The decision this reframes: how much to expect of lifestyle weight-loss maintenance, and why a structural or pharmacological prop may be the realistic route rather than a failure of resolve. The folk model treats regain after a diet as a lapse of willpower — old habits creeping back. Sumithran 2011 (NEJM) supplies the mechanistic alternative: after diet-induced loss the body’s appetite-regulating hormones shift in the direction that promotes eating and regain, and — the study’s contribution over prior acute work — those shifts are still present a full year later, after regain has already begun. The higher weight is being biologically defended. (inferred from Sumithran et al., 2011)
The finding — the hormonal shift persists at 12 months
Design (state it plainly, it bounds everything below): a single-arm, uncontrolled study. 50 overweight/ obese non-diabetic adults did an 8-week very-low-energy diet (VLCD, ~500-550 kcal/day meal replacement); hormones and appetite were measured at baseline, week 10 (post-loss), and week 62 (one year into a maintenance phase). Only the 34 completers are analysed (Sumithran et al., 2011). The 16 non-completers were not all dropouts: of the 50 who began, 4 withdrew on the VLCD and 5 during follow-up, but «an additional 7 participants did not lose the required 10% of body weight» — i.e. the analysed panel is conditioned on achieved large weight loss (responder selection), not merely on completion (corrected 2026-08-08) (Sumithran et al., 2011). There is no control group — every comparison is each participant against their own baseline.
Weight trajectory: mean loss «13.5±0.5 kg (14.0% of initial weight)» at week 10; participants then regained ~5.5 kg, so at week 62 they were still ~7.9 kg below baseline — i.e. partial regain was already under way when the hormones were re-measured (Sumithran et al., 2011).
At one year, relative to baseline, the appetite-regulating signals were displaced in the appetite-promoting / energy-conserving direction (each below is a within-subject change vs that person’s own baseline, significant at week 62):
| Signal | Direction at 12 mo | Effect on intake / storage |
|---|---|---|
| Leptin (adiposity / satiety) | down — still «35.5±4.7% below baseline» | less satiety, lower energy expenditure |
| Ghrelin (hunger) | up — «remained significantly higher at 62 weeks than at baseline», though it partially reverted («ghrelin levels fell between week 10 and week 62 (P<0.001)») | more hunger |
| Peptide YY (satiety) | down (further down at wk 62 than wk 10) | less satiety |
| Cholecystokinin (satiety) | down | less satiety |
| Amylin, insulin (satiety/meal signals) | down | less satiety |
| GLP-1 (satiety) | slightly down at wk 62 («significantly lower than baseline», P=0.005) — but the weakest signal in the panel: «did not change significantly between baseline and week 10» and absent from the abstract’s one-year still-significant list (corrected 2026-08-08) | less satiety |
| Pancreatic polypeptide, GIP | up (GIP may promote storage) | mixed |
| Subjective hunger / desire to eat | up at both wk 10 and wk 62 | conscious drive to eat rises and stays risen (Sumithran et al., 2011) |
Net: nearly every satiety signal is suppressed and the main hunger signal is elevated, a year out — a coordinated, sustained pressure toward eating more.
A review corroborates the defended-set-point framing (not independently). Hall & Guo 2017 (energetics review) reads current data as «most commensurate» with a set-point model with feedback control of both intake and expenditure, and records that adaptive thermogenesis «may continue for years after energy balance is reestablished at a lower weight» — but the same source flags the limb as contested, «although controversy remains regarding its persistence» (two attached citations, one arguing the negative) (corrected 2026-08-08), so this second limb is not settled even in the source that carries it (Hall & Guo, 2017). This is convergent framing on the energy-expenditure limb (Sumithran’s is the hormonal/appetite limb), but it is not independent type-E backing — both trace the thermogenesis limb to the same Rosenbaum/Leibel antecedent. It also settles a nearby question: the defended weight is defended against energy deficit, not carbohydrate specifically -> What Drives Fat Gain - Energy Balance vs the Carbohydrate-Insulin Model.
