Where fat sits matters more than how much of it there is
Two people can carry the same weight on the scale and face very different risk, because the fat that does the damage is the fat inside and around the organs — the liver and pancreas above all — not the total-body fraction. A person can be normal-BMI and metabolically ill, or heavier and, for a time, not. So the first move in this deliverable is to read depot — the waist, the intra-organ compartment — rather than a total-adiposity number like BMI or body-fat percentage, and to treat the familiar U-shaped “overweight is protective” mortality curve as mostly an artifact of confounding and reverse causation rather than a licence.
From there the questions unfold in order: what does losing fat actually buy — hard outcomes or only surrogates; what the ectopic-fat ladder drives; what the method of loss does to visceral fat and to muscle; which method moves which outcome, and why a calorie is a calorie for storage; and finally, below, whether the loss lasts, how the levers shift by stratum, and how the arrival of an effective drug re-sizes the whole decision.
Where fat sits carries the risk — read the depot, not the scale
The first decision about body fat is not how much but where. Cardiometabolic risk tracks the fat that overflows into the liver, pancreas and viscera once safe subcutaneous storage is exceeded, far more faithfully than it tracks total fat mass or the number on the scale. A normal-weight person can be metabolically ill and a heavier person can be, for a time, not — so a total-adiposity reading (BMI, or even body-fat percentage) is the wrong instrument, and the familiar U-shaped “overweight is protective” mortality curve is largely an artifact, not a licence. (Kramer et al., 2013; inferred from Taylor & Holman, 2014)
Three fat compartments are three different objects
Subcutaneous fat is the safe, expandable store; its capacity varies by person and ethnicity and is not itself pathogenic. Intra-organ fat — inside the liver and pancreas — is the pathogenic depot: liver fat drives hepatic insulin resistance, pancreatic fat suppresses insulin secretion. Visceral (intra- abdominal) fat is commonly cast as the villain, but Taylor demotes it to a marker: extent of visceral fat «is a surrogate marker for intra-organ fat ex- cess, but is not pathophysiologically related to adverse metabolic consequences». (Taylor & Holman, 2014) So waist circumference is a surrogate for a surrogate — useful because it tracks the intra-organ fat doing the damage, one step removed from it. (inferred from Taylor & Holman, 2014) -> Ectopic Fat and Depot-Specific Risk
BMI’s protective band is mostly an artifact — adjudicate the arm, not the curve
The observational BMI-mortality curve dips in the overweight range, and that dip is where the “obesity paradox” lives. The 10.6M-participant Global BMI IPD meta-analysis strips it away step by step: the overweight arm (BMI 25-30) walks from an apparent HR 0.96 (0.95-0.97) raw, to 0.99 after adjusting smoking and excluding baseline disease, to 1.03 after dropping the first 5 years of follow-up, to 1.11 (1.10-1.11) once restricted to never-smokers. (Global BMI Mortality Collaboration, 2016) Isolating smoking alone — same disease and follow-up exclusions — overweight is 1.07 (1.06-1.07) in never-smokers versus 0.94 (0.94-0.95) in ever-smokers. (Global BMI Mortality Collaboration, 2016) The protection was manufactured by confounding and reverse causation, not a real benefit of carrying extra weight -> The U-Shaped Association Artifact.
The genetic check converges. Wade’s Mendelian-randomization analysis in UK Biobank finds the J-shape survives but deflates: it remains «but with a smaller value of BMI at which mortality risk was lowest (~23 vs. ~26 kg/m2 with observational analyses) and apparently flatter over a larger BMI range». (Wade et al., 2018) Reading the nadir off the raw observational curve overstates the harm of being underweight and understates the harm of being obese — so the low arm is largely artifact, adjudicated by the strong (genetic) check, not merely argued -> BMI and All-Cause Mortality.
Metabolic status beats the BMI band
If risk tracks depot and metabolic state rather than mass, then obesity should not be safe merely because current metabolic markers are clean, and lean people with a bad profile should carry high risk. Kramer’s meta-analysis of observational cohorts shows exactly this: metabolically-healthy obesity is not durably benign (RR 1.24 (1.02-1.55) versus metabolically-healthy normal weight in studies with >=10 years’ follow-up — «there is no healthy pattern of increased weight»), while metabolically-unhealthy normal weight carries RR 3.14 (2.36-3.93). (Kramer et al., 2013)
Opio 2020 refines this on a larger base (23 prospective cohorts, n=4.49M) and answers the sub-question Kramer could not: the excess CVD risk holds even with zero metabolic risk factors — metabolically- healthy overweight RR 1.51 (1.21-1.88), metabolically-healthy obesity RR 2.18 (1.28-3.71) — and does not require a decade of latency (present at both <10y and >=10y, subgroup-difference p=0.98). Metabolically-unhealthy normal weight again dominates at RR 3.07 (2.27-4.15). (Opio et al., 2020) The honest bound: these are observational RRs with high heterogeneity and live fitness confounding, and the absolute excess is modest — the only held absolute anchor is Kramer’s ~0.7% over 10-11 years — so metabolically-healthy obesity licenses continued, not urgent, attention -> Baseline Risk and the Relative-Absolute Split.
On a hard endpoint, abdominal fat is the harmful depot — but the marker you pick barely changes the prediction
Kramer and Opio are metabolic-status sources; INTERHEART extends the depot distinction to a hard endpoint (first acute MI, 27,098 people, 52 countries) and reads the same shape. Adjusted for BMI, the top-versus-bottom-quintile odds ratio for MI is 1.77 (1.59-1.97) for waist (harmful) and 0.73 (0.66-0.80) for hip (protective); waist-to-hip ratio is the strongest single marker at 1.37 (1.34-1.41) per SD, while BMI is the weakest at 1.10 (1.07-1.13) and its whole MI association vanishes after adjusting for WHR (1.44 -> 1.12) then the other risk factors (-> 0.98). (Yusuf et al., 2005) Abdominal fat is harmful, lower-body fat protective — the depot claim, now on a cardiovascular event rather than a glycemic one.
INTERHEART read this as proof that clinics should abandon BMI and switch to waist-to-hip ratio. A larger prospective test overturns that marker-ranking claim while leaving the depot claim intact. ERFC 2011 pooled 58 cohorts (221,934 people) with BMI, waist and WHR measured in the same people at baseline, and found the three near-identical in strength — per-SD hazard ratios for cardiovascular disease of 1.23 (1.17-1.29) for BMI, 1.27 (1.20-1.33) for waist and 1.25 (1.19-1.31) for WHR — and «reliably refute[s] previous recommendations to adopt baseline waist-to-hip ratio instead of BMI as the principal clinical measure of adiposity» (Collaboration, 2011).
The ~3-fold WHR-over-BMI gap is a case-control artifact: acute illness before an MI strips skeletal muscle, lowering hip circumference and raising WHR in the cases measured after the event. BMI is also the more reproducible measure (regression-dilution ratio 0.95 vs 0.63 for WHR).
Two things survive the contest, and both bear on the decision. First, abdominal adiposity is still a real, modifiable determinant — ERFC is explicit that its null concerns prediction on top of the intermediates, not aetiology: once a person’s blood pressure, lipids and glucose are known, «simple adiposity measures provide little or no additional information on cardiovascular risk», because the fat acts through those markers.
