GH-METHODS

Math-Physical Medicine

NO. 014

Using Math-Physical Medicine and Artificial Intelligence Technology to Manage Lifestyle and Control Metabolic Conditions for Wellbeing and Longevity

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

Background and Aim
The author received an honorary PhD in mathematics and majored in engineering at MIT.  He attended different universities over 17 years and studied 7 academic disciplines including mathematics, engineering, computer science, and business administration.  He has also worked in various industries including defense, nuclear power, computer-aided-design, computer hardware, software engineering, and semiconductor design.

He developed Type-2 diabetes (T2D) in 1997 and by 2010 his diabetes and its complications became very serious.  Although he never received formal training in medicine, in order to save his life, he launched his own study and research on T2D.  First, he studied six metabolic diseases and food nutrition during 2010 – 2013, then conducted research during 2014 – 2018.  Thus far, he has spent 20,000 hours on his research and collected, processed, and analyzed ~1.5 million data to examine the relationship between metabolic conditions and lifestyle details.  He did not conduct his research using the traditional “bio-chemical” approach because he had no academic training in biology and chemistry, so instead he used a “math-physical medicine” approach which was based on mathematics, physics, engineering modeling, signal processing, computer science, big data analytics, statistics, machine learning, and artificial intelligence.  This approach could provide quantitative data proof and precise interpretation of certain biomedical phenomena.  His main focus was on preventive medicine for chronic disease control.  During the period of 2015 – 2017, he developed six prediction tools that he used to predict Metabolism, Weight, FPG, PPG, Adjusted Daily Glucose, and Estimated A1C.  He believed that the better the prediction, the more control one would have over chronic disease.

Glucose testing is invasive, troublesome, and costly.  Most T2D patients are not performing the measurement on a regular basis.  There is also an argument on the accuracy of glucose testing methods via either lab-tested A1C or finger piercing and testing strips.  Regardless of this argument, he has collected a complete set of FPG and PPG data using both lab-tested A1C and 9,328 finger prick testing strips.  The author spent 7.5 years researching and developing an effective way to help himself and other patients with diabetes control by predicting their glucose values, both FPG and PPG, accurately, easily, and instantly based on a math-physical medicine approach and artificial intelligence technology.

This particular paper was prepared to tell his story about how he developed his metabolism model and glucose prediction tools to manage his lifestyle, control his metabolic conditions, and lower his risk probability of having a heart attack or stroke via math-physical medicine and artificial intelligence.  Hopefully, through his knowledge, tools, and effort, he can improve his life wellbeing and achieve the goal of longevity.

Methods and Materials
1 – Data
All data was collected in its entirety from one patient only, himself, via a customized software over 7.5 years since 2012.  His extensive education and work experience have provided practical insight on how important it is to collect and categorize “clean data” from the beginning.  Otherwise, for many data analysis projects, research scientists spend 70% to 80% of their time and resources to clean up “dirty or contaminated data” before launching their real research work, which includes data process, analysis, and interpretation.  As a result, in 2010 he started his project by developing a software program using his invented “Software Robot”, and by using this  program the author was able to collect and process more than 95% of his data as “clean data” and needed very little data cleaning and organizing later on.   This project does not need to be concerned with “data interference” and “data contamination” problems due to different sources of genetic conditions, various lifestyles, and contradicting data source validations  and interpretations.  These data come from a consistent sample source, making it much easier for the author to dive into one variable and extract the buried information.

The author learned an important work ethic from Professor Norman Jones of MIT in the early 1970s about data integrity.  In this study, he used his measured data as the base for future data comparison and research.  He has safeguarded the integrity of his collected data and has never altered its original content or influenced its integrity.  All results from using his developed prediction tools are compared against the measured glucose and A1C values and plotted into data curves.