Hall also ranks the two limbs, and the ranking favours the appetite limb this page emphasises: after weight loss, intake rose by «about 100 kcal/d per kilogram of lost weight – an effect more than 3-fold larger than the corresponding energy expenditure adaptations to weight loss» (Hall & Guo, 2017). So the appetite/hormonal headwind is the larger driver, and the contested expenditure limb (above) is also the smaller one — a reason the D1 hedge costs the page’s core claim little.
The reframe — regain is defended physiology, not (only) a lapse of resolve
This is the page’s load-bearing move, and Sumithran states it directly:
«Whether these changes represent a transient compensatory response to an energy deficit is unknown, but an important finding of this study is that many of these alterations persist for 12 months after weight loss, even after the onset of weight regain, suggesting that the high rate of relapse among obese people who have lost weight has a strong physiological basis and is not simply the result of the voluntary resumption of old habits.» (Sumithran et al., 2011)
Two caveats the sentence itself carries, and they matter:
- It is associational, not a proven cause of regain. The hormones persist and regain happens; the
study shows persistence, and infers (suggesting) the physiological basis — it does not randomize the
hormones or demonstrate that correcting them prevents regain. is the wiki’s read of the
causal leap; the persistence itself is
[EXTRACTED]. - Not simply is doing work. The claim is not that behaviour is irrelevant — it is that a purely volitional account is incomplete, because a measurable biological headwind is present regardless of motivation. This is the corrective to the just eat less model, not a determinism.
Sumithran’s own decision-relevant conclusion: «Long-term strategies to counteract this change may be needed to prevent obesity relapse» (Sumithran et al., 2011) — i.e. maintenance is a distinct problem from loss, and one that unaided willpower is structurally disadvantaged against.
Starvation mode — the real kernel and the two over-statements
The folk starvation mode belief bundles a true mechanism with two false corollaries, and separating them is the decision-change. The kernel is real: metabolism does adapt downward with weight loss (adaptive thermogenesis), and the appetite-hormone panel above shifts toward eating and conserving — so a weight-reduced body faces a real appetite-and-thermogenesis headwind toward regain (the reframe above, carrying its still-lighter-body caveat: that the shift is defense beyond the weight difference is shown only for leptin — see Limits). What the folk version adds, and the held evidence does not support, are two stronger claims:
- Eating too little STALLS active fat loss. Not shown here — if anything the reverse: the most aggressive under-eating in this literature (the ~500 kcal/day VLCD) drove a 13.5 kg / 14.0% loss by week 10 (Sumithran et al., 2011) (in the responder panel conditioned on >=10% loss — see Limits). Severe under-eating produced large loss, not a stall: the adaptation is a partial headwind on the deficit, not an abolition of it — the deficit still works, it is merely defended, not broken.
- It permanently damages your metabolism. Unsupported on two separate counts. Permanence: persistence of the adaptation is measured only to 12 months here, so permanent is an extrapolation beyond the data (see Limits). Primacy: the energy-expenditure limb — the “slowed metabolism” the belief centres on — is the contested limb (Hall flags its persistence as contested, above) and the smaller one: Hall ranks the post-loss intake rise as more than 3-fold larger than the corresponding energy-expenditure adaptation (quoted above), so the dominant regain driver is appetite, not a crashed metabolism. Neither the metabolic slowdown’s permanence nor its primacy is supported.
Net: starvation mode names a real defended-set-point biology but mislabels both its mechanism and its severity. The corrective is neither eat even less (a deeper deficit does not overcome a headwind that scales with the loss and is largely appetite-mediated) nor your metabolism is ruined (unshown past 12 months) — it is the structural-leverage posture the reframe already licenses -> Layer 1 - Ranking Interventions for a Stratum.
