The Carbs-Obesity Hypothesis

Published: October 2015

What is causing the global increase in obesity rates, and what can we do about it? One popular view, associated with Gary Taubes and others, is the carbs-obesity hypothesis, which claims that carbohydrate consumption is the primary cause of obesity, and that limiting carbohydrate intake is the most reliably effective way to lose weight.

The Open Philanthropy Project plans to survey a wide range of potential cause areas, only some of which will turn out to look promising enough to warrant deeper investigation and potential grantmaking. For this reason I (Luke Muehlhauser) conducted a brief, surface-level investigation of the evidence for and against the carbs-obesity hypothesis.

My goals for this project were to learn what the major arguments are, what kinds of evidence those arguments appeal to, and what the current state of that evidence appears (from a superficial scan) to be. I did not attempt to evaluate any studies I found, except by evaluating “surface features” such as what methods the authors claim to have used. I also did not attempt to assess the effects of various diets on health outcomes besides obesity, such as cardiovascular risk or diabetes. Nor did I attempt to evaluate the effects of any property of a diet besides its mere macronutrient proportions (between carbs, fats, and proteins) — for example, I did not try to evaluate arguments about “good carbs” vs. “bad carbs.” I also chose to not evaluate several types of arguments, including the ubiquitous arguments over the biological mechanisms of fat accumulation. More details on my process are available in a later section.

To improve the transparency of our process, we have also included the output of a vet we conducted on an earlier draft of this document, and my response to the issues raised in that vet. See here. Update: On 10-02-2015, I added some comments on this report by Dr. Stephan Guyenet, along with some notes about how I updated this report in response to those comments: see here.

My tentative conclusions about the state of several kinds of evidence concerning the carbs-obesity hypotheses are here.

My overall tentative conclusion is this:

Though my investigation was brief and shallow, my impression is that the available evidence doesn’t fit the hypothesis that carbohydrates are the dominant cause of obesity. I suspect that whatever is true about the connection between diet and obesity — besides the laws of thermodynamics1 — must be more complicated than the hypothesis I’ve investigated here. Perhaps particular types of carbs, fats, or proteins — e.g. refined carbohydrates, or sugars, or something else — are reliably linked to obesity. But I haven’t investigated any of those hypotheses yet.

Naturally, a deeper and more thorough investigation could overturn this tentative conclusion. Here are some guesses at how it might do so: (a) it might be that a more detailed examination of the systematic reviews of diet RCTs turns out to indicate a clear “winner” among weight-loss diets (more likely low-carb or Mediterranean than low-fat), but if so then I expect one would only find a <2kg average marginal difference over the next best diet at the 1 year mark, which wouldn’t really vindicate the strongest claims of low-carb advocates and wasn’t an exciting enough prospect for us to more closely examine those systematic reviews ourselves; (b) perhaps there are strong arguments for or against the carbs-obesity hypothesis based on biological mechanisms, which I largely bracketed in this report for reasons described here; or (c) a more thorough investigation might conclude that our hunter-gatherer ancestors near-universally ate low-carb diets, which could at least provide some suggestive evidence for the carbs-obesity hypothesis. Another way to disprove my tentative conclusions would be, of course, to conduct additional original research — in particular, RCTs with longer intervention periods and follow-up periods than have been used for the RCTs conducted so far.

To increase the speed with which we can survey the evidence concerning many different potential cause areas, we decided not to invest as much time on exposition and thoroughness as we have for some other investigations. This document was originally prepared for internal use only, and then was lightly re-organized and edited for public consumption.

If you know of studies or reviews which seem like they should have been mentioned or cited in this report, or which are important but were published after the initial release of this report, please send them to socialscienceupdates+carbsobesity@openphilanthropy.org along with your comments, if any.

We originally published a differently-formatted version of this report as a Google Doc in August 2015. This page presents the most up-to-date version of the report.

Contents

My literature search process

Identifying the arguments

One key part of my search process was to find carbs-obesity proponents or opponents who discussed as many different kinds of argument and evidence as possible (randomized controlled trials aka RCTs, animal models, epidemiological studies, etc.), so that I could identify the major arguments in the debate and then look for recent literature reviews related to each argument separately. This was necessary because most literature reviews on macronutrient ratios or even obesity in general focus on only 1–3 types of evidence (e.g. only diet RCTs and cohort studies, or only epidemiological evidence).

First, I read perhaps the most famous recent book on the topic, Gary Taubes’ Good Calories, Bad Calories (2007, hereafter GCBC). I used GCBC and this handy outline to identify (what turned out to be) most of the major arguments in the debate.

Next, I briefly checked many other recent (post-2005) sources that came up when searching (Google, Google Scholar, and Amazon) for proponents or opponents of “low-carb” or “paleo” or “ketogenic” diets (including e.g. proponents of “low-fat” and “Mediterranean” diets), but only two other sources I found spent much effort arguing from many different angles like Taubes’ book does:

  • Volek & Phinney’s The Art and Science of Low Carbohydrate Living (2011) argues in favor of low-carb diets from multiple angles, but it doesn’t add much beyond what is already in GCBC.
  • Jaminet & Jaminet’s Perfect Health Diet (2012, hereafter PHD) pointed me to studies of obesity in pets,2 critiques of Cordain & colleagues on hunter-gatherer diets, an analysis of hunter-gatherer diets that I hadn’t encountered in other sources, and several new kinds of argument — re: “self-cannibalization,” breast milk, mammalian diets, and the food reward system — that I don’t recall reading in GCBC.

I also read some reviews of GCBC3 and PHD,4 and skimmed the list of sources which cited the sources listed above according to Google Scholar.

In general, I only looked at studies focused on adult obesity, rather than childhood obesity.

The full list of types of argument/evidence I found is:

  1. Observational studies. See my process and my conclusions.
  2. Quasi-experimental studies. See my process and my conclusions.
  3. Randomized controlled trials. See my process and my conclusions.
  4. Arguments about hunter-gatherer diets. See my process and my conclusions.
  5. Animal studies. See my process and my conclusions.
  6. Bio-mechanism arguments (ignored). My manager for this project, Holden Karnofsky, suggested I ignore arguments about bio-mechanisms for now, in part because they would take a long time to evaluate, and in part because Holden is skeptical that anybody understands the bio-mechanisms of fat accumulation. Some of this skepticism comes from his impression that scientists are almost never able to predict the efficacy a drug or other medical intervention from lab-based understandings of biomechanisms alone; empirical evidence about the effects of interventions on people is generally necessary to be confident in bio-mechanisms.
  7. Argument from common wisdom (ignored). Good Calories, Bad Calories claims the carbs-obesity hypothesis was common wisdom prior to the 1970s, but if true it would be weak evidence anyway, so I didn’t investigate it.
  8. Difficulty of getting fat on low-carb diets (ignored). Good Calories, Bad Calories claims it’s hard to get fat on low-carb diets. I haven’t checked, but I would guess this claim is false. E.g. see the studies Guyenet cites after the phrase “at least in studies lasting weeks” here.
  9. High-carb diets used for intentional fattening (mostly ignored). Good Calories, Bad Calories cites two examples high-carb diets being used for intentional fattening, and the one I checked is misleading.5 I didn’t look into this further.
  10. Clinician recommendations (investigated briefly). See my process and my conclusions.
  11. Self-cannibalization (ignored). Perfect Health Diet (ch. 3) makes an argument from “self-cannibalization,” which I didn’t investigate.
  12. Breast milk composition (ignored). Perfect Health Diet (ch. 4) makes an argument from breast milk composition, which I didn’t investigate.
  13. Mammalian diets (ignored). Perfect Health Diet (ch. 5) makes an argument from mammalian diets, which I didn’t investigate.
  14. Food reward system (ignored). Perfect Health Diet (ch. 6) makes an argument from the food reward system, which I didn’t investigate.

Below, I describe my process for investigating each type of argument/evidence I investigated. Later, I describe my tentative conclusions concerning each type of argument/evidence I investigated.

Hunter-gatherer diets (process)

On hunter-gatherer (HG) diets, Taubes and almost everyone else cites a series of papers by Eaton & Konner & Cordain (EKC). I skimmed the list of papers which cite those papers, and found replies by Milton and others, as well as more recent papers by EKC on HG diets. I skimmed the most promising6 of these papers for other references on HG diets they cited, and skimmed the list of studies which cited them. Almost all my discovered sources on HG diets — both those cited on the subject of HG diets below and dozens of others I don’t cite here — came from this search. One exception is Kaplan et al. (2000), which I found via the Perfect Health Diet blog. Skimming dozens more papers, and searching Google Scholar for a variety of terms related to hunter-gatherer diets and macronutrient ratios, didn’t turn up any new primary sources, so I stopped searching.

