DiRECT (Lancet 2018) is the first registered trial to set remission of type 2 diabetes as a primary outcome and to deliver the intervention in routine primary care rather than a research centre. It supplies two things the fabric did not hold: a quantified total-diet-replacement (TDR) protocol with a structured maintenance phase, and a remission-by-weight-loss dose-response — the %-body-weight -> outcome gradient that grounds the weight-loss lever with a magnitude.

Its pathophysiological premise is Taylor’s twin cycle hypothesis — that T2D of limited duration is driven by excess fat in liver and pancreas and is reversible by negative energy balance. Earlier mechanistic work had normalised liver insulin resistance within 7 days and pancreatic function over 8 weeks on a 600-700 kcal/day diet; DiRECT tests whether this is practicable at scale. (Lean et al., 2018) The premise is now held directly: «people with recent onset T2DM could regain normal glucose control and normal β-cell function when the fat content of the liver and the pancreas was decreased by a weight loss diet- ary regimen» — the depot-drawdown mechanism the remission gradient rides on -> Ectopic Fat and Depot-Specific Risk. (Taylor & Holman, 2014)

The stratum DiRECT’s BMI cut-off excludes — normal-weight T2D (a decision-change). DiRECT required BMI 27-45, so its evidence is silent on the normal-weight stratum (a third of the 1970s-80s UKPDS cohort; likely smaller today, as higher BMIs are now more prevalent per Taylor), and such patients are routinely not offered weight loss. Taylor’s personal-fat-threshold frame extends the lever there: T2D is one condition across BMI (36% of newly-diagnosed UKPDS patients had BMI <25), and reversal is «achievable equally readily by people with lower initial BMI» — indeed at a smaller required loss (UKPDS: ~13% at normal BMI vs ~21% for the whole cohort to reach normoglycaemia). Those UKPDS figures are fasting normoglycaemia (FPG <6.0 mmol/l) after a 3-4 month diet-only run-in, in people presenting at 8-10 mmol/l — not DiRECT-defined remission (HbA1c <6.5% off drugs), so they are not a remission dose for the lean stratum. (corrected 2026-10-06: scope narrowed from reversal to fasting normoglycaemia, Taylor chunk 01) So a normal-weight person with short-duration T2D is a candidate for the same energy-deficit lever, sub-threshold BMI notwithstanding — a recommendation DiRECT’s own population cannot license, which the hypothesised personal-fat-threshold mechanism supports only directionally (mechanism-grade; Taylor states it as a hypothesis — «We hypothesize that each individual has a personal fat threshold (PFT) which, if exceeded, makes likely the development of T2DM.» — and the equal-readiness claim rests on Taylor’s refs 30-31). (Taylor & Holman, 2014) (inferred from Taylor & Holman, 2014)

The intervention — an energy target, NOT a macronutrient one

The Counterweight-Plus programme, in three phases:

  • Total diet replacement: a low-energy formula diet, “825-853 kcal/day; 59% carbohydrate, 13% fat, 26% protein, 2% fibre”, for 3 months (extendable to 5). Note the 59% carbohydrate — this is high-carbohydrate by share of energy, but because energy is so low it is ~122-126 g/day of carbohydrate (825 x 0.59 / 4 to 853 x 0.59 / 4), just under (by 4-8 g) the <130 g/day low-carbohydrate cut-off Goldenberg uses. So during induction it is low in carbohydrate grams as well as in energy; the two cannot be separated in this phase. (Goldenberg et al., 2021; inferred from Lean et al., 2018)
  • Stepped food reintroduction: 2-8 weeks, “about 50% carbohydrate, 35% total fat, and 15% protein”.
  • Structured weight-loss maintenance: ongoing monthly visits, behaviour-change methods (including CBT elements), a step-count target.

«All oral antidiabetic and antihypertensive drugs were discontinued on day 1», reintroduced only per protocol if glucose or blood pressure required it. (Lean et al., 2018)

Population and design: aged 20-65, T2D diagnosed within 6 years, BMI 27-45, not on insulin. Open-label, cluster-randomised at 49 primary-care practices (the practice, not the patient, was the randomisation unit); 149 per group in the ITT population. Co-primary outcomes, analysed hierarchically: “weight loss of 15 kg or more, and remission of diabetes, defined as glycated haemoglobin (HbA1c) of less than 6.5% (<48 mmol/mol) after at least 2 months off all antidiabetic medications, from baseline to 12 months”. (Lean et al., 2018)

The headline result

Outcome (12 months, ITT)InterventionControlEffect
Weight loss >=15 kg36 (24%)0p<0.0001
Diabetes remission68 (46%)6 (4%)OR 19.7 (95% CI 7.8-49.8)
Mean weight change-10.0 kg (SD 8.0)-1.0 kg (3.7)adj diff -8.8 kg (-10.3 to -7.3)
HbA1c change-0.9%+0.1%adj diff -0.85% (-1.10 to -0.59)
QoL (EQ-5D VAS)+7.2-2.9adj diff +6.4 (2.5 to 10.3)

(Lean et al., 2018)

The 46% remission rate “greatly exceeded” the 22% deemed a-priori clinically important. On medication: at 12 months 73.6% of the intervention group were on zero oral antidiabetic drugs (vs 18.2% of control), and 68% remained off antihypertensives with no rise in mean blood pressure. (Lean et al., 2018)

The jewel — remission scales with weight loss (dose-response)

Pooling both randomised groups, remission rises monotonically with the weight loss maintained at 12 months. “Remission varied with weight loss in the whole study population”:

Weight change at 12 monthsRemission achievedRate
Weight gained0 of 760%
0-5 kg loss6 of 897%
5-10 kg loss19 of 5634%
10-15 kg loss16 of 2857%
>=15 kg loss31 of 3686%

(Lean et al., 2018)