Second, ERFC’s sample was 90% European descent, so INTERHEART’s finding that BMI carries no MI signal at all in South Asian, Arab and mixed-race African groups is untested, not refuted. Net: for a developed-country adult whose risk factors are known, BMI is an adequate and more-reproducible clinical measure, and the case for switching to WHR does not survive prospective design; the WHR case is strongest where lipids are unmeasured and in the non-European strata ERFC could not test. (Collaboration, 2011; inferred from Yusuf et al., 2005) -> Waist-to-Hip Ratio and Cardiovascular Risk, BMI vs Abdominal-Adiposity Markers - Which Predicts CVD.
Is visceral fat the pathogenic depot, or a marker for intra-organ fat? — a ladder, not a contradiction
Two readings sit side by side. INTERHEART and AASLD put central adiposity at the causal centre: INTERHEART calls waist and WHR «simple and crude surrogate measures for visceral obesity, which is probably the key determinant of metabolic abnormalities» (Yusuf et al., 2005), and AASLD states that «Visceral fat, which is more metabolically active and inflammatory than subcutaneous fat, mediates the majority of this risk.» (Rinella et al., 2023) Taylor goes one step past both, demoting visceral fat to a marker of the intra-organ (hepatic and pancreatic) excess that actually does the damage.
This is not a contradiction to adjudicate but a nested causal ladder — waist marks visceral fat marks ectopic fat — where the sources stop at different rungs. The issue is not fully joined: AASLD’s own mechanism routes visceral fat’s harm through intra-organ fat («insulin signaling is progressively impaired, promoting the inappropriate release of fatty acids leading to intrahepatic lipid accumulation») (Rinella et al., 2023) — compatible with Taylor’s marker reading if the two depots are tightly coupled.
The decision consequence is small: every rung points to the same lever (an energy deficit) and the same measurement — waist stays the right thing to read precisely as a marker, one step removed from the fat doing the damage. (Rinella et al., 2023; inferred from Taylor & Holman, 2014; Yusuf et al., 2005)
Body-fat percentage is better than BMI but still the wrong target — and its optimum is a named gap
Body-fat percentage measures the same thing BMI does — total adiposity — more accurately, but it is the same category of measurement, and sits subordinate to depot in this hierarchy. The risk carrier is where the fat is, not the total-fat fraction. No held gold source carries a body-fat-percentage optimum or nadir against all-cause mortality; asserting one would launder a total-adiposity number into the depot claim it cannot make. The BF%-versus-mortality threshold is an explicit named gap — its absence, not a number. Read depot (waist/WHR); treat any total-adiposity cutpoint, BMI or BF% alike, as a coarse screen, not a target. (inferred from Taylor & Holman, 2014; Yusuf et al., 2005)
The Personal Fat Threshold — and why reversal happens below BMI 25
Taylor’s model gives the individual-versus-population gap a name: each person has a personal fat threshold above which lipid spills into the organs, and «the hypothesized PFT is independent of BMI». (Taylor & Holman, 2014) The data bear the reframe out — 36% of newly-diagnosed T2DM occurred at BMI <25 (against 64% below 25 in the contemporaneous UK population), so diabetes does not require obesity, only carrying more fat than you can store safely. The threshold itself shifts by ethnicity (BMI 30 in a Caucasian is matched at 25.2 in South Asians, 27 in African/Caribbeans). (Taylor & Holman, 2014)
The reversal arm is the sharpest decision-change here: a normal-weight person with T2D is above their PFT even at BMI <25, and responds to weight loss — at a smaller required loss. In Taylor’s cohort, at a presenting glucose of 8-10 mmol/l, «normoglycaemia was achieved with a mean weight loss of 13% if body weight was normal, whereas a mean weight loss of 21% of body weight was required to achieve this in the whole cohort». (Taylor & Holman, 2014) This is the stratum the remission trials routinely exclude (DiRECT required BMI >=27), yet the depot lens says to offer them the lever -> Total Diet Replacement and Type 2 Diabetes Remission.
The decision
Rank cardiometabolic risk by central adiposity (waist, waist-to-hip ratio) and metabolic status, not by BMI and not by a body-fat percentage. Where the full metabolic panel is already in hand, it dominates the anthropometric marker — waist and WHR earn their keep mainly as a screen when blood pressure, lipids and glucose are not yet known, and as a reason to look harder (Collaboration, 2011). A normal-BMI person with a high WHR or a bad metabolic panel is above their personal fat threshold and carries elevated MI risk; a currently-clean panel in an obese person buys less durable safety than it appears to.
The lever in every case is negative energy balance, whatever delivers it, drawing the intra-organ and visceral depots back down — the arithmetic of who benefits how much then runs through baseline risk -> Baseline Risk and the Relative-Absolute Split. (Kramer et al., 2013; inferred from Taylor & Holman, 2014; Yusuf et al., 2005)
Losing fat reliably moves the markers; it moves hard outcomes only by some routes
The intuitive claim is that shedding fat prevents heart attacks and extends life. Tested head-on, the claim splits: fat loss moves surrogates — weight, HbA1c, blood pressure, lipids, inflammation, liver enzymes — almost every time, but it moves hard outcomes (major cardiovascular events, mortality) only by particular routes and in particular people. The demonstrated hard-outcome wins are narrow: bariatric surgery and a GLP-1 drug. Lifestyle weight loss has not cleanly shown one. Most of the “weight loss works” evidence lives on the surrogate side of that line, and reading it as an outcome win is the error this section guards against. (inferred from Look AHEAD Research Group, 2013; Ma et al., 2017)
The landmark null: every surrogate moved, the events did not
Look AHEAD is the largest, longest randomized test of the lifestyle route — 5,145 overweight adults with type 2 diabetes, intensive lifestyle intervention versus diabetes education. The intervention improved fitness and every cardiovascular risk factor except LDL cholesterol (Look AHEAD Research Group, 2013); a Look AHEAD CRP substudy separately found the lifestyle arm drove a ~40% relative reduction in C-reactive protein, similar in statin and non-statin users (Belalcazar et al., 2013).
Yet the primary cardiovascular composite was null: HR 0.95 (95% CI 0.83-1.09, P=0.505), and the trial was stopped for futility at 9.6 years median follow-up. (Look AHEAD Research Group, 2013) A bundle of favourable surrogate changes bought no measurable reduction in events — the Surrogate Outcomes warning made concrete. (inferred from Look AHEAD Research Group, 2013)
The null is narrower than it sounds. The two arms converged over follow-up, so the sustained contrast was a modest weight-loss dose rather than big-loss-versus-none; both arms received good diabetes care, shrinking any between-arm gap; and the trial was sized to detect only a large effect, leaving a smaller true benefit undetectable. (inferred from Look AHEAD Research Group, 2013)
The meta-analysis generalizes it: CV events null, all-cause mortality falls by a non-cardiac route
Ma 2017 pools 54 RCTs (30,206 mostly non-diabetic obese adults, weight-reducing diets) and splits cleanly. On the heart it confirms Look AHEAD: CV events RR 0.93 (95% CI 0.83-1.04) and CV mortality RR 0.93 (0.67-1.31), both null. But all-cause mortality falls: RR 0.82 (0.71-0.95), high quality, ~6 fewer deaths per 1000. (Ma et al., 2017) Since CV events are null, that mortality benefit is not routed through the heart — a puzzle the CV endpoint cannot resolve.
