GH-METHODS

Math-Physical Medicine

NO. 006

Using GH-Method: math-physical medicine to predict postprandial plasma glucose

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

Introduction
The author spent 8.5 years and 23,000 hours to research his diabetes conditions. Using GH-Method: math-physical medicine approach, he developed a postprandial plasma glucose (PPG) prediction model to evaluate and improve his type 2 diabetes.

Methods
The author examined correlations between PPG and three known major factors, carbs/sugar intake, post-meal exercise, ambient temperature (weather) and other secondary factors, including stress/tension, measurement time delay, traveling, illness, sleep disturbance, etc. A total of 19 influential factors were identified and analyzed and over 1 million data collected.

He applied optical physics at the front-end and wave theory at the back-end to develop an AI-based PPG prediction model and tool to assist him with controlling his PPG.

Results
During 1,309 days (from 6/1/2015 to 12/31/2018), he had 3,927 meals and collected ~80,000 PPG-related data.

The conclusions were:

  • ¬†Average PPG: 119 mg/dL (Figure 1);
Figure 1: Predicted vs. Measured PPG (6/1/2015-12/31/2018)
  • +48% correlation between PPG and carbs and sugar intake (15.0 gram and 38% contribution rate);
  • -66% correlation between PPG and post-meal walking (4,300 steps and 41% contribution rate), see Figure 2;
Figure 2: Relationship between PPG and carbs/sugar, exercise
  • Ambient temperature (weather) contributes ~10% of PPF formation;
  • Collectively, the remaining 16 secondary influential factors account for ~10% of PPG formation.

Those PPG values were analyzed using a cloud-stored food database containing ~8 million data. He collected and re-processed 6 million data from the Department of Agriculture, US government (USDA), 1.6 million data of 500 franchised restaurants, his own collected 4,354 meal phots with ~0.5 million data.

Conclusion
The predicted PPG (119.13 mg/dL) vs. measured PPG (119.05 mg/dL) has a 99.3% linear accuracy and 82% correlation coefficient.

Based on his developed two glucose prediction tools, the author was able to reduce his PPG from 380 mg/dL to 116 mg/dL, daily glucose from 279 mg/dL to 116 mg/dL, and A1C from 10% to 6.5%.