Cut the sugary drink, not the sugar molecule
This is a general, population-level appraisal for a healthy adult — a read of what added sugar does to most people. Where your own risk profile, diet, and goals change the answer, that belongs at the end, not the front. Added sugar carries a real health cost. For body weight that cost is an energy cost, and for diabetes and blood pressure it is plausibly partly one, though that is not shown. It concentrates in poorly-compensated liquid calories — the sugar-sweetened beverage (SSB) — rather than in the sugar molecule at equal calories Free Sugars Intake.
One question organizes every outcome below: does the harm survive when you swap sugar for other calories, gram-for-gram? Hold calories constant and the weight harm disappears. Trials that exchange free sugars for other carbohydrate find body weight barely moves — 0.04 kg (95% CI -0.04 to 0.13) Free Sugars Intake. The weight gain tracks the calories, not the molecule The Comparator Problem. The observational cardiometabolic harm plausibly runs partly through the same extra calories, but that route is likely, not shown.
Two harms are the exception: they are sugar- or fructose-specific and survive the calorie-for-calorie test. Dental caries is the classic one — moderate certainty, though no trial exists. Serum urate from SSBs is the second, at high certainty. Both are real, but serum urate is a surrogate, and its chain onward to blood pressure and hard events is not closed Uric Acid Lowering and Blood Pressure.
So the action is narrow: cut SSBs first — the dominant carrier of free sugars, and harmful both as poorly-compensated liquid energy and through the isocaloric urate channel. Sweets and desserts also raised urate in trials, on lower-certainty evidence. Keep whole fruit, which runs null-to-favourable on urate and vascular endpoints and sits outside the limit. Treat 100% fruit juice case by case: adverse on metabolic and dental outcomes, favourable on urate and vascular ones.
No guidance body has found a safe threshold. WHO’s 10%-of-energy limit is a dental recommendation, not a metabolic one. The four guidance families — WHO, SACN, NNR, EFSA — agree on the direction (less is better) and diverge only on where a number attaches and on what warrant supports it Free Sugars Intake.
Sugar behaves differently outcome by outcome. Three questions stay open exactly where the topic is discussed most: sugar-specific isocaloric liver fat rests on one study, sugar’s effect on hard cardiovascular events has no gold-standard evidence, and whether the type-2-diabetes signal persists net of weight is thin.
Before any per-outcome number, fix which sugar is the exposure — the word names several different objects.
Fix which sugar you mean before counting it
A nutrition label reports total sugars; a guideline limit is written against free sugars; a smoothie and a carton of juice made from the same orange are not the same exposure. These are different objects, and the number you should act on is the narrowest of them. WHO 2015 sorts all dietary sugar into three buckets: intrinsic sugars locked inside the cell walls of intact fruit and vegetables, milk sugars (lactose and galactose), and free sugars — everything added by a manufacturer, cook or consumer, plus the sugars in honey, syrups, fruit juices and fruit juice concentrates (World Health Organization, 2015). Only the third bucket is the decision exposure. WHO excludes the other two on the stated ground that no adverse effect of either has been reported, and the four guidance families that set a limit all write it against free sugars, not total sugars.
Free sugars, not total or added, is the exposure
Free sugars is the target because it isolates the sugar a diet can shed without giving up fruit, vegetables or milk. Added sugars — sugar added during processing or preparation — is a slightly narrower construct: it counts what a manufacturer or cook stirs in but not the intrinsic sugar a juicer frees from fruit. So a glass of 100% orange juice carries free sugars while carrying no added sugar at all. Total sugars is the widest and the least useful, because it folds the milk and whole-fruit sugar WHO deliberately left out back into the count (World Health Organization, 2015). The load-bearing boundary sits inside the fruit: fruit juice is in, whole fruit is out. A limit written as reduce sugar loses that line entirely -> Free Sugars Intake.
