For an adult setting out to lose fat, most methods take off similar amounts once the energy deficit is the same. They differ in what else leaves with the fat, whether they change a hard outcome, and whether the loss lasts.
- The size and persistence of the deficit set how much comes off. Diet labels, macronutrient splits and eating windows change the amount very little; the diet you can keep is the one that works (Named Diet Programs Compared). No held trial has tested deficit sizes to find an optimum for general dieters; the large drug trials gave their lifestyle arms 500 kcal a day below estimated needs. In type 2 diabetes, a very-low-energy diet lost more than a low-energy one at 3 months, on low-certainty evidence.
- Which foods fill the plate matters through intake. In one small, two-week inpatient trial, 20 adults ate about 508 kcal a day more on an ultra-processed diet than on a matched unprocessed one (Ultra-Processed Food and Health Outcomes).
- Exercise does more for the body than for the scale. Alone it moves weight by about a kilogram in people with obesity, partly because people make up some of what they burn, by eating more or moving less. In non-randomised comparisons, diet-plus-exercise programmes lost a little more than diet alone. It takes off some visceral fat and waist, and resistance training kept lean mass during a diet in older adults (Exercise Modality for Body Composition in Obesity, Exercise Energy Compensation).
- Formula diets and the newer obesity drugs produce the largest losses short of surgery. The drugs hold the loss only while you keep taking them (Comparing Obesity Drugs).
- Few methods have been shown to change a hard outcome. Intensive lifestyle loss did not lower heart attacks and strokes in its one large trial, though pooled trials point to fewer deaths. The drug trials that cut cardiovascular events enrolled people with type 2 diabetes or existing heart disease. Surgery holds a large loss for a decade, on non-randomised evidence (Does Weight Loss Reduce Cardiovascular Events).
- Diets and drugs take some lean mass with the fat. Resistance training and protein reduce that share. No held study has tested whether keeping that lean mass changes function or survival, or whether losing slowly spares it (GLP-1 and Lean Mass).
- Much of the lost weight returns over a few years. Appetite hormones were still shifted toward eating a year after a loss, and the thin evidence held suggests rising appetite pushes back harder than a slower metabolism. That is a reason to plan the maintenance, not to avoid trying (Weight-Loss Maintenance and Metabolic Adaptation).
Make the deficit big enough and sustainable; stop choosing between diet labels
How much weight comes off is set by the size of the energy deficit and how long a person keeps it up. The diets people argue about — low-carb, low-fat, the branded programmes, an eating window — are different ways of delivering that deficit. Head to head, they differ by around a kilogram, under the 2 kg the largest comparison set in advance as the smallest difference worth noticing. The choice between them comes down to which one a person will keep.
Low-carb against a balanced diet
A Cochrane review pooled 61 randomised trials and 6,925 people. In adults without type 2 diabetes, the low-carbohydrate arm lost about 1 kg more than the balanced-carbohydrate arm, a gap the review rated as little to no difference. The figures were 1.07 kg more (95% CI 0.59 to 1.55) at 3 to under 12 months across 37 trials, and 0.93 kg more (0.04 to 1.81) at 12 months or longer across 14 trials, both moderate certainty. In people with type 2 diabetes the 12-month gap was 0.33 kg, with an interval running from 2.13 kg in favour of low-carb to 1.46 kg against. (Naude et al., 2022)
The review placed its bar for a meaningful loss within an arm, not between arms: «reductions from around the middle of these ranges (about 4 to 6 kg in most trials) would start to become clinically meaningful» (Naude et al., 2022). That 4-6 kg is weight lost within each arm; the ~1 kg is the difference between arms. They are different quantities and should not be set against each other; the review judged the ~1 kg gap not clinically important.
Even that kilogram depends on how the trials set energy intake. At 3 to under 12 months, in the 27 trials that gave both arms a similar energy prescription, low-carb’s edge shrank to 0.48 kg (0.85 to 0.11), with no heterogeneity left (I2 = 0%). In the 7 trials that let the low-carb arm eat freely while the control arm restricted energy, the gap was 1.84 kg (3.07 to 0.62); in 3 small trials where neither arm had an energy prescription, it was 2.71 kg (4.20 to 1.22). (Naude et al., 2022) (Naude et al., 2022)
The review itself read the subgroup spread as no important clinical difference (Naude et al., 2022). These subgroups overlap heavily with the review’s split by depth of carbohydrate restriction, so no share of the gap can be pinned on either (Low-Carbohydrate vs Balanced-Carbohydrate Diets). Where energy intake was not matched, a larger loss cannot be credited to the food category over the larger deficit (Is the Food Category Doing Any Work).
Two further limits shrink the kilogram. Cutting carbohydrate sheds 2 to 3 kg of body water as glycogen is used up, and the water returns when carbohydrate is eaten again, so an early drop on the scale is partly hydration (Naude et al., 2022). And no trial in the review reported death, heart attack or stroke (Naude et al., 2022); none reported outcomes beyond two years, and 60% lasted six months or less (Naude et al., 2022).
A large trial built to match a diet to a person found that neither of the two predictors it tested worked. DIETFITS randomised 609 adults to healthy low-fat or healthy low-carb for 12 months; weight loss differed by 0.7 kg (−0.2 to 1.6), and neither a genotype pattern nor insulin secretion picked the better diet: «neither of the 2 hypothesized predisposing factors was helpful in identifying which diet was better for whom» (Gardner et al., 2018). On a gap this small, low-carb is not a meaningfully better weight-loss diet — and it is the poorer one for someone watching LDL, which it barely lowers (Named Diet Programs Compared).
The branded programmes
A network meta-analysis compared 14 named programmes (Atkins, DASH, Zone, Weight Watchers, Mediterranean and others), also grouped into three macronutrient patterns, against three control diets: 121 randomised trials, 21,942 adults with overweight or obesity. Against usual diet, every macronutrient pattern produced a loss that cleared the pre-set 2 kg importance bar at 6 months — low-carb and low-fat almost identically: «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)» (Ge et al., 2020).
Between diets, the differences typically fell under that bar. The authors’ summary: «Differences between diets were typically small to trivial and often based on low certainty evidence» (Ge et al., 2020). For instance, Atkins against Zone came to only 1.38 kg (95% credible interval 0.15 to 2.62) at six months (Ge et al., 2020).
The advantage over usual diet also shrinks with time. Across the named diets (all but paleolithic), the estimated loss relative to usual diet was on average 1.5 kg smaller at 12 months than at 6 months, and «At 12 months the effects on weight reduction and improvements in cardiovascular risk factors largely disappear» (Ge et al., 2020). The Mediterranean diet was the one label with a residual 12-month signal, on LDL rather than weight (Named Diet Programs Compared). Every outcome in this network is a surrogate, and none runs past 12 months.
Why the macronutrient split adds so little
Controlled-feeding studies test the claim that cutting carbohydrate burns extra energy at equal calories. Pooling 32 controlled-feeding studies, the edge ran the other way: «both energy expenditure (26 kcal/d; P <.0001) and fat loss (16 g/d; P <.0001) were greater with lower fat diets» (Hall & Guo, 2017). The author’s reading was that the effect is too small to matter in either direction: «for all practical purposes “a calorie is a calorie”» for body fat (Hall & Guo, 2017). Isocaloric feeding does not support the carbohydrate-insulin model’s prediction of a fat-loss advantage for low-carb. What survives of that model is an effect on appetite and intake, which acts through the deficit, not around it (What Drives Fat Gain - Energy Balance vs the Carbohydrate-Insulin Model).
Processing is where intake moves. In an inpatient crossover, 20 weight-stable adults ate two weeks of an ultra-processed diet and two weeks of a matched unprocessed one, as much or as little as they wanted: «Energy intake was greater during the ultra-processed diet (508 ± 106 kcal/day; p = 0.0001)» (Hall et al., 2019). Participants gained 0.9 kg on the ultra-processed diet and lost 0.9 kg on the unprocessed one, and the authors conclude: «Limiting consumption of ultra-processed foods may be an effective strategy for obesity prevention and treatment.» (Hall et al., 2019) That is a short trial in a small group, on intake rather than long-run weight, but it points at the same lever: the food a person chooses matters through how easily it lets them hold the deficit (Energy Density).
An eating window delivers a deficit; it is not a separate lever
One 12-month trial in 139 adults with obesity gave both arms the same calorie target and varied only whether eating was confined to 8:00 a.m. to 4:00 p.m. Both arms lost weight, 8.0 kg (95% CI 6.4 to 9.6) on the window and 6.3 kg (4.7 to 7.8) on plain calorie restriction. The window added no significant loss — «net difference, −1.8 kg; 95% CI, −4.0 to 0.4; P = 0.11» (Liu et al., 2022). Intake equality rested on prescribed targets and food logs, not measured expenditure (Time-Restricted Eating).
