Personalized Nutrition by Prediction of Glycemic Responses

“please stop asking gurus how many carbs you need to optimize health”


bananas cookies

An interesting paper came out recently by Zeevi et al. (2015), showing, in part, that we’re all unique snowflakes (in some contexts).





Mini-rant: this study is in line with a lot of my beliefs about individuality in human biology.  We don’t know all the mechanisms, but we do know that some people respond better to some interventions than others.  We learn a lot from studies on diet, light, sleep, physical activity, etc., but the findings rarely/never apply equally to everyone (and some people experience completely opposite effects; eg, see studies where individual data are reported).  LIGHT exposure can improve sleep quality in some but cause agitation in others.  Low carb diets can help weight loss in some people but low fat is better for others.  Dairy, wheat, protein, the ‘biome, and fibre/resistant starch all fall into this category.  Sleep ‘requirements’ vary by person, season, geography, etc., etc…  there’s no QED answers in many of these contexts.

anecdote: some people say they’ve never had better blood glucose than when they were having a few servings of beans/legumes per week; others just report bloating & farts (no bueno).

End rant.

Background reading:

  1. The Atlantic ran a decent piece on this study (certainly more colorful than my take)
  2. Reddit AMA with some of the people involved in the study


In this particular study (video summary below): they continuously monitored the blood glucose responses in 800 people to all of their meals for a week, including a variety of test meals.  Main result: many different responses, even to the same foods!  An oversimplified example: some people had smaller relative postprandial glucose excursions after 50g carbohydrate from rice compared to 50g carb from potatoes, and other people responded oppositely.  And friggin’ tomatoes?!

Translation: need to move beyond recommending #IIFYM.

Some foods were universally well-tolerated [in this population] in the context of mixed meals, like quinoa and salmon; other foods did the opposite, like chocolate chip cookies and sushi.  And lastly, some foods like cottage cheese and hummus were good for some people but others.


bananas cookies[participant 445 is winning]


*In general, I don’t believe in labeling foods as categorically good or bad, which is pretty much confirmed by this study, but some patterns emerged wrt postprandial glucose excursions in this population…




good and bad foods


To make a crude study design to test these findings, they classified foods as good or bad based on blood glucose responses (for each person, on an individual basis), assigned them a diet consisting primarily of their good or bad foods for a week, monitored blood glucose, then crossed them over to the opposite foods.

Lo and behold, it worked!

N.B. all follow-up studies were in much smaller subgroups.


The most interesting part, in my opinion, was the divergent responses to the same foods.  See the right side of the figure above (with mixed red and green boxes).   These are the foods that worked for some people but not others (eg, taters were cool for subjects E4 and P9, but not E3).




Take-home message: there really is no one-size-fits all.  Even seemingly healthy/non-offensive foods like liver, chicken, and taters caused opposite effects on blood glucose in different people (in the context of mixed meals).  If you’re prone to postprandial hyperglycemia, you really can’t take other peoples “n=1” and hope it applies equally well to yourself (and definitely not IIFYM macro-based “n=1’s”).  I don’t recommend obsessive blood monitoring for healthy people, but if you’re diabetic or need strict control over metabolites like glucose or ketones, then your own #context becomes critically important.


I really hope this study is stretched out into a few more publications, because they collected a TON of data, and other than the figure above, not many specifics were reported.

Qu’ils mangent de la brioche?

They also measured a slew of biomarkers and even analyzed the ‘biome (which was a big player)… and many of these things impacted how different people responded to specific foods.

Super-cool, right?  It’s moving in the right direction, imo, along the lines of Chris Gardner’s work.  They didn’t assess any potential correlations with sleep quality, season, or exposure to artificial light in this study, but they definitely matter, too.

correction: they measured sleep but didn’t report the results. Hopefully, this will appear in a future publication.


In conclusion, please stop asking gurus how many carbs you need to optimize health.


Video summary of this study:



For full access to all articles and much more (or if you just like what I do and want to support it), become a Patron! Five bucks a month for full access and there are many other options. It’s ad-free and you can cancel if it sucks 🙂

Also, I’m open to suggestions so feel free to leave a comment or email me directly at

Affiliate discounts: if you’re still looking for a pair of hot blue blockers, Carbonshade is offering 15% off with the coupon code LAGAKOS and Spectra479 is offering 15% off HERETrueDark is running a pretty big sale HEREIf you have no idea what I’m talking about, read this then this.