2 – Metabolism Model
Due to his mathematics and engineering background, the author views these data curves related to biomedical conditions and lifestyle management as a collection of various nonlinear input and output signal waves of the human body.  At first, he applied the “Finite Element” concept of structural engineering modeling to convert this “analog” human system into a “digitized” mathematical system in order to get an approximate solution of a real human system.  He spent the entire year of 2014 developing a mathematical governing equation of metabolism modeling which included lifestyle input and metabolic output. This equation contained 10 categories – 6 input and 4 output.  The input were 6 lifestyle categories of food, water, exercise, sleep, stress, and life pattern regularity.  The output were 4 metabolic categories of weight, glucose, blood pressure, and lipid.  In addition to food and exercise, he also investigated the impact his traveling patterns, water drinking, bowel movement, stress / tension / anxiety, daily life routine pattern disturbance, and psychological effect on physiology had on his body health and glucose.  For example, the stress category contains 34 elements for  people with “normal psychological profile: mainly from interpersonal relationships” or  “abnormal psychological profile: mainly self-induced from flashbacks”.  Overall, these 10 categories contain ~500 elements and ~1.5 million data over 7.5 years.  With such a big volume of data, a computer software program is necessary for handling the data collection and processing.

He also defined two new terms known as the Metabolism Index (MI) and General Health Status Unit (GHSU). MI is the total score reflecting body health condition (i.e. state of metabolism) which combines all of 10 categories.  GHSU is a moving average value of the past 90-days daily MI scores.  The graph of this data can be seen as a person’s “health state”.  The break-even line between a “healthy state” and an “unhealthy state” is 73.5%.  A value above this percentage is regarded as unhealthy and a value below is healthy.

Figure 2:Metabolism Index (MI) and General Health Status Unit (GHSU) from 2012 to 2018

3 – Glucose Prediction
The author started with a simple task of predicting tomorrow’s weight output from the previous 3-days weight, food quantity input, and bowel movement.  The weight prediction is the pre-processor for predicting FPG in the morning which leads into A1C estimation.  Although there are five influential factors for FPG creation, he discovered and proved that weight is the predominant one.

The prediction of PPG, however, is a much more complicated task since it involves about 15 influential factors that create PPG value.  He applied signal processing technology from geophysics, electronic and communication engineering to decompose the human body’s highly nonlinear biomedical signal curves, such as the glucose wave, into different sub-waves created by each influential factor.  He carefully checked each sub-signal waveform for its completeness, accuracy, and correlation with other curves, using time-series analysis, spatial analysis, and frequency-domain analysis (via Fourier Transform), etc.  Finally, he recombined them back to a predicted glucose curve to simulate the real measured one.  By developing and analyzing many mathematical models, he was able to identify primary, secondary, and tertiary factors according to their respective contribution margins and importance levels on glucose creation.

Over the past three years, he continuously explored and added some missing influential factors into the formation of the PPG signal.  His purpose was trying to improve the predicted PPG waveform’s contents and accuracy while maintaining high correlation with the measured PPG waveform.

For example, by the fall of 2016, the accuracy of his predicted PPG reached ~95%.  But, in September of 2017, he identified that weather temperature also had an impact on glucose value.  Therefore, he selected a 2-year period (6/2015 – 7/2017) to examine his past travel schedule in detail.  During the past 7 years, the author traveled via air, an average of every 13 days and entered each day’s local ambient temperature of the city where he stayed.  In this way, he was able to generate a new temperature sub-wave which brought the accuracy of the predicted PPG from ~95% to ~98%.

Another factor was that his glucose was quite high when he was sick with the flu for a month at the end of 2017.  After that experience, he further enhanced his prediction model with the inclusion of “physical sickness or wellbeing” which brought the prediction accuracy to 99.8%.

After analyzing each sub-wave in detail, he was ready to reintegrate these sub-waveforms into another nonlinear predicted PPG waveform.

He further improved his model via a “curve-fitting” trial-and-error engineering method which he learned from his defense work experience.  He has continuously compared these two sets of data and improved the accuracy until it reached a very high linear accuracy while still maintaining a high correlation.  High correlation means the trend of predicted curve moves along with the measured curve like its “twin”.