Two honesty notes bound this. Look AHEAD carried 54.6% of the CV-strand weighting, so “the MA generalizes Look AHEAD” is partly self-containing there; but the all-cause benefit is not a Look AHEAD artefact — without that trial it is larger (RR 0.78, 0.63-0.96, I^2 = 0%). And the trials were “usually low in fat and saturated fat,” so part of the benefit may travel the SFA -> LDL channel rather than weight loss per se — the design cannot separate the two. (Ma et al., 2017)
Where hard outcomes DO fall: the extreme surgical dose and the drug route
Bariatric surgery in severe obesity (SOS) sits at the opposite end of the dose-response and finds the mortality benefit the moderate-dose trials could not. It reports an unadjusted all-cause-mortality HR 0.76 (95% CI 0.59-0.99, P=0.04), strengthening to an adjusted HR 0.71 (P=0.01) because the surgery arm carried baseline survival disadvantages (heavier, more smokers). The design firewall is essential: SOS is not an RCT — patients self-selected (“desiring surgery”), so unmeasured motivational confounding is untouched by adjustment. (Sjöström et al., 2007) SOS also cannot attribute the benefit to weight loss per se (surgery alters gut hormones and satiety), and mortality did not track the degree of weight loss within the study — so it anchors a large-intervention point, not cleanly a large-weight-loss-dose one.
The drug route supplies the clean randomized hard-outcome win in obesity itself: in secondary-prevention obesity, semaglutide cut the CV composite (MACE HR 0.80, 95% CI 0.72-0.90). Full benefit-risk appraisal, per-agent ranking, and dosing live in the separate GLP-1 deliverable -> Semaglutide for Cardiovascular Risk in Obesity, GLP-1 Drugs; here it functions only as the demonstrated drug-route data point.
That win is not a one-off. A class meta-analysis of 11 cardiovascular-outcome trials (Badve 2024; 85,373 participants) finds the whole GLP-1 receptor-agonist class cuts hard events — «a 13% reduction in the risk of MACE (HR 0·87, 95% CI 0·81 to 0·93; high-certainty evidence)» in the type-2-diabetes trials, consistent regardless of diabetes status (Badve et al., 2025).
That sharpens the route-matters point rather than crediting weight loss: these were glucose-lowering trials with modest weight change, and the CV benefit separates early, before much weight is lost — a drug-specific pleiotropic effect, not a weight-loss effect. So the drug route to CV events is robust and class-wide, while the lifestyle weight-loss route’s CV-event null still stands -> GLP-1 Receptor Agonists and Cardiovascular and Kidney Outcomes. (inferred from Badve et al., 2025)
Tirzepatide reaches surgical-scale loss (SURMOUNT-OSA: body-weight treatment difference -16.1%, 95% CI -18.0 to -14.2) and moves the CV risk markers (systolic BP -7.6 mm Hg, -10.5 to -4.8; hsCRP down) plus the OSA-severity surrogates — but it is surrogate-only, with no powered CV endpoint, no deaths, and a 52-week horizon. Do not borrow SELECT’s benefit for it; its hard-outcome evidence AWAITS SURMOUNT-MMO. (Malhotra et al., 2024)
The lifestyle hard-outcome null in low-risk people — DPPOS over 21 years
The sharpest bound on the lifestyle route comes from diabetes prevention followed long enough to adjudicate events. DPPOS followed the three DPP arms a median 21 years and was null on the cardiovascular endpoint: metformin MACE HR 1.03 (95% CI 0.78-1.37) and lifestyle MACE HR 1.14 (0.87-1.50) versus placebo — point estimates trending the wrong way for lifestyle. (Goldberg et al., 2022) This is a bounded null, not a demonstration of no effect: the cohort became low-risk and heavily treated (statins 56-62%, antihypertensives 68-74%), leaving little absolute CV risk to remove. Note that preventing diabetes incidence — a disease-onset outcome one rung above a pure surrogate — was never in doubt in DPP; what DPPOS shows is that the incidence win did not carry to hard CV events in this stratum.
The tension resolved: route, baseline risk, and evidence grade — not a contradiction
The positive results (SOS all-cause HR 0.71; Ma all-cause RR 0.82) and the nulls (Look AHEAD HR 0.95; DPPOS MACE HRs crossing 1) are not a contradiction about whether weight loss “works.” Set against a parameter table they are different quantities:
| Axis | Positive results | Null results |
|---|---|---|
| Route | surgery (SOS), GLP-1 drug (SELECT) | lifestyle (Look AHEAD, DPPOS) |
| Baseline risk | severe obesity / higher-risk cohorts | lower-risk, well-treated (statins, BP drugs) |
| Endpoint / grade | all-cause mortality; observational (SOS) or RCT (SELECT) | CV-events composite; RCT |
The clinching evidence is baseline risk. The same diabetes-prevention question, asked in a higher-risk cohort (Da Qing, 30-year follow-up), did show a lifestyle CV benefit: MACE HR 0.74 (95% CI 0.59-0.92). (Goldberg et al., 2022) Absolute benefit scales with baseline risk (route (a) — no subgroup claim needed; Baseline Risk and the Relative-Absolute Split), so a low-risk cohort has little absolute risk to remove and even a real relative effect is hard to detect. The honest composite: hard-outcome benefit is demonstrated for surgery and a GLP-1 drug, plausible-but-unproven for lifestyle in a low-risk stratum, and scaling with baseline risk. (inferred from Goldberg et al., 2022; Ma et al., 2017; Sjöström et al., 2007)
One further separation the fabric draws: an energy-unrestricted Mediterranean pattern (PREDIMED), with little weight change, cut CV events HR 0.70 (95% CI 0.55-0.89) in high-risk primary prevention — so what you eat appears to carry a CV-event signal that how much you weigh did not (populations and comparators differ, so this is a reasoned contrast, not a head-to-head). (Estruch et al., 2018)
Surrogates that predict but are not targets
Even a strong predictor of events is not automatically a lever. The triglyceride-glucose (TyG) index, a cheap insulin-resistance proxy, predicts incidence — CAD HR 2.01 (95% CI 1.68-2.40) — yet is null on mortality: CV mortality 1.10 (0.82-1.47) and all-cause 1.08 (0.92-1.27), both CIs crossing 1. A predictor’s signal is outcome-specific, and TyG is silent on the outcomes people weight most; it flags the stratum, it does not name the lever. (Liu et al., 2022) The counter-case is LDL/apoB, a surrogate whose causal transmission to ASCVD is evidenced, so lowering it does reduce events — not all surrogates are equal -> Surrogate Outcomes.
Decision relevance
- Pursue fat loss for the outcomes it demonstrably moves. Its priority never rested on a proven hard-CV-event reduction: it rests on diabetes remission and prevention, glycaemic control, MASLD regression, and function -> Fatty Liver MASLD and Weight Loss, Lifestyle vs Metformin for Diabetes Prevention. The Look AHEAD null refines the claim rather than undermining it.
- Do not oversell a heart-attack reduction the largest lifestyle trial failed to show. For a low-risk person pursuing lifestyle weight loss, “this will lower your risk of a heart attack” is weakly evidenced; the demonstrated CV levers in that stratum are the direct ones (statin, BP control).
