Low-carb or low-fat? What a landmark year-long trial found

In short

In DIETFITS, adults randomised to a healthy low-fat diet lost 5.3 kg over 12 months and those on a healthy low-carbohydrate diet lost 6.0 kg — a between-group difference of 0.7 kg (95% CI −0.2 to 1.6).
The trial was built to find a personalisation effect. Neither a three-SNP genotype pattern (P = .20) nor insulin secretion measured 30 minutes after a glucose load (P = .47) modified which diet worked better.
Individual weight change within each group spanned roughly 40 kg, from about 30 kg lost to 10 kg gained. The spread between people on the same diet dwarfed the gap between the diets.
Tightly controlled ward studies do detect macronutrient effects, but small ones, running opposite to the popular claim: across 32 feeding studies, fat loss was 16 g per day greater on the lower-fat diets.

A trial built to remove the usual confound

For two decades the low-carbohydrate and low-fat camps have traded trials, and most of those trials shared a design problem: the arms differed in food quality as well as macronutrient ratio. Vegetables, nuts and olive oil were set against pasta and low-fat biscuits, and no result could say which difference produced the outcome.

DIETFITS was designed to strip that confound away.¹ Gardner and colleagues randomised 609 adults — aged 18 to 50, body mass index between 28 and 40, none with diabetes — to a healthy low-fat or a healthy low-carbohydrate diet for 12 months. Both arms received identical food-quality instructions: maximize vegetable intake, minimize intake of added sugars, refined flours, and trans fats, and focus on minimally processed whole foods prepared at home. Neither arm was given an energy target: No explicit instructions for energy (kilocalories) restriction were given.

The protocol was not gentle. For eight weeks participants cut total fat or digestible carbohydrate to 20 g a day, then added it back at 5 to 15 g a week to the lowest level they believed they could sustain.

What separated the two groups: 0.7 kilograms

The diets diverged and stayed diverged. Carbohydrate in the low-carbohydrate arm fell from 246.5 g a day at baseline to 132.4 g at 12 months; fat in the low-fat arm fell from 87.0 g to 57.3 g. Both groups reported cutting roughly 500 to 600 kcal a day. Of the 609 randomised, 481 (79%) completed.

Table 1. DIETFITS at 12 months. At baseline, mean age was 40 (SD 7), 57% were women and mean BMI was 33 (SD 3). The between-group difference in weight change was 0.7 kg (95% CI −0.2 to 1.6).
Measure at 12 months	Healthy low-fat	Healthy low-carbohydrate
Carbohydrate, % of energy	48	30
Fat, % of energy	29	45
Protein, % of energy	21	23
Weight change	−5.3 kg	−6.0 kg

An 18-percentage-point gap in carbohydrate and a 16-point gap in fat, sustained for a year, produced a difference the trial could not distinguish from nothing. The interval — −0.2 to 1.6 kg — contains zero, and even its favourable end is small beside the 5 to 6 kg both groups lost.

The personalisation test, and why its failure matters

The more interesting half of DIETFITS was pre-registered before anyone saw a result. The first hypothesis was genetic: three single-nucleotide polymorphisms sorted participants into patterns predicted to favour fat restriction, patterns predicted to favour carbohydrate restriction, or neither — 244 participants (40%) carried the low-fat genotype and 180 (30%) the low-carbohydrate genotype. The second was metabolic: insulin concentration 30 minutes after a glucose load, a proxy for insulin secretion, was said to identify people for whom carbohydrate restriction should work better.

Neither held. The trial reported no significant diet-genotype pattern interaction (P = .20) or diet-insulin secretion (INS-30) interaction (P = .47) with 12-month weight loss. This is a null from a study sized to find the effect: the design assumed roughly 100 participants in each of the four genotype-by-diet cells, giving 80% power to detect clinically meaningful differences in treatment effect by genotype.

What that establishes is narrow but real. It does not prove that no genetic or metabolic modifier exists; it shows that these markers, specified in advance and tested in a well-powered trial, did not do what had been claimed for them. The authors note that INS-30 is one of many possible indices of insulin–glucose dynamics.