The discussion states the composite the table implies: “86% of participants with at least 15 kg weight loss, and 73% of those with weight loss of 10 kg or more” achieved remission. (Lean et al., 2018)

A different design reports the same direction of per-kilogram gradient (type-F corroboration, independence CAPPED)

DiRECT’s dose-response is a within-trial association (responder confounding is live — see Limits). A separate observational design corroborates the same monotone gradient, reported across diets — which strengthens the direction without lifting the finding to [E-independent] (the independence is capped; see below). The diet-independence leg is not DiRECT’s: it is Churuangsuk’s judgement, plus ADDITION’s per-kg association as relayed by Churuangsuk (no composition contrast described). DiRECT’s gradient pools a single formula programme with usual care, so it cannot separate weight from composition. Reported in the Churuangsuk umbrella review: ADDITION-Cambridge, a population cohort of 867 newly-diagnosed T2D patients across 49 UK primary-care practices, recruited by screening (the study of «People with Screen-Detected Diabetes in Primary Care», so earlier-detected disease than a clinically-presenting population), followed 5 years — overall 30% remission — found «Every 1 kg of weight loss was associated with 7% higher chance of remission at 5 years, regardless of specific diet regimens or lifestyle interventions», and «Loss of >10% of baseline body weight in the first year after diagnosis was associated with 70% higher chance of remission at 5 years». (Churuangsuk et al., 2021)

Parameter table (op-weave 2a — the two are matched on the claim they jointly back, NOT on magnitude):

ParameterDiRECT (Lean 2018)ADDITION-Cambridge (via Churuangsuk 2022)Same quantity?
Designwithin-RCT dose-response, both arms pooled (conditions on achieved weight)population cohort, 49 practicesboth observational weight->remission associations; one nested in an RCT
PopulationT2D <6 yr, not on insulin, ~all whitenewly-diagnosed T2D, UK primary careoverlapping, both early-disease
Exposureweight-loss band at 12 moper 1 kg / >10% loss in yr 1different metric
Outcome / horizonremission (HbA1c<6.5% off meds), 12 moremission (definition not in held umbrella text), 5 yr~YES at best; different horizon
The claim backedremission rises with kg lost (one trial, formula arm and usual-care controls pooled)rises per kg lost, «regardless of specific diet regimens»PARTIAL — direction yes; diet-independence only as relayed by Churuangsuk for ADDITION (no composition contrast described)

The magnitudes are not pooled (12-mo band-rate vs 5-yr per-kg association); what is shared is the direction of the gradient (diet-independence comes only from ADDITION as relayed by Churuangsuk, with no composition contrast described; not from DiRECT). There IS a real independence gain on one axis — different datasets/institutions (Newcastle/Glasgow DiRECT vs Cambridge ADDITION). On design there is none: both are observational weight->remission associations (DiRECT’s is nested in an RCT but conditions on achieved weight, a post-randomisation variable). But the [E-independent] token is withdrawn (2026-08-08), for two reasons the earlier version missed:

  • Shared confounding. Both gradients are driven by the SAME dominant threat — responder confounding (people who lose more weight differ biologically, so the weight->remission slope is partly confounded in the RCT and the cohort). A second design that shares the incumbent’s main confounder does not neutralise it, so it cannot lift the finding the way a genuinely orthogonal design would.
  • The relay runs through DiRECT’s own PI. ADDITION-Cambridge is held ONLY via the Churuangsuk umbrella, whose byline is «…Simon J. Griffin … & Michael E. J. Lean» — Lean is DiRECT’s joint PI/first author and Griffin an ADDITION investigator. So the “independent corroboration” is read through a review co-written by the incumbent trial’s own lead; that is a lineage tell E forbids, disclosed nowhere before. Same shared-authorship class as the Fibre SACN⟂Reynolds demotion. So this is type-F corroboration by a different design, with independence capped — real but not [E-independent]; and transport beyond UK early-disease weight-centric-remission populations is corroborated by neither. (Churuangsuk et al., 2021; inferred from Lean et al., 2018)

Where DiRECT sits in the umbrella’s certainty-graded remission map

Churuangsuk’s GRADE table places total diet replacement (the two low-RoB RCTs, DiRECT + DIADEM-I) at the single GRADE-HIGH cell of the whole remission map — median 54% remission (range 46-61) vs 4-12% standard care, N=445 — above formula meal replacement (11%, moderate), Mediterranean (15%, low), and ketogenic (20%, very low). (Churuangsuk et al., 2021) So DiRECT is not an isolated strong trial: it anchors the one high-certainty conclusion in the graded T2D remission literature (graded by a review co-authored by DiRECT’s PI Lean — see the lineage note above), and that conclusion is a format (energy-controlling TDR), not a macronutrient class -> Diets for Weight Management in Type 2 Diabetes.

Why this is the high-value finding. A monotone gradient from 0% (weight gain) to 86% (>=15 kg), in a within-population comparison, is evidence that, within one trial, remission tracks kilograms lost. The gradient is drawn with «both randomised groups combined» (Lean et al., 2018), and the formula arm supplied every large loss: 36 intervention participants lost 15 kg or more and «no participants in the control group» did (Lean et al., 2018). So weight band and programme move together along it, and the gradient cannot separate weight from programme or composition. The diet-independence reading comes from elsewhere: Churuangsuk’s umbrella (co-authored by Lean) concludes «The main contributor to HbA1c reduction and remission appears to be weight loss, irrespective of diet type.» (Churuangsuk et al., 2021) — right after cautioning that «residual bias/confounding may remain» in the non-randomised studies — and ADDITION’s per-kg association is relayed «regardless of specific diet regimens» (no composition contrast described). The DiRECT gradient is a cross-group association (not the randomised contrast), so it carries confounding risk (a responder who loses more may differ), but the biological gradient is steep and coherent with the mechanistic twin-cycle work — work from the same research programme (Taylor is DiRECT’s senior author, and the twin-cycle papers DiRECT cites, refs 4-5, are Taylor’s), so coherence here is not an independent leg. Read it as name-the-curve: the remission rate is higher in each successive 5-kg band over the studied range (weight gain to >=15 kg), with no visible plateau below 15 kg. As an association across five bands, it does not establish that each added kilogram causes more remission. Whether it plateaus above 15 kg is unresolved (the top category is open-ended and the mechanism has a ceiling — remission is bounded at 100%). (inferred from Lean et al., 2018)

