Deep divefrom Episode 9Fat loss4 November 202618 min

Complete Meals, Low-Carb, and Four Weeks of Mostly Nulls

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: Optimized Nutri-Dense meals versus a low-carbohydrate high-fat diet: an exploratory randomized controlled trial (Scientific Reports)

MP3

Chapters

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

Cold open

Host 1You heard the trailer — this is the whole story. Welcome to the Hastrup Review of Science — where artificial intelligence reads real science, so you don't have to pretend you understood the paper. Today, three hundred adults in Tokyo, four weeks of meals built to hit twenty-six vitamins and minerals, and a low-carb version of the same meals.

Host 2The paper is Optimized Nutri-Dense meals versus a low-carbohydrate high-fat diet: an exploratory randomized controlled trial, by Hishikawa, Hayashi, Itoh and colleagues at Keio University, published in Scientific Reports on October third. We could only get the abstract and the article's front and back matter, not the full results, so everything here stays inside what those say. The short version is that the main outcomes did not move, and a few secondary ones did.

Host 1Then the first thing to get clear is who paid for it.

Host 2Nissin Foods funded the trial and supplied the meals, and six of the fifteen authors work for Nissin. The paper says the statistician was independent, the analysis plan was fixed before unblinding, and the sponsor had no veto over publication. The competing-interests line still reads none declared, which is worth keeping in mind.

Why this field exists

Host 1Let's start further back. Why does anyone run a trial like this at all?

Host 2As established background, nutrition science spent most of the twentieth century fighting about macronutrients, meaning the split between carbohydrate, fat and protein. Low-carbohydrate eating goes back at least to William Banting's pamphlet in eighteen sixty-three and came back with Robert Atkins in the nineteen seventies. Low-fat advice dominated official guidelines over roughly the same period.

Host 1And micronutrients?

Host 2Micronutrients, meaning vitamins and minerals, were the earlier story, about curing deficiency diseases like scurvy, rickets and beriberi. Once outright deficiency became rare in rich countries, they mostly dropped out of the obesity and heart-disease debate. The phrase people use now is hidden hunger, meaning people who eat enough calories but fall short on several micronutrients.

Host 1So this paper tries to bring that second story back into the first one.

Host 2That is the framing. The authors' opening line is that micronutrient adequacy, beyond macronutrient balance, may matter for health. The design tries to pull the two apart: hold the micronutrients fixed and change the macros, or hold the macros fixed and change the micronutrients.

What was known before

Host 1What was known before this paper?

Host 2On low-carb versus low-fat, the large trials mostly come out about even over the long run. As background, the DIETFITS trial at Stanford randomised six hundred and nine adults to healthy low-fat or healthy low-carb for a year, and weight loss did not differ meaningfully. Meta-analyses tend to find small, early advantages for low-carb that fade by twelve months.

Host 1And on blood lipids?

Host 2Low-carb, high-fat diets typically lower triglycerides and raise HDL, and LDL is the uncertain one. Depending on the fats used, LDL can rise substantially on some versions, especially those heavy in saturated fat. So an LCHF arm where LDL falls is not the default outcome, and that detail matters later.

Host 1What about feeding trials, where the researchers supply the food?

Host 2Those are the strongest designs in nutrition and also the rarest, because they are expensive. As background, Kevin Hall's inpatient studies at the National Institutes of Health fed people every meal under controlled conditions and showed, for example, that ultra-processed diets led people to eat about five hundred more calories a day. This trial sits in between, with meals provided but people living at home.

Host 1And epigenetic age, which turns up in the results?

Host 2As background, DNA-methylation clocks estimate biological age from chemical marks on DNA in blood. PhenoAge, published by Morgan Levine and colleagues in two thousand eighteen, was trained to track a composite of clinical blood markers that predict mortality. Diet trials have reported clock changes before, but those changes are usually small and noisy.

The open question

Host 1So what is the open question?

Host 2Whether simply making a diet nutritionally complete, without changing calories or macro ratios, does anything measurable to body fat, stress and sleep hormones, blood pressure or cholesterol. And separately, whether a low-carb, high-fat diet looks different when its micronutrients are equally complete. Most low-carb trials cannot answer that second one, because when you cut out grains and fruit, the micronutrient profile changes along with the macros.

Host 1So the comparison everyone usually makes is confounded.

Host 2Partly, yes. In a typical trial, low-carb versus low-fat is also a comparison of different vitamin and mineral intakes. This design tries to remove that.

What this paper did

Host 1Walk me through what they actually did. Start with the design.

Host 2Three arms, randomised, single-blind, four weeks, three hundred adults in total. It is labelled exploratory, and the abstract calls this a partial report, which matters for the primary outcomes. The registry number is UMIN zero zero zero zero five zero one zero three, and the ethics approval dates from January twenty twenty-three.

Host 1What were the three diets?