The empirical anchor — what sustained loss to expect across methods (Franz 2007)
Sumithran supplies the mechanism for why maintenance is hard; Franz 2007 supplies the magnitude — the outcome data the qualitative “regain is modal” claim above rests on. It is a systematic review + meta-analysis of 80 RCTs of overweight/obese adults with a minimum 1-year follow-up (26,455 enrolled; 18,199 completers, 69%), across eight intervention types (Franz et al., 2007). The headline durability shape, in the source’s own words:
«A mean weight loss of 5 to 8.5 kg (5% to 9%) was observed during the first 6 months from interventions involving a reduced-energy diet and/or weight-loss medications with weight plateaus at approximately 6 months. In studies extending to 48 months, a mean 3 to 6 kg (3% to 6%) of weight loss was maintained with none of the groups experiencing weight regain to baseline.» (Franz et al., 2007)
So the shape is loss -> plateau at ~6 months -> partial regain -> a modest sustained loss out to 48 months, but no group all the way back to baseline — the same trajectory Sumithran’s hormones explain, reached here from a completely different evidence base (80 RCTs’ weight outcomes vs one uncontrolled hormone study). The discussion sharpens the maintained band: food-strategy interventions stabilized to «approximately 4.5 to 7.5 kg (4.8% to 8%) at 12 months» and «approximately 3 to 4 kg (3% to 4.3%) … maintained at 24, 36, and 48 months» (Franz et al., 2007).
Per intervention type (mean kg / % of baseline; final = last reported timepoint), the anchor is that method changes the peak, not the eventual modest plateau — every food-attentive arm converges toward a few kg kept (Franz et al., 2007):
| Intervention | 6 mo (peak) | 12 mo | later (mo) | pattern |
|---|---|---|---|---|
| Diet alone | 4.9 kg (5%) | 4.6 kg (4.6%) | 4.4 kg @24; 3.0 kg (3%) @48 | shallow loss, shallow plateau |
| Diet + exercise | 7.9 kg (8.5%) | plateau | 3.9 kg (4%) @36 & @48 | deeper peak, same plateau |
| Meal replacements | 8.6 kg (9.6%) | 6.7 kg (7.5%) | — | structured, larger early loss |
| Very-low-energy diet | 17.9 kg (16%) | 10.9 kg (10%) | 5.6 kg (5%) @36 | biggest peak, biggest regain |
| Orlistat + diet | 8.3 kg (8%) | 8.2 kg (8%) | 7.8 kg @36; 5.8 kg (5.3%) @48 | drug flattens the regain arm |
| Sibutramine + diet | 8.2 kg (8.4%) | 8.2 kg (8.4%) | 10.8 kg (11%) @24 | (drug withdrawn — see below) |
| Exercise alone | 2.4 kg (2.7%) | ~2 kg | 1.0 kg (1%) @24 | minimal at any timepoint |
| Advice alone | minimal | minimal | minimal | minimal at any timepoint |
The efficacy meta-analysis (comparator-subtracted) shows the drug and food arms add only a few kg over their controls, with wide, heterogeneous intervals. Hedges’ g was pooled against a comparator arm (diet-alone vs advice; drugs vs diet-alone): diet-alone beat advice by «about 3.7 … 4.5 … 3.3 … 2.2 kg» at 6/12/24/36 mo; orlistat added «about 2.8 … 3.1 … 3.1 kg» over diet-alone at 6/12/24 mo; and even the (withdrawn) sibutramine added «about 2.3 … 5.1 … 4.0 kg» (Franz et al., 2007) — i.e. the added effect of any single lever over its comparator is small (a corroborating MA Franz cites found single-drug placebo-subtracted loss «never reached more than 4 kg»). The reported deltas carry large SDs and the effect sizes were «significantly heterogeneous across studies», so these point estimates are loosely bounded.