The maintenance phase — the structure the fabric lacked

DiRECT is unusual in designing for maintenance “from the outset”. Among engagers, weight fell sharply in TDR then partly regained: -14.5 kg (95% CI 13.4-15.5) during TDR, then +1.0 kg during food reintroduction and +1.9 kg during maintenance — phase means among engagers that sum to ~-11.6 kg (not a single-cohort mean), against the -10.0 kg ITT mean. (Lean et al., 2018)

Adherence: “79% completed the intensive total diet replacement phase”; the ~25% “dropout rate” (6 never engaged, 26 withdrew — 15 during TDR, 6 during reintroduction, 5 during maintenance) marks the programme’s non-acceptability ceiling. This is the missing-piece for the NICE TDR recommendation, which names a low-energy TDR but not the reintroduction+maintenance scaffold -> Diets for Weight Loss - What NICE Recommends. (DiRECT’s 825-853 kcal formula sits at the bottom of NICE’s stated 800-1200 kcal band, as held on that page; no source on this page states the band.)

If attempting it — two safety steps the sources state (trial protocol / author advice, not trial-tested as separate interventions). (i) Review blood-pressure and glucose-lowering drugs with the prescriber before starting an ~800 kcal diet. DiRECT withdrew both on day 1 under protocol (above); for the BP drugs the stated reason is: «Weight loss leads to a rapid and marked fall in blood pressure, with risk of postural hypotension if antihypertensive drugs are continued. The acute fall in blood pressure with a low energy formula diet is greater than anticipated from reduced salt intake alone.6 For that reason, all diuretic and antihypertensive medications were withdrawn at the start of the total diabetes replacement phase in participants in the intervention group, and only restarted if systolic blood pressure exceeded 140 mm Hg.» (Lean et al., 2018) Churuangsuk adds the glucose-lowering side: «Blood pressure falls with weight loss, and postural hypotension, common in older people and those with diabetes, is aggravated during rapid weight loss if diuretic or antihypertensive drugs are taken concurrently [73]. Hypoglycaemia is possible if hypoglycaemic drugs are also taken [74]. Diets other than nutritionally complete formula diets could incur vitamin and mineral deficiencies [75].» (Churuangsuk et al., 2021) (ii) If retinopathy is present at baseline, rescreen at about 6 months: «Even if diabetes recurs, there might be a legacy effect of a period of good glucose control, as suggested in the UK Prospective Diabetes Study.35 Sudden restoration of normoglycaemia can precipitate worsening of diabetic retinopathy, although this outcome is rare when early or no retinopathy is present.35 Nonetheless, if retinopathy is present at baseline, rescreening at 6 months is indicated.35 It will be possible to determine the effects of DiRECT intervention by future analysis of the national retinopathy screening databases.» (Lean et al., 2018) Both are review-with-the-clinician flags, not dosing; the drug changes themselves are the prescriber’s. (Churuangsuk et al., 2021; inferred from Lean et al., 2018)

Durability at 24 months — remission tracks the weight kept off

The 24-month report of the same trial (same 298 ITT participants, same definitions, same team — so type-F durability, never type-E) answers the 12-month paper’s open question. About a third of the arm were in remission at 24 months (not all the same people as at 12 months: 48 remitters held, some new), down by about a fifth from the 12-month level. The held text is the author-accepted manuscript (Draft 7, 7 Feb 2019), without main Tables 1-2; figures should be checked against the published version before a deliverable quotes them: «Findings: At 24 months, 53/149 (35·6%) of those commencing the intervention and 5/149 (3·4%) in the control group (adjusted odds ratio 25·8, 95% CI 8·3,80·8; p<0·0001) had remission» (M. E. J. Lean et al., 2019)

Parameter table — 12 vs 24 months (op-weave 2a):

Quantity12 mo (Lean 2018)24 mo (Lean 2019)Same quantity?
Remission, intervention (ITT /149)68 (46%)53 (35.6%)YES — same definition, same missing=not-remitted rule
Remission, control (ITT /149)6 (4%)5 (3.4%)YES
Weight loss >=15 kg, intervention36 (24.2%)17 (11.4%)YES
Adjusted between-arm weight difference-8.8 kg-5.43 kg (-6.87 to -3.99)YES
Remission in the >=15 kg band (pooled arms)31/36 (86.1%)14/20 (70.0%)NO — the band holds whoever is >=15 kg down at that visit, so it is a different, smaller set of people (true of every band row)
Remission, <5 kg band (pooled)6/165 (3.6%)8/154 (5.2%)NO as people (bands re-sort at each visit); same bin edges — and this <5 kg bin includes weight gain (the 12-month table above splits it into 0/76 + 6/89)
Per-kg OR for remission (Fig 2C)1.32 (1.23-1.41)1.25 (1.16-1.35)~YES — same figure and model, but fitted to a different weight-change distribution at each visit

(M. E. J. Lean et al., 2019) 12-month weight difference -8.8 kg (-10.3 to -7.3): (M. E. Lean et al., 2018)

