GH-METHODS

Math-Physical Medicine

NO. 050

Comparison of two postprandial plasma glucose measurements including finger-piercing and continuous sensor monitoring (GH-Method: Math-Physical Medicine)

Corresponding Author: Gerald C. Hsu, eclaireMD Foundation, USA.

Introduction
This paper discusses postprandial plasma glucose (PPG) measurement results from two different methods, finger piercing and testing strip (Finger) and Libre’s continuous glucose monitoring device (Sensor).

Method
The author has been collecting a total of 5,475 fasting plasma glucose (FPG) data by Finger in the early mornings from 1/1/2014 to 12/31/2018 (1,825 days) on a daily basis.

Recently, he has further collected 8,676 PPG data by applying a sensor on his upper arm to collect his glucose values continuously.  This sensor measurement is conducted in parallel with his routine finger measurement.  During the period of 5/5/2018 to 12/31/2018 (241 days), he has recorded his sensor PPG values 12 times per meal, totaling 36 times per day.  He has collected 723 PPG waveforms.

Results
These 723 PPG waveforms have different shapes which are formed via diet, exercise, weather, etc. A standard FPG waveform can be clearly observed from 12 data points per meal. In comparison, PPG waves are more aggressive than FPG waves, reflecting like tsunami waves caused by an earthquake (Figure 1).

Figure 1: Waveform Comparison between synthesized FPG and synthesized PPG

Using time-series analysis of predicted PPG curve (based on GH-Method: math-physical medicine approach), finger-piercing and test strip (Finger), and continuous glucose monitor device (Sensor), the results in Figure 2 reveal the following observed conclusions:

(1) The predicted PPG has 99% accuracy on predicting Finger PPG values, and their correlation coefficient (R) is quite high 67% (Figures  3).

(2) The predicted PPG has 87% accuracy on predicting Sensor PPG values, but their correlation coefficient (R) is moderately high 54% (Figures 3).

(3) Similar analyses between PPG and diet/exercise using Finger data.

Figure 2: Comparison PPG values among Predicted Finger, Measured Finger, & Measured Sensor
Figure 3: Prediction Accuracy and Correlation (R) between Predicted PPG and measured PPG values (both Finger and Sensor)

The analyses using Sensor data are (Figure 4 and Figure 5):

– Quite high correlation (52%) between Sensor Peak PPG vs. Carbs/Sugar intake
– Moderately high correlation (40%) between Sensor average PPG value of (Maximum minus Minimum) vs. Post-Meal Walking Steps (Exercise)

Figure 4: High correlation (52%) between Sensor Peak PPG and Carbs/Sugar Intake
Figure 5: Moderate High correlation (40%) between Sensor PPG Value if Max minus Min and Post-Meal Walking Steps (Exercise)

Conclusion
When investigating Sensor data, both prediction accuracy and diet/exercise correlation are high enough as expected by the author, but they are not as strong as Finger data.  The reason is that current prediction model was developed and fine-tuned based on the popular and pseudo-standard Finger-piercing method.  Nevertheless, in comparison with the author’s previous PPG findings, most of the conclusions regarding PPG basic characteristics are still valid.

The sensor method is relatively new for type 2 diabetes (T2D) patients.  There is limited information available regarding the difference between blood samples taken from finger vs. upper arm.  Besides, both Finger strip and arm Sensor device have a relatively high defect rate from the points of view of both mathematician and professional engineer.

The author’s present glucose prediction model is developed upon current mainstream glucose measurement device, i.e. finger-piercing and testing strip.  If and when some sort of non-invasive continuous monitoring sensor device becomes the standard for the majority of T2D patients, then another set of complete, rigorous data analysis, and model tuning will be required in order to provide a practical yet highly accurate mathematical model for glucose prediction.