The pooled evidence agrees, at a different scale

Two network meta-analyses put that single trial in context. Johnston and colleagues pooled 48 unique randomised trials in 7,286 people, and ended by supporting recommending any diet that a patient will adhere to in order to lose weight.²

Table 2. Johnston and colleagues' pooled estimates: kg lost versus a no-diet control, with 95% credible intervals.
Follow-up	Low-carbohydrate	Low-fat
6 months	8.73 (7.27–10.20)	7.99 (6.01–9.92)
12 months	7.25 (5.33–9.25)	7.27 (5.26–9.34)

Figure 1. The estimates in Table 2, drawn to scale. Each is against a no-diet control, not against the other diet. Read the bars, not the dots: the intervals overlap heavily at six months, and by 12 months the point estimates sit on top of each other.

Ge and colleagues later assembled a larger network — 121 trials and 21,942 participants across 14 named diets and three control conditions.³ Their six-month estimates run about half the size: Compared with usual diet, low carbohydrate and low fat diets had a similar effect at six months on weight loss (4.63 v 4.37 kg, both moderate certainty), with a 95% credible interval of 3.42 to 5.87 kg on the low-carbohydrate figure. The gap between the reviews is mostly the comparator — “no diet” against “usual diet” — plus a larger, more recent trial set; on the question at issue they agree. By 12 months, average weight loss across all macronutrient patterns was 1 to 2 kg smaller, on low-certainty evidence, with cardiovascular improvements largely gone.

The metabolic ward finds real effects — and they are tiny

Free-living trials measure what people manage to do. Ward studies, where every gram is weighed and served, measure what macronutrients do when intake is taken out of the participant's hands. That evidence does not return a null.

Table 3. Controlled-feeding evidence. Both lines favour fat restriction — the opposite direction to the carbohydrate–insulin hypothesis — at magnitudes far below what a free-living study could detect.
Study	Design	Result
Hall et al., 2015 Cell Metabolism	19 adults with obesity; randomised metabolic-ward crossover — each did a 5-day baseline, then 6 days at 30% energy restriction, once from carbohydrate, once from fat	Body fat loss 53 ± 6 g/day on carbohydrate restriction vs 89 ± 6 g/day on fat restriction (P = 0.002). 24-hour energy expenditure fell 97.7 ± 23 kcal/day on the reduced-carbohydrate diet (P = 0.0007) and was unchanged on the reduced-fat diet.
Hall & Guo, 2017 Gastroenterology	Meta-analysis of 32 controlled feeding studies with isocaloric substitution of carbohydrate for fat	Energy expenditure 26 kcal/day and fat loss 16 g/day greater with the lower-fat diets (both P < .0001).

Hall and Guo's summary is worth quoting exactly: both energy expenditure (26 kcal/d; P <.0001) and fat loss (16 g/d; P <.0001) were greater with lower fat diets.⁵ The earlier ward trial ran the same way over six days.⁴ These results contradict a strong prediction — that carbohydrate restriction confers a metabolic advantage. But the effect sizes are the point: 16 g of fat a day, or 26 kcal, is far smaller than the error in anyone's estimate of their own intake, and six days in a ward is not twelve months in a kitchen.

What did predict who lost weight

The most striking number in DIETFITS is not the 0.7 kg but the spread. Individual weight change covered a similar range for weight change of approximately 40 kg within each group (−30 kg to 10 kg). Diet assignment explained little about where in that range a person landed.

A later analysis went looking for what did.⁶ Hauser and colleagues took the 448 participants with complete 24-hour dietary recalls at baseline and 12 months and split each arm at the median on 12-month change in diet quality, scored with the Healthy Eating Index-2010, and in net carbohydrate or fat grams, the adherence measure. Against the low-quality, low-adherence group, those high on both lost more — a difference in BMI change of −1.15 kg/m² in the low-carbohydrate arm (95% CI −2.04 to −0.26) and −1.11 kg/m² in the low-fat arm (−2.10 to −0.11). High quality alone, or high adherence alone, did not differ significantly.

Two caveats are load-bearing. These subgroups were not randomised, so this is an association within a trial rather than a causal estimate, and the exposure is self-reported recall. Still, it points where the meta-analyses point: the axis that separates outcomes is not which macronutrient you cut.

The bottom line

Choose the diet you can actually keep. A year of sustained macronutrient separation moved the scale by 0.7 kg, with a confidence interval crossing zero, and the two markers most promoted for matching people to diets failed a pre-registered test in the trial built to give them their best chance. Diet quality and adherence tracked with outcomes inside both arms. The useful question is not low-carb or low-fat; it is which pattern you will still be following next spring.

Following any of these patterns means knowing roughly what you ate. Kettle logs food macros — protein, carbohydrates, fat and energy — and water on iPhone, with the data staying on the device unless you write it to Apple Health.