  • Absolute effect. The between-arm remission difference is about 32 percentage points at 24 months (35.6% vs 3.4%) against about 42 at 12 months — roughly 3 people started on the programme per additional 2-year remission vs usual care, in this stratum. The NNT inherits the remission definition’s medication asymmetry (22/119 control participants on drugs had HbA1c <48 mmol/mol but count as non-remitters; see Limits). (M. E. J. Lean et al., 2019) The odds ratios swing with the model’s adjustment set: imputed-outcome 25.8 in the results (M. E. J. Lean et al., 2019), vs 15.87 in Figure 2B (M. E. J. Lean et al., 2019). With only 5 control remitters, the counts are the stable quantity. (M. E. Lean et al., 2018; inferred from M. E. J. Lean et al., 2019)
  • The gradient survives. «For the entire study population, remissions at 24 months were achieved by 8/154 (5.2%) participants who lost less than 5 kg, 21/73 (28·8%) who maintained 5–10 kg loss, 15/25 (60·0%) who maintained» (M. E. J. Lean et al., 2019) 10-15 kg, and 14/20 (70%) at >=15 kg. Still monotone over the studied range (<5 to >=15 kg); the top band is lower than at 12 months, but that compares different people (row 5 above), not one person’s falling odds. (inferred from M. E. J. Lean et al., 2019)
  • Relapse went with more regain (descriptively; the stayer-vs-relapser contrast is p=0.073). Of the 63 intervention participants in remission at 12 months with weight at both visits, 48 stayed in remission and 15 relapsed; the stayers regained 4.25 kg (SD 3.68) in year 2, the relapsers 7.09 kg (SD 5.42), t-test p=0.073. Over the full two years the stayers were -10.4 kg net, the relapsers -3.7 kg. (M. E. J. Lean et al., 2019) The authors read this as «achieving and maintaining weight loss is the dominant factor behind remission of type diabetes» (M. E. J. Lean et al., 2019) — consistent with the dose-response, though the regain difference itself does not reach p<0.05. A different analysis, the post-hoc model over all 298 participants with both arms pooled, modelling 24-month remission (not relapse among the 63 remitters), found remission varied significantly «with weight- change from 12 to 24 months (adjusted odds ratio per kg gained 1.11 (1·03, 1·21), p=0·010)» (M. E. J. Lean et al., 2019) — the sign is ambiguous as printed (an OR above 1 per kg gained, listed among factors remission «increased with»), so read it as a significant regain–remission association of unclear coding. The stated adjustment set is «All models were adjusted for treatment, practice list size, centre and a random effect for practice.» (M. E. J. Lean et al., 2019) Weight loss from baseline (OR 0.83 per kg) is reported in the same sentence; the text does not say whether the two weight terms share a model. (inferred from M. E. J. Lean et al., 2019)
  • Keeping it off took repeated rescue. «about half required relapse management with brief total diet replacement and the offer of orlistat during the two years: 71/143 (49·7%) had not had any ‘rescue plan’, 49/143 (34·3%) had one, 15/143 (10·5%) had two and 8/143 (5·6%) had three or more rescue plan phases.» (M. E. J. Lean et al., 2019) The protocol behind that summary is two-tier: «In the event of weight regain >2kg, participants were offered a ‘rescue plan’ of 2-4 weeks partial meal replacement, and if >4kg a total diet replacement and food reintroduction, with the offer of orlistat treatment.» (M. E. J. Lean et al., 2019) Attendance averaged 7.7 of 12 monthly visits in year 2, and the intervention arm regained 2.6 kg (SD 5.0) between 12 and 24 months while control lost 1.3 kg. (M. E. J. Lean et al., 2019)
  • Within the stratum, BMI did not predict 2-year remission; duration was imprecise. «Likelihood of remission at 24 months was not influenced by baseline BMI (adjusted odds ratio per kg/m2 0·99 (0·92, 1·06), p=0·77) or duration of diabetes within the 6-year range included (adjusted odds ratio per year 0·92 (0·76, 1·11), p=0·39).» (M. E. J. Lean et al., 2019) Men (female OR 0.44) and older participants (OR 1.08 per year) remitted more often, and the arm-by-factor interaction tests were null (sex p=0.31, duration p=0.11). (M. E. J. Lean et al., 2019) These are post-hoc, pooled-arm prognostic associations (route (a)), not evidence of effect modification (route (b)), and not a basis to triage people in or out. The duration OR (0.92 per year, about 0.6 across the 6-year range) points the same way as the 2018 caution in Limits — insufficient evidence, not a null. Part of the age gradient may be built in: non-attenders were younger and the 26 untraceable were counted as non-remitters. (inferred from M. E. J. Lean et al., 2019)
  • Secondary outcomes held, smaller. HbA1c adjusted difference -4.82 mmol/mol with antidiabetic-drug use 39.5% vs 83.9% of control; SBP -3.43 mmHg with fewer on antihypertensives (47.3% vs 60.1%); QoL (EQ-5D VAS) adjusted difference 4.6 (0.4-8.9) (M. E. J. Lean et al., 2019), down from +6.4 (2.5-10.3) at 12 months (M. E. Lean et al., 2018).
  • Which bias hits which estimate. In the ITT contrast, contamination and attrition push against the intervention; the remission definition’s medication-withdrawal asymmetry pushes the other way (control participants kept their drugs, so some with HbA1c <48 mmol/mol count as non-remitters — see the NNT note above and Limits). Contamination shrinks it: «A proportion of the control group took personal action to lose weight (9 participants in the control group lost >10 kg during the second year compared to 2 during the first year).» (M. E. J. Lean et al., 2019) Attrition was unequal (77.9% vs 94.6% attended), handled by counting the 26 untraceable participants as non-remitters (M. E. J. Lean et al., 2019); non-attenders were younger, with shorter diabetes duration and more microalbuminuria (HbA1c higher, p=0.075) (Table S1). (M. E. J. Lean et al., 2019) Counting the intervention arm’s extra non-attenders as non-remitters also lowers its ITT rate. The selected analyses run the other way: the available-data rate (53/129, 41.1%, incl. GP-record data), the weight bands and the maintainer-vs-relapser contrast all condition on who stayed and how much they lost. (inferred from M. E. J. Lean et al., 2019)