SRs/MAs of RCTs (process)

Systematic reviews (SRs) and meta-analyses (MAs) of relevant RCTs were relatively easy to find. I simply searched Google scholar for terms like “systematic meta-analysis” and “low-carbohydrate,” restricted the publication date range, and several of the somewhat-recent systematic meta-analyses came up in the results. I then scanned through the list of studies which cited each of these meta-analyses, and this led me to the very most recent meta-analyses. I also checked which RCTs were most cited by the first several recent SRs/MAs I found, and searched for SRs/MAs citing those studies. I looked at only systematic meta-analyses which (1) seemed to have used fairly reasonable methods,7 (2) included at least 5 RCTs, (3) were published between January 2011 and April 2015, and (4) mostly compared the performance of non-usual diets to the performance of other non-usual diets which differed in their macronutrient ratios (rather than to e.g. the performance of the standard American diet or to baseline body weight values, because such studies cannot distinguish the effects of going on any diet from the effects of going on a particular diet).8 I also did not include meta-analyses which mainly reported on differences between lower-protein and higher-protein diets (as opposed to e.g. low-fat vs. low-carb diets).

Observational studies (process)

My first paragraph in the section on observational studies below mentions the apparent rarity of large populations which adhere to a low-carb diet. While reading or skim-reading the hundreds of studies I reviewed for this project — including several books and dozens of articles by low-carb advocates — I only recall encountering the few example populations I list in that paragraph. In contrast, examples of populations with high-carb diets seem to be abundant. I cited the examples provided by Guyenet (2011) because he conveniently listed them in reply to Taubes, but I expect one could find additional examples if one searched the literature for this purpose.

My second paragraph about observational studies concerns between-population ecological studies. I found Bray & Popkin (1998) cited in Popkin’s more recent work on nutrition transition, and it was the first time I had encountered a regression of an obesity measure on a population diet measure. I skimmed the list of studies which cite Bray & Popkin (1998) for more recent reviews which might cite similar studies, and found a few more population-level analyses. Willett & Leibel (2002) describes the results of two additional between-populations ecological studies to which I do not have easy access and have not read, but which Willett reports as being in tension with Bray & Popkin’s result.9 Next, I conducted a series of searches (1, 2, 3, 4) for studies citing one of these 4 studies and which also use some other terms. I skim-read the most promising10 results, used the bibliographies to find additional promising studies, skimmed the list of papers citing those studies, and so on until I seemed to not be finding any promising new sources. None of these searches, nor a keyword search, revealed any additional large-scale between-population studies of macronutrient ratios and obesity.

Quasi-experimental studies (process)

Due to time constraints, the only type of quasi-experimental study I surveyed was the prospective cohort study. By this point I already knew of USDA (2015), which summarizes data from many prospective cohort studies, but draws its own data from Jensen et al. (2013) [RCTs only], USDA (2014), Esposito (2011) [RCTs only], and Ambrosini et al. (2014) [which focuses on childhood dietary patterns]. So I reviewed USDA (2014)’s review of prospective cohort studies and added it to the section on quasi-experimental studies below. USDA (2014, PDF-page 25) found 10 prospective cohort studies on adults that met their search criteria, published between 2006 and 2012. To find additional recent systematic reviews including prospective cohort studies, I searched for studies which cite the 8 “positive-quality” studies on adults identified by this USDA report — Buenza et al. (2010), Gao et al. (2008), Kesse-Guyot (2009), Lassale et al. (2012), Mendez et al. (2010), Romaguera et al. (2010), Runawas et al. (2009), Zamora et al. (2010) — and which also included the term “systematic review.” (E.g. one search was this.)

Skimming titles and (in some cases) abstracts, the most promising studies I found this way were: Fogelholm et al. (2012), Aljadani et al. (2013), Garcia-Fernandez et al. (2014), Aljadani et al. (2015), Grosso et al. (2014), and Kastorini et al. (2011). Aljadani et al. (2015) is an update to Aljadani et al. (2013), so I didn’t look further at Aljadani et al. (2013). Fogelholm et al. (2012) didn’t directly report macronutrient ratios. Garcia-Fernandez reported only three cohort studies with obesity-related outcomes. Kastorini et al. (2011) included only 2 prospective cohort studies. So, I only added Aljadani et al. (2015) to the section on quasi-experimental studies and stopped there.

Clinician guidance (process)

I stumbled upon Bray & DeLany (1992) while studying other topics. Then, PubMed ‘related articles’ and ‘cited by’ searches, plus ‘cited by’ searches on Google scholar — of that paper, and then of each subsequent promising paper I list below, up to the asterix below — led me to a few more studies surveying expert opinion on dieting which, at a glance, I thought might contain information about experts’ opinions about a diet’s ideal macronutrient ratios: Bray & DeLany (1995), Foster et al. (2003), Park et al. (2005), Harvey & Hill (2001), Loomis et al. (2001), Warner et al. (2008), Harvey et al. (2002), Starostina & Dreval (2001), Dixon et al. (2008), Somannavar & Appajigol (2014), McConnon et al. (2013), Shifman et al. (2009), Bleich et al. (2012), Epling et al. (2011), Huber et al. (2011),* Sack et al. (2009), Ashby et al. (2012), Alexander et al. (2012), Bocquier et al. (2005), Yo et al. (2008), Bleich et al. (2015), Glauser et al. (2015), van Dillen et al. (2013), Engstrom et al. (2013), Hartley (2013), Kristeller & Hoerr (1997), Campbell et al. (2000), Lang et al. (2013).

Skimming these, I found that very few of them asked about experts’ recommendations about macronutrient ratios. Or if they did, the survey asked about one option (e.g. low-fat) but not about others (e.g. low-carb). The only useful study in this batch was van Dillen et al. (2013), which summarized results from many other surveys but only pointed to a single study (Phelan et al. 2009) that asked about macronutrient recommendations by general practitioners (GPs). In that study (n=101), a statistically significantly greater proportion of GPs recommended low-fat than recommended low-carb. But it’s just one study.

I checked studies which cite Phelan et al. (2009), but this didn’t turn up any promising11 results. Next I checked for studies which cite, or are ‘related articles’ for, van Dillen et al. (2013), but this didn’t turn up anything helpful either. Next I searched PubMed using two MeSH terms applied to Phelan et al. (2009), but this didn’t turn up any promising studies either (among the first 40 results). At this point I stopped looking.

Animal studies (process)

I investigated the literature on animal models only briefly. Non-systematic literature reviews of animal models used to study human obesity tend to focus on their use for investigating the biomechanisms of obesity, so I didn’t look much at those studies. Nearly all my investigation into the literature on animal models was conducted before I was keeping detailed records of my searches, and unfortunately I can’t remember much about my process except that it involved lots of keyword searches on Google Scholar.

My tentative conclusions

What do observational studies suggest?

(My process described here.)

Low-carb diets are rare ‘in the wild,’ at least in recent decades and perhaps centuries.12 The only examples I’ve seen cited so far are the Inuit, the Tokelauns, Masai herdsmen in Central Africa, and the “Bison People of the North American Great Plains.”13 Meanwhile, Guyenet (2011) cites examples of populations with high-carb diets and low rates of obesity: the New Guinea tribe at Tukisenta, the people who lived in the West Nile district of Uganda in the 1940s, the Massa tribe of Northern Cameroon, the Kitavans in the 1990s, the Pima before ~1890, and, well, 20th century Asians generally.

What about between-population ecological studies? The largest such study I’ve seen is found in Bray & Popkin (1998), which found a positive relation between percent of energy from fat and a country’s obesity rate, using data from 20 countries (adjusted R2 = 0.78). These data weakly support low-fat rather than low-carb recommendations, because I have the impression that one’s proportion of energy from protein usually doesn’t vary widely, so a high-fat diet usually implies a low-carb diet, and vice-versa.

However, Willett & Leibel (2002) critiques Bray & Popkin’s conclusions, and points to other between-population ecological studies which showed no apparent correlation between dietary fat and obesity.14 I searched fairly extensively for additional large-scale between-population ecological studies of macronutrient proportions and obesity, and couldn’t find any besides these. Such observational studies are of limited use for causal inference, anyway.

What about quasi-experimental studies?

(My process described here.)