But free sugars itself names two different constructs
The four guidance families (WHO, SACN, NNR, EFSA) all point the same direction — less free sugar is better — and they diverge only on where a number attaches and on what warrant justifies it. At the level of the definition, though, the divergence starts one step earlier: the phrase free sugars does not name one quantity across them. WHO’s 2015 wording includes fruit juice concentrates; NNR adopts that same 2015 wording; but SACN adopts WHO’s older 2002 definition, which stops at fruit juices and omits concentrates (Scientific Advisory Committee on Nutrition, 2015). Fruit juice concentrate is the sweetener of choice in products marketed as no added sugar, so this is not a boundary case — it is a whole category that moves between counted and uncounted depending on whose definition is in force.
The consequence bites when the numbers appear to agree. WHO’s conditional <5% and SACN’s primary
<=5% are numerically identical, but they are computed on different numerators — SACN’s excludes
juice concentrates. So the match is not two bodies corroborating one figure; it is two bodies
measuring two things and landing on the same digit (Scientific Advisory Committee on Nutrition, 2015). The phrase names at least two objects, and none of the bodies that use it flags the
difference — a textbook disambiguation problem. State which definition is in force whenever
concentrates are load-bearing -> Is the Food Category Doing Any Work.
Juice, smoothie, and whole fruit are three exposures, not one fruit
The fibre matrix separates them, and it drives WHO’s own decision to split fruit down the middle. Juicing removes the insoluble-fibre matrix, leaving a poorly-compensated liquid dose of free fructose — which is exactly why 100% juice sits inside free sugars and whole fruit sits outside it (World Health Organization, 2015).
Blending is not juicing. A blended whole fruit disrupts the cell walls but removes nothing, so a smoothie retains all its fibre and does not inherit juice’s fibre-stripped profile — the common worry that a smoothie is just as bad as juice conflates the two under the loose heading of processing.
One residual question remains: does cell-wall disruption raise the glycaemic response modestly against intact fruit? That magnitude is not held (a blended-fruit glycaemic-response review is queued), but it is bounded well below the juice case because the fibre is still there -> Dietary Fibre and Health.
Whole fruit therefore stays outside the limit on every axis. 100% juice is the genuinely case-by-case exposure: net-adverse on the metabolic and dental channel where its free-sugar load dominates, but not uniformly harmful across outcomes — its urate and vascular signals run the other way, taken up per outcome below. The practical upshot of the whole split: a get your five a day from juice instruction and a cut free sugars instruction pull in opposite directions on the same carton, and only naming the form resolves it -> Free Sugars Intake.
GI and GL: the sugar, or the carbohydrate pattern it stands in for?
One more construct is often folded into the sugar question and should be kept separate. Glycaemic index measures how much 50 g of carbohydrate in a food raises postprandial blood glucose against a glucose standard; glycaemic load is that index times the carbohydrate amount. Jenkins’ 2024 pooled analysis of ten mega-cohorts finds real, consistent associations: high GI tracks a 27% higher risk of type 2 diabetes (RR 1.27, 1.21-1.34) and a 15% higher risk of total CVD (1.15, 1.11-1.19), with smaller signals for mortality and cancer. But these are extreme-quantile relative risks, mostly graded GRADE-low, over a narrow measured contrast (mean GI 58 in the lowest quantile versus 67 in the highest) (Jenkins et al., 2024).
The open question is whether GI is a lever of its own or a proxy. GI, GL, fibre and whole-grain content ride together in whole-food diets, and the mutual adjustment that would separate them is almost absent: only 12 of 69 GI assessments (17%) controlled for fibre, and no fibre assessment controlled for GI (Jenkins et al., 2024). On this evidence, steering by fibre and whole-grain content captures most of what steering by GI would, so GI/GL is not a separate exposure to fix here — it is the same is the category doing any work? question applied to a construct rather than a food -> Is the Food Category Doing Any Work, Glycaemic Index and Glycaemic Load and Chronic Disease.
