PPG Glucose Prediction

An 8-Year Journey — Truth from 5,000+ Finger Pricks

8
Years
5000+
Blood Samples
r=0.97
Correlation

Research Journey

Eight years pursuing the "holy grail" of non-invasive glucose monitoring. From near-infrared sensors to PPG, from glucose numbers to vascular response. A record of paradigm shift.

Phase 1 — Near-Infrared Approach
Chasing the Holy Grail
Non-invasive glucose measurement using InGaAs sensors and proprietary wavelength LEDs. Achieved MARD below 10% with error grid within clinical range. However, commercialization required billions of yen in additional investment. When Apple withdrew, supporters disappeared.
Phase 2 — Shifting the Question
"So what if we measure glucose?"
I questioned the meaning of chasing glucose numbers. So what if it reads 140mg/dL? What's wrong with glucose spikes? The answer is clear: each spike damages vessel walls, cholesterol accumulates to repair damage, eventually causing atherosclerosis. Heart attack, stroke, dementia in 10 years. Gangrene, blindness, dialysis. Today's postprandial spike is one step toward that future. What we really need to know isn't a number—it's whether our blood vessels are screaming right now.
Phase 3 — Focus on PPG
5,000+ Finger Pricks
Breaking free from mainstream research trends, I focused solely on PPG. To analyze vascular response and glucose relationships, I collected data from 5,000+ finger pricks on myself. However, with 5-minute intervals, I could only manipulate equations to barely fit rapid rises, falls, and the difficult hypoglycemic range below 70mg/dL—the essence remained hidden.
Phase 4 — High-Frequency Sampling
100 Consecutive Pricks at 2-Min Intervals
Wanting to know what happens within 5 minutes, I tried smaller time gaps. Switching to 1-2 minute intervals, 100 consecutive pricks per session revealed complex vascular mechanisms invisible at 5-minute intervals. Even for a strong constitution, the mental toll was immense. Finally, 500 short-time-gap data points were ready.

The Importance of Sampling Intervals

How much information is lost when observing the same glucose fluctuation at different sampling intervals. True fluctuation is invisible at CGM's 5-minute or OGTT's 30-minute intervals.

Original Data (~2 min)
104 data points — Vascular "turbulence" clearly observable
5-min Sampling (CGM)
40 data points — Captures peaks, loses details
15-min Sampling
17 data points — Patterns simplified
30-min Sampling (OGTT)
9 data points — Completely misses spike peaks
Sampling Data Points Duration Clinical Application
~2 min (Original) 104 pts 0-268 min Vascular turbulence observable
~5 min (CGM) 40 pts 0-265 min Captures peaks, loses details
~15 min 17 pts 0-265 min Trends only
~30 min (OGTT) 9 pts 0-251 min Misses true peaks

Key Discoveries

The relationship between vascular response and glucose fluctuation discovered from high-frequency data. An approach tracking "magnitude of change" rather than "values," transcending linear regression limits.

Turbulence Index
r = 0.97
Cumulative |Δ Vascular| vs |Δ Glucose|
PPG-Temp Crossover
r = 0.89
Glucose Trend Prediction
Inflection Match
97.8%
Glucose & PPG/Temp Inflections
PPG Inverse Correlation
67-80%
PPG↑ → Glucose↓ Rate

With every glucose spike, vessels are damaged, inflamed, and hardened.
While CGM watches "the delayed shadow of interstitial fluid,"
blood vessels are screaming in real-time.
PPG is the technology that hears that scream.

Truth Revealed by Short Time Gaps

The true nature of glucose fluctuation, invisible at 5-minute intervals, finally revealed by 2-minute sampling. Two discoveries: "plateaus" and "initial velocity of rapid rise."

