05/16/2026
Great reading— looks like inexpensive, continuous glucose monitors (typically you wear these on your shoulder) will become more popular/recommended by the medical community overtime.
In 2015, a team at the Weizmann Institute connected 800 adults to continuous glucose monitors and watched what their blood sugar did for a full week. Each participant ate the same standardized meal four times: 50 grams of carbohydrate from white bread. Identical portion. Identical timing. The blood sugar curves that came back were not identical at all.
The average 2-hour glucose response across the cohort was 44 mg/dL·h, measured as incremental area under the glucose curve. The bottom 10% of responders averaged under 15. The top 10% averaged over 79. Same bread. Same dose. More than a fivefold spread in how much their blood glucose rose.
Zeevi and colleagues weren't measuring fringe cases. The 800 participants were broadly representative of a Western adult population: 54% overweight, 22% obese, 24% with HbA1c in the prediabetic range. None had been diagnosed with type 2 diabetes, but the cohort wasn't strictly "healthy" in any rigorous sense. The responses partly reflect underlying differences in insulin sensitivity, beta cell function, and metabolic state. People with higher BMI, higher HbA1c, and higher waking glucose tended to spike more. That part wasn't surprising.
What surprised was that the variability extended into the normoglycemic subgroup too. Two adults with the same fasting glucose, the same age, the same body composition could still produce post-bread curves that looked like they belonged to different studies. Most of the residual variation traced to microbiome composition, sleep duration the night before, physical activity around the meal, and what the person had eaten at the previous meal. Each was independently predictive after the standard clinical variables were accounted for.
This is the part that doesn't fit on a nutrition label. Glycemic index, the number assigned to white bread (around 71 in standard tables), comes from averaging responses across a small group of test subjects. By design, it tells you about the food, not about you. A "high GI" food predicts a high response on average. It predicts nothing about your response.
A few things this finding is not. It is not "GI is useless." For population-level diet research and food-labeling shorthand, it still works as a coarse signal. It is not "carbohydrates are deceptive." Carbohydrate amount remained the single strongest predictor of glucose response in the Zeevi data, it just didn't tell the whole story. And it is not "one big spike will harm you." Healthy human physiology handles post-meal glucose excursions all day, every day.
The concern with high responses isn't acute. It's chronic. Repeated large postprandial spikes, year after year, drive glycation, oxidative stress, and accelerated vascular damage. They drive insulin demand. They are independent risk factors for type 2 diabetes, cardiovascular disease, and all-cause mortality, distinct from fasting glucose and HbA1c. If you spike high to a food you eat three times a week, that food matters more than a label can say.
For most people, the actionable move is simple. You can now buy a continuous glucose monitor without a prescription. Stelo (Dexcom), Lingo (Abbott), and Libre Rio (Abbott) are all available over the counter, run roughly $50 to $90 per sensor for two weeks of data, and need no clinical justification. Wear one for two to four weeks. Eat the foods you normally eat. The patterns are usually obvious within a few days. The foods that consistently spike you are not necessarily the foods that spike anyone else, and they are not necessarily the ones with the worst labels.
If a CGM feels excessive, paired finger-stick testing works too. Same meal, multiple mornings, glucose at fasting and at 30, 60, 90, and 120 minutes. A few weeks of structured testing produces enough signal to identify your real outliers.
One honest caveat. The same person eating the same food on different days will produce somewhat different responses depending on sleep, prior meal, hormone cycle, and activity. Within-person variability is real, and Tom Wolever has argued, fairly, that part of what looks like between-person variability in studies like Zeevi's is actually day-to-day noise. But the fivefold spread Zeevi documented is too large to be explained by noise alone. The signal is real. The eater shapes the response as much as the food does.
Averages aren't destinies. A food's glycemic index was always a population description, never a personal prediction. We finally have tools cheap enough to find out what your own metabolism actually does. Use them on the foods you eat most often. The ones that show up over and over are the ones worth knowing.
References:
Zeevi et al., Cell, 2015 (PMID 26590418)
Wolever, Eur J Clin Nutr, 2016