Decision consequence. For the DiRECT stratum (age 20-65, BMI 27-45, T2D <6 years, not on insulin, UK primary care), the 12-month 46% figure overstates what a person should expect: the 2-year ITT rate is about one in three on the programme. Remission tracked weight kept off (post-hoc, pooled arms, among those with data: 29/45, 64.4%, at >=10 kg), but it was neither guaranteed by it nor impossible without it — 4 of the 50 who gained weight, each with a small (0-2 kg) gain and baseline HbA1c 6.5-6.63%, were in remission at both 12 and 24 months. DiRECT’s maintenance phase included monthly visits and rescue TDR, used by about half the arm; whether remission lasts as well without them is untested. The fabric holds no evidence beyond 2 years for this route; the trial was funded to 3 years with consent for 5. The plain action: for adults with T2D, overweight, and a recent diagnosis (the DiRECT stratum above), a structured TDR programme with food reintroduction, maintenance visits and rescue for regain is worth offering — expect about 1 in 3 in remission at 2 years, roughly 3 people started per extra 2-year remission vs usual care, with the safety steps above. Basis: one randomised ITT contrast at 12 and 24 months, its twin RCT (DIADEM-I) in Churuangsuk’s GRADE-high cell; not tested outside this stratum or beyond 2 years. (M. E. Lean et al., 2018; inferred from M. E. J. Lean et al., 2019)

Does a fast, large formula-diet loss itself raise regain? (added 2026-11-02). A meta-regression of regain after mostly diet-induced loss (drug, surgical and moderate-to-vigorous-exercise loss excluded) names this setting. Its authors read an amount-by-rate interaction as suggesting that someone planning a large loss «may wish to consider a more conservative method, whereas if only a small amount of weight loss is planned, the rate at which it is lost may not affect subsequent regain», and add: «In particular, this could be have implications in areas where large amounts of weight lost by VLCD are recommended, such as in diabetes treatment [89].» The interaction was «(p=0.042) (figure 1), although this reduced to a trend after adjustment for BMI (p=0.09).» Faster loss went with more regain on its own (p=0.049), but rate was no longer a significant predictor of regain once amount was adjusted for (p=0.42). The larger signal is amount: more weight lost went with more regain (beta 0.50 kg regained per kg lost, 0.25 to 0.74, p<0.001; net loss still grows with the amount lost), still significant with rate in the model (p=0.001). The only covariate is baseline BMI; the authors could not adjust for exercise, diet composition, or behaviour-change interventions and their dose in some studies. (Turicchi et al., 2019) (inferred from Turicchi et al., 2019) That is a between-study association in group-level data from studies selected for having regained. Its pool admitted adults with type 2 diabetes (Turicchi et al., 2019) but reports no diabetes subgroup and no remission outcome, so it does not move the plain action above. DiRECT’s own within-trial regain and rescue data are the more direct evidence on regain and remission in this stratum, though DiRECT has no slower-loss arm and so cannot test rate. Its diet is «total diet replacement (825–853 kcal/day formula diet for 3–5 months)» (M. E. Lean et al., 2018). Whether a slower loss to the same weight would keep more people in remission is untested in the held fabric -> Weight-Loss Maintenance and Metabolic Adaptation.

Updated 2026-11-03. One randomised trial now compares a fast formula loss with a slower one to the same target, but outside this stratum: Purcell 2014 excluded diabetes («the presence of clinically signifi cant disease (including diabetes)»). A 12-week VLCD (450-800 kcal/day) got 78% of starters to a >=12.5% loss against 51% on a 36-week diet with one or two shakes a day (6 vs 18 dietitian visits in the loss phase, so the arms differ as packages, not by pace alone), and among those who reached the target and completed 144 weeks of maintenance, regain was 10·3 vs 10·4 kg (difference 0·09 kg, -2·8 to 2·6); both arms had regained most of the loss (70·5% vs 71·2% of it; 76·3% in each arm by intention to treat). (Purcell et al., 2014) (inferred from Purcell et al., 2014) For non-smoking adults with BMI 30-45 and no clinically significant disease it gives no support to fast formula loss raises regain, within a post-randomisation subgroup that the two arms filled at different rates. It has no diabetes participants and no remission outcome, and its fast arm (450-800 kcal/day for 12 weeks) is deeper than DiRECT’s (825-853 kcal/day) and about as long as DiRECT’s 3-month minimum (DiRECT ran 3-5 months), so it does not move the plain action above; transport to type 2 diabetes is an assumption, not a finding.

Synthesis — is remission driven by WEIGHT LOSS or by CARBOHYDRATE RESTRICTION?

The fabric holds a low-carbohydrate remission result (Carbohydrate Restriction and Type 2 Diabetes Remission, Goldenberg 2021). DiRECT achieves remission on a formula diet that is 59% carbohydrate by energy — so the two routes must be reconciled. Parameter table (op-weave 2a):

ParameterDiRECT (Lean 2018)Goldenberg 2021 (BMJ MA)Same quantity?
Outcomeremission = HbA1c <6.5%, off antidiabetic meds >=2 months, 12 moremission = HbA1c <6.5% or fasting glucose <7.0 mmol/L, with or without medication (both reported), 6/12 moYES only on Goldenberg’s med-free definition (pooled RD 0.05 NS at 6 mo; -0.04 NS at 12 mo; non-insulin-trial subgroup 0.20, 0.03 to 0.38, at 6 mo only); the RD 0.32 headline permits meds
Intervention825-853 kcal TDR formula, 59% carbohydrate = ~122-126 g/day<130 g/day or <26% of energy from carbohydrate (of a 2000 kcal reference, where 26% = 130 g)NO as a share of the formula’s own energy (59% vs 26%); on Goldenberg’s 2000 kcal basis both are <130 g/day during induction (DiRECT just under)
Mechanism claimweight loss / negative energy balance (twin cycle)carbohydrate restrictionNO — the contested variable
Weight lossadj diff -8.8 kg vs usual care (-10.3 to -7.3; within-arm -10.0 kg), dose-response to remissionMD -3.46 kg vs comparator diets at 6 mo (18 RCTs); +0.29 kg, NS, at 12 mo (7 RCTs, 6 of them also in the 6-month pool)NO — different comparators; DiRECT’s difference much larger

(Goldenberg et al., 2021) for the Goldenberg column (remission definition, RDs, weight MDs); DiRECT weight difference (M. E. Lean et al., 2018); grams/day and the 26%-of-2000-kcal equivalence are arithmetic.