Host 2The first arm got Opti meals, built to the Japanese Dietary Reference Intakes for twenty-six micronutrients with a standard macronutrient ratio. The second was a control diet with the same macronutrient ratio but deliberately insufficient in micronutrients. The third was LCHF Opti, with a lower carbohydrate-to-fat ratio and identical micronutrients to the Opti meals.

Host 1What are the Japanese reference intakes?

Host 2As background, they are Japan's official nutrient targets, published by the Ministry of Health, Labour and Welfare and revised every five years, much like the reference intakes used in the United States or Europe. They set recommended amounts and upper limits for each vitamin and mineral by age and sex. The paper's abbreviation list runs from vitamin A, B and D through zinc, selenium, iodine and molybdenum, so this covers essentially the full set.

Host 1So the clean comparisons are which?

Host 2Opti against control tests micronutrients with macros held equal. Opti against LCHF Opti tests macros with micronutrients held equal. That is a neat design on paper.

Host 1Who was blinded?

Host 2The abstract does not say. With meals this different in carbohydrate and fat, participants could probably tell the low-carb arm from the others, so the blinding most plausibly applied to assessors or to the micronutrient contrast. That is my inference, not something the paper states.

Host 1Who were the participants?

Host 2Adults in Tokyo, recruited and managed by a contract research organisation and a clinic in the Ueno area. The abstract gives no age range, body mass index cutoff or sex split for the full sample. All we know is that older men were analysed as a subgroup.

Host 1What did they measure?

Host 2The primary outcomes were visceral fat area, meaning fat around the organs, and salivary cortisol and melatonin, which are markers of stress and of the sleep cycle. Prespecified secondary outcomes were LDL cholesterol, systolic and diastolic blood pressure, gut microbiota, the serum metabolome and DNA-methylation age. The authors say the secondary outcomes were analysed in an exploratory, hypothesis-generating way.

Host 1You said partial report. Partial how?

Host 2Part of the primary outcome set came from wearable devices, and those streams are being published separately. The abbreviation list includes heart rate variability and REM sleep, which hints at what those are. So we are not seeing the whole primary picture in this paper.

A word from our sponsor

Host 1Before the results, a short break.

Host 1Do you have something at home that has been divided into identical portions with no regard for what it contains?

Host 2The Hastrup Review of Science is brought to you by ice cube trays. Fourteen compartments, each optimised to the same reference intake of water, with an identical micronutrient profile across all arms.

Host 1I randomised one tray to the freezer and one to the counter. The freezer arm reached its primary endpoint, and the counter arm was hypothesis-generating.

Host 2Ice cube trays are sold in ordinary shops at ordinary prices. No contract research organisation, no single-blinding, no sponsor. The link is in the show notes.

Host 1Back to the paper. Results.

Results

Host 1Start with the primary outcomes.

Host 2Neither primary outcome differed between the groups. Visceral fat area did not separate between Opti, control and LCHF Opti, and neither did salivary cortisol or melatonin. The authors state it directly: the reference-intake meals did not outperform the comparators on the primary endpoints.

Host 1Is four weeks long enough to see visceral fat change?

Host 2Probably not, unless people also lost weight, and the design held macros and presumably energy roughly comparable. Visceral fat does respond to calorie restriction within weeks, but a change in micronutrient density at similar calories would be expected to produce a small effect, if any. So the null is not surprising, and it is informative.

Host 1Then the secondary outcomes.

Host 2LDL cholesterol fell in both the Opti and the LCHF Opti groups compared with control. Systolic blood pressure fell in the LCHF Opti group compared with control. The abstract gives no sizes for either, so I cannot tell you whether that is two milligrams per decilitre or twenty.

Host 1LDL fell on the low-carb, high-fat arm. You flagged that earlier.

Host 2Yes, and the comparison is the catch. Both drops are against the control, and the control was deliberately micronutrient-poor, not someone's usual diet. So the finding is that both complete diets beat an incomplete one, not that low-carb lowers LDL in general.

Host 1And Opti against LCHF Opti directly?

Host 2The abstract never reports that head-to-head contrast, even though it is in the title. We do not know whether the two optimised arms differed from each other on anything. That is one of the main things the full results would need to show.

Host 1Gut bacteria and metabolites?

Host 2The abstract only says exploratory analyses showed shifts in gut microbiota and metabolites. There is no direction, no named taxa and no effect sizes there, and the abbreviation list mentions false discovery rate correction without telling us which results survived it. With that many features, some shifts are almost guaranteed by chance.

Epigenetic age

Host 1Now the epigenetic age result, which is the one people will quote.

Host 2In a small subgroup of older male participants, ten per group, the Opti group showed lower PhenoAge. That is thirty people out of three hundred, selected by age and sex, after four weeks. The abstract does not say how much lower, and does not say whether the subgroup was prespecified.

Host 1How much weight can that carry?