20% off some delish stocks and broths from Kettle and Fire HERE

If you want the benefits of  ‘shrooms but don’t like eating them, Real Mushrooms makes great extracts. 10% off with coupon code LAGAKOS. I recommend Lion’s Mane for the brain and Reishi for everything else.

calories proper


Become a Patron!



Be Sociable, Share!
  • Jo De Koek

    I still wonder, if they pick a random 800 people one may assume approx half of them may be metabolically unhealthy so their responses to food may actually be a result of that – not so much the result of an individual, healthy response to food for which a human was evolved to eat. Fix their environment and life style for a while, give their bodies time to adapt and I bet responses will be different. Perhaps not so varied…… Just thinking out loud…

    • Hi Jo,
      I see where you’re coming from, but what about Participant #445! The glucose response to a banana suggests they’d be in your “metabolically unhealthy” group, but the response to a cookie suggests “metabolically healthy.”

      • Jo De Koek

        It is much more complex than I can grasp, and you’re in a better position than me to understand much of these complexities. Anyhow, haven’t read the details, and they have been monitoring glucose levels continuously, but still I’d be interested in how genetic, environmental and metabolic factors of the subjects were taken into account. Did they?
        Per my n=1, I know that I respond differently to the same type of foods eaten at different times, circumstances. Sometimes BG goes up after eating them, and sometimes not. Perhaps it depends on time of day, stress levels, amount of sleep, exercise I had or did not have, and a dozen other health markers at that specific time of eating….

  • Luis Villasenor

    As always Bill, #context is key 😉

    Awesome post.

  • Ketard Aesthetic Bodybuilder

    This article implies that the blood glucose response to a meal is the major determinant of whether or not a food is healthy (or indirectly, whether or not a person can sustain a healthy weight/cardio-metabolic factors). If that’s the case, then wouldn’t keto be the “correct” diet 100% of the time?

    • weilasmith

      you could have a lower bg than someone else because your body had a stronger insulin surge. that wouldn’t necessarily be a bad thing if the insulin pushed food nutrients into muscle and not fat. high bg that hangs around for over an hour (over 140) i believe is always a bad thing because it promotes inflammation of the wall of the blood vessel.

      • Sky King

        Right! Hyperinsulinemia (excessive insulin secretion) should be considered independently to insulin resistance (the glucose uptake rate) even tho the 2 conditions are intertwined and can co-exist under normal circumstances.

        I think having a constant condition of hyperinsulinemia can lead to many pathological conditions independent of insulin resistance.

    • 1) nope — see participant 445

      2) I don’t think blood glucose response is the only determinant of whether or not a food is healthy

      • This Old Housewife

        What about nutritional density? I know sometimes that quality can set off a huge response, such as with Omega-3 foods. The food is OBVIOUSLY healthy, but it still got a response.

  • sushi is rice w/vinegar, right? sashimi is raw fish.

      • I think most people imagine raw fish as sushi. So who knows what that means in Israel and in this study.

        • Eve

          Sushi in Israel is with rice, as in the US. But there is a world of difference between raw fish, rice and avocado, and specialty rolls that are deep fried or have tons of mayo/sweet sauces.

    • Yaroslav Fedevych

      I remember eating sashimi at a sushi place in Poland. I don’t know what do they soak their fish in to deodorize it and stuff, but the net result for me was blood glucose jumping from 85 to 155.

      NB: the kind of rice they cook for rolls does always have added sugar.

      Well, no more sashimi for me, I guess.

  • TechnoTriticale

    «I really hope this study is stretched out into a few more publications, because they collected a TON of data, and other than the figure above, not many specifics were reported.»

    Concur. Based on one of the HbA1c charts, it appears that they might have had some low carbers in the participant pool. Zooming in on them might be interesting.

    This wasn’t specifically a low carb trial, although that’s what the tailoring may have delivered in many cases. The standardizing diet was 150 grams net carb per day, and included wheat (bread), which I consider to be a microbiome antagonist among a long list of other problems.

    «they measured sleep but didn’t report the results.»