Pooled across 99 trials of intermittent fasting, alternate-day fasting was the only form to beat continuous calorie restriction, by 1.29 kg (0.59 to 1.99), below the review’s 2.0 kg importance bar (Semnani-Azad et al., 2025). In trials of 24 weeks or more, no fasting form beat continuous restriction: «with no differences between intermittent fasting strategies and CER in these moderate-to-long term studies». Only 17 trials reached that length, and the authors say the lost difference «may be due to an insufficient number of studies available», or to adherence fading and metabolic adaptation in longer trials (Semnani-Azad et al., 2025). Whether a window costs lean mass is taken up in the section on muscle.
Guidance reached the same place
England’s NICE reviewed the same trial literature and recommended no specific macronutrient-defined diet, while noting that lowering a specific macronutrient is one way to reach a deficit. It found that, for almost all outcomes, low-carbohydrate diets «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» — very-low-carbohydrate diets did reduce weight at one year, but not at two (National Institute for Health and Care Excellence, n.d.). NICE also dropped its old 600 kcal/day deficit target, «because they felt this was an arbitrarily specific number», and now says only that an approach «should maintain an energy deficit» (National Institute for Health and Care Excellence, n.d.). The committee had queried the evidence base for the 600 kcal figure, so it was withdrawn as arbitrary, not disproved (Diets for Weight Loss - What NICE Recommends).
No held trial has tested deficit sizes to find an optimum for general dieters. The closest, a very-low-energy against a low-energy diet in type 2 diabetes, found the larger deficit lost more at 3 months, on low-certainty evidence (see the diabetes section below). For a starting point, use the figure the large drug trials gave their lifestyle arms: STEP-1 counselled both arms to eat «a reduced-calorie diet (500-kcal defi- cit per day relative to the energy expenditure estimated at the time they underwent random- ization)», with 150 minutes a week of activity encouraged (Wilding et al., 2021). SURMOUNT-1 used the same 500-calorie deficit (Jastreboff et al., 2022).
That is a trial protocol, not a tested dose. On it, with regular counselling, the placebo arms lost 2.4% of body weight at 68 weeks and 3.1% at 72 (Semaglutide for Cardiovascular Risk in Obesity, The Disease Model of Obesity). Refalo’s authors, writing about trained non-obese dieters, advise that «indi- viduals seeking to avoid FFM losses at all costs should likely implement mod- est energy restriction.» They cite Murphy and colleagues’ suggestion that a deficit of about 500 kcal a day prevented lean-mass gains (Refalo et al., 2025). That figure is second-hand, from a different research group, so read it as a hint, not a ceiling.
This is a case where current guidance and the trial evidence agree, for the same reason. The one diet NICE does recommend, a formula-based total diet replacement, is defined by energy, not macronutrients; it belongs to the section on larger losses.
The missing term is adherence
The diet that works is the one a person keeps, and the trials say little about which diets people keep. The named-diet network had little adherence data, and its authors suggest that, if adherence was low, «our results describe what is likely to happen for average adherence by patients» (Ge et al., 2020). In the Cochrane review, about a quarter of trial arms reported nothing on whether the assigned diet was followed (Low-Carbohydrate vs Balanced-Carbohydrate Diets). So the numbers above are the effect of being assigned a diet at the adherence trials achieved. So pick the pattern a person will sustain and size the deficit; searching for the right label buys little.
If the deficit is what counts, the obvious move is to widen it by burning more energy. On the scale, that works less well than the arithmetic suggests.
Exercise for where the fat comes from and what stays, more than for the scale
On its own, exercise is a weak tool for weight. Added to a diet, it adds somewhat to the loss, and it does more for waist and visceral fat than for the scale. Every claim below about its weakness is a claim about weight.
Exercise alone moves weight by about a kilogram
A network meta-analysis of 45 RCTs in 3,566 adults with a BMI of 30 or more ranked six exercise programmes against each other and against no exercise. No programme moved weight by much more than a kilogram: «exercise as an independent intervention, irrespective of which prescription, induced at best, a small reduction in weight loss (mean weight loss ranged from −0.05 to −1.01 kg).» (O’Donoghue et al., 2020) The trials ran a mean of about 20 weeks, mostly three 45-minute sessions a week. The result holds for that dose and that group, not beyond it (Exercise Modality for Body Composition in Obesity).
Waist and body-fat percentage responded better than weight. Combined aerobic plus resistance training ranked at or near the top on waist and body fat, and low-to-moderate-load resistance training alone ranked last; on waist, programmes with an aerobic component beat resistance-only. The best arm for waist, combined low-to-moderate training, took off 2.76 cm (95% CI 1.00 to 4.52 cm) against no exercise (O’Donoghue et al., 2020). The finer order among the leading programmes is not established: their confidence intervals overlapped, and none of the programmes produced a significant gain in VO2max. «Taken together, more than 70% (n = 33) of the included trials were judged as being at moderate to high ROB.» (O’Donoghue et al., 2020) Only 27% used intention-to-treat analysis. Every outcome here is a surrogate (Surrogate Outcomes).
The body makes up part of what exercise burns
Weight barely moves partly because people compensate. A review of 61 exercise studies found that «The overall mean energy compensation was 18% ± 93%.» (Riou et al., 2015) The ±93% is a standard deviation across study groups, not a range: some groups compensated more than they burned and some not at all, so the average says little about any one programme. The studies also did not measure enough to tell whether the compensation came from eating more or from moving less. Part of that 18% is not biology at all — under-delivered or over-estimated exercise inflates it. The share compensated grows with the length of the programme and «approached 84%» at about 80 weeks (Riou et al., 2015), but that figure comes from a fitted trend resting on a few long studies (Exercise Energy Compensation).
A doubly-labelled-water study of 1,754 free-living adults found that part of the compensation is metabolic, not only extra eating: «energy compensation by a typical human averages 28% due to reduced BEE» — basal energy expenditure (Careau et al., 2021). People at the 10th percentile of BMI compensated 27.7% of activity calories; people at the 90th compensated 49.2% (Careau et al., 2021). The heaviest people may get the least weight change per calorie burned. Treat that as a caution, not a law: it is a comparison between people, and the exercise-trial review found the opposite direction for initial fat mass among younger participants in short studies (Riou et al., 2015) (inferred from Careau et al., 2021; Riou et al., 2015).
Compensation is usually partial in short studies
The two averages measure different things. Careau’s reading of the 28% figure is that «only 72% of the extra calories we burn from additional activity translates into extra calories burned that day» (Careau et al., 2021). That 28% counts only the drop in basal expenditure, not any extra eating. Riou’s figure is inferred from the change in body fat and lean mass, so it nets every route, and its fitted trend approached 84% at about 80 weeks (above) (inferred from Careau et al., 2021; Riou et al., 2015). Careau’s one within-person analysis, in 68 people aged 70 to 90, found compensation close to total (Careau et al., 2021). That subset is small and elderly, but an individual can fare worse than the average (Exercise Energy Compensation). Franz’s pooled arms show the scale result, though the two sets of arms used different diets. In Franz’s review of 80 trials, diet-plus-exercise arms lost 7.9 kg at about six months against 4.9 kg for diet-alone arms, and kept 3.9 kg against 3.0 kg at 48 months (Franz et al., 2007).
Those figures are simple averages of separate study arms, not a randomised test of adding exercise to the same diet. The arms also ate differently: the diet-alone studies cut about 500 kcal or more below estimated need, while the diet-plus-exercise studies generally prescribed 1,200 kcal for women and 1,500 for men (Franz et al., 2007), so part of the 3 kg peak gap may be the diet. Franz’s own 24-month comparison favoured diet plus exercise by 1.3 kg (Franz et al., 2007). In this non-randomised arm comparison the direction favours diet plus exercise, and the size is modest (Weight-Loss Maintenance and Metabolic Adaptation). The one randomised add-on held tested resistance training, not aerobic work: added to a diet in older adults, it kept lean mass but did not add weight or fat loss (Sardeli et al., 2018) (Resistance Training During Caloric Restriction). Exercise does not replace the deficit; its case in a diet rests on lean mass, visceral fat and fitness more than on the scale (inferred from Franz et al., 2007; Sardeli et al., 2018).
Adjusted for the size of the deficit, exercise takes off more visceral fat
Does the source of the deficit matter for where the fat comes from? A meta-analysis of 40 RCTs in 2,190 adults with overweight or obesity gives two answers that point opposite ways (Exercise vs Caloric Restriction for Visceral Fat). Unmatched, the caloric-restriction arms cut visceral fat more than the exercise arms (standardized effect −0.53 vs −0.28 against control); larger deficits in the diet trials may explain this. Once the size of the weekly deficit was adjusted for, across trials, exercise came out ahead: «exercise had a superior dose–response effect on reducing visceral fat compared with caloric restriction (ES −0.18 (−0.33 to −0.04); p=0.012).» (Recchia et al., 2023) Each extra 1,000 kcal a week spent through exercise bought more visceral-fat loss. The diet slope was flat, but it rested on 16 effects against 46, so it may reflect too few studies.