- A widely-cited >=10%-responder CV signal exists but is not held as a magnitude. A Look AHEAD post-hoc analysis (Gregg 2016) is reported — via SELECT’s and DiRECT’s discussions — to show a ~21% cardiovascular-event reduction in those losing >=10%, but the primary paper is not held; it is an awaited figure, not a fact this cut can assert. (Lean et al., 2018)
- Where hard outcomes are the concern, route and baseline risk size the choice. A mature GLP-1 drug with a demonstrated MACE benefit, or a high enough baseline risk, is where the hard-CV rock is largest (Baseline Risk and the Relative-Absolute Split).
The pathogenic fat is inside the organs — draw it down and the liver ladder reverses in dose
Where the excess fat sits is what does the damage, and the most dangerous place is inside the liver and pancreas. Taylor’s model puts intra-organ fat at the centre: liver fat is «pivotal» to hepatic insulin resistance, and he demotes the more-famous visceral depot to a readout — «Extent of visceral fat accumulation is a surrogate marker for intra-organ fat excess, but is not pathophysiologically related to adverse metabolic consequences». (Taylor & Holman, 2014) Fatty liver (MASLD, renamed from NAFLD in 2023) is the clinically visible edge of that intra-organ depot, so its weight-loss lever is the same depot-drawdown the cardiometabolic pages already rank first -> Ectopic Fat and Depot-Specific Risk.
The actionable finding is a dose-response histology ladder: how much sustained weight loss reverses each rung of liver disease. EASL states «a bodyweight reduction of >=5% is required to reduce liver lipid content, 7-10% to improve inflammation, and >=10% to improve fibrosis», with «a dose-dependent association between the amount of weight loss and the extent of improvement in biomarkers of liver damage». (European Association for the Study of the Liver, 2024) A second national body, AASLD, draws the same-shaped curve — «Weight loss of 3%-5% improves steatosis, but greater weight loss (>10%) is generally required to improve NASH and fibrosis» — differing only in where the middle rung sits. (Rinella et al., 2023)
| Rung (histological outcome) | EASL 2024 (GRADEd) | AASLD 2023 (un-GRADEd) |
|---|---|---|
| Steatosis (liver fat) | sustained >=5% | 3-5% |
| Steatohepatitis (NASH) | 7-10% | folded into >10% |
| Fibrosis | >=10% | >10% |
These are guideline rungs, not interval-bearing effect estimates: EASL traces the ladder to a single histology-endpoint RCT (Vilar-Gomez 2015), so read the numbers as name-the-curve thresholds whose studied support is one trial, not a pooled dose with confidence intervals. (European Association for the Study of the Liver, 2024)
Three caveats bound the lever, all from EASL: «a limited proportion of individuals achieve a weight reduction of >=5%»; loss peaks at 6 months then regains to «about 5% at 12-24 months» with partial liver-fat regain; and «evidence for an effect of weight reduction by lifestyle modification on advanced fibrosis or cirrhosis is insufficient» — the lever works on the reversible stages, not once bridging fibrosis or cirrhosis is established. (European Association for the Study of the Liver, 2024)
Composition moves liver fat beyond calories — but it is the fat type, not the carbohydrate fraction
Once a person is losing weight, what they eat at a given calorie level is a second-order lever on liver fat — and the fabric can separate the real composition signal from folklore, because a gold SR-MA pooled only isocaloric RCTs (26 trials, 32 comparisons), holding energy equal between arms. Its effects are standardized mean differences (SMD; 0.2 small, 0.5 medium, 0.8 large — negative = liver fat fell), never percentages of liver fat removed. (Winters-van Eekelen et al., 2020)
| Swap (isocaloric) | SMD (95% CI); n comparisons | Read |
|---|---|---|
| Unsaturated vs saturated fat | -0.80 (-1.09 to -0.51); 4 | large reduction (unsaturated favoured) |
| Carbohydrate -> protein | -0.33 (-0.54 to -0.12); 5 | moderate reduction (protein favoured) |
| Total fat <-> carbohydrate | 0.01 (-0.36 to 0.37); 12 | null — the fat/carb ratio is not the lever |
(Winters-van Eekelen et al., 2020)
This resolves the “fructose and carbohydrates uniquely drive liver fat” claim. At equal calories the carbohydrate story does not hold: the total-fat<->carbohydrate swap is null (SMD 0.01), and a separate gold SR-MA of controlled-feeding trials found isocaloric fructose vs glucose «did not alter IHCLs (+0.11% ± 2.1%)» and «did not differ in any hepatic outcome measure». (Chung et al., 2014) The liver-fat rise Chung found sits in the added-energy arm — hypercaloric fructose raised intra-hepatic lipid 54% (95% CI 29-79%) vs weight maintenance, but at equal excess energy fructose ~= glucose, so the effect «appear[s] to be confounded by excessive energy intake». (Chung et al., 2014)
What survives is not a special carbohydrate mechanism but two composition moves — replacing saturated with unsaturated fat (large SMD) and cutting excess liquid energy: sugar-sweetened soda is the one dietary factor Peng’s umbrella graded HIGH, RR 1.53 (1.34-1.75), where «the quality of evidence was only high for the association of sugar-sweetened soda with increased NAFLD risk». (Peng et al., 2022)
Caveat — the ceiling on all three arms is that they move a surrogate. Every effect above is on liver-fat content, and AASLD flags that even histology is «two surrogate steps» from the endpoints that matter: «Additional studies are needed to better understand the long-term association among changes in liver fat, histological response, and clinical outcomes.» (Rinella et al., 2023)
The risk MASLD carries is mostly extrahepatic — it feeds diabetes and cardiovascular disease
For most people with fatty liver, the danger is not the liver. EASL is explicit that «the presence of steatosis in the general population is not associated with a clinically meaningful increase in the risk of liver-related outcomes» — liver risk tracks fibrosis stage, not fat. (European Association for the Study of the Liver, 2024) The load sits downstream: MASLD carries «higher risk of non-fatal cardiovascular disease (HR 1.40)», coronary heart disease (OR 1.33), heart failure (OR 1.5), chronic kidney disease, and «a more than two-fold increased risk of an incident diagnosis of T2D». (European Association for the Study of the Liver, 2024)
Two honesty notes bind these figures. First, EASL reports them as bare point estimates — the guideline text carries no confidence intervals for the HR 1.40, OR 1.33, or OR 1.5, so they enter as guideline-cited associations, not interval-bearing effects. Second, the causal direction is unresolved — a named [G] gap. The >2x incident-T2D and HR 1.40 figures are observational associations; no held Mendelian-randomization umbrella settles whether fatty liver drives T2D or the two share an upstream cause (insulin resistance), so the direction NAFLD/MASLD <-> T2D is a hole in the evidence, not a resolved arrow. (inferred from European Association for the Study of the Liver, 2024)
So for the metabolic-syndrome stratum MASLD adds no separate intervention — it adds a reason and a target (>=7-10% loss for the inflammatory stage) to the weight loss already indicated for cardiometabolic risk. The one lever here that reaches a patient-important endpoint is bariatric surgery, the aggressive form of the same weight-loss lever: «Resolution of NASH without worsening of fibrosis occurred in 80% of patients 1 year following bariatric surgery, which was maintained at 5 years». (Rinella et al., 2023) Full drug/surgery benefit-risk appraisal lives in GLP-1 Drugs; here the point is that the mortality signal, where it exists, tracks large sustained loss, not a distinct mechanism.