The fourth column is NO on mechanism and only partly matched on intervention and outcome — so this is NOT a like-for-like tension to file. It is an F-refinement on the causal question, with a narrower reach than a first reading suggests. Both routes produce weight loss. On DiRECT’s medication-free definition, Goldenberg’s pooled low-carb effect is non-significant at 6 and 12 months (RD 0.05, then -0.04, GRADE low). In the trials that excluded insulin users, as DiRECT did, it is significant at six months: «Studies that included patients using insulin had fewer remissions for both definitions of remission (HbA~1c~ <6.5%; HbA~1c~ <6.5% and no diabetes medication) at six months (risk difference 0.14, 0.03 to 0.25; 0.00, —0.07 to 0.07) compared with studies that did not (risk difference 0.51, 0.36 to 0.65; 0.20, 0.03 to 0.38) (test for subgroup difference P<0.001; P=0.03).» (Goldenberg et al., 2021) Goldenberg counts this among its credible subgroups: «Most subgroup observations were not deemed credible; however, three credible subgroups were identified on the basis of meeting four of five credibility criteria.» (Goldenberg et al., 2021) (The sentence after the subgroup result concerns the very-low-carbohydrate weight subgroup, not insulin.) The split is reported at six months only, and it compares trials, not patients within a trial; the pooled 12-month medication-free estimate (-0.04) is the only long-term read, and it says little about the non-insulin stratum: two of the three 12-month remission trials (Tay, Sato) included insulin users (Goldenberg et al., 2021). On the medication-free definition the insulin split may also coincide with Goldenberg’s post-hoc trial-length split, which gives the same 0.20 / 0.00 figures (0.20 in trials of 3 to under 6 months’ length, 0.00 in trials of 6-9 months), so it may be one finding, not two; with medication permitted the figures differ (insulin split 0.51 vs 0.14, trial-length split 0.49 vs 0.25) (Goldenberg et al., 2021), so the two partitions are not identical. So low-carb shows a six-month drug-free remission signal in trials that, like DiRECT, excluded insulin users, not an established effect -> Carbohydrate Restriction and Type 2 Diabetes Remission. (inferred from Goldenberg et al., 2021) But:

  • DiRECT’s within-population dose-response ties remission to kilograms lost within one trial, formula arm and usual-care controls pooled (it does not vary composition, so it cannot itself show diet-independence — that leg rests on Churuangsuk’s judgement and its relay of ADDITION, which describes no composition contrast; above). What DiRECT can show about carbohydrate: remission was achieved on a formula with a high carbohydrate share, and it was measured at 12 and 24 months, after structured food reintroduction at «about 50% carbohydrate, 35% total fat, and 15% protein» and the maintenance phase (M. E. Lean et al., 2018). What it cannot show: that remission occurs without carbohydrate restriction in grams. During induction the formula supplied ~122-126 g/day (59% of ~840 kcal), 4-8 g under Goldenberg’s 130 g/day cut-off; energy and absolute carbohydrate restriction are confounded in this phase. Whether carbohydrate restriction adds a remission effect beyond the weight it produces is untested here — insufficient evidence, not shown absent.

  • Goldenberg’s own data are consistent with weight-mediation but do not discriminate it: its low-carb remission and weight advantage are both non-significant by 12 months (weight MD -3.46 kg at 6 mo -> +0.29 kg at 12 mo; remission on the medication-permissive definition RD 0.32 at 6 mo -> 0.10, NS, at 12 mo; on the medication-free definition, pooled, RD 0.05 NS at 6 mo, 5 studies, and -0.04 NS at 12 mo, 2 studies, both GRADE low). (Goldenberg et al., 2021) The 12-month pools are smaller sets, largely nested in the 6-month pools (7 vs 18 RCTs for weight, 6 of the 7 also in the 6-month pool, all except Yancy; 3 vs 8 for remission, all 3 in it), so this is a between-pool contrast, not observed regain. The two 12-month pools also barely overlap each other: only Tay is in both the weight and the remission pool, and two of the three remission trials (Morris, Sato) list 3- and 6-month interventions in Table 1. (Goldenberg et al., 2021) So weight and remission are not observed fading together in the same trials. At a near-zero 12-month weight gap, HbA1c, triglycerides and medication reduction still point toward low-carb, from pools of 8, 7 and 3 trials whose membership the held text does not list, so not shown to match the weight pool (figures on Carbohydrate Restriction and Type 2 Diabetes Remission). A remission effect that fades alongside the weight advantage fits a weight-loss effect delivered via a low-carb route, and fits equally well a carbohydrate effect that fades as adherence to restriction fades; the fading pattern does not discriminate them. (inferred from Goldenberg et al., 2021) Goldenberg itself leaves the cause open: «Of note, the limited number of studies with 12 month outcome data providing differing levels of support and having highly adherent versus less adherent intervention arms precluded subgroup analyses that explicitly explored the effects of adherence at 12 months. Although improvements noted at six months diminished by 12 months, determining with any certainty whether this is related to intensity of intervention and/or dietary adherence beyond six months is difficult.» (Goldenberg et al., 2021) (The sentences before it report that very-low-carbohydrate diets underperformed less restrictive ones for six-month weight loss, a difference negated among highly adherent patients; that concerns restriction level and adherence, not the weight-vs-carbohydrate question.)