Three limits keep this modest. The numbers are standardized effect sizes, not litres or square centimetres of fat; the only absolute figure is waist, 3.15 cm with exercise and 4.67 cm with diet. The deficit was prescribed, not measured, so compensation could have shrunk the exercise arms’ real deficit. And the authors call the effect small: «the overall effect of exercise as well as its dose–response effect on visceral fat were small, which limits the interpretation of our results.» (Recchia et al., 2023) Visceral fat is a strong predictor of cardiometabolic disease, but it is a surrogate; nobody here counted heart attacks or deaths.
How the two findings fit together: compensation refills part of the calorie gap, which blunts exercise’s effect on weight. Exercise’s effect on visceral fat appears partly independent of weight lost, so the same refilled calories do not erase it (Exercise Energy Compensation).
Training in the fat-burning zone has not been shown to beat other exercise
FATmax training sets aerobic intensity at the point where a person burns fat fastest, measured in a lab test; in adults with overweight or obesity that lands at a low intensity, about 42-49% of VO2max. A meta-analysis of 19 RCTs found −3.66 kg of body weight against non-exercise or diet-only controls (Jiang et al., 2026) — much more than the kilogram above. The two results used different comparisons, so they do not contradict each other. Some trials included a diet co-intervention, the population included adolescents and people with diabetes or fatty liver, and no trial pitted FATmax against ordinary moderate aerobic exercise (FatMax Training for Body Composition and Cardiometabolic Health).
The prediction intervals for BMI, body-fat percentage and waist «crossed zero» (Jiang et al., 2026), so a new group might see no effect. FATmax beats doing nothing, as other aerobic exercise does. Whether picking the fat-burning intensity adds anything is untested in this pool.
What the evidence does not say
Exercise’s case for keeping muscle during a diet rests elsewhere. The visceral-fat review offers it only as background from earlier literature («exercise might stimulate fat loss while maintaining muscle mass» (Recchia et al., 2023)), not as a measured result. No held trial compares diet against exercise at a matched deficit on lean mass — only on visceral fat. A step-counter or wearable can help people move more, but the held evidence on it reports activity, not weight (Wearable Activity Trackers and Physical Activity).
Some people need a much larger loss than a diet change or exercise can deliver. For them, three bigger tools exist: a formula diet, a drug, and surgery.
Reach for formula diets, drugs or surgery when the loss needed is large — and read their numbers on one comparator
A change of diet or more exercise typically takes off a few kilograms. Three tools take off much more: a formula diet that replaces all food for a few months, the newer obesity drugs, and bariatric surgery. Each comes with its own costs, and the published numbers are measured against different comparators, so they cannot be lined up as one league table. Comparing Obesity Drugs
Total diet replacement
DiRECT put adults with type 2 diabetes on a 825-853 kcal/day formula diet for 3 to 5 months, followed by stepped food reintroduction and a monthly maintenance programme. At 12 months the intervention group had lost a mean 10.0 kg against 1.0 kg in usual care (adjusted difference -8.8 kg, 95% CI -10.3 to -7.3). Diabetes remission reached 46% against 4% (about 42 more remissions per 100 people started (inferred from Lean et al., 2018)), in a group aged 20-65 with BMI 27-45, diagnosed within 6 years and not on insulin. (Lean et al., 2018)
Remission tracked the kilograms, not the formula. Pooling both arms, none of 76 people who gained weight went into remission, against 7% of those who lost 0-5 kg, 34% at 5-10 kg, 57% at 10-15 kg and 86% (31 of 36) at 15 kg or more. The formula itself was 59% carbohydrate, so the remissions did not require a low-carbohydrate diet. This is a within-trial association, not the randomised contrast, and people who lose more may differ in other ways. (Lean et al., 2018) Total Diet Replacement and Type 2 Diabetes Remission
In people who already have type 2 diabetes, an umbrella review finds that no macronutrient profile beats another: differences between intervention and control diets are commonly 0-2 kg, which the review calls of little clinical significance. Formats that control energy do better. A very-low-energy diet (VLED, 400-500 kcal for 8-12 weeks) lost 6.6 kg more than a low-energy diet at 3 months (95% CI -9.5 to -3.7), but the review grades that low certainty: two trials, completers only. Formula meal replacement lost 2.4 kg more (-3.3 to -1.4) over 12-52 weeks, at moderate certainty. (Churuangsuk et al., 2021) (Churuangsuk et al., 2021) On remission, total diet replacement is the one diet the review grades at high certainty: «Remissions at 1 year were reported for a median 54% of participants in RCTs including initial low-energy total diet replacement (low-risk-of-bias study, GRADE high), and 11% and 15% for meal replacements and Mediterranean diets, respectively (some concerns for risk of bias in studies, GRADE moderate/low).» (Churuangsuk et al., 2021)
Rapid loss aggravates postural hypotension in people taking diuretics or blood-pressure drugs, and hypoglycaemia is possible if glucose-lowering drugs continue (Churuangsuk et al., 2021) — so anyone on those drugs needs them reviewed first. Diets for Weight Management in Type 2 Diabetes
NICE says to consider a low-energy total diet replacement (800-1200 kcal/day) for at most 12 weeks, for people with obesity or with overweight plus type 2 diabetes, inside a specialist programme with long-term support. It made the recommendation weak because rapid regain and weight cycling could cause harm, and it is explicit about duration: «the committee recommended that low-energy diets should not be used as a long-term strategy to manage obesity.» (National Institute for Health and Care Excellence, n.d.) Diets for Weight Loss - What NICE Recommends DiRECT tested the formula as the front end of a maintenance programme, not as a one-off 12-week course. Total Diet Replacement and Type 2 Diabetes Remission
The drugs, one comparator per row
Drug trials report weight loss in three different ways, and the three are not interchangeable. Each figure below is labelled with what it was measured against. Comparing Obesity Drugs Semaglutide for Cardiovascular Risk in Obesity
| Comparator | Tirzepatide | Semaglutide (weekly injection) |
|---|---|---|
| vs lifestyle (network meta-analysis, ~1 yr) | -14.9% (-16.0 to -13.9) | -9.8% (-10.6 to -9.1) |
| placebo-adjusted (SURMOUNT-1 wk 72; STEP-1 wk 68) | -11.9 pts (5 mg) to -17.8 pts (15 mg) | -12.4 pts (-13.4 to -11.5) |
| within-arm (drug arm vs placebo arm) | -15.0% (5 mg) to -20.9% (15 mg) vs -3.1% | -14.9% vs -2.4% |
Sources: vs-lifestyle row (Nong et al., 2026); tirzepatide trial rows (Jastreboff et al., 2022); semaglutide trial rows (Wilding et al., 2021).