The only type of quasi-experimental study design I checked was the prospective cohort study. I found two recent (last 5 years, i.e. 2011 and later) systematic reviews which included more than 5 prospective cohort studies:

  • USDA (2014) found 10 recent (2006–2012) prospective cohort studies on adults matching its search criteria, and concluded from these studies that there is moderate evidence in favor of a Mediterranean-style diet (but not, e.g., in favor of a low-carb diet).
  • Aljadani et al. (2015) found 16 prospective cohort studies15 published from 1970 to 2014, and found that greater adherence to national healthy diet indices, and/or to Mediterranean-style diets, predicted lower weight gain.

My impression is that more prospective cohort studies in the last decade have specifically tested for the effects of Mediterranean-style diets than have tested for the effects of other specialized diets, such as low-fat or low-carbohydrate. If prospective cohort studies typically tested a different specialized diet instead of Mediterranean diets, would systematic reviews of prospective cohort studies find evidence favoring that specialized diet instead? Perhaps.

I only conducted a semi-thorough search for post-2010 systematic reviews including at least 5 prospective cohort studies, and I wouldn’t be terribly surprised if there were a few I missed.

Did our hunter-gatherer (HG) ancestors typically eat low-carb diets?

(My process described here.)

Many authors argue that we will be healthiest if we consume what our hunter-gatherer ancestors ate, i.e. the diet of our environment of evolutionary adaptedness.

On the question of macronutrient proportions and ancestral diets, most authors depend on the work of Eaton & Konner & Cordain (EKC) and colleagues, whose analyses rely on ethnographic studies of 20th century HGs. Eaton & Konner’s 1985 study suggested that most 20th century HGs ate a relatively high-carb diet (~%45 carb, similar to current standard American diet),16 while their latest studies suggest that most 20th century HGs ate a low-carb diet (58% of HG groups ate 22%–40% carb, depending on which plant:animal subsistence ratio is assumed).17 One recent non-EKC study also came to a low-carb conclusion via essentially the same calculations as EKC, but with high variance between populations (“≈3%-50% of the total energy intake; median and mode, 16%-22% of the total energy”).18 Another recent non-EKC study also suggested HGs ate a low-carb diet, again based on ethnographic studies (7 of 9 HG groups ate <20% carbs, according to this interpretation).19 Lindberg and others explain in detail why one must be cautious about taking such estimates too seriously, though Lindberg also has a low-carb conclusion (citing just EKC).20

I haven’t yet found any serious recent analysis that argues directly that HGs generally ate a high-carb diet, though it does seem to be a common view that, contra EKC, hunter-gatherers typically ate very little meat, which may imply a high-carb diet for HGs.21 It’s also a common view that estimating the typical diet of hunter-gatherers is very difficult.22

I haven’t evaluated these studies, but I’ll share my impressions of the state of the evidence from my surface-level readings. I’d be surprised if further investigation justified confident conclusions about the diets of ancient HGs, except perhaps via a methodological breakthrough, e.g. a way to infer macronutrient ratios from analyses of stable isotopes (see Schoeninger 2014). I wouldn’t be surprised if further investigation (1) demonstrated that Cordain et al. have made a reasonable case for the diets of 20th century HGs, (2) demonstrated that we can’t even have a “best educated guess” at this point, or (3) demonstrated that our “best educated guess” right now should be something other than what Cordain et al. conclude.

What do randomized controlled trials suggest?

(My process described here.)

Results are mixed.

The most recent (past 5 years, i.e. post-2010) credible-looking systematic reviews (SRs) and meta-analyses (MAs) of randomized controlled trials (RCTs)23 comparing multiple specialized diets are:24

Review Aim of most relevance to us Included RCTs Key outcome Notes
Bueno et al. (2013) ≤50g/day carb vs. <30% fat 13 low carb = -.91kg WMD, n=1415 Searched MEDLINE, CENTRAL, ScienceDirect, Scopus, LILACS, SciELO, ClinicalTrials.gov, OpenGrey.eu, DissOnline.de, NYAM.org, ClinicalEvidence.com up to Aug 2012, any language. Inclusion: ≥18yo participants, ≥12mo followup, mean BMI >27.5. Exclusion: concomitant pharma intervention.
Wycherley et al. (2012) High-protein vs. low-protein, within <30% fat, isocaloric, %fat-matched diets 23 high-protein (thus lower carb) = -0.79kg WMD, n=~1000 Searched MEDLINE, EMBASE, PubMed, CENTRAL for 1947-May 2011, English. Inclusion: ≥4wk followup, ≥18yo participants. Exclusion: concurrent exercise intervention, nonparallel study design, very low energy intakes, pregnant or breastfeeding participants, concurrent weight-loss medication, participants who had undergone surgical procedure affecting weight loss. Here, I’m only counting studies from Figure 2.
Ajala et al. (2013) Compared low-carb, veggie, low-GI, Med., & others with control diets incl. low-fat, high-GI, & others 16 Only Med. had greater weight loss (-1.84kg WMD) compared to control diets, n=2229 Searched PubMed, EMBASE, Google Scholar up to July 2011. Inclusion: ≥18yo participants, ≥6mo intervention, type 2 diabetes patients.25
Hu et al. (2012) ≤45% carb vs. ≤30% fat 23 No difference, n=2788 Searched MEDLINE, EMBASE, Web of Science, Cochrane Database of SRs, from 1966–June 2011, any language. Inclusion: ≥18yo participants, ≥6mo intervention. Exclusion: differences other than macronutrient & energy intake between the 2 diets.
Clifton et al. (2014) increase protein or lower carb vs. other control diets 32 increase protein or lower carb = -0.39kg WMD, n=3492 Searched PubMed, EMBASE, Cochrane Library through July 2013. Inclusion: ≥18yo participants, ≥12mo followup. Exclusion: meal replacement treatments.
Rehnquist et al. (2013) Compared several diets against each other, for weight loss in obese people 23 No meta-analysis performed; conclusion is “In the short term (six months), advice on strict or moderate low carbohydrate diets is a more effective means of achieving weight loss than advice on low fat diets. In the long term, there are no differences in the effect on weight loss between advice on strict and moderate low carbohydrate diets, low fat diets, high protein diets, Mediterranean diets, diets aimed at achieving a low glycaemic load or diets containing a high percentage of monounsaturated fats.” n=56808 Searched PubMed, EMBASE, Cochrane Library, Cinahl, PsycInfo up through May 2013. Inclusion: ≥6mo followup, ≥10 participants in each treatment group. Exclusion: participants “selected on the basis of factors such as cancer, hereditary diseases, psychosis, polycystic ovary syndrome, endocrine disorders, serious renal or liver disease, pregnancy with concomitant obesity or women up to one year after childbirth.” Here, I’m only counting RCTs from Table 1.26
Jensen et al. (2013) Compared several diets against each other, for weight loss in overweight or obese people 12 No meta-analysis performed; conclusion is “A variety of dietary approaches can produce weight loss in overweight and obese adults.” n=2025 Searched CENTRAL, from Jan 1998–Dec 2009, English. Inclusion: ≥100 participants in each treatment group, ≥6mo followup, ≥3mo intervention, ≥18yo participants, BMI ≥25. Exclusion: rated poor quality e.g. due to intention-to-treat and attrition rates, outcomes by measure of self-report, non-Western countries, unpublished.27
Esposito et al. (2011) Med. vs. other (mostly not standard) diets 16 Med. = -1.75kg WMD, n=3436 Searched PubMed, EMBASE, Scopus, CENTRAL through Jan 2010. Inclusion: ≥15 participants, ≥4wk followup.
Nordmann et al. (2011) Med. vs. ≤30% fat 6 Med. = -2.2kg WMD, n=2650 Searched MEDLINE, EMBASE, Biosis, Web of Science, CENTRAL to Jan 2011, any language. Inclusion: ≥6mo followup. Exclusion: no intention-to-treat analysis.
Castañeda-González et al. (2011) ≤130g/day carb vs. other (mostly not standard) diets 8 No meta-analysis performed; studies show inconsistent results. Searched PubMed, Cochrane, EBCOhost, Jan 2000–Dec 2009, English. Inclusion: ≥18yo participants, ≥4mo followup, type 2 diabetes patients. Exclusion: type 1 diabetes patients, pharma interventions, interntions in hospitalized individuals, “variations in contents and quality of carbohydrates such as increase of fiber.”
Schwingshacki & Hoffman (2013) ≤30% fat vs. >30% fat 32 No difference if including all trials. Including only trials of hypocaloric diets, >30% fat = -0.59kg WMD, n=~2800? Searched MEDLINE, EMBASE, CENTRAL up to Mar 2013, any language. Inclusion: ≥12mo followup, assessment of cholesteral-related outcomes. Exclusion: RCTs on “secondary prevention of CVD.”28
Dutton et al. (2014) Several diets against each other 22 No meta-analysis performed; studies show inconsistent results. Searched PubMed, from Jan 2003–Apr 2014, English. Inclusion: free-living adults, ≥15 participants per treatment group, ≥2mo treatment. Exclusion: attrition >40%, focused mostly on overweight or obese subjects, not focused specifically on weight or cardiovascular risk.
Johnston et al. (2014) Several branded diet programs against each other 48 Low-fat and low-carb diet programs outperformed the others at 6mo and 12mo, n=7286 Searched AMED, CDSR, CENTRAL, Cinahl, EMBASE, MEDLINE, ClinicalTrials.gov, metaRegister of Controlled Trials through Apr 2014. Inclusion: ≥18yo participants, ≥3mo followup, BMI ≥25. Exlusion: diet treatment wasn’t mostly whole foods, pharmacological agents.
Naude et al. (2014) <45% carb vs. ~isocaloric “balanced” diets 19 No difference, n=3209 Searched MEDLINE, EMBASE, CENTRAL through Mar 2014, English. Inclusion: ≥10 participants in each treatment group, ≥18yo participants, ≥3mo intervention. Exclusion: pregnant or lactating women, treatments combined with other interventions (exercise, pharma, surgical), >500 kilojoules difference between diets, ad libitum energy restriction, interventions focused on specific food groups or components, meal replacements or supplements used, test meal responses were assessed.