  • The closest held test of carbohydrate beyond energy is null and imprecise. Goldenberg’s prespecified calorie-matching subgroup: «For this reason, as part of our a priori planned subgroup analysis, we investigated the effect of calorically matched controls (as assessed by follow-up dietary questionnaires). On the basis of 18 studies providing adequate data, we identified no evidence of credible effect modification based on caloric matching or lack thereof. However, self-reported dietary intake data are prone to measurement error, particularly in dietary trials in which participants are not blinded.» (Goldenberg et al., 2021) The preceding sentence gives the reason for the analysis: because carbohydrate restriction tends to reduce hunger, it was unclear whether any benefit came from carbohydrate or from calorie restriction. The test matches calories, not weight, and rests on self-reported intake, so it fits insufficient evidence, not a shown absence.

  • One held RCT datum leans against a pure weight reading. Churuangsuk relays Esposito: 15% vs 4% remission at 1 year with a Mediterranean diet vs low-fat, on mean weight loss of 6.2 vs 4.2 kg (GRADE low). (Churuangsuk et al., 2021) The contrast is Mediterranean vs low-fat, so carbohydrate is not isolated; it is weighed on Carbohydrate Restriction and Type 2 Diabetes Remission and does not move the verdict.

  • DiRECT’s own authors leave composition open: «Effective long-term weight management with a resetting of long-term energy consumption is clearly essential, but other factors contribute and there remain unanswered questions and debates about dietary approaches, and the optimal ratio of macronutrients.» (M. E. J. Lean et al., 2019) Their next sentences cite a non-randomised very-low-carbohydrate study with substantial weight loss, then state that meta-analyses of controlled trials show no important differences between high- and low-carbohydrate diets for weight control or HbA1c (their ref 31). (M. E. J. Lean et al., 2019)

The defensible composite: T2D remission (of short-duration disease) appears to be driven mainly by the magnitude of sustained weight loss. That is Churuangsuk’s hedged judgement, which notes residual confounding; DiRECT’s gradient is confounded with programme; ADDITION’s per-kg association is relayed without a composition comparison. Carbohydrate restriction is one lever for achieving weight loss. Whether it adds a remission effect beyond the weight loss it produces is insufficient evidence, and DiRECT’s authors list the optimal macronutrient ratio as unresolved. This refines — does not contradict — Goldenberg: the low-carb result holds at six months on HbA1c <6.5% with medication allowed (not significant by 12 months), and on the drug-free definition shows only a six-month signal in trials excluding insulin users, but its framing (low-carb for remission) is at least as well read as weight loss for remission, low-carb being one delivery route. A person choosing a diet for remission should aim first at the weight loss they can sustain; the held evidence does not show that the route matters beyond that, nor that it does not (Low-Carbohydrate vs Balanced-Carbohydrate Diets). (Goldenberg et al., 2021; inferred from M. E. Lean et al., 2018)

DiRECT states the same logic for surgery: «The essential mechanisms behind bariatric surgery are weight loss and decrease in body fat content, rather than any direct surgical effect.27,28,29,30,31 The very large weight losses targeted by bariatric surgery are not essential for achievement of remission of type 2 diabetes, as shown by the present data. Changes in intra-organ fat content and β-cell function in a subgroup of the DiRECT cohort will be reported separately.» (M. E. Lean et al., 2018) Lean 2018 does not discuss low-carbohydrate diets. Extending the point across three routes (formula diet, low-carb, surgery), a common thread is weight/fat loss; the route-independence of that thread is this page’s generalisation, not a source’s.

Remission is a real benefit weight loss moves — but it is intermediate

DiRECT is a counterweight to the cardiovascular-event null (Does Weight Loss Reduce Cardiovascular Events): where Look AHEAD showed lifestyle weight loss did not cut CV events, DiRECT shows the same lever delivers a patient-important benefit it demonstrably does move — freedom from diabetes and its drugs, plus a measured QoL gain. Weight loss is indicated for the outcomes it moves; remission is squarely one of them.

But remission is an intermediate outcome, not a hard one. HbA1c normalisation is expected (if maintained) to reduce microvascular risk — Lean’s words: «if maintained, can reasonably be expected to reduce microvascular complications» — but DiRECT was “not designed or powered to evaluate effects on complications” and remission can relapse. DiRECT itself reaches for the hard-outcome bridge only by citing a Look AHEAD post-hoc: “a 10% weight loss in the first year… was associated with a 21% decrease in occurrence of cardiovascular outcomes” — the same >=10%-responder signal that page holds as reported, not proven -> Does Weight Loss Reduce Cardiovascular Events. (M. E. Lean et al., 2018)