STEP-1 and SURMOUNT-1 both enrolled adults with a BMI of 30 or more, or 27 or more with a weight-related condition, and no diabetes, with lifestyle advice in both arms (Wilding et al., 2021) (Jastreboff et al., 2022). The SURMOUNT-1 authors warn that the two trials cannot be compared directly, since populations and designs differed (Jastreboff et al., 2022). SELECT’s -9.39% against -0.88% at 104 weeks comes from an older population with heart disease, in a cardiovascular trial, and is not a weight-trial result. (Lincoff et al., 2023)
On the top agents, a third to a half of people lose 20% or more within a year, against about 2% on lifestyle alone. (Nong et al., 2026) In STEP-1, 32.0% of the semaglutide arm reached a 20% loss against 1.7% on placebo; in SURMOUNT-1, 57% on the 15 mg dose did against 3%. (Wilding et al., 2021) (Jastreboff et al., 2022)
Bigger losses come with more side effects and more stopping
Across 19 drugs, Nong’s network meta-analysis states the pattern plainly: «Greater weight loss was consistently accompanied by higher rates of adverse events and treatment discontinuation, indicating a clear benefit- harm trade-off.» (Nong et al., 2026) Within that pattern, drugs with similar weight loss can differ widely in how many people stop. Against lifestyle, stopping because of side effects rose by 28 per 1,000 people over a year on tirzepatide (RR 1.9, 1.5 to 2.5), the same as oral semaglutide, and by 20 per 1,000 on injected semaglutide (RR 1.7, 1.4 to 2.0). That 20 is the lowest among the drugs with substantial weight loss at moderate-to-high certainty. Orforglipron, with similar weight loss to injected semaglutide (-9.9% vs -9.8%), added 94 per 1,000. (Nong et al., 2026)
The single trials show the same signal. In STEP-1, 7.0% on semaglutide against 3.1% on placebo stopped because of adverse events, 4.5% against 0.8% because of gastrointestinal ones; serious adverse events were 9.8% against 6.4%. (Wilding et al., 2021) In SURMOUNT-1, 4.3%, 7.1% and 6.2% stopped on 5, 10 and 15 mg tirzepatide against 2.6% on placebo. (Jastreboff et al., 2022) Real-world data suggest roughly half of patients stop within the first year. (Nong et al., 2026)
Quality of life barely moved. On a 36-item scale where 10 points is the smallest difference patients notice, every drug with moderate-to-high certainty scored under 5: tirzepatide 3.9 (2.3 to 5.5), injected semaglutide 2.9 (1.7 to 4.0) (Nong et al., 2026). Across the GLP-1 drug class, an umbrella of 60 meta-analyses finds gastrointestinal effects the most consistent harm (nausea OR 2.47, 1.84 to 3.34); gallbladder or biliary disease may rise (OR 1.34, 1.16 to 1.55), a signal the umbrella rates exploratory. Fracture odds were, if anything, lower (OR 0.67, 0.52 to 0.87), a possible signal that did not reach the umbrella’s credibility threshold. (Yang et al., 2026) GLP-1 Non-Cardiometabolic Effects and Safety The strongest agents also take the most lean mass with the fat, which the muscle section takes up. GLP-1 and Lean Mass
Surgery
In the held evidence, bariatric surgery is the one route with a large loss held for a decade. In the Swedish Obese Subjects study, weight loss stabilised at 25%, 16% and 14% of baseline after 10 years for gastric bypass, vertical banded gastroplasty and banding. The study enrolled people with severe obesity (BMI 34 or more for men, 38 or more for women) and was not randomised. (Sjöström et al., 2007) Does Weight Loss Reduce Cardiovascular Events
No test picks the right drug for a given person
Nong’s analysis looked for differences in relative effect by obesity severity, type 2 diabetes, baseline cardiovascular disease and trial length, and found none credible except larger semaglutide losses in longer trials. (Nong et al., 2026) So who gains most is decided by baseline risk, not by a known responder type: the same relative effect buys more for someone at higher risk. Without individual patient data, this is no modifier found, not none exists. Baseline Risk and the Relative-Absolute Split Ranking the agents for a prescription is outside this appraisal. Comparing Obesity Drugs
No held trial compares a drug, an intensive lifestyle programme and surgery head to head, so their relative merits rest on separate trials with different comparators. Semaglutide for Cardiovascular Risk in Obesity
A larger number on the scale is not yet a longer or healthier life. The next question is which of these methods has been shown to change a hard outcome.
Separate the methods that reached a hard outcome from those that moved a marker
Weight is a surrogate. The outcomes people care about are hard outcomes (heart attacks, strokes, death), with intermediate outcomes (type 2 diabetes remission, fatty-liver resolution) in between. Only a handful of trials of weight-loss methods counted hard outcomes at all, and they rank the methods differently from the scale (Surrogate Outcomes).
| Method | Highest outcome reached | Design | Population |
|---|---|---|---|
| Intensive lifestyle loss | all-cause death down; CV events not | RCTs, meta-analysis | obese adults; T2D in Look AHEAD |
| Bariatric surgery | all-cause death down | matched, non-randomised | severe obesity |
| Semaglutide | CV events down | one RCT | established CVD, no diabetes |
| Tirzepatide | weight; heart-failure signal | RCTs on weight, network meta-analysis | obesity |
| Mediterranean diet, no calorie target | CV events down | one RCT | high-risk primary prevention |
— the table condenses the sections below; each row is sourced there.
Intensive lifestyle loss: fewer deaths, but not fewer heart attacks
Look AHEAD randomised 5,145 adults with type 2 diabetes and overweight or obesity to an intensive lifestyle programme or to diabetes education. The programme won on weight (8.6% vs 0.7% within-arm at one year), but cardiovascular events did not fall: «hazard ratio 0.95; 95% CI 0.83 to 1.09, p=0.505», or 18.3 vs 19.2 events per 1,000 person-years. «The trial was stopped early based on a futility analysis when median follow-up was 9.6 years.» By then the weight gap between arms had narrowed to about 2.5 points (6.0% vs 3.5%), so the trial ended up testing a modest sustained difference, not large loss against none (Look AHEAD Research Group, 2013) (Does Weight Loss Reduce Cardiovascular Events).
A meta-analysis of 54 RCTs (30,206 adults with obesity, almost all on low-fat weight-reducing diets, follow-up of at least a year), which includes Look AHEAD, agrees on cardiovascular events and adds a benefit on death. New cardiovascular events: «risk ratio 0.93, 95% confidence interval 0.83 to 1.04». All-cause death: «risk ratio 0.82, 95% confidence interval 0.71 to 0.95», «six fewer deaths per 1000 participants (95% confidence interval two to 10)», rated high quality. Look AHEAD carried 54.6% of the weight, but removing it left the death benefit in place: «risk ratio 0.78, 95% confidence interval 0.63 to 0.96; I2=0%» (Ma et al., 2017).
The two results are not in conflict; they measure different endpoints. Lifestyle weight loss probably lowers deaths without measurably lowering cardiovascular events (Ma’s authors: it «may reduce» premature death), so the death benefit should not be read as a heart benefit, nor the heart null as no benefit. Because the diets were mostly low in saturated fat, the trials cannot separate weight loss from a change in fat quality (Does Weight Loss Reduce Cardiovascular Events).
A larger loss may matter. A post-hoc Look AHEAD analysis, reported by the DiRECT authors, found that «a 10% weight loss in the first year … was associated with a 21% decrease in occurrence of cardiovascular outcomes» (Lean et al., 2018). This is a responder subgroup, not a randomised contrast, and the primary paper is not held — it unlocks whether the ≥10% threshold survives its own analysis.
Bariatric surgery: fewer deaths, on non-randomised evidence
The Swedish Obese Subjects study followed 2,010 people who chose surgery and 2,037 matched controls with severe obesity for about 11 years. There were 101 deaths after surgery and 129 among controls; the adjusted «hazard ratio … was 0.71 (P = 0.01)», 95% CI 0.54 to 0.92 in the study’s adjusted table (Sjöström et al., 2007). Crudely, that is about 50 vs 63 deaths per 1,000 over the follow-up (arithmetic on the death counts, unadjusted).
Two limits travel with this number. «The main limitation of our study was the ab- sence of randomization»: people who wanted surgery may differ in ways matching cannot remove. And mortality showed no significant difference by how much weight people lost, though the authors note that the low death count hindered detection of modest differences: «we did not find significant differences in mortality (starting at year 2) according to the degree of weight loss during the first year» (Sjöström et al., 2007). The benefit is shown for the operation, not provably for the kilograms.
Still, surgery and the lifestyle meta-analysis reach the same answer on all-cause death by unrelated routes: different designs, different authors, and no shared trials [E-independent]. That convergence makes a death benefit from these interventions in obesity more credible than either source alone. It does not show that the kilograms are the cause, and it says nothing new about cardiovascular events (Does Weight Loss Reduce Cardiovascular Events).
Semaglutide: fewer cardiovascular events, in people who already had heart disease
SELECT gave semaglutide or placebo to 17,604 adults aged 45 or over with a BMI of 27 or more, established cardiovascular disease and no diabetes. A first heart attack, stroke or cardiovascular death occurred in «569 of the 8803 patients (6.5%) in the semaglutide group and in 701 of the 8801 patients (8.0%) in the placebo group (hazard ratio, 0.80; 95% confidence interval, 0.72 to 0.90; P<0.001)» (Lincoff et al., 2023). That is about 15 fewer events per 1,000 people over a mean 3.3 years, a number needed to treat of about 67 (arithmetic on the two event rates) (Semaglutide for Cardiovascular Risk in Obesity).
SELECT tested secondary prevention only. The trialists say so: «The effects of semaglutide on primary prevention of cardiovascular events in persons with overweight or obesity but with- out previous atherosclerotic disease were not studied.» (Lincoff et al., 2023)
Only the main composite is confirmed. Cardiovascular death came next in the testing order and missed (hazard ratio 0.85; 95% CI 0.71 to 1.01), so «superiority testing was not performed for the remaining confirmatory secondary end points». All-cause death (0.81, 0.71 to 0.93) is therefore a point estimate, not a confirmed effect, and non-fatal stroke was not significantly reduced (0.93, 0.74 to 1.15) (Lincoff et al., 2023). A 2026 network meta-analysis reports the same death estimate, «largely informed by cardiovascular outcome trials in high risk populations» — in effect it re-pools SELECT (inferred from Nong et al., 2026), so it does not lift that caveat (Nong et al., 2026) (Comparing Obesity Drugs).
Tirzepatide: the most weight, no cardiovascular-outcome trial yet
In the network meta-analysis, tirzepatide produced more weight loss than semaglutide against lifestyle (-14.9% vs -9.8%) (Nong et al., 2026), but its pivotal trials measured weight and risk markers, not events. In two 52-week trials in sleep apnoea and obesity it lowered systolic blood pressure and CRP, and the authors state that «the design and shorter duration of the current trials does not support the assessment of long-term cardiovascular outcomes» (Malhotra et al., 2024). The network meta-analysis finds a heart-failure reduction («0.49, 0.27 to 0.88») but no demonstrated reduction in death or heart attack (Nong et al., 2026). SELECT’s benefit should not be borrowed for a different molecule; the dedicated trial is pending — it settles whether the dual agonist cuts cardiovascular events at all (Semaglutide for Cardiovascular Risk in Obesity).