Downstream disease — the same fat drives cancer and loads the knee
Cancer: broadly graded, but the wiki holds no per-person magnitude. WCRF grades greater body fatness a convincing or probable cause of many cancers and puts it first because the evidence is «particularly strong … and has grown stronger over the last decade»; «12 of the 17 cancers reviewed by the CUP are linked to greater body fatness», and IARC independently adds thyroid cancer, multiple myeloma and meningioma. (World Cancer Research Fund & American Institute for Cancer Research, 2018)
Carry this as a graded qualifier, not a number: WCRF does not quantify a per-person effect, so there is no CI-bearing body-fatness -> cancer risk estimate to state — the honest form is convincing/probable across 12-of-17 sites, an absence of magnitude, not a hazard ratio. Two edges: risk rises «even within the so-called ‘healthy’ range» (graded, not a BMI-25 threshold), and body fatness in young adulthood runs opposite — a probable-graded protective arm with «no stated mechanism» that the wiki treats as a candidate artifact until it survives a genetic or referent check, not a sign-flip -> The U-Shaped Association Artifact. (World Cancer Research Fund & American Institute for Cancer Research, 2018)
Knee osteoarthritis: here the fabric does hold CI-bearing function outcomes. The IDEA RCT (454 overweight/obese adults, K-L 2-3 knee OA, 18 months) showed diet-plus-exercise beat exercise alone on pain — «the D + E group had less pain relative to the E (mean score, 1.02; 95% CI, 0.33-1.71; P = .004) and D (1.13; 95% CI, 0.44-1.82; P = .001) groups» — with 38% of D+E reporting little or no pain versus ~20% in either single arm, and a monotone dose-response favouring >=10% loss over the studied 0-32% range. (Messier et al., 2013)
Exercise adds function safely: the EULAR SR-MA found «moderate effects … of aerobic exercises and resistance training on cardiovascular fitness (SMD 0.56 (95% CI 0.38 to 0.75)) and muscle strength (SMD 0.54 (95% CI 0.35 to 0.72))», with no effect on flexibility (SMD 0.12, 95% CI -0.16 to 0.41 — insufficient evidence, not a proven null). (Rausch Osthoff et al., 2018) On the incidence side the same lever dominates: per BMI unit, adjusted OR 1.17 (1.10-1.24) for radiographic knee OA, with overweight/obesity plus prior injury (aOR 2.67, 1.41-5.05) accounting for 14% of cases. (Duong et al., 2025)
The synthesis for U3. One depot — intra-organ and central fat — sits upstream of a liver ladder, a diabetes risk, a cardiovascular load, a cancer footprint and a mechanically-loaded knee, and one sustained weight-loss lever draws it down across all of them. That shared upstream is a structural-leverage point: for the overweight metabolic patient the same loss buys liver, cardiometabolic, cancer-risk and joint benefit at once — which is why depot-drawdown, not any single-disease diet, is the move.
How you lose the fat changes what you lose — and the body fights the deficit
A calorie deficit reliably removes weight, but the body does not sit still while it happens: it compensates, spending fewer calories than the arithmetic predicts. Riou’s SR of 61 exercise studies put the average energy compensation at 18% (SD ±93%) — meaning roughly a fifth of the exercise deficit is offset, but the huge dispersion means the average barely predicts any one person, who may compensate over 100% or not at all (Riou et al., 2015). Compensation also grows with time: in the few long trials it «approached 84%» at about 80 weeks, a thin extrapolation but a solid direction (Riou et al., 2015). Careau’s doubly-labelled-water database (n=1,754) reached the same conclusion by a different route — «energy compensation by a typical human averages 28% due to reduced BEE» (Careau et al., 2021).
The twist that matters for a fat-loss decision is that compensation rises with adiposity. In Careau’s data, people at the 10th BMI percentile compensate 27.7% of activity calories while those at the 90th compensate 49.2% (Careau et al., 2021) — so exercise “counts” least toward the deficit in exactly the person who most wants to burn fat off. (Careau reports these as cross-sectional points without confidence intervals, and Riou’s own adiposity term runs the other way short-term, so treat the direction as a caution, not a settled within-person law.) The practical read: anchor weight change on intake; use exercise for fitness, function and the visceral-fat benefit below, which do not run through the compensated calories.
Exercise strips visceral fat per unit deficit; diet magnitude drives total loss
How the deficit is created changes where the fat comes off. Recchia’s SR-MA (40 RCTs) found exercise cut visceral fat with a significant dose-response of -0.15 SD per 1000 kcal/week of prescribed deficit (95% CI -0.23 to -0.07), whereas caloric restriction’s dose-response per unit deficit was non-significant after outlier removal (-0.04, 95% CI -0.17 to 0.08; p=0.49) (Recchia et al., 2023). Caloric restriction removes more visceral fat in total — because it produces more weight loss — but exercise’s visceral benefit is partly weight-independent, so the compensated calories blunt the scale without erasing the depot benefit -> Exercise vs Caloric Restriction for Visceral Fat.
For total loss, magnitude beats composition. In people with T2D, the format that best controls energy wins: very-low-energy diets lost -6.6 kg (95% CI -9.5 to -3.7) and formula meal replacement -2.4 kg (95% CI -3.3 to -1.4) versus a self-administered low-energy diet, while macronutrient profiles sat within 0-2 kg of each other (Churuangsuk et al., 2021). The lever is how much energy the deficit actually delivers, not which macronutrient is cut.
Lean mass: the ratio improves, the absolute kilograms fall
Every large weight loss sheds some muscle, and the drug case makes the two-faced pattern explicit. Laverde’s SR-MA of 7 obesity-dose GLP-1 trials found lean mass as a proportion of weight rose — +1.81% (95% CI 1.1 to 2.52) — because fat is lost faster than lean, so composition improves as a ratio (Laverde et al., 2026). Yet absolute lean mass fell -1.74 kg (95% CI -3.04 to -0.45), with semaglutide the outlier at -5.44 kg (95% CI -7.07 to -3.81), or -9.9% (Laverde et al., 2026). About 30% of the class’s weight loss is lean — inside the ordinary 20-30% band for any rapid loss, «comparable to … bariatric surgery» (Laverde et al., 2026). Rapid loss sheds muscle whatever drives it; the drug is not special on average, semaglutide the exception.
Is lean-mass loss a harm? Not straightforwardly. Lean mass is a surrogate for function, not the outcome itself, and the fat-to-lean ratio typically improves. Laverde’s own verdict is that «lean mass loss should not be considered a limitation for the use of these drugs in patients with obesity» — but the same paper measured no strength or physical capacity, so function is unmeasured on both sides (Laverde et al., 2026). Crucially, no RCT shows that preserving muscle mass during weight loss changes a hard outcome — that is a type-G gap. Muscle mass does independently predict mortality (de Santana SR-MA: appendicular ASMI SMD -0.18, 95% CI -0.23 to -0.12) (de Santana et al., 2021), but a predictor is not a proven target.
Steer toward strength, not the scale
Because mass is a weak surrogate and function is the thing that matters, EWGSOP2 puts strength first: «strength is better than mass in predicting adverse outcomes» (Cruz-Jentoft et al., 2018). Screen on grip strength (cut-offs <27 kg for men, <16 kg for women) — but read those as normative thresholds set at roughly -2 SD against a healthy-young reference, not outcome-validated knees in a dose-response curve (Cruz-Jentoft et al., 2018).