Three method caveats bound every number above [inferred from @franz2007]:
- The descriptive weight figures are simple pooling, not a random-effects estimate. «By using simple pooling across studies» the mean loss per timepoint was computed; only the efficacy analysis used Hedges’ g (Franz et al., 2007). So the per-intervention trajectories are a weaker synthesis than a proper pooled estimate — no weighting for study size or precision.
- The maintained-loss numbers are completer figures, and dropout is high. «Most weight-loss outcomes were based on study completers» (Franz et al., 2007); attrition averaged 29% at one year and 31% at study end. If dropouts regained more than completers (the usual direction), the maintained loss is optimistic — a person’s expected sustained loss is likely below these means. Franz concedes the publication/enrollment bias runs the same way: «bias, if any, would likely favor reporting successful interventions and enrolling participants with the greatest potential to complete the trial» (Franz et al., 2007).
- It closes the loop back to the mechanism above. Franz notes the counterintuitive fact the hormone data explain: continuing (or deepening) the diet does not resume loss — «this appears not to happen even when weight-loss interventions are continued. However, if weight-loss interventions are discontinued entirely, weight regain is likely to occur» (Franz et al., 2007). That is the defended set-point seen from the outcome side: the deficit stops paying off at the plateau, yet must be held indefinitely to prevent regain.
The drug arm is a pre-GLP-1 snapshot — do NOT read it as the current expectation [inferred from @franz2007]. Franz’s medications are orlistat and sibutramine only (2007). Sibutramine was withdrawn in 2010 for cardiovascular harm, so its row is of historical interest only. And the modern incretin/GLP-1 agents (semaglutide, tirzepatide) achieve far larger sustained loss than anything in this table — Franz’s «the addition of weight-loss medications somewhat enhances weight-loss maintenance» (a ~2-5 kg edge over diet-and-exercise at 24 months) badly understates the current drug landscape -> Comparing Obesity Drugs, Semaglutide for Cardiovascular Risk in Obesity. Franz bounds the old lifestyle-vs-modest-drug decision, not the lifestyle-vs-GLP-1 one.
Exercise-alone is weak for weight loss — a statement about scale, not about exercise as a health lever. Franz found exercise-alone «did not result in successful weight loss, although no further weight gain was observed» (Franz et al., 2007) — ~2.4 kg at 6 months decaying to ~1 kg. This is the substitution/net-effect point, not a dismissal: Franz itself flags that «aside from weight, physical activity has important positive effects on lipid levels … insulin sensitivity … and all-cause mortality and cardiovascular disease mortality» (Franz et al., 2007). Exercise is simply not the weight-loss lever, which does not make it a weak health lever -> Exercise vs Caloric Restriction for Visceral Fat, Physical Activity Dose and Mortality.
Cross-source: the plateau-then-partial-regain SHAPE is corroborated by the MASLD literature, but the percentage ladders are a different quantity (a distinction, not a weld). Before the contrast, matched by quantity:
| Parameter | Franz 2007 (weight outcome) | EASL/MASLD ladder (histology outcome) | Same quantity? |
|---|---|---|---|
| Peak-loss timing | plateau at «approximately 6 months» | «maximal weight loss at 6 months» | YES — both body-weight, 6-mo peak |
| Post-peak shape | partial regain, no return to baseline | «gradual weight regain to a net weight loss of about 5% at 12-24 months» | YES — both body-weight partial regain |
| The % figures | % of baseline body-weight kept | 5% / 7-10% / >=10% weight-loss required for steatosis / NASH / fibrosis | NO — Franz’s % is weight maintained; EASL’s is the dose needed for a liver endpoint |
So the durability shape is genuine independent corroboration (RCT weight outcomes ↔ MASLD cohort follow-up reaching the same 6-month-peak-then-partial-regain trajectory), strengthening confidence in the shape. But Franz’s «3 to 6 kg (3% to 6%)» maintained and the MASLD ladder’s «5% … 7-10% … >=10%» thresholds answer different questions — how much a person keeps vs how much a person must lose to move a histological marker -> Fatty Liver MASLD and Weight Loss. Do not read the ladder’s rungs as achievable-maintenance targets: Franz’s data say the modal sustained loss (~3-6%) sits at or below the ladder’s lowest (steatosis) rung, so the fibrosis rung (>=10% sustained) is above what most lifestyle maintainers hold.