At a glance, each of these recent systematic meta-analyses seem to have been conducted using fairly reasonable methods, but because each SR/MA asks such different questions, it’s hard to declare a “winner.” About the best I can do at a succinct summary is:

In Bueno et al., low-carb won over low-fat. In Wycherley et al., higher-protein/lower carb won over lower-protein/higher-carb among low-fat diets. In Ajala et al. and Esposito et al., Mediterranean won over several control diets, including low-carb and low-fat. In Clifton et al., advice to increase protein or lower carb intake beat other control diets. In Rehnquist et al., low-carb won over other diets in 6mo but not at 12mo. In Nordmann et al., Mediterranean beat low-fat. In Johnston et al., low-fat and low-carb branded diets beat other branded diets at 6mo and 12mo. Several other SRs/MAs found no difference between diets.

That is: four studies looked for, and didn’t find, an advantage for low-carb (or relatively high-fat) diets relative to low-fat diets or some other diet. Four studies found an advantage; all were less than a 1kg overall average effect. Two studies found an advantage for the Mediterranean diet (the weight loss advantages were larger but still under 3kg). The rest of these studies didn’t do summary analysis that would allow them to be summarized in this way. None of these meta-analyses found an advantage of more than 1kg for a low-carb-type diet.

Even if I found some reason to focus on the systematic reviews that found a slight advantage for low-carb diets, and ignore the other systematic reviews, such a small observed advantage would be hard to reconcile with the hypothesis that carbohydrate consumption is the dominant driver of obesity.

I wouldn’t be surprised if there are 1-3 SRs/MAs (which meet my criteria29) that I failed to list here, but I would be very surprised if there are six such SRs/MAs that I failed to list here.

I haven’t evaluated any of these SRs/MAs, but I’ll share my impressions of the state of the evidence from my surface-level evaluation. I would be somewhat surprised, but not very surprised, if a thorough evaluation of these recent SRs/MAs revealed that some of them show serious methodological bias or error, such that the RCT evidence concerning non-usual diets with differing macronutrient ratios pointed moderately strongly in favor of a particular non-usual diet. I would not be surprised at all if a thorough evaluation of these SRs/MAs resulted in the conclusion that the available RCT evidence does not suggest a diet “winner” (from mere macronutrient proportions, anyway).

What do animal studies say about the carbs-obesity hypothesis?

(My process described here.)

Non-systematic literature reviews of animal models used to study human obesity tend to focus on their use for investigating the biomechanisms of obesity, so I didn’t look much at those studies.

Studies on the effects of different diets on obesity in rats and mice tend to more commonly support low-fat rather than low-carb conclusions.30 But I didn’t survey this literature very thoroughly.

I found only one study (Klimentidis et al. 2011) tracking obesity changes in animal populations over time, and anyway this doesn’t tell us much about causes.

What do clinicians typically recommend?

(My process described here.)

Good Calories, Bad Calories suggests that clinicians typically recommend low-carb diets. In a brief search, I found only one survey of health care practitioners which reported on how frequently practitioners recommended one macronutrient proportion compared to others. Phelan et al. (2009) reported that low-fat recommendations were slightly more common than low-carb recommendations among New England area physicians. But this is unlikely to be a representative sample of U.S. physicians in general, or especially physicians around the world. I decided not to look further.

Tentative conclusions (overall)

See here.

A vet of an earlier draft of this document

Holden Karnofsky asked two summer research analysts to vet an earlier version of the document above. This earlier version had similar substantive content, but was organized differently. To shed more light on our process, we reproduce below the exact output of one of these vets.

Holden’s instructions for the vet process were provided as comments on the original draft of the vet; the vet below addresses them explicitly.

The output of the vet, addressed to Holden, is below. The author of this part is Isabel Arjmand.