The defense against muscle loss under any deficit is the same regardless of route: resistance training as the primary lever, protein a modest adjunct. At energy balance the muscle-building benefit of added protein plateaus in the region of ~1.6 g/kg/day — a break-point with a wide interval, not a settled target (the pinned figure and its CI live on Protein and Resistance Training for Muscle and Strength) — and beyond it added protein brings no further resistance-training gains in fat-free mass, so the training, not the supplement, does the work (Morton et al., 2017). A deficit raises the target: protein’s benefit for fat-free-mass retention under energy restriction is a linear dose-response, stronger the leaner the person (lower baseline body-fat %) (Refalo et al., 2025) -> Sarcopenia Definition and Diagnosis.
Who this bites depends on the stratum. For a young, well-muscled obese adult the ratio face governs — composition improves, the absolute loss is minor, and RT plus protein is prudent rather than urgent. For an older or sarcopenia-risk adult the absolute face governs: the muscle cost becomes a front-line patient-important harm (falls, fractures, lost independence), sharpened by a GLP-1’s appetite suppression cutting protein intake exactly when the muscle-sparing target has gone up -> GLP-1 and Lean Mass, Big Rocks (Elderly). Monitor function directly, not the scale.
Named gap — do not fabricate a magnitude. How much extra muscle is lost by losing weight faster is not held: the muscle-loss-by-speed-of-loss quantity has no gold MA, and the diet-deficit-versus-exercise-deficit body-composition partition is reasoned from mechanism, not measured.
What each method moves: the drug takes off the most weight, total diet replacement reverses diabetes, and a calorie is a calorie for storage
The methods that shed body fat do not all buy the same outcome. Line them up by what they demonstrably move — a hard endpoint, an intermediate one, or only a surrogate — and the ranking stops being a contest between diets. A GLP-1 drug is the only route with a proven hard-outcome win; total diet replacement drives type 2 diabetes into remission on a clean dose-response; carbohydrate restriction improves glycaemia but its edge fades; and for fat storage itself, the macronutrient split barely matters once calories are matched. Full drug appraisal, per-agent ranking, and dosing live in the separate GLP-1 Drugs and Comparing Obesity Drugs deliverables — here each method is named for what it moves.
The GLP-1 / dual-agonist drugs: the largest sustained loss, and (semaglutide) the one hard-outcome win
Semaglutide is the best-evidenced drug route to the weight lever, and SELECT is the landmark. In 17,604 adults with obesity and established cardiovascular disease but no diabetes, semaglutide cut major adverse cardiovascular events to a hazard ratio of 0.80 (95% CI 0.72-0.90, P<0.001) over ~3.3 years on top of statins and antiplatelets — an absolute reduction of ~1.5 percentage points, NNT ~67 (Lincoff et al., 2023). This is the first demonstration that treating obesity pharmacologically moves a hard endpoint. Its weaker signals are not confirmed: CV death was 0.85 (95% CI 0.71-1.01), P=0.07, which failed the pre-specified hierarchical gate, so all-cause mortality 0.81 (0.71-0.93) and every estimate downstream are point estimates, not confirmed effects (Lincoff et al., 2023).
On the surrogate SELECT and STEP-1 agree the drug is a powerful weight lever: SELECT lost -8.51 percentage points placebo-adjusted; STEP-1, in a younger primary-prevention obese population, lost -12.4 pp (95% CI -13.4 to -11.5) with half the treated arm losing >=15% (Wilding et al., 2021). But SELECT’s benefit was proven only in secondary prevention; for a low-risk obese person the hard-outcome benefit is unproven and small in absolute terms even if it transported -> Semaglutide for Cardiovascular Risk in Obesity.
Tirzepatide, a dual GIP/GLP-1 agonist, takes off more weight than any prior non-surgical agent — placebo-adjusted -11.9 to -17.8 percentage points by dose in SURMOUNT-1, with 57% of the top-dose arm losing >=20%, approaching the bariatric range (Jastreboff et al., 2022). But its endpoint is weight, a surrogate, and no hard-outcome trial exists for tirzepatide — the SELECT-equivalent has never been run. That is a named gap, not a null: the larger weight number does not carry a CV benefit with it, and must not be read as though it did.
A 19-drug network meta-analysis (Nong 2026, 262 RCTs) places the class cleanly: subcutaneous semaglutide is «the only drug associated with reduced all cause mortality (risk ratio 0.81, 95% confidence interval 0.72 to 0.93)» and myocardial infarction (0.72, 0.61 to 0.85), while tirzepatide — the bigger weight lever — reaches only heart-failure signals (Nong et al., 2026). Across the class, the biggest weight loss is not the drug with the hard-outcome evidence.
The drugs also cost muscle in proportion to the fat they remove: the two biggest weight-loss agents are the two hardest on lean mass — tirzepatide -8.3% and subcutaneous semaglutide -5.8% lean-mass change (both moderate certainty) — a cost that inverts from cosmetic to patient-important in older and sarcopenia-risk strata (Nong et al., 2026). The full composition-of-loss treatment is on GLP-1 and Lean Mass.
Total diet replacement reverses diabetes — an intermediate outcome, on a clean dose-response
DiRECT delivered an energy-target (not a macronutrient-target) total-diet-replacement programme in routine primary care and put 68 of 149 patients (46%) into diabetes remission versus 6 of 149 (4%) in usual care — odds ratio 19.7 (95% CI 7.8-49.8) at 12 months, on a formula diet that was 59% carbohydrate (Lean et al., 2018). The value is the gradient: pooling both arms, remission rose monotonically with weight lost — 0% among those who gained weight, 7% at 0-5 kg, 34% at 5-10 kg, 57% at 10-15 kg, and 86% at >=15 kg (Lean et al., 2018). The umbrella review places total diet replacement at the single GRADE-HIGH cell of the whole remission map — median 54% remission (range 46-61) across its two low-bias RCTs (Churuangsuk et al., 2021).
Remission is a real benefit weight loss demonstrably moves — freedom from diabetes and its drugs, plus a measured quality-of-life gain — and it is the counterweight to the cardiovascular-event null. But it is an intermediate outcome, not a hard one: HbA1c normalisation is a strong surrogate for microvascular risk, DiRECT was not powered for complications, and remission can relapse. Label it as intermediate, and note that the operative variable is kilograms lost, not the diet’s composition -> Total Diet Replacement and Type 2 Diabetes Remission.
Carbohydrate restriction improves glycaemia — but the medication-free remission edge fades by a year
Cutting carbohydrate (<26% of energy or <130 g/day) produces a real short-term glycaemic effect: remission defined as HbA1c <6.5% with medication still allowed reached a risk difference of 0.32 (95% CI 0.17-0.47) at 6 months (Goldenberg et al., 2021). But under the definition that also requires coming off medication, the effect is never significant at any timepoint, and the 12-month point estimate is RD -0.04 (-0.16 to 0.09) — gone (Goldenberg et al., 2021). The one durable-looking rescue is stratum-specific: in trials that excluded insulin users, medication-free remission is significant at RD 0.20 (0.03-0.38), NNT 5 — but that is a six-month result with no 12-month durability estimate (Goldenberg et al., 2021).