The yardstick reorders the methods
Judged by the scale, tirzepatide and surgery lead and lifestyle trails. Judged by hard outcomes, the order changes. Among the drugs, the one that removes the most weight has the least outcome evidence. Semaglutide has confirmed event reduction only in people who already had heart disease, and lifestyle and surgery share a death benefit without a proven cardiovascular one (Does Weight Loss Reduce Cardiovascular Events).
Weight change is not the only route to fewer events. PREDIMED gave high-risk adults a Mediterranean diet with no calorie target — «No total calorie restriction was advised, nor was physical activity promoted» — and cardiovascular events fell (hazard ratio 0.69, 95% CI 0.53 to 0.91, with olive oil; 0.72, 0.54 to 0.95, with nuts); five-year risk was 36 and 40 vs 57 per 1,000 (Estruch et al., 2018). Diet composition can lower events where weight loss did not, though the populations differ, so this is a cross-trial contrast (Mediterranean Diet and Cardiovascular Events).
No weight-loss method has been shown to cut cardiovascular events in people with obesity who have neither heart disease nor type 2 diabetes. The lifestyle trials were null on events (their benefit was on all-cause death), and no held randomised drug or surgical trial has tested events in that group (Semaglutide for Cardiovascular Risk in Obesity, Does Weight Loss Reduce Cardiovascular Events). For someone at low baseline risk, any benefit that did transfer would also be smaller in absolute terms (Baseline Risk and the Relative-Absolute Split). How BMI relates to death is a separate question from what losing weight achieves (BMI and All-Cause Mortality).
What the loss does reach: intermediate outcomes
Large enough losses reach some intermediate outcomes. Type 2 diabetes remission scales with kilograms lost (see the formula-diet section above). The proposed mechanism is fat leaving the liver and pancreas: Taylor’s group reports that people with recent-onset diabetes «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» (Taylor & Holman, 2014) (Ectopic Fat and Depot-Specific Risk).
Fatty liver follows a dose ladder. European guidance says loss «should aim at a sus- tained reduction of >−5% to reduce liver fat, 7-10% to improve liver inflammation, and >−10% to improve fibrosis» (European Association for the Study of the Liver, 2024). The ladder rests on one histology trial (European Association for the Study of the Liver, 2024), and the same guidance says its effect on liver-related outcomes and death is still unknown: «Further follow-up studies are needed to determine the long- term effectiveness … on clinical liver-related outcomes and liver-related mortality» (European Association for the Study of the Liver, 2024) (Fatty Liver MASLD and Weight Loss). Remission and liver fat are real gains; they are not yet hard outcomes.
For the liver, unlike the scale, the type of fat matters at equal calories. A meta-analysis of isocaloric trials found: «Replacing dietary fat with carbohydrates did not result in changes in liver fat (12 comparisons, SMD 0.01 (95% CI −0.36; 0.37)). Unsaturated fat as compared with saturated fat reduced liver fat content (4 comparisons, SMD −0.80 (95% CI −1.09; −0.51)).» (Winters-van Eekelen et al., 2020) These are standardized effects, not a percentage of liver fat removed, and the fat-type result rests on four comparisons (Fatty Liver MASLD and Weight Loss).
Count the muscle that leaves with the fat, and protect it with training and protein
Diets and drugs that take fat off also take some lean mass — the cost the scale does not show. The studies below measure lean mass by DXA or as fat-free mass, which includes water and organ tissue, so a kilogram of lean is not a kilogram of muscle. Lean mass is a surrogate: it stands in for muscle, which in turn stands in for strength and independence, and none of the weight-loss trials here measured those directly (Surrogate Outcomes, GLP-1 and Lean Mass) (Laverde et al., 2026; inferred from Lowe et al., 2020).
How much of the loss is lean, by method
On an ordinary diet, roughly a fifth to three-tenths of the weight lost is lean. The TREAT authors give the band: «Loss of lean mass during weight loss typically accounts for 20% to 30% of total weight loss.» (Lowe et al., 2020). It is the authors’ reference figure from earlier studies, not a pooled estimate, but it is the yardstick the other numbers here are read against (Protein and Resistance Training for Muscle and Strength).
GLP-1 drugs sit near that band on average, with semaglutide reaching above it in some trials. A meta-analysis of 7 RCTs at obesity doses (821 patients) found two things at once. Lean mass rose as a share of body weight, by 1.81% (95% CI 1.1 to 2.52), because fat went faster. Absolute lean mass fell, by 1.74 kg (95% CI −3.04 to −0.45; I² = 98%), and by 5.44 kg over 52 weeks on semaglutide (Laverde et al., 2026) (Laverde et al., 2026). In the included trials, lean mass made up 14-22% of the loss on liraglutide, about 26% on tirzepatide and up to 45% on semaglutide; the authors call this consistent with an earlier meta-analysis that put the class at about 30% (Laverde et al., 2026) (GLP-1 and Lean Mass).
The authors add that the gain in lean share, though statistically significant, «may not be clinically significant» (Laverde et al., 2026), and they conclude that lean-mass loss «should not be considered a limitation for the use of these drugs in patients with obesity» (Laverde et al., 2026). A 19-drug network meta-analysis gives the same ranking against lifestyle comparators: the drugs that take off the most weight take off the most lean mass — tirzepatide −8.3% of lean mass (95% CI −12.9 to −3.7) and subcutaneous semaglutide −5.8% (−8.7 to −2.9), both moderate certainty (Nong et al., 2026). These are a different comparator and scale from the placebo-controlled figures above, so they confirm the order, not the size (Comparing Obesity Drugs).
One eating pattern pushed the lean fraction well above the band. In the TREAT trial, adults told to eat freely within an 8-hour window lost 1.70 kg on average in the DXA sub-cohort, and 1.10 kg of it (about 65%) was lean. Total lean mass did not differ significantly from the control arm (−0.75 kg; 99.7% CI −1.96 to 0.45), but appendicular lean mass, the arms and legs, did: −0.47 kg (95% CI −0.82 to −0.12) (Lowe et al., 2020).
Read the 65% with care: it sits on a small loss in one sub-cohort. When Liu’s 12-month trial matched calories between a window and a daily deficit and gave both arms a protein shake for six months, lean mass did not differ between groups (Liu et al., 2022). Low protein intake inside a short window is the leading explanation, but neither trial tested it (Time-Restricted Eating) (Liu et al., 2022; inferred from Lowe et al., 2020).
Resistance training keeps lean mass during a diet
In older adults, adding resistance training to a diet keeps about 0.8 kg of lean mass that the diet alone would remove — and leaves fat and weight loss unchanged. A meta-analysis of 6 RCTs in older adults with obesity (mean age above 57) compared calorie restriction with and without RT. The diet-only arms lost 0.76 kg of lean mass; the RT arms held steady (+0.05 kg). The pooled difference was 0.819 kg (95% CI 0.364 to 1.273), with «the LBM loss in the CRRT group was 93.5% less than the CR group» (Sardeli et al., 2018). Fat loss (−3.86 vs −3.73 kg) and weight loss (−4.16 vs −4.73 kg) did not differ (Resistance Training During Caloric Restriction).
Two limits bound that result. It is a tested finding for older adults, not for dieters in general: the one trial in young adults was left out of the pool, so it says nothing about them directly (the case for younger dieters is taken up below). And function stayed a trend: the authors saw «a trend towards the increase in whole body muscle quality when RT was added to CR (p = 0.07)», in a study they say «was not designed to test muscle function» (Sardeli et al., 2018).
In older adults who already have sarcopenic obesity and are not dieting, RT against no exercise, mostly for 12 weeks, improved a composite of body-composition measures (SMD 0.35, 95% CI 0.12 to 0.57; moderate certainty) and physical performance (SMD 0.36, 0.03 to 0.69; low to very low certainty) (Polo-Ferrero et al., 2025). The body-composition gain came mostly from lower body fat; the muscle-mass measures did not change significantly (Resistance Training for Sarcopenic Obesity). This is a different question from RT during a diet, and its numbers do not add to the 0.8 kg above.