  • “Vetter: Can you falsify this, using Luke’s method (described in process section) or otherwise?”
    • Luke’s claim: “I haven’t yet found any serious recent analysis that argues directly that HGs generally ate a high-carb diet.”
    • My overview: Found some sources that were unsure how many carbs HGs ate, or claim different HG societies ate diets with vastly different amounts of carbs (most promising is here: Link), but (as Luke found), none that claim that HGs in general ate high-carb diets.
    • Luke’s process:
      • Skimmed the list of papers citing the series of papers by Eaton, Konner, and Cordain
      • Found the most “promising” of those papers, then looked at papers they cite and papers that cite them
      • One more iteration of this
      • Searched Google Scholar for related terms
    • My process:
      • Google search for “hunter gatherers ate high carb diet” – 1 scientific paper on the first three pages of results.
        • Link – really just argues for diversity of diets/macronutrient breakdowns, but doesn’t specifically talk about high carb diets. Also not a “serious recent analysis” – reads more like an editorial.
      • Google Scholar search for “high carb diet hunter-gatherer” and for “high-carbohydrate diet hunter-gatherer.” Looked at first 3 pages of results, found one relevant paper (which I only read the abstract of).
        • “Diets of modern hunter-gatherers vary substantially in their carbohydrate content depending on ecoenvironments: results from an ethnographic analysis” Link: Claims carb intake ranged from 3% to 50% of energy intake. However, seems like it was generally found to be 35% or less, which I guess would be classified as low-carb.
      • Quickly repeating Luke’s process:
        • Started with this paper: Link
        • Looked at the articles that cite it (there are 551). Ran a search for “high-carbohydrate,” which narrowed it down to 111. Most were about high-carbohydrate diets being bad for health. Also most take “high-carbohydrate” as a synonym for “Western” and in opposition to hunter-gatherer diets. Haven’t seen any articles that go further than saying we don’t know exactly what proportion of HG diets was carbs – none argue that it was definitely high-carb.
        • I did find this: “Brand Miller & Colagiuri (1994) suggest that an insulin-resistant genotype evolved to provide survival and reproductive advantage to populations adapted to high meat, low carbohydrate diets. They argue that primates evolved on a high carbohydrate diet, with dietary glucose being the main fuel for reproductive tissues and the brain. However, during the ice age over, approximately, the past 2 million years, the diet changed to high protein, low carbohydrate, and metabolic adaptations were needed to accommodate the low glucose supply.” Link
        • The book The Nutrition Transition: Diet and Disease in the Developing World includes this paragraph: “In general the diet was more varied in these traditional, preagricultural societies (Truswell, 1977). Seasonality and the need to combine hunting and gathering activities ensured this greater variety…. Fat content in the diet was low because of its high carbohydrate content and the low proportion of saturated fat in the meat consumed.” Doesn’t expand on this claim, or say whether the diet was really high-carb or just high-protein, medium-carb, and low-fat.
        • Looked up Truswell 1977: Could only find an abstract, but it’s specifically about the !Kung Bushmen of Southern Africa. Link
  • “Vetter: Can you find important sources Luke didn’t include, in say 20 mins of searching? Is Luke’s outline consistent with the sources he does list?”
    • My overview: The four promising books that I found are listed below, arranged approximately in order of how useful I think they might be. Also, I went through the 13 questions in the outline to make sure the conclusions Luke draws make reasonable sense. All of them do.
    • Needed to ask Lynette/Holden: Not sure I understand how to approach the second question of whether Luke’s outline is consistent with the sources he lists. (Response from Holden: Confirm that Luke’s conclusions are consistent with his cited sources.)
    • Luke’s process:
      • Read Good Calories, Bad Calories
      • Run Google, Google Scholar, and Amazon searches (post-2005 results only)
      • Read 2 more books: The Art and Science of Low Carbohydrate Living and Perfect Health Diet
    • My process:
      • 20 min. on Amazon
        • Started by looking at books frequently bought with the 3 books Luke used.
        • Taubes has another, newer book called Why We Get Fat: And What to Do About It – probably some of the same arguments as his previous books
        • The Big Fat Surprise: Why Butter, Meat and Cheese Belong in a Healthy Diet by Nina Teicholz got great reviews from The NYTimes and The Economist, was published in 2014, focuses more on the pitfalls of a low-fat diet than the pitfalls of a high-carb diet, but they’re 2 sides of the same coin, to a degree.
        • Low Carb, High Fat Food Revolution: Advice and Recipes to Improve Your Health and Reduce Your Weight by a Swedish family practice physician – not sure how scientific it is.
        • Real Food: What to Eat and Why by Nina Planck – argues that fats like beef and butter are healthy after all.
        • Some books about the Paleo diet came up, which I didn’t list here. I also tended to avoid books called “The X Diet” where X is the name of the “expert” who wrote the book.
      • My process for making sure Luke’s outline is consistent with sources:
        • I went number by number through Luke’s outline.
        • 1. What do observational studies suggest?
          • Luke isn’t really able to draw any conclusions here, which makes sense given the available evidence.
        • 2. Did our hunter-gatherer (HG) ancestors typically eat low-carb diets?
          • Also unclear based on the evidence; this conclusion makes sense based on the limited research I did. I do wonder what answering this question would establish, other than that *maybe* our ‘natural state’ is to eat a low-carb (or high-carb) diet. Doesn’t seem like it would prove much about modern-day obesity.
        • 3. Was the carbs-obesity hypothesis the common wisdom prior to the 1970s?
          • Found the section in GCBC that claims this.
        • 4. What do we know about the biomechanisms of fat accumulation? Does what we know support the carbs-obesity hypothesis?
          • Luke doesn’t check, for reasons that make sense. Still seems like a question worth including in the outline.
        • 5. Do intervention studies of low-carb diets show they work better than alternative non-usual diets?
          • I didn’t check the studies Luke cited to make sure the way he described their conclusions was accurate – I can do that if you’d like. Including the studies he includes makes sense, based on my recollection of coming across most of them at various points during the vet.
        • 6. Are high-carb diets generally used for intentional fattening?
          • Doesn’t look like Luke spent much time looking at it; from what he wrote, it sounds like an argument not worth looking at more.
        • 7. Is it hard to fatten people on high-fat, high-protein diets?
          • Looked at some of the studies Guyenet cites; seems like solid evidence.
        • 8. Do clinicians generally prefer to advise low-carb diets?
          • Doesn’t look like Luke has looked into it enough to draw conclusions. Seems like it might not be a very important question anyway.
        • 9. What does work on animal models say about the carbs-obesity hypothesis?
          • Doesn’t look like Luke has looked into it enough to draw conclusions.
        • 10. What about the evidence from “self-cannibalization”?
          • Looked at ch. 3 of PHD to confirm that this is the question laid out.
        • 11. What about the evidence from breast milk composition?
          • Looked at ch. 4 of PHD to confirm that this is the question laid out.
        • 12. What about the evidence from mammalian diets?
          • Looked at ch. 5 of PHD to confirm that this is the question laid out.
        • 13. What about evidence from the food reward system?
          • Looked at ch. 6 of PHD to confirm that this is the question laid out.
  • “Vetter: Using similar methods to Luke, can you find additional studies he should have included? [Systematic reviews and meta-analyses of RCTs]”
    • My overview: Only found 1 study where I wasn’t sure why Luke didn’t include it. (Link Schwingshackl and Hoffman 2013 – compared low-fat high-protein diets to low-fat low-protein diets.) Luke includes a different study by the same authors, though.
    • Luke’s process:
      • Search Google Scholar for terms like “low-carbohydrate” and “systematic review,” restricted to somewhat recent SRs/MAs
      • Looked at studies that cited those meta-analyses
    • My process:
      • Searched Google Scholar for “low-carbohydrate” + “systematic review.” Looked at the first 30 results.
        • Link Noto, et al. 2013 – looks at mortality not weight loss
        • Link Santos, et al. 2012 – doesn’t seem to compare low-carb diet to another non-standard diet
        • Link Clifton, et al. 2014 – is included by Luke
        • Link Hu, et al. 2012 – is included by Luke
        • Link Bueno, et al. 2013 – is included by Luke
        • Link Naude, et al. 2014 – is included by Luke
        • Link Schwingshackl and Hoffman 2013 – is included by Luke
      • Searched Google Scholar for “low-carbohydrate” and “meta-analysis.” Looked at first 30 results.
        • Link Wycherley, et al. 2012 – is included by Luke
        • Link Chowdhury, et al. 2014 – focuses on fatty acids and unsaturated v. saturated, not micronutrient ratios
        • Link Fung, et al. 2010 – doesn’t seem to be intervention-based
        • Link Kastorini, et al. 2011 – same authors as the 2011 Esposito study Luke includes, but this one seems to only look at Med diet
        • Link Hooper, et al. 2012 – looks at high-fat v. low-fat diets, but doesn’t seem to measure carb intake consistently
        • Link Nordmann, et al. 2011 – is included by Luke
        • Link Esposito, et al. 2011 – is included by Luke
        • Link Schwingshackl and Hoffman 2013 – compared low-fat high-protein diets to low-fat low-protein diets. Includes 15 RCTs. Uses obesity/weight loss as an outcome. Not sure why Luke didn’t include it.
        • Link Yancy, et al. 2010 – compares low-carb diet to low-fat diet plus diet pills. Presumably not included because of the diet pills.
        • Link Santesso, et al. 2012 – presumably not included because it compares low-protein to high-protein diets, without a focus on fat v. carb breakdown.
        • Link Johnston, et al 2014 – is included by Luke
      • I ignored studies that focused only on subjects with diabetes, since I think Luke’s aims may have been more general than to assess the impact of low-carb diets on people with diabetes.
      • I came across more than half (9/14) of the studies Luke included in my searches, which I think is a sign that I was looking in the right places. I only found one study, highlighted in red above, where I was unsure why Luke wouldn’t have included it. (He did include another study by the same authors, for what that’s worth.)
      • I didn’t look at the studies that cite studies Luke includes. I can go back and do that if it would be helpful.
  • “Vetter: Using similar methods to Luke, can you find additional [large-scale between-population studies of macronutrient ratios and obesity] he should have included? (This might be easier after “promising” is more defined.)”
    • My overview: Didn’t find anything Luke should have included.
    • Luke’s process
      • Searched Google Scholar for: obesity correlation “macronutrient distribution” OR “low-fat” OR “low-carbohydrate” OR “macronutrient ratios” OR macronutrients AND “ecological study” OR “ecologic study” OR “between populations”
      • Found 2: Bray & Popkin 1998 and Willett & Leibel 2002
    • My process
      • Repeated Luke’s keyword search, looked at first 30 results
        • I’m not convinced that any of the studies I found ought to be included – I did list some of the more interesting results below.
        • Almost every study that came up evaluates something more broad/complex than ratio of carbs to other macronutrients.
        • Link This isn’t a study, but it’s a critique of Bray & Popkin that might be interesting
        • Link I don’t think this study (on page 11) meets the criteria Luke would be looking for, but it does look at macronutrient ratios and finds no relationship between macronutrient percentages and obesity when controlling for total energy intake.
      • Ran my own search with similar keywords: obesity “macronutrient distribution” OR “low-fat” OR “low-carbohydrate” OR “macronutrient ratios” OR macronutrients AND “between-population” OR “between populations”
        • Willett’s papers were the 3 top results
        • Found no other promising papers
  • “Vetter: Using similar methods to Luke, can you find additional studies he should have included? [Prospective cohort studies]”
    • My overview: Only found one study where I wasn’t sure why Luke didn’t include it. (“Systematic review of diet quality indices and their associations with health-related outcomes in children and adolescents” Link.) Other than that, I didn’t turn up anything interesting.
    • Luke’s process:
      • Started with USDA (2015), looked at where its data came from, looked at USDA (2014) and where its data came from.
      • Took the 8 high-quality studies identified and looked for systematic reviews citing them.
      • Skimmed titles and abstracts for promising reviews of prospective cohort studies.
    • My process:
      • Looked at articles citing Aljadani 2013 (nothing promising)
      • Looked at USDA 2014, page 19, which finds 10 prospective cohort studies of positive quality: Berz, 2011; Beunza, 2010; Cheng, 2010; Gao, 2008; Kesse-Guyot, 2009; Lassale, 2012; Mendez, 2006; Romaguera, 2010; Rumawas, 2009; Zamora, 2010 – a fair number of these focus only on the Med diet
      • I looked for systematic reviews citing Berz, et al 2011 (Link) – no promising results
      • I looked for systematic reviews citing Cheng, et al 2010 (Link)
        • Found this paper: “Systematic review of diet quality indices and their associations with health-related outcomes in children and adolescents” Link and am not sure why Luke didn’t include it although it does seem to focus on childhood nutrition and something called “diet quality indices.” I can only read the abstract.
      • Looked at USDA 2015 – wasn’t able to figure out where Luke had found the list of places he says its data comes from (Esposito, Jensen, USDA 2014, & Ambrosini)
      • Repeated Luke’s searches for the other 8 prospective cohort studies
        • Beunza 2010: The Fogelholm, Aljadani, and García-Fernández reviews come up. Nothing else promising.
        • Gao 2008: Nothing promising.
        • Kesse-Guyot 2009: Indirectly came across this: “Obesity and the Mediterranean diet: a systematic review of observational and intervention studies” Link. Possibly excluded by Luke because it only uses 3 cohort studies.
        • Lassale 2012: Nothing promising.
        • Mendez 2010: Nothing promising not already included.
        • Romaguera 2010: Turns up Fogelholm, Aljadani and Grosso, already included. Found this: “Consumer response to healthy eating, physical activity and weight-related recommendations: a systematic review” Link which focuses on dietary guidelines in general, not particular dietary adjustments, so probably not relevant.
        • Runawas 2009: Turns up Kastorini and Grosso, already included. Nothing else promising.
        • Zamora 2010: Turns up Aljadani and Fogelholm, already included. Nothing else promising.