Discovery 1 — Glucose "Plateaus"
Temporary Leveling During Rise/Fall
A phenomenon where glucose temporarily levels off during rise or fall. At 5-minute sampling, this appears as "straight lines connecting dots" and is completely missed. Plateaus are evidence of homeostasis mechanisms fighting in real-time, serving as crucial signs for predicting trend reversals.
Discovery 2 — Initial Velocity of Rapid Rise
Most Glucose Rise Occurs in the First Few Minutes
With high-GI foods, glucose peaks around 45 minutes, but most of the rise occurs in the first 15-30 minutes. GLUT2 transporters relocate to intestinal epithelial apical membranes within 5-10 minutes in response to high glucose, rapidly accelerating absorption. Capturing this "initial velocity" requires 2-3 minute sampling intervals.
Physiological Mechanism — Why Plateaus Occur
Pulsatile Insulin Secretion at 5-15 Min Intervals
Insulin secretion from pancreatic β-cells is not continuous but pulsatile, released every 5-15 minutes. This pulsatility arises from coordinated synchronous release from over 1 million islets in the pancreas. The moment glucose rise and insulin pulse temporarily equilibrate appears as a "plateau." In type 2 diabetes, pulse regularity is lost and amplitude decreases—observing plateau patterns may lead to pancreatic function assessment.
Plateau Position Physiological Meaning Clinical Implication
Plateau during rise Insulin secretion catching up May turn to suppression with next pulse
Plateau before peak Equilibrium of rise and suppression Reversal point
Plateau during fall Liver applying brakes Evidence of glucagon secretion
Plateau before hypoglycemia Defensive mechanism activated Glucagon & adrenaline working
Insulin Pulse Period
5-15 min
Coordinated Islet Secretion
GLUT2 Translocation
5-10 min
Rapid Intestinal Adaptation
High-GI Food Peak
~45 min
17 min earlier than low-GI
Initial Absorption Peak
30-40 min
Depends on gastric emptying
Evidence References
Supporting Academic Literature
Pulsatile Insulin Secretion: Pørksen N, et al. "Pulsatile Insulin Secretion: Detection, Regulation, and Role in Diabetes." Diabetes 2002;51(suppl_1):S245-S254. — Insulin concentrations oscillate at 5-15 min periods, with pulsatility arising from coordinated secretory bursts from over 1 million islets.

Rapid GLUT2 Translocation: Gromova LV, et al. "Mechanisms of Glucose Absorption in the Small Intestine." Nutrients 2021;13(7):2474. — GLUT2 translocation to apical membranes occurs within 5-10 minutes and becomes prominent at high glucose (>30mM).

High-GI Food Peak Time: Louie JC, et al. "Timing of Peak Blood Glucose after Breakfast Meals of Different Glycemic Index." Nutrients 2013;5(1):116-134. — High-GI breakfast peaks at ~45 min, approximately 17 min earlier than low-GI.

Initial Absorption Peak: ScienceDirect Topics "Glucose Absorption" — Typical initial peak occurs 30-40 min after ingestion, largely dependent on gastric emptying rate.

Paradigm Shift

From traditional "value-based prediction" to "fluctuation-based prediction." Avoiding multicollinearity by tracking "magnitude of change" rather than "values."

Conventional Approach
Glucose = f(Vascular Response A, B)
Looking at value relationships
Multicollinearity issues
r = 0.22
→
New Approach
Cumulative |Δ Glucose| ≈ Cumulative |Δ Vascular|
Tracking magnitude of change
Multicollinearity irrelevant
r = 0.97
Clinical Implications
"Effective When It Matters" — An Ideal Property
The "variance" in turbulence index slope (0.05-0.43) is not a model flaw but meaningful differences reflecting clinical characteristics of each session. Effective during intense fluctuation (clinically important) and weak during calm periods (prediction unnecessary). This is an ideal property for preventive medicine.

Time Resolution Beyond CGM

Item CGM PPG
Measurement Target Interstitial Glucose Direct Vascular Response
Delay 5-15 min None
Spike Detection Blunted Sharp
Sampling Interval 5 min Under 2 min possible
Turbulence Observation Impossible Possible