Protein helps less than training, and its number depends on who is dieting
Higher protein during a deficit keeps a little more lean mass and does not change how much weight comes off. The evidence comes from three groups of dieters, and each figure belongs to its own group (Protein Intake During Energy Restriction):
| Who | Protein contrast | Lean mass kept (higher vs lower protein) | Source |
|---|---|---|---|
| Adults, mostly with overweight or obesity, no structured exercise | ~1.25 vs ~0.72 g/kg/d, same calories | +0.43 kg (95% CI 0.09 to 0.78) | (Wycherley et al., 2012) |
| Adults aged 50+, mostly overweight or obese | ≥1.0 vs <1.0 g/kg/d | +0.83 kg (95% CI 0.47 to 1.19); body weight no different | (Kim et al., 2016) |
| Non-obese, resistance-trained dieters | dose-response across 0.8-3.2 g/kg/d | lean loss reaches zero, on average, near 1.9 g/kg/d (2.5 g/kg of fat-free mass) | (Refalo et al., 2025) |
The first two rows are randomized contrasts at modest intakes. Kim’s review, of 20 RCTs, concludes that it supports «older overweight and obese adults to consume protein intakes 1.0 g/kg/d to help preserve lean mass» (Kim et al., 2016). In Wycherley’s 24 trials, the higher-protein arms also lost 0.87 kg more fat and 0.79 kg more weight, small differences that may partly reflect misreported intake (Wycherley et al., 2012).
The 1.9 g/kg figure is not a target for people with obesity. It is the point where a straight trend line crosses zero, not a threshold: the line kept rising up to 3.2 g/kg with no knee shown, and the slope’s own interval reaches zero. The authors call their findings «exploratory rather than confirmatory», and the review included only non-obese, trained people (Refalo et al., 2025). For a dieter with obesity, the evidence supports the direction (more protein keeps a little more lean mass) but no optimum, because no held source tested a range of doses in that group (The Underivable Optimum).
A usable default is about 1.0 to 1.25 g of protein per kg of body weight a day. Those are the higher intakes that beat lower ones in the two randomized pools: about 1.25 g/kg in Wycherley’s trials (no BMI restriction; mostly overweight or obese adults; structured exercise excluded), and at least 1.0 g/kg in Kim’s adults aged 50 and over (Wycherley et al., 2012) (Kim et al., 2016). The figure is directional: it shows that this much did better than the lower arms (about 0.72 g/kg in Wycherley’s pool, under 1.0 g/kg in Kim’s), not that it is enough or that more would not help (Protein Intake During Energy Restriction) (Kim et al., 2016; inferred from Wycherley et al., 2012).
Training comes first, protein second. Refalo’s authors judge a robust RT stimulus likely more important for keeping fat-free mass than protein intake, and say the size of the deficit also matters (Refalo et al., 2025). The two may stack. In Sardeli’s pool, the one trial whose diets were higher in protein (30% of energy, against 15% in the other trials) kept more lean mass with RT (1.3 kg) than the authors’ stated main effect of 0.6 kg. That 0.6 kg does not match the 0.819 kg pooled result above, and the paper does not explain the gap. This is one trial compared with the others, not a randomised test of protein (Sardeli et al., 2018).
Low muscle predicts early death; keeping it during a diet has not been shown to help
Older adults with less muscle die sooner, but no trial shows that adding muscle extends life. In 9 studies of non-frail adults aged 65 and over (10,028 people), those who died had less appendicular muscle at baseline than those who survived — a small difference, SMD −0.18 (95% CI −0.23 to −0.12) (de Santana et al., 2021). The design is observational: low muscle may mark an illness that does the killing. It makes muscle mass a predictor, not a proven target (Low Muscle Mass and Mortality).
Whether keeping lean mass during weight loss changes strength, falls or survival has not been tested in the held trials. The GLP-1 meta-analysis states that its trials «did not evaluate functional outcomes, such as muscle strength or physical capacity» (Laverde et al., 2026); the RT-during-diet pool found only a non-significant trend; and the protein reviews measured soft tissue alone. The case for protecting muscle rests on its value for function and independence, and on low muscle predicting death — not on a trial that closed the loop.
Who should care most
Older adults and people on a GLP-1 drug carry the largest stake. Older adults are already losing muscle with age, and their muscles need more protein per meal to respond (Anabolic Resistance). A GLP-1 drug cuts appetite, and probably with it protein intake, at the moment the need goes up (inferred from Laverde et al., 2026). The GLP-1 trials mostly enrolled «relatively young populations with a low burden of comorbidities», so the older, multimorbid group is untested; the authors conclude «it is essential to accompany drug treatment with nutritional and physical exercise interventions to preserve or improve muscle mass» (Laverde et al., 2026) (Laverde et al., 2026).
For a younger adult with obesity and plenty of muscle, the stake is smaller and the improving lean share is the more relevant finding (GLP-1 and Lean Mass) (inferred from Laverde et al., 2026). Resistance training is still worth doing at any age while dieting. It is cheap, low-harm and easy to stop. The diet-trial evidence comes from older adults; in younger dieters it is an extrapolation by mechanism. Refalo’s authors, writing about non-obese dieters who already train, judged a robust training stimulus likely more important than protein for keeping fat-free mass (Refalo et al., 2025) (Resistance Training During Caloric Restriction) (Refalo et al., 2025; inferred from Sardeli et al., 2018).
If some methods cost more muscle than others, the natural suspect is speed. The evidence on speed is thinner than the advice.
Treat speed of loss as unanswered, not as a known muscle cost
No study the wiki holds compares fast with slow loss of the same amount by the same method. So whether a slower rate of loss spares lean mass is an open question, not a finding. The familiar advice to lose slowly to keep muscle leans on nearby evidence, and each piece measures something other than rate:
- Deficit size. In Refalo’s meta-regression of non-obese, resistance-trained dieters, the link between protein intake and lean-mass change looked the same with deficits under and over 300 kcal/d. An earlier meta-regression of 52 studies, reported second-hand through Refalo, found that larger deficits cost more lean mass even with RT. Refalo’s authors advise a modest deficit for anyone who must avoid lean-mass loss (Refalo et al., 2025) (Protein Intake During Energy Restriction). A bigger daily deficit does mean faster loss, but none of these studies held the total loss fixed, so they cannot separate speed from amount.
- Method and protein. The lean fraction of weight lost differs by method and by protein intake, as the muscle section above shows (GLP-1 and Lean Mass, Time-Restricted Eating); none of those trials varied the rate.
- The appetite defence. The one held study that tracked appetite hormones a year after loss used a 500-550 kcal/day formula diet (Sumithran et al., 2011). Whether a slower, food-based loss provokes a weaker defence has not been tested (Mitigations of the Defended Setpoint, Weight-Loss Maintenance and Metabolic Adaptation).
If keeping lean mass matters to you, a modest deficit with resistance training and enough protein (the 1.0 to 1.25 g/kg default above) is the low-regret choice. It costs little and fits the deficit-size evidence above. The only lean-mass deficit threshold held is that second-hand 500 kcal figure; Refalo’s own 300 kcal split found no difference and is not a threshold. As a labelled default, that puts modest at about 500 kcal a day or less. It is a hedge against an unanswered question, not a tested answer about speed.
Speed is still open. What happens after the loss is better known: for most people, much of it comes back.
Plan for regain from the start, because the body defends the lost weight
Most people regain much of the weight they lose, and no diet method changes that very much. Weight loss peaks at around six months, then drifts back to a modest net loss that lasts. The method changes the peak far more than the plateau. The drugs are the exception: they hold the loss while you take them and stop holding it when you stop. The body pushes back through appetite and, less certainly, a slower metabolism; the thin evidence held points to appetite as the larger push, and shows nothing that wears it down over time. That is a reason to plan the maintenance before the loss, not a reason to skip the attempt (Weight-Loss Maintenance and Metabolic Adaptation, Mitigations of the Defended Setpoint).
Every lifestyle method peaks early, then settles at a few kilograms
Franz pooled 80 RCTs of overweight and obese adults followed for at least a year (26,455 enrolled). Diet, diet plus exercise, meal replacements and the drugs of the day all produced 5 to 8.5 kg (5-9%) by six months, then levelled off: «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.» The review’s next sentence sets two arms apart: advice alone and exercise alone lost little at any time point (Franz et al., 2007).
| Method (Franz) | Peak, ~6 months | Latest reported |
|---|---|---|
| Diet alone | 4.9 kg (5%) | 3.0 kg (3%) at 48 mo |
| Diet + exercise | 7.9 kg (8.5%) | 3.9 kg (4%) at 36 and 48 mo |
| Very-low-energy diet (VLED, formula) | 17.9 kg (16%) | 5.6 kg (5%) at 36 mo |
| Orlistat + diet (pre-GLP-1 drug) | 8.3 kg (8%) | 5.8 kg (5.3%) at 48 mo |
| Exercise alone | 2.4 kg (2.7%) | 1.0 kg (1%) at 24 mo |
The formula diet makes the biggest peak and loses the most of it. Exercise alone stays smallest throughout, as a weight tool; that says nothing about its value for fitness or health. Added to a diet, exercise deepened the peak and left about a kilogram more at 48 months, across separately pooled, non-randomised arms (see the exercise section above). The orlistat row predates the GLP-1 drugs and understates what current drugs do (Weight-Loss Maintenance and Metabolic Adaptation).
