NICE NG246 Evidence Review F, January 2025. The wiki’s first source that names an intervention, a dose, a duration and a population — rather than an exposure and an association.
The recommendation
Low-energy total diet replacement (800-1200 kcal/day): “consider” — a deliberate weak recommendation, for people living with obesity, or with overweight plus type 2 diabetes.
Bounds, all stated:
- nutritionally complete
- maximum 12 weeks
- «should not be used as a long-term strategy to manage obesity»
- only within “a multicomponent overweight and obesity management strategy with long term support… offered within specialist overweight and obesity management services”
- offered “irrespective of diabetes duration” — extended beyond the trial evidence on committee reasoning, since “they were not aware of wider evidence on the relationship between the duration of type 2 diabetes and the likelihood of remission with weight loss”
[EXTRACTED (NICE - NG246 Evidence Review F) 1.1.10]
Why weak rather than strong, in the committee’s own words: “the committee were aware that, sometimes, weight regain could occur rapidly and be harmful if people experience weight cycling. Therefore, a weak recommendation was preferred to a stronger one as, in certain cases, a low-energy TDR might not be the most appropriate intervention.” Note this is strength lowered by a harm the evidence base did not measure — not by certainty.
The 600 kcal deficit was deleted for arbitrariness, not disconfirmation
NICE CG189 had recommended a 600 kcal/day deficit. NG246 removes it — “because they felt this was an arbitrarily specific number”, replacing it with “dietary approaches to support weight loss should maintain an energy deficit and this can be achieved by lowering specific macronutrient content.” [EXTRACTED (NICE - NG246 Evidence Review F) 1.1.10.3]
No evidence in the review tested 600 against any other deficit. A number can be withdrawn for having no warrant, without anything having shown it wrong — a distinct move from a refuted claim, and one the telos’s four evidence-states should record as such.
Low carbohydrate: assessed, and left without a recommendation
NICE’s entire low-carbohydrate verdict:
“Evidence on low-carbohydrate diets demonstrated that for almost all outcomes, the diet did not result in improvements in weight, BMI, waist circumference, or HbA1c, relative to conventional (usually low-fat) diets, apart from a small improvement in waist circumference for females. Evidence on very-low carbohydrate diets did show a reduction in weight at 1 year, but this was not sustained at 2 years and there was no improvement in quality of life for people with type 2 diabetes.” [EXTRACTED (NICE - NG246 Evidence Review F) 1.1.10.3]
No recommendation is attached to it — for it or against it. But NICE does make a research recommendation covering it, which is a different thing from silence: Appendix K research recommendation 3 (adverse events of dietary approaches) names both “Low carbohydrate diet (defined as under 130g of carbohydrates)” and “Very low carbohydrate (defined as under 50g of carbohydrates)” in its PICO. [EXTRACTED (NICE - NG246 Evidence Review F) Appendix K, RR3]
The evidence-vs-practice move, stated openly
“While evidence on macronutrient diets (low carbohydrate, very low carbohydrate, or low-fat diets) did not demonstrate an improvement in a number of outcomes including change in weight, in practice, variations of these diets are commonly used and have shown to be effective.”
“Have shown to be effective” carries no citation and contradicts the review’s own finding — and it is the immediate premise for the only change the committee made in this section. The same appeal to unreferenced practice experience appears four more times, but only here is it used to assert effectiveness against the review’s result. [EXTRACTED (NICE - NG246 Evidence Review F) 1.1.10.3]
Recorded as a process observation, not a verdict on the recommendation. Under the telos’s five divergence classes this is a candidate class-5 (process defect) — and the wiki cannot currently establish class 5, because that requires an independent institutional standard it does not hold. AWAITS NASEM - Clinical Practice Guidelines We Can Trust.
PARAMETER TABLE — rebuilt 2026-07-26 after the first version was falsified
The first version of this table was decoration, and it laundered errors rather than catching them.
It contained two fabricated quotations (a NICE cell reading identical in both arms — zero hits
in the source — and a Naude threshold that dropped the source’s >50 g floor and turned a conjunction
into a disjunction), an inverted comparator row, a delivered-carbohydrate range that omitted the
one trial the argument turned on, and no row for follow-up horizon, which is the parameter that
actually breaks the comparison. Rebuilt below with located quotes only.