Figure 4: Decomposition of 4 Sub-Waveforms of PPG

4 – A1C Prediction
For A1C estimation, he utilized all of his historical test data to determine a “customized” glucose-to-A1C conversion ratio.  He also utilized statistical algorithms to automatically modify it when new test data was available.  Finally, he specifically added in a “safety margin” which he learned from his nuclear power work experience.  He inserted a +15% margin on top of his originally predicted A1C value for the purpose of providing a numerical safety buffer.   This predicted A1C value can serve as a daily “early warning” to T2D patients before they have a chance to get their A1C tested.  The “adjusted A1C” is defined as a combination of both FPG and PPG with their respective weighting factor to create the “estimated A1C”.  Both the Adjusted Glucose and Estimated A1C models also utilized another layer of “self-adjusting” machine-learning algorithms in order to correct or compensate for the built-in “error” from chemical process of various lab tests and glucometers.

Figure 9: Estimated Daily A1C Curve (with 15% safety margin) and Lab-Tested A1C Data Since 2010

5 – Environmental Harmony
The author has learned much about the importance of keeping a harmonious relationship with his environment from his self-study and medical research work.  He realized that food, the most important input factor of T2D control, is a double-edged sword with regards to the human body in the same way that glucose is a double-edged sword to our internal organs.  Both food and glucose are essential elements of life; however, too much of each can cause severe risks to our health.  Therefore, the author has totally changed his diet by reducing the amount of red meat and greatly increasing the amount of vegetables eaten.  He also avoids consuming factory processed, chemically altered foods.  He has learned how much damage to his body can occur by eating processed foods containing excessive amounts of sugar, salt, fat, artificial coloring, chemical additives, or hormones and antibiotics from modern farming and processing methods.  He tries to eat natural, organically produced foods.  Each day he also drinks 6 bottles (3,000 cc) of water, instead of other beverages.

Human beings have evolved over millions of years and many traditional lifestyle patterns have been shaped through adaptation, change, and evolution to be in harmony with our environment.  Traditional society’s food intake and exercise were more in balance with our body’s healthy metabolic process.  Unfortunately, modern society’s lifestyle often pushes people out of harmony.  This harmony is not only important for elements, such as “food patterns”, but also holds true for other elements, such as “exercise patterns”.  We need regular and routine exercise.  The author’s daily life is no longer sedentary and defined by sitting.  He walks about 3 hours which is about 8 miles each day.  He chooses to walk in comfortable places such as along beaches, in shopping malls, in large stores, and city parks so that he can walk, think and read during his exercise.

In addition, he keeps his stress level low.  He tries to remain calm and in control of his emotions even in severely stressful situations.  He also maintains good sleep habits by making sure he gets a minimum of 6 to 7 hours sleep each night.  In addition, he tries his best to match his sleep cycle with the natural diurnal rhythm of day and night.

Through his research, he understands the influence of weather and ambient temperature on his glucose.  He has been fortunate to be able to live in locations where the temperature has minimal impact on his T2D.

He has finally realized the importance and truth of the old saying, “Treat your body like a temple.”  He tries to avoid bad habits and ingesting any chemical or foreign substances.  He understands and respects the importance of modern medicine, especially with severe medical conditions and in emergency situations.  However, in dealing with some chronic diseases, he avoids taking certain medications which cannot cure them but only suppresses disease symptoms.  The author recognizes the importance of living in harmony with environmental rhythms and life cycles.

From his research and analysis of his decomposition of the PPG signal waveform, he observed clearly the intensity and impact different influential factors had on each sub-waveform.  Mathematically, it can be seen that some external intruding “shock waves”, such as sickness and stress, create significant unbalanced force on the natural bio-rhythms of our body, causing disruption of our glucose metabolism.  This is what he has learned from his own 8-years of math-physical medicine research about “keeping an environmental harmony”.

nce on PPG.  On the other hand, Weight is the primary factor of FPG.  Weight is directly proportional to the total “quantity” of food consumption while PPG is directly related to food “quality”, specifically the intake amount of carbs and sugar.  Of course, a person who eats a large quantity of food will likely take in more carbs and sugar.  However, a knowledgeable and well-disciplined T2D patient can control both quantity and quality of food.  The above conclusion should be re-verified for light-weight and obese patients.  Nevertheless, a strict weight reduction will be a very effective way for obese patients to put their glucose (both FPG and PPG) under control.