Limits and bias — read honestly

  • Open-label, and allocation was visible to participants (cluster design). Baseline balance suggests limited bias, but blinding of a lifestyle programme is impossible.
  • Asymmetric medication handling: antidiabetic drugs were stopped in the intervention arm but not control. The trial compares the whole programme vs standard care, so the remission definition (off-meds HbA1c) is favourable to the arm whose meds were deliberately withdrawn — a design choice, disclosed, that inflates the between-arm remission gap relative to a med-matched comparison.
  • 12-month horizon; durability is unaddressed here (follow-up planned to >=4 years). Remission of short-duration T2D only — “remission is less likely with longer durations of disease”. (Superseded in part 2026-10-06: the 24-month report is held — see Durability at 24 months above; nothing beyond 2 years is held. (inferred from M. E. J. Lean et al., 2019))
  • No hard outcomes at 24 months either. Year-2 serious adverse events were fewer in the intervention arm (9 in 6 participants vs 22 in 16), but «DiRECT was not powered to assess ‘hard’ clinical outcomes, but seeing fewer serious adverse events in the second year of weight management is reassuring, given the past anxiety over safety of older formula diets.» (M. E. J. Lean et al., 2019) Read it as a safety signal, not evidence of complication reduction.
  • Scottish/Tyneside population; no unqualified transport to e.g. South Asian groups who develop T2D at lower weight: «The racial and ethnic characteristics, while typical of the populations of Scotland and Tyneside, do not allow for unqualified extrapolation to other groups» (M. E. Lean et al., 2018) Baseline Table 1 gives the share: «Ethnicity (White) 147 (98·7) 146 (98·0)» (control, intervention). (M. E. Lean et al., 2018) (corrected 2026-10-06: an earlier same-day note said no ethnicity percentage was located; Table 1 has it)
  • The dose-response is a within-population association, not the randomised contrast — responder confounding is possible.
  • The twin-cycle / personal-fat-threshold mechanism is a HYPOTHESIS, not a settled cause. Taylor’s depot-drawdown account is mechanistically corroborated (liver/pancreas fat falls with weight loss and tracks glucose recovery) but the personal fat threshold — that each person has an individual fat level above which T2D triggers — remains an inference under test, not an established quantity; the remission gradient is consistent with it but does not prove it. That corroboration comes from the same research programme as DiRECT (Taylor is DiRECT’s senior author; the held PFT paper is Taylor’s own), so it is coherence within one lineage, not an independent leg.
  • Sponsor / vendor COI: the programme (Counterweight-Plus) and the formula are commercial products. The trial was funded by Diabetes UK and «The formula diet was donated by Cambridge Weight Plan. Neither organisation had any input into the study design, data analysis or interpretation.» That disclaimer names Diabetes UK and Cambridge Weight Plan, not Counterweight Ltd — yet «HMR reports employment by Counterweight during the study» and «LM reports employment by Counterweight during the conduct of study», and «MEJL reports personal fees from Counterweight during the conduct of the study». (M. E. Lean et al., 2018) Disclosed here for symmetric standards — it does not overturn the RCT, but the honest-bias inventory must name it. (corrected 2026-10-06: generic vendor-stake line -> the source’s funder / donor / employment disclosures)
  • Adverse events during TDR were common, mostly mild or moderate, and transient (constipation, cold sensitivity, headache, dizziness); 9 serious AEs in 7 (4.5%) intervention participants, 2 possibly related (biliary colic + abdominal pain, same participant), none causing withdrawal.
  • Coherence, not validity (R1): a strong single trial, not proof the effect transports or persists.

(inferred from M. E. Lean et al., 2018)

References

Churuangsuk, C., Hall, J., Reynolds, A., Griffin, S. J., Combet, E., & Lean, M. E. J. (2021). Diets for weight management in adults with type 2 diabetes: an umbrella review of published meta-analyses and systematic review of trials of diets for diabetes remission. Diabetologia, 65(1), 14–36. https://doi.org/10.1007/s00125-021-05577-2
Goldenberg, J. Z., Day, A., Brinkworth, G. D., Sato, J., Yamada, S., Jönsson, T., Beardsley, J., Johnson, J. A., Thabane, L., & Johnston, B. C. (2021). Efficacy and safety of low and very low carbohydrate diets for type 2 diabetes remission: systematic review and meta-analysis of published and unpublished randomized trial data. BMJ, m4743. https://doi.org/10.1136/bmj.m4743
Lean, M. E. J., Leslie, W. S., Barnes, A. C., Brosnahan, N., Thom, G., McCombie, L., Peters, C., Zhyzhneuskaya, S., Al-Mrabeh, A., Hollingsworth, K. G., Rodrigues, A. M., Rehackova, L., Adamson, A. J., Sniehotta, F. F., Mathers, J. C., Ross, H. M., McIlvenna, Y., Welsh, P., Kean, S., … Taylor, R. (2019). Durability of a primary care-led weight-management intervention for remission of type 2 diabetes: 2-year results of the DiRECT open-label, cluster-randomised trial. The Lancet Diabetes &amp; Endocrinology, 7(5), 344–355. https://doi.org/10.1016/s2213-8587(19)30068-3
Lean, M. E., Leslie, W. S., Barnes, A. C., Brosnahan, N., Thom, G., McCombie, L., Peters, C., Zhyzhneuskaya, S., Al-Mrabeh, A., Hollingsworth, K. G., Rodrigues, A. M., Rehackova, L., Adamson, A. J., Sniehotta, F. F., Mathers, J. C., Ross, H. M., McIlvenna, Y., Stefanetti, R., Trenell, M., … Taylor, R. (2018). Primary care-led weight management for remission of type 2 diabetes (DiRECT): an open-label, cluster-randomised trial. The Lancet, 391(10120), 541–551. https://doi.org/10.1016/s0140-6736(17)33102-1
Purcell, K., Sumithran, P., Prendergast, L. A., Bouniu, C. J., Delbridge, E., & Proietto, J. (2014). The effect of rate of weight loss on long-term weight management: a randomised controlled trial. The Lancet Diabetes &amp; Endocrinology, 2(12), 954–962. https://doi.org/10.1016/s2213-8587(14)70200-1
Taylor, R., & Holman, R. R. (2014). Normal weight individuals who develop Type 2 diabetes: the personal fat threshold. Clinical Science, 128(7), 405–410. https://doi.org/10.1042/cs20140553
Turicchi, J., O’Driscoll, R., Finlayson, G., Beaulieu, K., Deighton, K., & Stubbs, R. J. (2019). Associations between the rate, amount, and composition of weight loss as predictors of spontaneous weight regain in adults achieving clinically significant weight loss: A systematic review and meta‐regression. Obesity Reviews, 20(7), 935–946. https://doi.org/10.1111/obr.12849