My response to this vet

My response to the vet above, addressed to Holden, is below.

  • Isabel didn’t find any exceptions to my finding that “I haven’t yet found any serious recent analysis that argues directly that HGs generally ate a high-carb diet.”
  • Re: my search for “recent (post-2005) sources that came up when searching… for proponents or opponents of ‘low-carb’ or ‘paleo’ or ‘ketogenic’ diets (including e.g. proponents of ‘low-fat’ and ‘Mediterranean’ diets)… [which] spent much effort arguing from many different angles like Taubes’ book does.” Isabel’s four findings were Why We Get Fat, The Big Fat Surprise, Low Carb, High Fat Food Revolution, and Real Food.
    • I had read Why We Get Fat, but it just repeats stuff from GCBC.
    • I had looked at The Big Fat Surprise earlier, but as Isabel says, at a glance it seems to focus more on the pitfalls of low-fat diets than on the evidence for or against high-carb diets, so I didn’t investigate it further.
    • I hadn’t encountered Low Carb, High Fat Food Revolution in my searches. Glancing at it today, it seemed like it might fit my search criteria, so I purchased and skimmed it. The tone is polemical and casual. As with Taubes, it spends a lot of time on topics not immediately relevant to my investigation (e.g. diabetes, the effects of sugars specifically, cholesterol). But I think it might have a couple points or sources that are useful, but it would take quite a while to evaluate, since it is not remotely organized argumentatively, it has few footnotes, and the references at the end are just organized by chapter and then by general topic, so there’s no indication in the text what he does and doesn’t give sources for, and for most things he doesn’t give sources. My takeaway is that it probably fits the criteria for my search, though upon skimming it I’m not sure it’s worth to trying to extract additional points or sources from it because it would be so time consuming given how it’s organized. I’m not sure whether I should have come across this book given the speed at which I was searching, since Isabel doesn’t specify which Amazon searches she used. For comparison, the books I found (incl. WWGF and Big Fat Surprise) are all Amazon-ranked #6517 or better except for Perfect Health Diet which is ranked #16,981. This book and the next one (Real Food) are ranked around #35000.
    • I hadn’t encountered Real Food in my searches. As Isabel notes, at a glance is focuses on the healthiness of fats like beef and butter, not on the evidence for or against low-carb diets. So I don’t think this book fits my search criteria.
  • In response to your question “Is Luke’s outline consistent with the sources he does list?”, Isabel didn’t find any inconsistencies.
  • As for additional SRs or MAs of RCTs, Isabel found only one study for which she wasn’t sure why I hadn’t included: Schwingshackl & Hoffmann (2013). I had indeed found this study during my search, but had excluded it (along with 1 or 2 similar MAs) for the following reason given in my takeaways document: “I also did not include meta-analyses which mainly reported on differences between lower-protein and higher-protein diets (as opposed to e.g. low-fat vs. low-carb diets).”
  • Re: large-scale between-population studies of macronutrient ratios and obesity, Isabel didn’t find anything I should have included.
  • Re: SRs of prospective cohort studies. Isabel found only one study for which she wasn’t sure why I hadn’t included: Marshall et al. (2014). I don’t recall encountering this study in my searches. Checking the supplemental info, it turns out that only 8 of the included studies are prospective cohort studies, and all of them are on children rather than adults, which doesn’t fit my search criteria.

The main thing I might want to edit in my doc in response to this vet is to drop the exclusion criteria of “I also did not include meta-analyses which mainly reported on differences between lower-protein and higher-protein diets (as opposed to e.g. low-fat vs. low-carb diets)” for SRs and MAs. I added that exclusion criteria to save time and because the original project was more specifically focused on carbs-obesity rather than macronutrients more generally, but dropping this exclusion criteria would make that section of the review more comprehensive and less arbitrary-seeming.

I could also mine Low Carb, High Fat Food Revolution for additional points or sources, but I won’t unless you say so, because it doesn’t seem likely to be good ROI.

Comments to Isabel about the vet: The vet document was well-organized and useful. In retrospect, my only major request for future vets is that all web searches be listed precisely via hyperlinks. E.g. you could just say “On this question, I conducted a variety of Google Scholar and Amazon keyword searches (1, 2, 3, 4, 5, 6, 7, 8, 9)” — where each number in the parenthetical is hyperlinked to the exact Google Scholar or Amazon search URL, which includes e.g. date restrictions. I also wish the vet would have been more thorough (more time searching), but I presume there was some kind of time limit placed on you.

Dr. Stephan Guyenet’s comments on this report

One source I found particularly helpful in preparing this report was the blog of Dr. Stephan Guyenet, called Whole Health Source. After this report was published, I contacted Dr. Guyenet and said I’d be grateful to hear his feedback on my report if he had the time to give it. In September 2015 he sent me some comments on this report. I asked for and received his permission to publish those comments below.

COH = carbohydrate-obesity hypothesis.

I think it’s important to point out that researchers’ views on the COH depend very much on how it is stated. If you asked “do you think refined carbohydrate and sugar contribute to the development of obesity”, probably 95% of researchers would answer “yes”. If you asked “do you think carbohydrate and sugar are the dominant drivers of obesity”, probably at least 90% would answer “no”. The point is that there is good evidence, and I would argue a reasonably strong consensus among researchers, that carbohydrate CAN contribute to obesity, but that it is not the singular or dominant cause of obesity. The secondary point is that the carb-obesity debate is located mostly outside the scientific community. Those who are the most knowledgeable about obesity, human physiology, human dietary trials, and insulin biology generally find this debate to lack scientific substance.

I like that you ignored “bio-mechanism”-based arguments. I think one of the biggest pitfalls in the alt diet-health sphere is that people locate some mechanism that confirms their beliefs, then base advice on it without actually having empirical evidence to support the advice. This often gets pretty absurd.

Also good that you ignored common wisdom arguments, although for the record it was not traditional common wisdom in the US that carbs but not fat promote obesity. Just like today, some people avoided fats, others avoided carbs. To argue otherwise requires careful cherry picking.

You state that “low-carb diets are rare ‘in the wild’”. That depends on your definition of low-carb and “in the wild”. Most hunter-gatherers ate/eat less carbohydrate than we do in affluent nations today, and more protein. But they did eat some carbohydrate; typically more than what we would consider “very low-carbohydrate”. Most high-carbohydrate traditional cultures are agriculturalists and horticulturalists rather than HGs, although there are (or were) a few high-carb HGs. These cultures, as you noted, did/do not typically develop obesity. Also, there is good evidence that HGs typically ate substantial quantities of meat on a regular basis, although this varies by group and is not usually the case among agriculturalists. Katherine Milton has argued otherwise but she has very little human evidence on her side (she often argues from the perspective of non-human primate diets).