Figure 6: Impact of on PPG and What-If Analysis

6 – Risk Probability of Heart Attack and Stroke
In 2014, he researched and built a metabolism model (MI & GHSU) to measure the multiple interactions of four metabolic disease outputs and six lifestyle inputs.  In his research, he noticed the close relationship between chronic diseases and heart attack /stroke, and the high percentage of death caused by heart attacks and strokes.  Therefore, he extended his math-physical medicine research to cover the risk probability of cardiovascular diseases and stroke.

Initially, he chose age, gender, race, family history, smoking, drinking, substance abuse, personal medical history, and waistline to establish a “static” baseline. He then applied the hemodynamics concept to develop a “dynamic” macro-simulated model for blood blockage and artery rupture.

He utilized 368,513 data which included 72,893 metabolic conditions (obesity, diabetes, hypertension, hyperlipidemia) and 295,620 lifestyle conditions (food, exercise, water, sleep, stress, daily life routine) within 2,274 days (1/2012 – 3/2018) to  separately compute three different sets of risk probabilities.  Finally, he integrated them into one overall risk probability.  He also conducted data sensitivity analyses to cover the probability variance by using a wide range of different weighting factors.

Results
1. Metabolism
The results showed in Figure 2: MI & GHSU that he was very unhealthy (MI and GHSU score of 80% – 110%) before 2013.  The curve went through a sharp decline in 2014 due to the knowledge he learned from his research.  After 2015, he was “healthy” (MI and GHSU score of 60% – 70%).  As of 5/19/2018, his MI is 52.7% and GHSU is 55.6% due to his disciplined lifestyle management.  All of his current health examination results also confirmed the fact that his chronic disease conditions are well under control.  In 2000, he could not climb more than five steps in a flight of stairs; however, in 2017, he climbed 520 steps (~33 stories) without stopping.  He completed his first 5K marathon at the end of 2017 in Abu Dhabi and finished a 10K marathon in the spring of 2018 in Silicon Valley.

It should be mention here that, he recognizes one of key factors for longevity is keeping a regular and healthy life patterns; therefore, his 10th category of metabolism is “Life pattern regularity”.  The detailed 14 input data list can be seen in Figure 14: Element List of Daily Life Pattern Regularity of Metabolism.

Figure 14: Element List of Daily Life Pattern Regularity of Metabolism

2. Glucose – FPG
In 2015 – 2016, he spent ten months investigating FPG.  Initially, he exhausted all avenues to find possible connecting factors, including a very low correlation of ~9% between FPG and PPG.  Although his 50-years of engineering training taught him to always look for relationships between input and output, now he must think “out-of-box” to seek for a suitable solution.  In the early morning of 3/17/2016, he had a dream about searching for the relationship among different body output categories.  He then discovered that there was a high correlation of 84% between FPG and Weight.  In the attached Figure 3: FPG and Weight Relationship, he used ~26,000 FPG-related data from 1,436 days, (4/1/2014 – 3/7/2018), to conduct statistical analyses.  In the time-series diagram, there are 6 high periods and 6 low periods of Weight, and the FPG curve followed the Weight curve like its “twin”.  In the spatial analysis diagram of BMI vs. FPG (without time factor, in Figure 12), there is a “quasi-linear” equation existing between two coordinates of BMI and FPG from point A (24.5, 102) to point B (27.0, 142).  The stochastic (random) distribution of data has two clear “concentration bands” stretched from lower left corner toward upper right corner.  The +/- 10% band covers 67% of the total data and the +/- 20% band covers 94% of the total data.  Only the remaining 6% of the total data is influenced by other secondary factors.

The predicted FPG vs. measured FPG achieved a linear accuracy of 99.8% (118.42 mg/dL vs. 118.62 mg/dL) and 98.6% correlation.