    That too. If the raw data is available to other investigators, much might be learned.

    «In conclusion, please stop asking gurus how many carbs you need to optimize health.»

    I’m following a named diet that seems to work for nearly everyone under the metabolic bell curve: 50 net carbs/day (so VLC, borderline keto); targets of PPBG under 90mg/dl, with no spikes above 100, HbA1c 5.0% or less; grain-free, moderate prebiotic, guidance on specific fats & supplements for common deficiencies and RDAs considered too low.

    The blog for this approach seems to hear from pretty much everyone who doesn’t get textbook results (usually weight loss stall), but idiosyncratic BG responses to arbitrary foods, interestingly, hasn’t so far been one of them. There may be a net carb threshold below which the individual variations seen in this study just go quiet.

    That said, I agree with the implication that the future of diet is personalized. My guess is that it’s going to be based on much more than personal PPBG response number crunching; more likely genotype, phenotype, goals (longevity, vitality, performance), incep status (multiple factors, including microbiome), epigenetic factors (some perhaps alterable), age, gender, geographic location, lifestyle preferences, and doubtless other factors to be named later.

  • Yaroslav Fedevych

    Now that’s a good post.

    There are things clearly bad and unsafe, not a big lot, really.
    There are things clearly safe, also not a big lot.
    And there is a huge gray zone where your mileage may vary different.

    For example, I don’t keep my diet to primarily get lean (a nice bonus, but not all). I have T2DM, so I want primarily to increase my life expectancy, and having a brain-heavy job, I want the mental clarity ketosis gives (or is it high blood glucose not fogging me? I can feel the effect whichever way it goes). The concerns of athletes and body builders seeking gainz, obese-but-otherwise-healthy people who want to get rid of excess body fat, bipolar/epileptic people who need ketosis to unglitch their brains, simply don’t apply to me.

    Also, I observe the response to different foods has actually changed after a month into steady ketosis. I thus must revisit and revalidate my own observations after a while, it turns out.

    So much for absolute numbers and one-size-fits-all attitudes, I think.

    • Thanks, Yaroslav!

      “So much for absolute numbers and one-size-fits-all attitudes, I think.”


      • Yaroslav Fedevych

        That said, I still crave any good reading material on low carb/ketosis specifically in the context of long term T2DM management (is a full reversal/remission even possible?). Looking forward to the Volume 2 of your work and any good pointers you might have to share.

  • Kindke

    its really hard to know what to make of these results,

    the question is are the variations in blood sugar due to different absorption rates from the intestine or different uptake rates into peripheral tissues ( muscle is responsible for clearing over 85% of glucose that enters circulation )

    my feeling is that it is due mainly to different uptake rates from the intestine mainly because its already been shown that obese and diabetics have more glucose transporters in the duodenum. Also This is the only way to explain why the cookie curve was flat for #445

    either way you are correct, everyone needs to tailor diet to themselves infact this should of been obvious to anyone studying diet/nutrition closely.

    Ive long observed skinny people are happy with small servings of carbs where as I need huge servings to feel satiated. That right there tells me IIFYM is stupid concept and I didn’t even need to measure blood sugar.

    However to my physiology/body, all carbs are fattening, so the study and blood sugar measuring is moot.

    there may well be some carbs I don’t spike blood sugar too but that is irrelevant as they all make me fat. The only exception would seem to be berry type fruit and maybe yoghurt.

    • “Also This is the only way to explain why the cookie curve was flat for #445”

      but if it was strictly due to glucose transporters, we should’ve seen similar responses with cookie and banana (I think)

  • Eve

    Interesting findings about the role of gut bacteria, specifically low levels of bifidobacterium adolescentis being associated with weight loss. I wonder how taking prebiotic and probiotics could manipulate results, and whether any of the subjects were taking them.

    • they measured a ton of variables — I’m hoping more associations are reported in future publications

      • Eve

        Based on your own readings, do you think that prebiotic or probiotic supplements could contribute to weight grain?

        • there are a lot of different pre- and pro-biotics out there… and I wouldn’t expect everyone to respond similarly… so, it’s possible

      • James


        I’ve read your blog for a long time, and I’ve gotten a lot of useful information from it. I’m sorry my first comment will be a bit critical.