The composite across DiRECT and Goldenberg resolves the mechanism: T2D remission of short-duration disease is driven by the magnitude of sustained weight loss, and carbohydrate restriction is one delivery route for that weight loss, with no evidence it adds a remission effect beyond the weight it produces. DiRECT achieves 46% remission at 59% carbohydrate — the opposite of low-carb — so carb restriction is not necessary; Goldenberg’s own remission and weight advantage both decay to null by 12 months as weight is regained (Goldenberg et al., 2021; inferred from Lean et al., 2018). Optimize for the loss a person can sustain, by whichever route they will hold to -> Carbohydrate Restriction and Type 2 Diabetes Remission.
A calorie is a calorie at the storage step: the isocaloric tests bound the carbohydrate-insulin model
The remaining question is whether the macronutrient source of calories changes how much fat is stored — whether cutting carbohydrate is a distinct lever or a re-description of eating less. State both models in their own terms first.
The carbohydrate-insulin model (CIM) proposes that dietary carbohydrate, via insulin, drives fat into storage and lowers energy expenditure, so that restricting carbohydrate confers a metabolic advantage per calorie — «a shift in substrate partitioning favoring fat storage drives a positive energy balance», not the reverse (Ludwig et al., 2021). The energy-balance model (EBM) holds that fat balance follows energy balance, with the carbohydrate fraction close to irrelevant once calories and protein are matched (Hall et al., 2022).
On the quantity that decides it — body-fat change under isocaloric carbohydrate-for-fat substitution with protein held equal — controlled feeding refutes the CIM in direction. Hall and Guo pooled 32 controlled-feeding studies (563 subjects): energy expenditure was +26 kcal/d and fat loss +16 g/d greater on the lower-fat diet — the sign reversed from the CIM’s prediction, and the magnitude trivial. Hall’s verdict: «for all practical purposes “a calorie is a calorie” when it comes to body fat and energy expenditure differences between controlled isocaloric diets varying in the ratio of carbohydrate to fat» (Hall & Guo, 2017).
The CIM’s one surviving whole-organism prediction — that high insulin-secretors lose more on low-carb, so insulin status tells you which diet to assign — got a pre-specified, well-powered RCT test in DIETFITS and did not appear: weight change was -5.3 kg (low-fat) vs -6.0 kg (low-carb), difference 0.7 kg (95% CI -0.2 to 1.6), with the diet x insulin-secretion (INS-30) interaction P=.47 [@gardner2018].
What survives is not a per-calorie storage advantage but two real channels that both live inside energy balance. The first is intake and adherence: Hall’s inpatient trial drove +508 kcal/day of ad-libitum intake on an ultra-processed diet at matched presented macronutrients, through energy density and eating rate, not glycaemia (Hall et al., 2019). The second is the anatomic location of fat (refined-carbohydrate load on visceral and liver fat), an outcome-specific effect judged on its own evidence, not on the metabolic-advantage claim. The level at which the two models actually disagree is storage-per-calorie, which is settled, not intake-and-organ-fat, where composition genuinely matters [inferred from @hallguo2017; @gardner2018].
For fat loss, then, the decision moves off which diet and onto energy intake and whatever makes a lower intake sustainable. The macronutrient split is close to irrelevant at equal calories.
One real trade-off survives the near-equivalence, though: a 121-trial network meta-analysis found that at 6 months low-carbohydrate patterns barely move LDL cholesterol (a 1.01 mg/dL reduction, below the 5 mg/dL meaningfulness threshold) where low-fat and moderate patterns lower it (7.08 and 5.22 mg/dL) — so low-carb is interchangeable on weight and blood pressure but worse on the atherogenic-lipid axis, a gap that (like the weight difference) largely fades by 12 months (Ge et al., 2020) -> LDL ApoB and Cumulative Exposure.
The method that matters is the one whose deficit a person can hold — and, where a demonstrated route (a GLP-1 drug) or a high enough baseline risk is present, the one that also moves a hard outcome.
Durability is the binding constraint: most regain, and the body defends the higher weight
The number that matters is not peak loss but sustained loss, and sustained loss is where lifestyle weight management runs into a wall. Franz’s 2007 meta-analysis of 80 RCTs (26,455 enrolled, minimum one-year follow-up) fixes the shape: a mean 5 to 8.5 kg (5-9%) lost in the first six months, a plateau around six months, then partial regain to a modest few kilograms held out to 48 months — with no intervention group returning all the way to baseline (Franz et al., 2007).
The method changes the peak, not the eventual plateau: a very-low-energy diet drove the biggest peak (17.9 kg, 16%) and the biggest regain (down to ~5.6 kg by 36 months), while diet-alone lost a shallow 4.9 kg and kept ~3.0 kg (3%) at 48 months (Franz et al., 2007). Read the honest caveats with the figures: the maintained-loss numbers are completer figures with ~29-31% attrition, so a person’s expected sustained loss is likely below these means (Franz et al., 2007).
Sumithran 2011 supplies the mechanism behind that trajectory. After an 8-week very-low-energy diet drove a 13.5 kg (14.0%) loss, participants regained ~5.5 kg, yet a full year out — still ~7.9 kg below baseline — their appetite-regulating hormones remained displaced toward eating and energy conservation (leptin still 35.5% below baseline, ghrelin still elevated, subjective hunger still raised) (Sumithran et al., 2011). The study’s own reading is that relapse “has a strong physiological basis and is not simply the result of the voluntary resumption of old habits” (Sumithran et al., 2011).
Two limits keep this from hardening into determinism. Sumithran is single-arm, uncontrolled, n=34 completers conditioned on achieving >=10% loss, and every week-62-versus-baseline hormone difference is confounded with a genuinely lighter body — the “defense beyond the weight difference” reading is established only for leptin, and the persistence is measured to 12 months, so permanent metabolic damage is an extrapolation past the data (Sumithran et al., 2011). The practical consequence is a posture, not a verdict: maintenance is a distinct problem from loss, an unaided-willpower plan is mis-specified against a real biological headwind, and the rational response is structural leverage — a standing environmental or pharmacological prop -> Weight-Loss Maintenance and Metabolic Adaptation, Layer 1 - Ranking Interventions for a Stratum.
The drug route makes the durability problem concrete rather than solving it. A 19-drug obesity network meta-analysis reports the class-wide pattern from a secondary source: after stopping treatment, participants regain weight at ~0.4 kg per month, projected to return to baseline within ~1.7 years, with cardiometabolic gains reverting (Nong et al., 2026). Semaglutide’s own off-treatment extension shows the same shape — ~two-thirds of the loss regained within a year of stopping -> Semaglutide for Cardiovascular Risk in Obesity. So the drug does not escape the defended set-point; it holds it at bay only while taken, which makes the realistic decision a lifetime one.
Weight cycling and hard outcomes is a NAMED GAP. Whether the loss-regain-loss cycle itself harms — beyond carrying no benefit — is not answered by any gold meta-analysis the fabric holds. The one source that speaks to it, Montani 2015, is a labelled-moderate narrative review, a counterweight and not a magnitude anchor, and it is reverse-causation-prone: intercurrent illness drives both the cycling and the outcome, so an unadjusted association would read as harm that the underlying illness actually caused -> Weight Cycling and Cardiometabolic Risk, Does Weight Loss Reduce Cardiovascular Events. State the absence; do not infer a direction.
The levers shift by stratum — and an effective drug re-sizes the rock
The intervention ranking is not a fixed list; it moves with the stratum, and three cases show how.