Read these maintained figures as a best case. Most of Franz’s outcomes come from people who finished the trials — «Most weight-loss outcomes were based on study com- pleters.» — and about 30% dropped out. If dropouts regained more, which the review did not measure, a typical person keeps less than the table shows. The descriptive trajectories are also simple averages across studies, not weighted pooled estimates (Franz et al., 2007).
Franz also records why the plateau frustrates people. Holding or deepening the diet does not restart the loss — «This appears not to happen even when weight-loss interventions are contin- ued.» — yet dropping the programme altogether brings the weight back (Franz et al., 2007). The deficit stops paying out at the plateau but still has to be held.
Newer trials show the same fade: the named-diet network’s advantage over usual diet shrinks between six and twelve months, as the diet section above sets out (Named Diet Programs Compared). In type 2 diabetes, low-carbohydrate diets beat comparison diets by 3.46 kg (95% CI 1.67-5.25) at six months; by twelve the difference was trivial and non-significant, with low-carb 0.29 kg behind (95% CI 1.60 kg behind to 1.02 kg ahead), and the remission advantage faded with it (Goldenberg et al., 2021) (Carbohydrate Restriction and Type 2 Diabetes Remission).
Total diet replacement held better, with repeated rescue
DiRECT, a trial of total diet replacement (TDR) in people with type 2 diabetes, followed the same 298 people to two years. Remission fell from 46% of the intervention arm at twelve months to 35.6% at twenty-four, against 3.4% in usual care: about 32 more remissions per 100 people started on the programme (inferred from M. E. Lean et al., 2018; M. E. J. Lean et al., 2019) (M. E. Lean et al., 2018) (M. E. J. Lean et al., 2019). The share keeping off at least 15 kg roughly halved, from 24% (36 of 149) to 11.4% (17 of 149), and the weight gap over usual care narrowed from 8.8 kg to 5.43 kg (95% CI 3.99-6.87) (Total Diet Replacement and Type 2 Diabetes Remission).
Holding that result took active rescue. About half of participants needed at least one relapse plan — a short return to TDR, with orlistat offered — and only 71 of 143 (49.7%) never needed one (M. E. J. Lean et al., 2019). Across the entire study population (both arms pooled), remission at two years was reached by 8 of 154 people (5.2%) who lost less than 5 kg and in 14 of 20 (70.0%) who lost 15 kg or more (M. E. J. Lean et al., 2019).
That is an association, not proof that remission lasts only while the weight stays off. Four of the 50 people who gained weight (gains of 0-2 kg) were still in remission at both 12 and 24 months, though all four started with HbA1c only just above the diabetes threshold (6.5-6.63%). Among people in remission at 12 months, those who relapsed regained 7.09 kg on average against 4.25 kg in those who stayed in remission, a difference that was not statistically significant (p=0.073) (M. E. J. Lean et al., 2019). The two-year figures come from the authors’ final manuscript, and the published version may differ slightly (Total Diet Replacement and Type 2 Diabetes Remission).
The drugs hold the loss only while you take them
STEP-1 stopped semaglutide 2.4 mg at week 68 and followed a subset of 327 participants for a further year, in analyses the authors call exploratory. In that subset, weight loss went from 17.3% at week 68 to a net 5.6% at week 120: «participants regained two-thirds of their prior weight loss, with similar changes in cardiometabolic variables» (Wilding et al., 2022). Blood pressure returned to baseline, and most other markers drifted back (Semaglutide for Cardiovascular Risk in Obesity).
The trial withdrew the lifestyle programme at the same time, and its authors say that «may also have contributed to the trajectory of weight regain» (Wilding et al., 2022). So STEP-1 fits a defended weight reasserting itself, but cannot separate the drug’s withdrawal from the loss of support. A class-wide figure — about 0.4 kg regained per month after stopping, back to baseline in roughly 1.7 years — comes from a 37-study review (West 2026) that the wiki holds only as reported inside Nong’s network meta-analysis, not first-hand (Nong et al., 2026) (Nong et al., 2026) (Comparing Obesity Drugs).
Appetite may do most of the defending
In one study Hall reports, intake rose much more than metabolism slowed. Hall’s review reports a study of patients with type 2 diabetes losing weight on an SGLT2 inhibitor, which spills glucose into the urine; they «progressively increase average energy intake by about 100 kcal/d per kilogram of lost weight – an effect more than 3-fold larger than the cor- responding energy expenditure adaptations to weight loss.» (Hall & Guo, 2017) That intake was calculated from repeated weight measurements with a mathematical model, not measured, in a drug-driven loss the patients did not notice (Hall & Guo, 2017). It is one study, held only second-hand, so the ranking of appetite over metabolism is a lead, not a settled finding.
The hormones point the same way. In a study that enrolled 50 adults, the 34 who finished lost 13.5 kg in a 10-week programme built on an eight-week formula diet; a year later their leptin was still 35.5% below baseline, hunger hormones stayed up and satiety signals stayed down (Sumithran et al., 2011). That study had no control group and analysed only completers who had lost at least 10%. Participants were also still about 8 kg lighter, so only the leptin drop is shown to exceed what a smaller body would produce (Weight-Loss Maintenance and Metabolic Adaptation).
Some people read that pull as an addiction. A systematic review concludes: «The results of the current systematic review generally support the validity of food addiction as a diagnostic construct, particularly as it relates to foods high in added sweeteners and refined ingredients.» (Gordon et al., 2018) The same review opens by calling the construct «a highly controversial subject» (Gordon et al., 2018). It tested whether the construct is valid, not what it does to weight or health, and no held study links it to weight change apart from ordinary overeating. The label adds no step beyond eating fewer highly processed, energy-dense foods (Food Addiction).
Whether metabolism stays slowed is not settled. The slower-metabolism arm of the defence, adaptive thermogenesis, «may continue for years», Hall writes, «although controversy remains regarding its persistence» (Hall & Guo, 2017). On the one estimate held, it is also the smaller arm. Two common beliefs fit this evidence poorly. One is that eating too little stalls fat loss, yet the formula diets above produced the largest losses. The other is that dieting permanently damages metabolism, yet persistence is contested and the slowdown looks like the smaller arm (Weight-Loss Maintenance and Metabolic Adaptation). Body weight behaves as a regulated quantity, but regulated does not mean fixed. Genes explain much of why people in the same environment differ, while the rise in obesity over recent decades tracks the food environment (The Disease Model of Obesity).
Waiting does not wear the defence down; a few things counter it
The held evidence sorts the counters by what they do (Mitigations of the Defended Setpoint):
- Drugs override the defence by resupplying satiety signals, and lose the effect when stopped.
- Bariatric surgery durably lowers weight, in severe obesity, on non-randomised evidence (figures in the surgery section above). Whether it resets the defended level is not shown.
- Holding a lower weight for months shows no sign of weakening the defence. The one measurement, at twelve months, found the hormonal shift still present.
- Protein and resistance training may blunt one arm by keeping lean mass, and a gentler deficit might provoke less defence; both are inferred, not tested.
- Avoiding the gain in the first place sidesteps the defence entirely — a mechanism-based expectation, not a trial result.
on the sorting; surgery and drug figures from Mitigations of the Defended Setpoint.
Trying and partly regaining is not shown to be worse than not trying
Weight cycling does carry an association with harm in cohorts. Zou’s review of 23 cohorts (441,199 people) found, across the 20 that reported deaths, all-cause mortality RR 1.18 (95% CI 1.05-1.32) after correcting for publication bias, down from 1.41 (Zou et al., 2019). But in the few reports that separated the two (three each), intentional weight fluctuation showed no significant link with all-cause (P=0.49) or CVD mortality (P=0.12). Unintentional fluctuation, which often signals illness, did show a link (all-cause P=0.016), though on only 2,975 people, which the authors say limits that conclusion. A second review by the same group, on diabetes, points the other way for intentional loss. In three reports, people with intentional weight cycling had a higher risk of diabetes (P < 0.001), though «the small number of studies limits the conclusion drawn based on these studies». In people with BMI 30 or more, that review found no significant association (Zou et al., 2020) (Weight Cycling and Cardiometabolic Risk).
The authors draw the decision line themselves: «Hence, we cannot conclude whether obese individuals should be deterred from efforts to control their body weight by concerns about the hazards of weight cycling.» (Zou et al., 2019) Two caveats come before that sentence. Most studies cannot tell why weight was lost, which narrows the harm signal. And the CVD-mortality link was absent at BMI 30 or more, which the authors read as obesity carrying a similar risk of its own. Stopping a GLP-1 drug and regaining is, in effect, a planned cycle; no held source has studied it as one. (inferred from Zou et al., 2019)
What the evidence cannot yet say
- How weight holds over five to ten years on a GLP-1 drug: not in the held evidence.
- Lifestyle durability beyond about four years: only Franz, and only from completers.
- Regain after stopping the drugs, from primary data: held only secondhand, via Nong.