Regarding observational studies, in general they find that higher-carb diets are linked with leanness, while higher-fat diets are linked with weight gain (although I’m not sure how well that holds if you limit the analysis to prospective studies only). Most modern observational studies don’t ask this question so you have to go back to the older ones. This review by Hill and Prentice covers some of the evidence:

http://ajcn.nutrition.org/content/62/1/264S.abstract

These observational findings come with the usual caveats and I find them difficult to interpret. But at the very least, they are hard to reconcile with carbohydrate being the dominant player in obesity.

Regarding the RCT reviews you reviewed, there is an important point that emerges from this literature that you didn’t mention: even if LC does cause a modestly greater weight loss than LF, that still doesn’t offer convincing evidence that carbohydrate is the primary cause of obesity. In fact, if the difference is so small as to be difficult to detect in systematic reviews, it suggests to me that there’s nothing magical about carbohydrate.

I think another ball of wax you should consider separately is metabolic ward studies. One of the arguments that LC diet advocates such as Taubes make is that carbs cause fat gain independently of calorie intake, by slowing metabolism. This has been tested and falsified repeatedly in metabolic ward studies.

I think you are right to ignore the animal literature in this context. It’s important but it requires a substantial background to be able to navigate and interpret effectively. Also, as you said it tends to be more mechanism-focused, which is beyond the scope of your review.

Overall, I agree with your conclusions. I think the COH is a minimally scientific hypothesis that is mostly useful for selling books and cruises. That doesn’t mean low-carb diets are useless or that there is nothing to be learned from further research into carbohydrate restriction. I think there is. But the COH has been wrong for a long time, will continue to be wrong, and isn’t worth additional funding in my view.

I made several decisions about how to edit the report in response to Dr. Guyenet’s comments. Those decisions were:

  1. In response to Dr. Guyenet’s comments about how it’s important to be clear about the difference between (1) the hypothesis that carbohydrate consumption can contribute to obesity and (2) the hypothesis that carbohydrate consumption is the dominant cause of obesity, I changed the first sentence of my “overall tentative conclusion” paragraphs near the top of this report from “Though my investigation of the carbs-obesity hypothesis was brief and shallow, my impression at this point is that there isn’t good reason to think it is true” to “Though my investigation was brief and shallow, my impression is that the available evidence doesn’t fit the hypothesis that carbohydrates are the dominant cause of obesity.” This also fixed a problem I had noticed earlier, which is that my conclusion paragraphs originally said that there “isn’t good reason to think [the carbs-obesity hypothesis] is true,” when really the thing I had concluded, and suggested throughout the report, is stronger: that is, that the evidence doesn’t fit the carbs-obesity hypothesis. Especially, we have many relatively long-term RCTs on this question, and they probably should have turned out differently if the carbs-obesity hypothesis is true.
  2. In response to Dr. Guyenet’s comments about the ambiguity of saying “low-carb diets are rare ‘in the wild,’” I changed the sentence which originally read “Low-carb diets are rare ‘in the wild’ ” to “Low-carb diets are rare ‘in the wild,’ at least in recent decades and perhaps centuries.” At the end of this sentence, I added a footnote which reads: “Cordain et al. (2000) estimated that 58% of 20th century HG groups, nearly all of which are now gone (I suspect), ate 22%–40% carb. So, probably somewhat less than half of 20th century HG groups ate ≤35% carb, even if we believe low-carb enthusiasts like Cordain et al., and a negligible portion of of the world’s population belonged to those HG groups.”
  3. I didn’t make any edits in response to Dr. Guyenet’s comments on observational studies. The paper he linked, Hill & Prentice (1995), reviews some observational studies on carbohydrate consumption, but it does so in a non-systematic way, and the paper is mostly about sugars in particular rather than carbohydrates in general. I hadn’t discovered this paper during my carbs-obesity investigation, but I had discovered it later during my investigation of the health outcomes of sugar (still in progress as of October 2015). Though I found Hill & Prentice (1995) informative, I decided it wasn’t worth the effort to add a discussion of it to my report.
  4. I added a paragraph in response to Dr. Guyenet’s comment that “if the difference [in weight loss between low-carb and low-fat diets] is so small as to be difficult to detect in systematic reviews, it suggests to me that there’s nothing magical about carbohydrate.” This was also one of my own conclusions when I originally released the report, but this conclusion wasn’t made clear in the “What do randomized controlled trials suggest?” section. Thus, I added a paragraph to that section reading: “Even if I found some reason to focus on the systematic reviews that found a slight advantage for low-carb diets, and ignore the other systematic reviews, such a small observed advantage would be hard to reconcile with the hypothesis that carbohydrate consumption is the dominant driver of obesity.”
  5. Dr. Guyenet made a comment about metabolic ward studies. These are interesting, but I had decided not to discuss them in my report due to my report’s bracketing of bio-mechanism arguments.