| Parameter | NICE NG246 F | Naude 2022 | Goldenberg 2021 | Same quantity? |
|---|---|---|---|---|
| Low-carb definition | ”under 130g of carbohydrates" | "> 50 g to 150 g per day or < 45% of total energy intake” (non-ketogenic) | “less than 130 g/day or less than 26% of the 2000 kcal/day diet” | partly — NICE and Goldenberg share the 130 g arm |
| Very-low / ketogenic band | ”under 50g of carbohydrates" | "≤ 50 g per day or < 10% of total daily energy intake" | "<10% daily calories or <50 g/d” | YES — all three co-extensive |
| Follow-up horizon | 1 and 2 years | the restriction-depth subgroup exists only at 3 to <12 months | 6 and 12 months | NO — and this is decisive |
| Comparator | active diet arm and “No intervention” (Lim 2010 pooled) | “no dietary intervention” control excluded | mostly low-fat; one no-treatment arm | NO — inverted from what I first wrote |
| Outcome unit, HbA1c | ”% Change in HbA1c” | change in HbA1c (%) | change in HbA1c (%) | NO — NICE reports % CHANGE, not points |
What the honest table forces — and it reverses my first conclusion
Row 2 is the row I ignored, and it is the one that matters. All three sources define the deep band identically (<=50 g / <10%). NICE therefore has a stratum at Goldenberg’s exposure — and it is not null:
weight at 1 year
MD -2.38 (-4.22 to -0.55)“Favours very low carb diet”;% Change in HbA1c MD -0.38 (-0.66 to -0.1)“Favours very low carb diet” [EXTRACTED (NICE - NG246 Evidence Review F) Tables 25-26]
So the claim I first wrote — NICE’s result says nothing about Goldenberg’s exposure — is false, and it was reached by silently narrowing “NICE’s trials” to the low-carbohydrate stratum while the page’s headline drew on the whole review. NICE stratified by depth of restriction exactly as its protocol required, found a null in one band and a signal in the other, and I read the null as the review’s verdict.
Row 3 blocks the comparison anyway. Naude’s restriction-depth subgroup exists only at 3 to <12 months; every NICE low-carb row is at 1 or 2 years. There is no ≥12-month restriction-extent analysis in Naude to compare against. No cell, no claim — so no cross-source conclusion is drawn here.
And it would not be independent if it were drawn. NICE lists Naude 2022 among its six source systematic reviews — “Naude 2022 - 61 RCTs [19 matched our inclusion criteria]” — so 19 of NICE’s low-carb and very-low-carb primary studies come from the Cochrane review already held. Shared backing, never type-E. NICE separately excluded Goldenberg 2021 for “Incorrect analysis VLCK and LCD analysed together” — the exact lumping this section was worried about. [EXTRACTED (NICE - NG246 Evidence Review F) Table 2; excluded-studies table]
A shared trial sorted into opposite bands settles it: NICE classes Layman 2009 as very low carbohydrate; Naude places the same trial in its non-ketogenic subgroup. The bands are not co-extensive in application even where their definitions match.
What the TDR recommendation rests on — DiRECT supplies the trial and the maintenance scaffold
NICE’s low-energy TDR “consider” is the recommendation form of the DiRECT-type evidence the wiki now holds directly. DiRECT (Lean 2018) delivered an 825-853 kcal/day formula diet — at the bottom of NICE’s stated 800-1200 kcal band — for a matching population (overweight/obese + T2D), and achieved 46% diabetes remission with -10 kg mean loss -> Total Diet Replacement and Type 2 Diabetes Remission. Two points of contact:
- DiRECT fills the maintenance gap NICE names but does not specify. NICE bounds the TDR at 12 weeks inside a “multicomponent… strategy with long term support” but leaves the structure unstated; DiRECT’s stepped food reintroduction (2-8 weeks) + monthly-visit maintenance phase is the concrete scaffold, and its phase-wise regain data (+1.0 kg reintroduction, +1.9 kg maintenance after -14.5 kg TDR) quantify what maintenance must hold against.
- The weight-cycling harm NICE weighted its recommendation down for is visible in DiRECT — regain begins immediately after the TDR trough, and 24% withdrew — so NICE’s “weak not strong” call reads as calibrated, not over-cautious.
Decision relevance
- The only diet NICE will positively recommend is defined by ENERGY, not macronutrient — total diet replacement at 800-1200 kcal, for 12 weeks, with specialist support. But “every macronutrient-defined diet came out null” would be false — the very-low-carbohydrate stratum favours the intervention on weight at 1 year and on HbA1c. What is true is that no macronutrient-defined diet earned a recommendation, which is a different statement and the one to carry.
- “Consider” with a 12-week ceiling is a different instruction from eat this way. It is a time-boxed clinical intervention inside a service, not a dietary pattern.
- Weight cycling is treated as a real harm and is what lowered the recommendation’s strength — though no trial in the review measured it.
- A quoted low-carb null must be checked against its delivered exposure, not its label. NICE’s trials averaged ~35% energy carbohydrate; that is not what most readers mean by “low carb”.
Limits
- The numbered recommendation text (1.16.1-1.16.12) is not in this document — it lives in NG246 itself, so what is recorded here is the committee’s operative reasoning, not the recommendation wording.
- Ethnic-minority and low-socioeconomic-status subgroups: evidence absent, stated as such.
- The committee lists six trials lacking a control-arm deficit and the study tables flag Layman 2009 as a seventh; the recovered table truncates one row, so treat “seven” as unverified.