        The way you’ve described this study raises several red flags:

        1.) If they’ve measured a ton of variables, what kind of multiple comparison procedures did they use to decide if their results were significant? This will affect the reproducibility of these results, especially if they didn’t decide which results to report first until after they analyzed the data.

        2.) In the world of predictive modeling, we often say, “Counterintuitive results are always wrong.” Of course, they’re only ALMOST always wrong. There are typically three possibilities to consider: (a) there’s a problem with the data, (b) there’s a problem with the model, or (c) a previously hidden insight has been unearthed. Possibilities (a) and (b) should be thoroughly and convincingly ruled out before possibility (c) should be seriously entertained. Do the researchers discuss performing any data quality audits? Do they discuss how they avoid overfitting their models?

        3.) It sounds like the personalized algorithm takes data from this week to predict data from next week. Would the personalized algorithm still beat the glycemic index if it used this week’s data to predict 6 months from now? How often would the model need to be refit to continue to be useful? How much data collection would be necessary for frequent refits, and how much of a compliance burden would that impose on people?

        4.) My concerns about how quickly the personalized algorithm will lose its predictive power makes me think about the post you wrote about having diabetics include protein in calculating how much insulin to inject. (For people who didn’t read that post, it turned out to be harmful rather than helpful. I suspect that this personalized nutrition algorithm would have similar risks.

        Although I am deeply skeptical about the reproducibility and usefulness of this study, I share your belief that there is not one perfect diet for all of us. And for most of us, we can try things for ourselves and discover what works best for us.

        • interesting point about how frequently they’d have to re-do the algorithm (if ever)

        • Drew

          First things first: Thank you, Dr. Lagakos, for bringing the science and skepticism to the critical thinkers on the interwebs: this is a true public service.

          I think everyone’s immediate reaction to the overwhelming data was that they left no stone unturned, but instead of being critical to you or James, I’ll add to the conversation using James’ points for the most part.

          1. and 2.) For the microbiome data, they used a false discovery rate p value of .15 which is a stretch (multiple comparisons) for multiple samples from 800 for experimental study and 100 in the validation study. In their PPGR model there were 137 variables (6 categories, including meal features, meal timing, GCM data/trends, labs, personal features [sample was age 18-70], and microbiome KEGG data); how they were weighted probably indicates far fewer variables can reliably predict glycemic response in the average person. As far as reproducibility goes, the supplemental info is thorough, but they would have to share the PPGR model, which they would probably prefer to patent instead. I don’t think these results are counterintuitive: the unique snowflake figures above are from two random 20g cho “meals” from two participants at random times (controlling for nothing that might affect baseline or postprandial glucose response)–great conversation starter, though.

          3) If your diet, exercise, sleep, stress, meds, adiposity, etc. never change, the algorithm would probably only change every few years (with aging). In the intervention group, several microbiota changes were reported, so these would require an update. Based on these microbiota changes, it appears the intervention groups on the good food diet were probably eating less calories, which was not monitored; more importantly, the caption on the good food/bad food heat map states “Only dominant food components are shown, defined as foods whose carbohydrate content was more than 50% of the entire Caloric content of the meal in which they were prescribed.” The only way to make a meat serving have more than 25g CHO is to bread it and fry it or add a nice little sugary sauce (or both). You certainly will not reproduce their carb-rich meat concoctions without more data.

          4.) This won’t help anyone who isn’t willing to spend a nice sum of money on personalized nutrition, which would be the bare minimum for microbiome analyses, CGM, other labs, processing data for all of the 137 components, etc. So fear not, this was merely a thorough debunking of IIFYM (with 150g CHO).

          • James

            Thanks for diving into this paper, Drew. You’ve provided some very useful details.

  • Let’s feed them all Oreo cookies. That’s a constant. Why test with bananas and pizza or sushi where there’s so much variability,

    • they tested a variety of foods on multiple occasions, even pure glucose and fructose (all controlled for carbohydrate content). Results were surprisingly consistent.

      IOW, I don’t think the difference between 50g carbs from pepperoni pizza or broccoli pizza would really matter that much in this context.

      • Thomas Hemming Larsen

        Did people react just as differently to pure glucose and fructose?