The newer drugs take off as much as any non-surgical method but hold it only while taken. The last question is what lifestyle effort still adds for someone who could take one.
Rank lifestyle against the drug on each outcome, net of what the drug costs
An effective drug weakens the case for a lifestyle lever only on the outcomes the drug itself moves, and only after you subtract what the drug costs: side effects, stopping, lifelong use, and the regain that follows when it ends. This appraisal ranks every lever that way (Layer 1 - Ranking Interventions for a Stratum). Applied outcome by outcome, the GLP-1 drugs and lifestyle effort split the territory rather than compete for all of it. No trial has compared them head to head, so this ranking is assembled from separate bodies of evidence.
Heart, kidney and survival: the drug can stand in, mainly in type 2 diabetes
A pooled analysis of 11 placebo-controlled cardiovascular-outcome trials (85,373 people, ten of the trials in type 2 diabetes plus SELECT) found that the GLP-1 drug class lowers hard outcomes at high certainty. In type 2 diabetes, major cardiovascular events fell by HR 0.87 (95% CI 0.81-0.93) and death from any cause by HR 0.88 (0.83-0.93); median follow-up across the 11 trials was 25.2 months (range 15.9-64.8) (Badve et al., 2025). The reported numbers needed to treat, 74 for an event and 101 for a death over 25.1 months, work out to roughly 14 fewer cardiovascular events and 10 fewer deaths per 1,000 people treated for about two years (arithmetic on Badve’s NNTs). The class also lowered hard kidney outcomes, though the absolute gain is small outside people at high kidney risk (GLP-1 Receptor Agonists and Cardiovascular and Kidney Outcomes).
The result does not reach everyone. SELECT is the only trial in the pool without diabetes, and it enrolled people who already had heart disease. The result is strong for type 2 diabetes and rests on one trial for anyone else (Semaglutide for Cardiovascular Risk in Obesity). Tirzepatide was excluded by design. And these were glucose-lowering trials with modest weight loss. The benefit may owe as much to the drug as to the weight lost, and the meta-analysis does not separate the two, so it does not show that losing weight prevents heart attacks (Does Weight Loss Reduce Cardiovascular Events).
Where the drug does stand in, for someone with type 2 diabetes and high cardiovascular risk, a lifestyle programme no longer has to carry the heart and survival case alone. On type 2 diabetes remission and liver fat, nothing held shows that lifestyle does what a drug cannot. Both track the kilograms lost (Total Diet Replacement and Type 2 Diabetes Remission), and no held trial compares drug and lifestyle on remission. For the liver, AASLD reports a phase 2b trial of daily semaglutide in 320 people with NASH resolved it in 59% against 17% on placebo, without a significant reduction in fibrosis, although «a dose-dependent decrease in progression was observed» (Rinella et al., 2023). For neither route does the held evidence show an effect on liver-related outcomes or death (Fatty Liver MASLD and Weight Loss). The channel the drug clearly does not replace is muscle and function.
Muscle and function: no prescription replaces training
On these outcomes the drug does not substitute; it adds a cost. The agents that take off the most weight also take off the most lean mass, as the muscle section above shows. None of those with moderate-to-high certainty evidence lifted quality of life by a margin patients notice (Comparing Obesity Drugs).
Exercise alone moves weight by about a kilogram in people with obesity; added to a diet it adds somewhat to the loss, and it shifts the loss slightly toward visceral fat. Resistance training kept lean mass during a deficit in older adults (Exercise Modality for Body Composition in Obesity, Resistance Training During Caloric Restriction). On muscle and function, training keeps its full rank whether or not someone takes a drug, though the held trials measured lean mass more often than function, and the obesity exercise trials found no significant gain in VO2max. Someone on a drug has more reason to train, not less (GLP-1 and Lean Mass) (inferred from Laverde et al., 2026).
The drug’s costs enter the ranking
Stopping shrinks the drug’s value as a substitute. Across the pooled trials (not split by diabetes status), 12.7% of people on a GLP-1 drug stopped for adverse events against 9.2% on placebo, RR 1.51 (1.18-1.94), about 35 more per 1,000 (arithmetic on the two rates), though heterogeneity was very high (I² 96%) (Badve et al., 2025). The obesity-drug figures in the section on formula diets and drugs come from a different population and measure; only their direction is shared (Comparing Obesity Drugs).
The side effects are set out with the drugs above (GLP-1 Non-Cardiometabolic Effects and Safety), and the benefit does not appear to last without continued treatment (Nong et al., 2026); the regain after stopping is set out under Plan for regain from the start. Price and lifelong use sit on an affordability axis this appraisal does not weigh. The trade-off exists, and the person weighs it (Which Objective Moved This Recommendation).
This ranking sizes the levers; it does not choose between them. Whether to take the drug, train and eat to a deficit, or do both depends on which outcomes a person cares about most, how well they tolerate the drug, and whether they expect to keep taking it (Comparing Obesity Drugs). That leaves a short list of what to do.
Act on the deficit, protect muscle, plan the maintenance
The levers that do the most are plain, and the open questions are about muscle and the long run, not about which diet to pick.
Choose the eating pattern you can keep, and size the deficit. Switching between diet labels buys little (Named Diet Programs Compared); where calories were not matched, credit the deficit, not the food (Is the Food Category Doing Any Work). A starting figure is 500 kcal a day below estimated needs, the deficit the large drug trials prescribed; it is a trial convention, not a tested optimum (Wilding et al., 2021) (Semaglutide for Cardiovascular Risk in Obesity).
Eat fewer ultra-processed, energy-dense foods. In one inpatient crossover, 20 adults ate about 508 kcal a day more on an ultra-processed diet and gained 0.9 kg in two weeks. The trial was small, short and run on a metabolic ward, and it measured intake and weight, not long-run loss (Hall et al., 2019) (Ultra-Processed Food and Health Outcomes, Energy Density).
Train for the body first and the scale second. Add resistance training at any age (tested in older dieters, extrapolated to younger ones) and about 1.0 to 1.25 g of protein per kg of body weight a day, the higher intakes that kept more lean mass in dieting trials of mostly overweight or older adults; treat the figure as directional (Wycherley et al., 2012) (Kim et al., 2016). In non-randomised comparisons, diet-plus-exercise programmes lost a little more than diet alone; exercise does more for visceral fat and waist (Resistance Training During Caloric Restriction, Protein Intake During Energy Restriction, Exercise Energy Compensation).
Match the tool to the size of the loss you need. If it is large or your risk is high, formula diets, drugs and surgery are the methods with the size to match (Total Diet Replacement and Type 2 Diabetes Remission). Weigh a drug’s hard-outcome evidence by whether you resemble the trial population, and weigh its side effects and the regain when you stop (GLP-1 Receptor Agonists and Cardiovascular and Kidney Outcomes).
Plan the maintenance from the first week. Keep the eating pattern going after the loss stops, because the lower intake still has to be held. Franz’s review warns that «if weight-loss interventions are discontin- ued entirely, weight regain is likely to occur» (Franz et al., 2007). Set a regain trigger in advance, as DiRECT did: 2 to 4 weeks of partial meal replacement after more than 2 kg regained, a return to the formula diet after more than 4 kg; about half its participants needed at least one (M. E. J. Lean et al., 2019). That trigger was part of DiRECT’s protocol in type 2 diabetes and was not tested on its own (inferred from M. E. J. Lean et al., 2019). Keep regular check-ins (DiRECT’s were monthly) (M. E. Lean et al., 2018). Expect appetite to push back, and treat stopping a drug as a decision with a known regain (Weight-Loss Maintenance and Metabolic Adaptation, Total Diet Replacement and Type 2 Diabetes Remission, Mitigations of the Defended Setpoint).
Do not let fear of regaining stop you from trying. In the few studies that separate it out, weight cycling from intentional loss was not linked to higher mortality, though three reports linked it to more diabetes (Weight Cycling and Cardiometabolic Risk).
Three questions remain open: whether losing slowly spares muscle, whether keeping muscle during a loss changes function or survival, and what five to ten years on a GLP-1 drug look like (Weight-Loss Maintenance and Metabolic Adaptation).
Evidence box
Question ’For an adult who has decided to reduce body fat: how do the available methods — diet composition, calorie restriction, exercise, total diet replacement, and pharmacotherapy (the GLP-1 class) — compare on how much weight comes off, what else is lost (lean mass, function), whether the loss reaches a hard outcome or only a surrogate, and how durably it lasts; and does the rate of loss or the source of the deficit (diet-created vs exercise-widened) change body composition over and above the amount lost?‘ Evidence included 46 sources — 22 gold, 22 high, 2 moderate Overall certainty Medium (see Rating Certainty of Evidence) Source-selection note 2 source(s) below the gold/high evidence bar feed this page: Lowe (RCT, moderate); Taylor (mechanism, moderate). Their roles and limitations are described in the text. Last updated 2026-10-20 · Independently reviewed: No · Full edit history