Why this is the mechanistic keystone for the drug route (GLP-1) — a synthesis
(inferred from Sumithran et al., 2011) The signals Sumithran shows are chronically deficient after weight loss — GLP-1, PYY, amylin, leptin (satiety) down, ghrelin (hunger) up — are precisely the axis obesity pharmacotherapy now targets. A GLP-1 receptor agonist (semaglutide) does not persuade a patient to want to eat less; it pharmacologically resupplies one of the depleted satiety signals, overriding the defended set-point from the outside. So the biology here is the mechanistic why behind two facts held elsewhere: (a) lifestyle maintenance regain is the norm, not the exception -> Diets for Weight Loss - What NICE Recommends (DiRECT’s phase-wise regain; NICE’s weak consider); and (b) the drug route produces durable loss while taken and regain on stopping -> Semaglutide for Cardiovascular Risk in Obesity. The efficacy-while-taken / regain-on-cessation pattern of GLP-1 drugs is the pharmacological mirror image of the defended set-point: remove the exogenous satiety signal and the endogenous deficit Sumithran measured reasserts itself. This is a directional mechanistic synthesis, not an outcome finding — no held trial has tested correct these specific hormones -> prevent regain.
The STEP-1 off-treatment extension shows limb (b) as a descriptive pattern — but it does not cleanly demonstrate the physiological cause (corrected 2026-08-08): one year after stopping semaglutide, participants regained ~two-thirds of their prior weight loss and their cardiometabolic gains reverted toward baseline (fuller magnitudes on Semaglutide for Cardiovascular Risk in Obesity → Durability). The regain is consistent with the defended set-point reasserting once the exogenous satiety signal is withdrawn, but the trial’s own authors decline the clean-cause reading: they attribute the regain «driven by physiological and behavioural factors» and note that «the absence of structured lifestyle intervention following semaglutide withdrawal … may also have contributed to the trajectory of weight regain» (Wilding et al., 2022). So STEP-1 cannot separate biological rebound from behavioural drift and loss of support — it is corroboration of the mechanism this page holds, not an independent test of it, and not a discriminating one.
The class-wide regain rate that bounds the discontinuation problem [2026-08-22, Nong]. STEP-1 gives
one drug’s regain trajectory; a 19-drug obesity NMA (Nong 2026) reports the class-wide rate from a
secondary source (West et al. 2026, a systematic review of 37 studies): «participants regained weight at
an average rate of about 0.4 kg per month after stopping treatment, with a projected return to baseline
weight within approximately 1.7 years, accompanied by loss of cardiometabolic improvements»
(Nong et al., 2026). So «benefits do not
appear to be sustained without continued treatment»
(Nong et al., 2026) — the STEP-1
two-thirds-in-a-year pattern is not semaglutide-specific but a property of withdrawing any of these
agents, which is why the realistic decision is a lifetime one (indefinite therapy vs behavioural
maintenance vs surgery) -> Comparing Obesity Drugs. The number is a secondary citation inside Nong,
not Nong’s own analysis; held as Nong reporting West.
(inferred from Nong et al., 2026)
What it licenses, and the limits that bound it
Decision-relevant, for an identified stratum (the person maintaining a large diet-induced loss):
- Recalibrate the expectation. Maintenance failure is the modal outcome and has a physiological driver — inferred, not proven: Sumithran’s own sentence (block-quoted above) only goes as far as suggesting a strong physiological basis, and the causal step to regain is associational (above) (corrected 2026-08-08). So a maintenance plan that assumes willpower will close the gap is mis-specified. This licenses someone to stop reading regain as personal failure and to seek structural leverage — a standing environmental / pharmacological prop that does not depend on daily resolve -> Layer 1 - Ranking Interventions for a Stratum (structural > point-optimization).