Figure 3: FPG and Weight Relationship
(time-series analysis & spatial analysis)
Figure 12: Weight Reduction vs. Constant PPG and weight Change vs. FPG Change from 2012 to 2018 (annually accumulated glucose data)

3. Glucose PPG
The author has collected a complete set of PPG data including his lifestyle detailed data during a period of 1,075 days with 3,225 meals (6/1/2015 – 5/11/2018).  This PPG-related data set, size of ~400,000 data, is only a small portion of his entire ~1.5 million data.

As shown in the attached Figure 5: PPG and its Influential Factors, his average PPG values are:

  1. Predicted PPG: 119.82 mg/dL
  2. Measured PPG: 119.98 mg/dL

with 99.8% linear accuracy and a high correlation of 84%.

Figure 5: Predicted vs. Measured PPG and Correlation Between Influential Factors and PPG

It should be noted that an overlapping period of 953 days (10/1/2015 – 5/11/2018) was used for calculating the 90-days moving average for easy viewing of the PPG trend (similar to the concept of “dynamic daily A1C”).  The first 90 -120 days data were not included in the calculation due to the consideration of data stability.

The daily PPG values contributed amount by each key influential factor and individual contribution margins are listed as follows:

  1. Carbs/Sugar: +14.5 mg/dL, 38%
  2. Post-meal walking: -15.8 mg/dL, 41%
  3. Temperature: +3.7 mg/dL, 10%
  4. All others: +1.9 mg/dL, 11%
  5. Net gain on PPG: +4.3 mg/dL.

In addition, correlation coefficients between key influential factors and measured PPG (119 mg/dL) are:

Carbs/sugar intake (14.8 gram per meal): +55% (high positive value means higher carbs/sugar intake  pushes PPG higher)

Post-meal Exercise (4,200 steps per meal): -66% (high negative value means higher exercise amount brings PPG lower).

Through the continuous use of his AI software program as shown in Figure 10: AI Glucometer and Meal Photos, the author was able to track and analyze all meals using optical physics and signal processing, making meal data collection and PPG prediction much simpler.  The 3,225 meal photos were analyzed against 6 million food nutrition content data collected from the US Department of Agriculture (USDA) and stored in a cloud server.  All food data were also sorted based on countries, franchise restaurants, individually owned restaurants, home-cooked meals, airline food, etc.

Here are some summarized post-meal glucose results:

  1. Airline food PPG – 136 mg/dL
  2. Restaurant food PPG – 127 mg/dL
  3. Home cooking PPG – 111 mg/dL
Figure 10: AI Glucometer Screen Design to Predict Glucose via Meal Photos

From the attached Figure 7: PPG and Temperature Record, the temperature impact on PPG is quite obvious, especially in warmer weather >77℉. PPG value would increase 0.9 mg/dL due to temperature increase of each degree above 77℉.  This phenomenon is due to increased energy demand and metabolism creation.  It should be noted that the FPG value would decrease 0.3 mg/dL due to temperature decrease of each degree below 67℉.  This phenomenon is due to “hibernation” effect.

For an overweight patient (BMI 25 – 30), the correlation coefficient between PPG and Weight is a low 11% in time-series analysis.  In the spatial analysis diagram, Figures 8 and 12: PPG and Weight, his PPG values stay within a “constant band” regardless of his weight reduction.  These two diagrams prove that PPG is not influenced by Weight.  Also shown in the same Figure 8, the correlation coefficient between PPG and FPG is a mere 0.9% which means they are not related at all.

Figure 7: Influential Factor’s Contribution to PPG and Temperature Record
Figure 8: Low Correlation Existed Between PPG vs. FPG and PPG vs. Weight

In summary, both FPG and Weight have no relationship and influence on PPG.  On the other hand, Weight is the primary factor of FPG.  Weight is directly proportional to the total “quantity” of food consumption while PPG is directly related to food “quality”, specifically the intake amount of carbs and sugar.  Of course, a person who eats a large quantity of food will likely take in more carbs and sugar.  However, a knowledgeable and well-disciplined T2D patient can control both quantity and quality of food.  The above conclusion should be re-verified for light-weight and obese patients.  Nevertheless, a strict weight reduction will be a very effective way for obese patients to put their glucose (both FPG and PPG) under control.

