Everyday Difficulties

[Best viewed on a widescreen]

Below are examples of difficulties arising from everyday situations and how RADD eventually deals with them.

Unfortunately there is a great amount of difficult situations in a life with T1D, which are impossible to completely depict here. These situations were selected for the diversity of their progression, not for what caused them.

1. Sensor Noise - Fake Highs and Lows

The Dexcom G6 sensor has this notorious feature where it suddenly jumps up or down by between 1-2 mmol/L.

In the example below you can observe 3 jumps (up, down and down) in an hour starting at the 01:00 mark.

If you were to follow the BGC readings from left to right and block out the 'future' part, you can see how difficult it is to interpret these readings.

RADD does not overreact to these sudden jumps and interim highs, yet it delivers a fair amount of insulin despite the noisy data, ensuring BGC doesn't deviate much higher than target levels.

2. Late dinner on a high-day

The CGM sensor had failed earlier during the day [at school] a manual BG reading following a meal was relatively low, so insulin was administered manually and apparently inadequately. When the CGM sensor was finished with the 2 hours warm-up, the BGC was already over 13mmol/L and kept rising.

Even though the BGC keeps rising and reaches almost 20 mmol/L, RADD keeps it cool and does not go over the board with insulin administration. Evidently this was the right decision, as dinner was at 18:00. Had it not been for the carb rich dinner, the BGC would have sunken to significantly low levels.

You can see the post-prandial effect of dinner until 21:45, a significantly slow absorbed meal.

Due to the unstable insulin/glucose response, RADD adopts a 'careful' approach and leads BGC slowly to target, yet without allowing BGC to rise after the meal-high.

3. Infusion site going bad

This is a demonstration of how RADD detects that the infusion site has gone bad.

BGC seems to be stuck at about 10mmol/L and RADD keeps infusing roughly the same amount of insulin.

That is until 00:17, where RADD boluses 2.15U 'out of the blue' which breaks the cycle.

This is the result of the optimization process kicking in with a suggestion, so it can subsequently analyze the effects to determine the level of deterioration at the infusion site.

Back to samples index

Technical Details
  • Single infusion site (back of the arm)
  • Unfiltered CGM sensor data, mostly noisy during highs.
  • No additional insulin injection.
  • No glucagon infusion.
  • Pump programmed with 0.8U/h basal schedule over 24h at all times.
RADD Setup:
  • Target: 5.5 mmol/L (100 mg/dL)
  • Low target: 4.4 mmol/L (80 mg/dL)
  • Extreme low: 3.3 mmol/L (60 mg/dL)
  • Targeting mode: Balanced
  • Analytic data going back ~2 years
Hardware Setup:
  • Omnipod Eros with Fiasp
  • Dexcom G6 Sensor & Transmitter
  • OmniCore Oyster Prototype for reading CGM data
  • OmniCore Oyster Prototype for expedited pod comms