Menopause redistributes fat centrally rather than adding total fat. The intuitive story — “menopause makes women gain weight” — puts the emphasis in the wrong place. Ambikairajah’s meta-analysis (201 cross-sectional studies, ~1.05 million women; 11 longitudinal) finds that the quantity of fat gained across midlife tracks aging, with no significant additional menopause influence; what menopause more likely does — on hedged, mostly cross-sectional evidence the source frames as possible, not established — is shift the fat centrally. The visceral-fat increment is +26.90 cm2 (13.12-40.68) on the cross-sectional pooling, but the more precise longitudinal estimate is under half that, +12.95 cm2 (8.65-17.25) (Ambikairajah et al., 2019).
The decision-change is diagnostic: BMI and scale weight are confounded post-menopause by bone loss, sarcopenia and height shrinkage, so a normal BMI does not clear a menopausal woman of central-adiposity risk — measure the waist. The levers are the ones already ranked (visceral-fat reduction, resistance training plus protein for muscle, weight-bearing work for bone); HRT is a bone adjunct and a small body-composition adjunct, not the muscle lever. This is route-(a) stratification — a shifted baseline, not a new subgroup effect -> Menopause and the Shifting Levers, Baseline Risk and the Relative-Absolute Split.
In prediabetes, an intensive lifestyle program out-prevents metformin, and does so broadly. DPP randomized prediabetic adults to lifestyle, metformin, or placebo: lifestyle cut diabetes incidence by 58% (48-66) and metformin by 31% (17-43), with lifestyle 39% (24-51) better than the drug head-to-head (Knowler, 2002).
The two arms behave differently by stratum, and this is the decision-relevant structure. Lifestyle works across essentially every subgroup. Metformin’s effect is genuine effect modification (route b): near-null at the lean, near-normal-fasting end — 3% (-36 to 30) at BMI 22 to <30 and 15% (-12 to 36) at fasting glucose 95-109 mg/dl — rising to 53% (36-65) at BMI >=35 and 48% (33-60) at fasting glucose 110-125 mg/dl (Knowler, 2002). So metformin is a stratum-specific drug (worth most in the more obese and more hyperglycemic, near-useless in the lean), while lifestyle is a broad-spectrum lever -> Lifestyle vs Metformin for Diabetes Prevention.
That prevention win is outcome-scoped, and the scoping is the honest edge. DPP prevents the diagnosis of diabetes; whether that averts heart attacks or extends life is a separate claim. Over a median 21 years, DPPOS found neither metformin nor lifestyle reduced cardiovascular events — metformin MACE HR 1.03 (0.78-1.37), lifestyle MACE HR 1.14 (0.87-1.50), the lifestyle point estimate trending the wrong way (Goldberg et al., 2022). This is a bounded null in a low-risk, heavily statin- and BP-treated cohort, not a demonstration of no effect: a higher-risk population does show the benefit — Da Qing’s 30-year follow-up gave a lifestyle MACE HR 0.74 (0.59-0.92) (Goldberg et al., 2022). Absolute CV benefit scales with baseline risk (route a), so a low-risk person has little absolute risk to remove -> Baseline Risk and the Relative-Absolute Split.
Meal timing is a small lever, and the drug landscape now dwarfs it. Time-restricted eating adds little once the calorie deficit is matched: Liu’s 12-month RCT prescribed both arms the same deficit and varied only the 8-hour window, and the window added no significant weight loss (net -1.8 kg, 95% CI -4.0 to 0.4, P=0.11), with every metabolic secondary null between groups (D. Liu et al., 2022).
Pooled, even the best intermittent-fasting form — alternate-day fasting — beats continuous restriction by only -1.29 kg (-1.99 to -0.59), below the 2.0 kg threshold of clinical meaningfulness (Semnani-Azad et al., 2025). A short ad-libitum window can also quietly cost muscle: TREAT recorded a between-group appendicular-lean-mass deficit of -0.47 kg (-0.82 to -0.12) (Lowe et al., 2020), so the practical instruction is to keep protein up inside any window. Against this, the same meta-analysis notes GLP-1 receptor agonists such as semaglutide “result in substantial weight reductions of 10-15% body weight” (Semnani-Azad et al., 2025) — roughly 3-8x any fasting effect -> Time-Restricted Eating.
That contrast is not only about weight — it is the drug-landscape-sizes-the-rock principle, a Layer-1 input and not merely a Layer-3 choice. A lever’s marginal rank is the patient-important benefit it adds over the best realistic alternative a person could reach instead, and a mature, low-harm GLP-1 with a demonstrated MACE benefit is now such an alternative for the cardiovascular outcome (magnitude carried on Semaglutide for Cardiovascular Risk in Obesity). Where that drug captures most of the reachable CV benefit, the marginal “lose weight by lifestyle for hard CV outcomes” rock shrinks for that stratum.
But the shrinkage is outcome-specific and net of the drug’s own costs: the drug must be taken for life (regain on stopping, above), carries its own side-effect and dependency burden, and does not remove the upstream cause, whereas lifestyle change carries structural and pleiotropic leverage a single-channel drug cannot substitute for. So the drug re-sizes the rock without dissolving it — and which to reach for stays the person’s decision at Layer 3.
The decision that remains: reach the organs, and make the loss last
For most people the useful question is not which diet but whether the loss is the kind that reaches the organs and lasts. Losing intra-organ and visceral fat — by whatever sustained route a person can actually hold to — is what moves liver disease and diabetes remission, and, where a demonstrated route (bariatric surgery, a GLP-1 drug) or a high enough baseline risk is present, hard cardiovascular outcomes. The composition of the diet and the timing of meals are second-order to the magnitude and durability of the loss. And the honest edges are two: lifestyle weight loss has not cleanly shown a hard-outcome benefit in already-low-risk people — the DPPOS 21-year cardiovascular null is the worked instance — and the long-run harm of weight cycling is unmeasured, held only by a moderate narrative source that reverse causation can explain.
That is the calibrated confidence this deliverable carries, and it is what defeats the guidance null. The hard-outcome benefit of losing fat is demonstrated for surgery and for a GLP-1 drug, plausible but unproven for lifestyle in a low-risk stratum, and scales with baseline risk everywhere. Most of the “weight loss works” evidence sits on surrogates and remission — weight, HbA1c, blood pressure, liver enzymes, muscle mass — and is labelled as such rather than blended into the hard-outcome column. The stratum where the advice fails is the already-lean, active, well-treated, low-risk person: the levers are largely pulled, the marginal cardiovascular rock is small, and reporting that the remaining gains are small and uncertain is itself the decision-change — it licenses that person to stop optimizing. The loop stays open: everything here grades coherence and fidelity to the sources, never a recommendation against a realized outcome.
Evidence box
Question ’For an adult across the body-fat range: what is the effect of body-fat level and its modification, via modifiable exposures, on each patient-important outcome — does where the fat is stored change the effect, does losing fat change hard outcomes rather than only surrogates, how do the rate, composition and durability of loss change the answer, and how does it vary by stratum?‘ Evidence included 53 sources — 21 gold, 28 high, 4 moderate Overall certainty Medium (see Rating Certainty of Evidence) Source-selection note 4 source(s) below the gold evidence bar feed this page: Taylor (mechanism, moderate); Lowe (RCT, moderate); Hall (narrative review, moderate); Ludwig (narrative review, moderate). Each labelled by tier; none load-bearing for the core claims. Last updated 2026-09-05 · Independently reviewed: No · Full edit history