4. A1C
He utilized optical physics, signal processing, big data analytics, statistics, machine learning, and AI to create prediction models for FPG and PPG, achieving >98% linear accuracy with >80% correlation between predicted and measured glucose.  He also developed an easy-to-use AI tool for T2D patients to instantly predict and control their glucose conditions.  A screen shot of this AI tool is attached in Figure 10: AI Glucometer.

Combining the knowledge gained from his research, convenience from his AI prediction tools, and persistent lifestyle maintenance efforts, he has brought his A1C value from 10.0% in 2010 to 6.5% in 2018, as shown in Figure 1: Health Data Comparison.

It is not surprising to notice that his diabetes is under control, and at the same time, his other two chronic conditions, hypertension and hyperlipidemia, are also no longer health concerns.

Figure 1: Health Data Comparison Between 2010 and 2017
Figure 10: AI Glucometer Screen Design to Predict Glucose via Meal Photos

5. Risk Probability of Heart Attack or Stroke
As shown in Figure 11: Risk Probability of Heart Attack and Stroke Using 4 Models, especially in the metabolic conditions case, his risk probability of having a heart attack or stroke has dropped from 74% in 2000 (followed by three cardiac episodes from 2001 to 2006) to 62% in 2012 and finally decreased to 26.4% in 2017 (compatible with 26.7% by the Framingham Study equation)

It should be noted that his weighting factor sensitivity results are within the range of +/- 10% to +/- 18%.

Figure 11: Risk Probability of Having a Heart Attack Using 4 Models

Conclusion
As shown in Figure 13: Flow Diagram of T2D Control, the quantitative results from the developed prediction models including metabolism, Weight, FPG, PPG, A1C, reflect the accuracy and applicability for Type-2 diabetes control via a guided scientific lifestyle management.  The utilization of math-physical medicine is also proven quite effective for this investigation.  As shown in Figure 1: Health Data Comparison Between 2010 and 2017, the author’s health condition has been improved significantly due to his own efforts based on his research.

Figure 13: Flow Diagram of T2D Quantitative Control
Figure 1: Health Data Comparison Between 2010 and 2017

The author firmly believes that for chronic diseases, prevention is more important and effective than treatment; therefore, if you can predict your disease condition accurately by a scientific method, then you can control it in a correct and effective manner.

This same big data dynamic simulation approach using math-physical medicine could also provide an early warning to patients with chronic disease of having a heart attack or stroke in the future.

The author has not only saved his own life, but also wants to offer his findings, results, methodologies, and tools to other patients with chronic disease.  Hence, they can understand their environment interactions, lifestyle improvements, and disease control with an ultimate goal of improving their well-being and achieve their longevity.

Acknowledgement
First and foremost, I wish to express my sincere appreciation to a very important person in my life, Professor Norman Jones at MIT.  Not only did he give me the opportunity to study at MIT, but he also trained me extensively on how to solve problems and conduct scientific research.

I would also like to thank Professor James Andrews at the University of Iowa.  He helped and supported me tremendously when I first came to the United States.  He believed in me and prepared me to build my engineering foundation during my undergraduate and master’s degree work.

References and Other Declarations
The author created math-physical medicine himself in order to save his life.  Although he has read many medical books, journals, articles, and papers, he did not specifically utilize data or methodology from other medical references.  All of his knowledge, information, technique, and methodology about mathematics, physics, engineering, and computer science came from his lifelong learning.  He has never hired any scientific assistant or associate to help with his research work except for a part-time programmer.  He applied his own invention of a “Software Robotic” concept and methodology to produce his needed computer software for this research project.  He was self-funded, spending his own money which he earned from a successful high-tech venture in Silicon Valley.  He did not receive financial assistance or grants from any institution.