        • the fructose data were pretty tight; glucose, not so much

  • Nagar J
  • This Old Housewife

    I read the article and the study, and they still have about 4,000 volunteers to work through. I would DEARY LOVE to sign myself and Hubby up, but there was no contact info (they probably wouldn’t have room anyway). The study (or was it the article?) says you can do the same thing with a continuous glucose monitor, but how do you get one if you aren’t Type 1 diabetic?

    For now, fasting’s been a light at the end of a long, dark tunnel for him. Just last night, there was no change in pre-prandial and 1-hr. post-prandial numbers after a week of fasting (yes, while you guys were FEASTING, we chose to fast all week). 2-hr. post-prandial only dropped 2 pts. Hubby was pleased with last night’s numbers, and agreed to extend the fast until Wednesday.

    We’re also planning another fast during Christmas week. I want to see how long he can go with food before he starts climbing back into danger territory–it may turn out that he needs to fast for a week each month, or maybe every other week. Hey–whatever it takes to keep him off insulin, right?

    Meanwhile, my grocery bill has been slashed to ribbons. I even found a free source of “meat”–the stuff I’ve been boiling for homemade bone broth (pig feet, chicken feet, chicken necks/backs, osso bucco, ox tails, etc) still has some meat on it, and after boiling and straining, amounts to about 2.5 lbs worth. I’ve been picking out bones, claws, knuckles, and hard cartilage (anything unchewable), subdividing the remaining meat, and freezing this to use for when we come off the fast. Most people wouldn’t bother, and just throw this stuff out.

    After making broth about every 3-4 days, I have quite a supply of this “meat” stocked up in the freezer. From now on, an actual cut of meat will be a treat around here.

    What can I say? I’m an old housewife (not really that old, but still).

    • iris for enrollment

  • Thomas Hemming Larsen

    Great post Bill! We have talked about the individuality of LCHF/LFHC before and I think this is just the next layer of it. I hope that we more and more can individualise nutrition by thinking about these things (and then people will start researching in how to change it so that we can all eat cookies :)).
    I really hope that they will release more data from the study and that you’ll write more about it!

  • 64 magpies

    So if carb intake is an inaccurate proxy for insulin in some people, I wonder how accurate blood glucose would be as a proxy for insulin?

    • “So if carb intake is an inaccurate proxy for insulin in some people”

      I agree, but a better interpretation of this study may be that just saying “carbs,” without specifying specific foods, isn’t a good proxy for postprandial blood glucose.

      “I wonder how accurate blood glucose would be as a proxy for insulin?”

      Blood glucose is a poor proxy for insulin.

  • Pingback: Latest in Paleo 148: Derailing Evolution | – Humans Are Not Broken()

  • Larry Rotenberg

    i discovered this a few years ago when i was diagnosed with t2d……. i test every meal see how i react…..oatmeal even steelcut spikes my bg……while tonnes of red kidney beans no effect…..

    • I’ve hear this a lot, about beans/legumes. Cool.

  • so so so interesting. Love it. The more I learn about this, the less I know.

  • Lyle genyk

    I really think we need to trust our (gut) instincts more when it comes to eating. If certain foods don’t sit well with you don’t eat them regardless of their “healthy” label. If we listen to what our bodies are saying it can become pretty easy to eat a very personalized healthy diet. Stop worrying about what all the experts are saying to the masses and trust yourself.

    • further, I think this study shows that “healthy” label can’t really be applied to foods (in general) bc even something like white bread can be perfectly fine for someone but totally spike BG in someone else

  • Pingback: Tedenski pregled #30()

  • keithlovesfood

    Hey Bill,

    Would you agree that this study basically dismisses the standardization of glycemic index values of foods that is widely used by folks trying to reduce their glycemic response to food?

    • My take-away: for each food, there’s a bell curve and someone might be above average on some foods but below it on others.

      If someone is having trouble with blood glucose control, then they may want to record the foods (not just carb count) and monitor blood glucose response.

  • Mindbody Medic

    Have the guys who authored it shared the algorithm openly so the lay-obsessive could use it?

    I’m no scientist but this is the coolest study I can remember looking at /thinking of implications

    • it wouldn’t be of much use because you need to input a ton of blood tests, microbes, etc.

      • Mindbody Medic

        I was thinking from the perspective of the maniacally obsessed biohacker! lol