- The lever is the biggest one and it is defended — which is exactly why the drug route earns its place in the ranking for this stratum rather than being dismissed as the easy way out.
Limits (all real, keep them visible):
- Single-arm, uncontrolled, n=34 completers. No comparator; regression-to-baseline and time effects cannot be separated by design. Small n. And the analysed panel is conditioned on ≥10% achieved loss (7 of 50 excluded for not reaching it, above), so the hormone profile is measured only in strong responders — transportability is narrower than VLCD-specific alone conveys.
- The still-lighter-body confound — the central diagnosticity gap (corrected 2026-08-08). At week 62 participants were still ~7.9 kg (~8%) below baseline, so every week-62-vs-baseline hormone difference is confounded with a genuinely smaller body. The competing explanation — that these are simply the hormone levels appropriate to a lighter person, not a persisting adaptation over and above the weight difference — is not excluded except for leptin: Sumithran ran the fat-mass adjustment for leptin alone («Reductions in leptin levels from baseline at weeks 10 and 62 remained significant when adjusted for fat mass»), and even there notes «leptin levels and body weight rose concurrently» (Sumithran et al., 2011). For ghrelin, PYY, CCK, GIP, PP, amylin, GLP-1 and the appetite ratings, “adaptation beyond the weight difference” is assumed, not established. This is the difference between the body defends the lower weight (licenses a permanent structural prop) and hormones track the smaller body you now have (licenses much less).
- VLCD-specific. The exposure is an aggressive ~500 kcal/day liquid-formula loss. Whether the same hormonal persistence follows a slower, food-based, more moderate deficit is not shown here — the depth and speed of the deficit could modulate the adaptation. — transportability to gradual loss is a genuine open question; do not assert the finding generalizes to every weight-loss method. What would confirm/refute: a comparable hormone-panel study after gradual food-based loss.
- Persistence measured to 12 months only — permanent defense is an extrapolation beyond the data.
- Population: overweight/obese, mostly older (mean age ~56), non-diabetic; not lean or young dieters.
- The energy-expenditure arm is corroborated but not by this study. Sumithran cites Rosenbaum/Leibel (via (Sumithran et al., 2011)) for persistent adaptive thermogenesis (reduced energy expenditure) in weight-reduced people — a second, independent limb of the same set-point defense measured on a different variable. An independent replication of the metabolic-rate arm would strengthen the defended set-point synthesis, but no adequate one is held: the candidate long-term cohort (Biggest Loser 6-year resting-metabolic-rate follow-up) was assessed and cut as below the quality bar (n=14, self-selected, uncontrolled), so this arm stays single-study rather than type-E.
Synthesis
(inferred from Sumithran et al., 2011) The unit-of-analysis shift this page installs: weight-loss maintenance is not weight loss, continued — it is a distinct problem against an active biological headwind. The same fact reads three ways across the cluster — it is why lifestyle regain is modal (NICE’s calibrated weak recommendation), why the drug route works while taken and rebounds off it (SELECT/STEP), and why structural leverage over point-optimization is the right Layer-1 posture for a defended lever. Confidence is medium, and split by limb: the empirical durability pattern (loss -> ~6-month plateau -> partial regain -> a modest few-kg sustained loss, no full return to baseline) is now well-backed — a high-tier 80-RCT meta-analysis (Franz 2007), convergent with the MASLD literature’s independent 6-month-peak-then-regain shape. The causal mechanism (that hormonal defense drives the regain) stays weak: one small uncontrolled VLCD study carries it, and the causal step to regain plus the generalization past VLCD are both inferred, not held. The dated drug arm (Franz’s orlistat/sibutramine) does not transport to the current GLP-1 era.