array}{l}I{G}_S(t)=(a_0+a_1\cdot t+a_2\cdot t^2)\cdot IG(t)+b_0\\ CGM(t)=I{G}_S(t)+v(t)\end{array}$

with $a_0$, $a_1$, $a_2$, and $b_0$ (min${}^{-1}$) model coefficients; and $v(t)\sim N(0,{ϵ}_v)$ random white noise. Particularly, in order to mimic real CGM systems, $CGM(t)$ is saturated between 40 and 400 mg/dL.

How to choose the CGM error model to use during replays

The replay method of the ReplayBG class accepts a sensor_cgm parameter, which can be set to the class of the CGM to be used.

import MyCustomCGM
results = rbg.replay(..., sensor_cgm=MyCustomCGM)

If no sensor_cgm is provided, the default CGM error model (Vettoretti19CGM) is used.

Example of use

This example shows how to use a custom CGM error model in a ReplayBG simulation. The example uses a fake CGM model that doubles the interstitial glucose values for demonstration purposes. The sensor has a maximum lifetime of 10 minutes.

The full example is available in the example/code/ folder in the fake_CGM.py and replay_map_fakecgm.py files.

class FakeCGM(CGM):
    def __init__(self):
        super().__init__()
        self.max_lifetime = 10

    def measure(self, ig: float, past_ig: list[float], t):
        return ig * 2

    def connect_new_cgm(self, connected_at=0):
        super().connect_new_cgm()
        print(f"New CGM sensor connected at {connected_at}")
        
def main():
  ...
  # Replay the twin with the same input data used for twinning
  replay_results = rbg.replay(..., sensor_cgm=FakeCGM)
  ...