Sources

Document Source
Ajala et al. (2013) Source (archive)
Alexander (2013) Source (archive)
Alexander (2014) Source (archive)
Alexander et al. (2012) Source (archive)
Aljadani et al. (2013) Source (archive)
Aljadani et al. (2015) Source (archive)
Ambrosini et al. (2014) Source (archive)
Ashby et al. (2012) Source (archive)
Berz et al. (2011) Source (archive)
Beunza et al. (2010) Source (archive)
Beunza et al. (2010) search Source
Bleich et al. (2012) Source (archive)
Bleich et al. (2015) Source (archive)
Bocquier et al. (2005) Source (archive)
Boylan et al. (2012) Source (archive)
Bray (2008) Source (archive)
Bray and DeLany (1995) Source (archive)
Bray and Popkin (1998) Source (archive)
Bray and Popkin search Source
Bray et al. (1992) Source (archive)
Buckland et al. (2008) Source (archive)
Buckland et al. (2011) Source (archive)
Bueno et al. (2013) Source (archive)
Campbell (2014) Source (archive)
Campbell et al. (2000) Source (archive)
Casper et al. (2008) Source (archive)
Castañeda-González et al. (2011) Source (archive)
Chen et al. (1990) Source (archive)
Cheng et al. (2010) Source (archive)
Chowdhury et al. (2014) Source (archive)
Clifton et al. (2014) Source (archive)
Cordain et al. (2000) Source (archive)
Cordain et al. (2002) Source (archive)
Corsica and Spring (2008) Source (archive)
de Garine and Koppert (1991) Source (archive)
Dixon et al. (2008) Source (archive)
Dutton et al. (2014) Source (archive)
Eaton and Konner (1985) Source (archive)
Engstrom et al. (2013) Source (archive)
Epling et al. (2011) Source (archive)
Esposito et al. (2011) Source (archive)
Fairweather-Tait (2003) Source (archive)
Fogelholm et al. (2012) Source (archive)
Foster et al. (2003) Source (archive)
Freeman et al. (2006) Source (archive)
Fung et al. (2010) Source (archive)
Gao et al. (2008) Source (archive)
Garcia-Fernandez et al. (2014) Source (archive)
Gary Taubes Source (archive)
Glauser et al. (2015) Source (archive)
Grosso et al. (2014) Source (archive)
Guyenet (2011) Source (archive)
Guyenet (2014) Source (archive)
Guyenet, Whole Health Source Source (archive)
Hagen (1999) Source (archive)
Hariri and Thibault (2010) Source (archive)
Hartley (2013) Source (archive)
Harvey and Hill (2001) Source (archive)
Harvey et al. (2002) Source (archive)
Hill and Prentice (1995) Source (archive)
Hooper et al. (2012) Source (archive)
Hu et al. (2012) Source (archive)
Huber et al. (2011) Source (archive)
Jaminet (2011) Source (archive)
Jaminet and Jaminet (2013) Source (archive)
Jensen et al. (2013) Source (archive)
Johnston et al. (2014) Source (archive)
Junshi et al. search Source
Kaplan et al. (2000) Source (archive)
Kastorini et al. (2011) Source (archive)
Katsuura et al. (2014) Source (archive)
Katz and Meller (2014) Source (archive)
Kesse-Guyot et al. (2009) Source (archive)
Keyword search Source
Klimentidis et al. (2011) Source (archive)
Ko et al. (2008) Source (archive)
Konner and Eaton (2010) Source (archive)
Kristeller and Hoerr (1997) Source (archive)
Kuipers et al. (2010) Source (archive)
Lang et al. (2013) Source (archive)
Lassale et al. (2012) Source (archive)
Lindberg (2009) Source (archive)
Lissner and Heitmann (1995) Source
Loomis et al. (2001) Source (archive)
Marshall et al. (2014) Source (archive)
Marshall et al. (2014) supplemental information Source (archive)
Masterjohn (2010) Source (archive)
McConnon et al. (2013) Source (archive)
Mendez et al. (2010) Source (archive)
Milton (2000) Source (archive)
Naude et al. (2014) Source (archive)
Nestle (2000) Source (archive)
Nordmann et al. (2011) Source (archive)
Noto et al. (2013) Source (archive)
Park et al. (2005) Source (archive)
Phelan et al. (2009) Source (archive)
Phelan et al. (2009) PubMed search Source (archive)
Rehnqvist et al. (2013) Source (archive)
Romaguera et al. (2010) Source (archive)
Rumawas et al. (2009) Source (archive)
Sack et al. (2009) Source (archive)
Santesso et al. (2012) Source (archive)
Santos et al. (2012) Source (archive)
Schoeninger (2014) Source (archive)
Schwingshackl and Hoffmann (2013) Source (archive)
Schwingshackl and Hoffmann (2013)(2) Source (archive)
Search for systematic reviews citing Berz et al. (2011) Source
Search for systematic reviews citing Cheng et al. (2010) Source
Shifman et al. (2009) Source (archive)
Somannavar and Appajigol Source (archive)
Starostina and Dreval (2001) Source (archive)
Ströhle and Hahn (2011) Source (archive)
Taubes (2008) Source (archive)
Taubes (2008)(2) Source (archive)
Truswell (1977) Source (archive)
US Department of Agriculture (2014) Source (archive)
US Department of Agriculture and US Department of Health and Human Services (2015) Source (archive)
van Dillen et al. (2013) Source (archive)
Volek and Phinney (2011) Source (archive)
Warner et al. (2008) Source (archive)
Wiebe (2012) Source (archive)
Wilkins et al. (2009) Source (archive)
Willett (1998) Source (archive)
Willett and Leibel (2002) Source (archive)
Willett and Leibel search Source
Wycherley et al. (2012) Source (archive)
Yancy et al. (2010) Source (archive)
Zamora et al. (2010) Source (archive)
  • 1. The “calories in, calories out” model, assuming it is true, is not necessarily helpful. The problem is that for many people in many circumstances, human biology may make “reduce calories in order to reduce weight” very difficult to maintain in practice. Taubes uses the example of a growing child. Is it fair to say that the child is growing taller “because” she is consuming more calories than she is burning? Certainly she must be consuming more calories than she burns in order to grow, but that doesn’t mean the growth process is under her control — short of locking herself in a room devoid of food and drink and having her family pass her a limited number of calories each day, and probably feeling too hungry and exhausted to get anything done for most of each day. Her body is growing, and thus her body makes her feel very hungry when she isn’t consuming enough calories to grow, and makes her feel very tired when she is in danger of burning so many calories that she won’t have sufficient calories to grow. Similarly, many who try simple calorie reduction might find that hunger and exhaustion make it very difficult for them to reliably consume fewer calories than they burn. The question is: Are there particular diets which reliably allow most people to successfully consume fewer calories than they burn over the long run, resulting in sustained weight loss and thus better health outcomes?
  • 2. Freeman et al. (2006); Klimentidis et al. (2011).
  • 3. See e.g. Alexander, Bray, Guyenet, and Campbell. Taubes replied to Bray’s review here.
  • 4. See e.g. Alexander, Masterjohn.
  • 5. GCBC searchable phrases: “induce extensive fattening”, “sumo wrestlers of Japan”. Taubes’ two cited examples are sumo wrestlers and the Massa tribe of Northern Cameroon. On the latter, Taubes writes: “The Massa are cattle herders, and their staple diet is primarily milk [i.e. a low-carb diet]. This fattening, therefore, is accomplished by the addition of carbohydrates almost exclusively…” But as Guyenet points out, Taubes’ source for this, Garine & Koppert (1991), actually says the Massa diet is normally 81% carbs (Table VIII), and suggests that the villagers (males, anyway) tend to be neither overweight nor obese (Table IX).
  • 6. By “promising,” I typically mean (in the context of a literature search): “promising according to my intuitions about which studies might yield novel discoveries useful to this project, given the project’s goals and limitations.” So in this particular case, I was looking for things like: (1) sources which, given their titles or abstracts, probably reply directly to EKC or Milton on the subject of HG diets, rather than sources which, given their titles or abstracts, appear to be talking about something else entirely and probably just citing EKC or Milton to support some random aside; (2) sources which appear to be overview articles on the topic of HG diets in general, and thus which might cite other useful papers alongside their citations of EKC and/or Milton; and/or (3) sources which might be otherwise useful to the project, e.g. a a paper which appears to mostly be about “ecological” studies of macronutrient ratios and rates of obesity in large human populations.
  • 7. E.g. they explained their literature search strategy, provided their eligibility criteria, described their data extraction process, assessed risk of study bias according to a previously published protocol, and detailed their data analysis process.
  • 8. By “mostly” I mean that the majority of studies selected for the systematic meta-analysis had to compare at least one non-usual diet to at least one other non-usual diet (e.g. low-carb vs. low-fat, but not low-fat vs. the standard American diet).
  • 9. Lissner & Heitmann (1995); Chen et al. (1990).
  • 10. In this case, “promising” means “studies which seemed likely to report a new-to-me large-scale between-populations ecological study including measures of macronutrient proportions in diet and also rates of obesity.”
  • 11. In this case, “promising” means “studies that report or systematically review clinicians’ or other health experts’ recommendations in a way that can compare low-carb to other diets, for weight loss.”
  • 12. Cordain et al. (2000) estimated that 58% of 20th century HG groups, nearly all of which are now gone (I suspect), ate 22%–40% carb. So, probably somewhat less than half of 20th century HG groups ate ≤35% carb, even if we believe low-carb enthusiasts like Cordain et al., and a negligible portion of of the world’s population belonged to those HG groups.
  • 13. The first two are cited by Taubes in GCBC; search the ebook for “nothing but meat among the Inuit” and “Polynesian atolls of Tokelau”. The Inuit and the last two examples are cited by Volek & Phinney (2011).
  • 14. Lissner & Heitmann (1995); Chen et al. (1990).
  • 15. Well, I think so. The paper does not make it easy to distinguish the case-control studies from the prospective cohort studies, but it looks like they initially searched for case-control or prospective cohort studies, but all the studies they included were prospective cohort studies. I haven’t checked this.
  • 16. Eaton & Konner (1985)
  • 17. Cordain et al. (2000), Kuipers et al. (2010), and Konner & Eaton (2010). The calculations in the parenthetical are from Cordain et al. (2000), table 4.
  • 18. Strohle & Hahn (2011).
  • 19. Kaplan et al. (2000).
  • 20. E.g. see Lindberg (2009); Nestle (2000);
  • 21. E.g. see Milton (2000) and the other studies cited in Cordain et al. (2002) after the phrase “of food energy consumed.”
  • 22. E.g. see Milton (2000); Katz & Meller (2014); Lindberg (2009).
  • 23. Throughout this document I use “randomized controlled trial” to mean “experiment in which subjects are randomly allocated to two or more treatment conditions.” Some researchers use the term more narrowly.
  • 24. Note that Santos et al. (2012) doesn’t compare low-carb diets to other diets, but rather to baseline values. Similarly, Hooper et al. (2012) mostly compared low-fat diets to normal, unrestricted diets rather than to other non-usual diets such as a low-carb diet. Also, it must be noted that diet RCTs are not “blinded” as in drug trials, since experimenter’s can’t do much to hide the identity of the foods subjects consume. There are a few rare exceptions, e.g. Corsica & Spring (2008).
  • 25. Search: “The 16 studies included” & “Six studies were excluded from meta-analyses”. Count for n done by manually summing participant counts from the tables.
  • 26. Table 1 describes Buckland et al. (2011) as a “Multi-centre RCT” but it is a prospective cohort study, and thus not included here.
  • 27. Search: “Twelve studies described in 18 reports… provided evidence about different dietary interventions and weight loss.” Count for n done by manually summing participant counts from Summary Table 3.1.
  • 28. I’m not sure if by “hypocaloric” (in the body text) they mean the same thing as “energy restricted” (in the tables). If so, then summing the participant counts for the energy-restricted RCTs comes to roughly 2800, though I’m unclear on how to count the apparently partially energy-restricted studies, e.g. Stern et al. (2004), which in the “Energy restricted” column reads both “-500” and “no”. I didn’t take the time to understand this better.
  • 29. For my criteria, see this section.
  • 30. E.g. see Katsuura et al. (2014); Casper et al. (2008); Hariri & Thibault (2010).