Host 2Very little. With ten people per arm, a methylation clock's ordinary measurement noise is in the same range as the effects these trials report, and once you start looking at subgroups the chance of a false positive rises quickly. The authors call it hypothesis-generating, and that is the right label.

Host 1Why older men in particular?

Host 2The abstract does not explain it. One plausible reason is that methylation measurement is expensive, so it may have been run on a subset only. Either way, it is not a finding about the whole trial.

Predicting responders

Host 1And the machine learning part?

Host 2Post hoc, they trained models to predict who would be a cardiometabolic responder, and reported a ROC-AUC of zero point seven one to zero point seven six. The abbreviation list names light gradient boosting, random forests and support-vector classifiers. An AUC of zero point seven one means that if you pick one responder and one non-responder at random, the model ranks the responder higher about seventy-one percent of the time.

Host 1Is that good?

Host 2It is moderate. Whether it means anything depends on validation, meaning whether it was tested on people the model had not seen, and the abstract does not say. Post-hoc models on a few hundred people, with many candidate features and several algorithms, tend to look better than they will on the next cohort.

Why it matters

Host 1So what is the principle here?

Host 2The useful idea is the design itself, which separates micronutrient completeness from macronutrient ratio. Most diet comparisons mix the two, and this one at least tries to untangle them. That is worth copying, ideally for longer and with a usual-diet comparator.

Host 1What changes for whom?

Host 2For researchers, it is a template and a list of hypotheses, LDL and blood pressure especially. For anyone eating, nothing changes yet. Eating a nutritionally complete diet was already sensible advice, and this trial adds no new evidence that it reduces visceral fat or stress hormones in a month.

Host 1And what does it not mean?

Host 2It does not show that these meals reverse ageing; that rests on thirty older men and an unreported effect size. It does not show that low-carb diets lower LDL, because the comparator was a deficient diet. And it does not show that micronutrient-optimised meals beat low-carb, since the head-to-head is not reported in the abstract.

Host 1There is a product angle here, isn't there?

Host 2As background, Nissin sells meals marketed as nutritionally balanced, under the Kanzen Meal name in Japan. The abstract does not name a product, and the trial meals were described only as test meals, so I would not tie this paper to any specific item on a shelf. But it is reasonable to read the trial as a company testing the premise behind that category.

Honest caveats

Host 1Then the caveats, all of them.

Host 2One, both primary outcomes were null, and the trial says so. Two, four weeks is short for body composition and very short for anything called ageing. Three, the comparator was built to be deficient, which makes the optimised arms look good relative to a diet nobody would choose.

Host 1Keep going.

Host 2Four, there are many secondary outcomes, including microbiome and metabolome data with thousands of features, so some differences will appear by chance. Five, the epigenetic result is a ten-per-group subgroup, and the responder models are post hoc with no stated external validation. Six, this is a partial report, and part of the primary outcome set is published elsewhere, so the full picture is split across papers.

Host 1And the funding.

Host 2Seven, the sponsor makes food, provided the meals, and helped design the trial and write the paper, while declaring no competing interests. The safeguards they describe, an independent statistician and a locked analysis plan, are real and they help. Industry funding does not invalidate a trial, but it is a reason to want independent replication.

Host 1Anything about who was studied?

Host 2Eight, it is one site in Tokyo, with a Japanese reference standard and a Japanese baseline diet. As background, Japan's average diet differs a lot from European or American ones, including in fat intake and salt. And since we only have the abstract, there may be effect sizes and dropout numbers that change this reading in either direction.

What to watch next

Host 1What should we watch for next?

Host 2First, the companion paper with the wearable and sleep data, since that completes the primary outcomes. Second, the direct Opti versus LCHF Opti contrast and the actual LDL and blood pressure effect sizes. Third, any prespecified, adequately powered trial of micronutrient density on methylation clocks, ideally with a usual-diet control and three months or more.

Host 1And the machine learning?

Host 2Look for the responder model tested in a separate cohort. If it holds up at around zero point seven there, it starts to be useful. If it drops toward zero point five, it was fitting noise.

Bottom line

Host 1Bottom line?

Host 2Three hundred people, three diets, four weeks: nutritionally complete meals did not change visceral fat or stress and sleep hormones more than a deficient diet did. LDL fell on both complete diets and blood pressure fell on the low-carb one, against a control designed to be poor. The epigenetic age result in thirty older men is a reason to run a bigger trial, which is what the authors say too.

Sign-off

Host 1That is the deep dive: Optimized Nutri-Dense meals versus a low-carbohydrate high-fat diet: an exploratory randomized controlled trial, by Hishikawa, Hayashi, Itoh and colleagues, in Scientific Reports. The link is in the show notes.

Host 1Every Friday this same feed carries the roundup episode: five new papers as trailers, and you pick which become deep dives. This has been the Hastrup Review of Science.