With the exposure fixed — free sugars, carried above all by beverages, with juice a case apart and whole fruit outside the line — take each patient-important outcome in turn, starting where the evidence is strongest.
Start where the evidence is strongest: three harms the data actually size
Three outcomes carry a real, sized signal against free sugars: dental caries, the sugar-sweetened beverage gradient to cardiometabolic disease, and body weight. Each is measured as sugar is consumed — sugar eaten on top of the diet, adding its own calories — and each curve the evidence draws is monotone over the studied range, with no knee or plateau located. There is no intake below which the harm switches off inside the range anyone has studied The U-Shaped Association Artifact. Take them in order of how well the evidence pins them down.
Dental caries: the one harm specific to the sugar molecule
Caries is the harm the free-sugars limit is actually built to prevent, and it is sugar-molecule-specific: oral bacteria ferment free sugars to acid regardless of the calories involved, so this harm does not run through energy the way the others do.
WHO rates the evidence moderate certainty, and does so from an unusual base — no randomised trial exists, in children or adults, and no adult cohort study either (World Health Organization, 2015). The rating climbed to moderate only by an upgrade for large effect from the observational floor of low Upgrading Observational Evidence, warranted by how consistent the studies are: 42 of 50 child studies and 5 of 5 adult studies reported a positive association (World Health Organization, 2015).
A pooled magnitude does exist but never reached the guideline: the commissioned review pooled caries severity at SMD 0.82 (95% CI 0.67 to 0.97) across designs (Moynihan & Kelly, 2013).
This is what the 10% limit protects — not body weight, not cardiometabolic disease
Free Sugars Intake. And the threshold below it is soft in a specific, decision-relevant way.
The <5% recommendation rests on three ecological studies (correlations r = 0.6 to 0.8), rated
very low by WHO itself (World Health Organization, 2015). No study measured a knee in the caries curve. The <5% number marks the
edge of the observed data, not a point where the dose-response bends — the first thing to
suspect of any guideline threshold, and true here. Read <10% and <5% as where the evidence
thins, not as a curve feature.
Sugar-sweetened beverages: a monotone cardiometabolic gradient
The beverage is the dominant free-sugars carrier, and a dose-response meta-analysis of 39 cohort articles (Qin) sizes what it does to cardiometabolic outcomes per 250 mL/day of extra intake. Each outcome rests on a subset; the type-2-diabetes dose-response pools 19 studies (Qin et al., 2020):
| Outcome | RR per 250 mL/day (95% CI) | Shape |
|---|---|---|
| Type 2 diabetes | 1.19 (1.13 to 1.25) | monotone-linear (P non-lin 0.71) |
| Obesity | 1.12 (1.05 to 1.19) | monotone-linear |
| Hypertension | 1.10 (1.06 to 1.14) | monotone-linear |
| All-cause mortality | 1.04 (1.01 to 1.07) | monotone-linear |
Every curve rises linearly across the whole studied range — no threshold, no plateau, every extra serving adds risk. The type-2-diabetes estimate draws on the most studies (19) and has a tight interval, but it also carries the highest heterogeneity (I2 82%) and clear small-study asymmetry (Egger P<0.001); trim-and-fill shrinks the high-versus-low RR from 1.27 to 1.10 (1.03 to 1.18), which Qin reads as leaving the main result intact (Qin et al., 2020). Placed against eleven other food groups in the DIfE/Boeing series, sugar-sweetened beverages sit among the strongest cardiometabolic-harm levers. Their all-cause-mortality cell there is imprecise (RR 1.03, 0.91 to 1.18, low, 4 studies) — not evidence of safety (Schwingshackl et al., 2017) — and Qin’s larger pool, in the table above, gives a small rise (1.04, 1.01 to 1.07). The clear harm is cardiometabolic -> Food Groups and Health Outcomes - A Dose-Response Matrix.
These are cohort estimates, so read them as association, not effect. Qin’s authors say as much: the analysis rests on observational cohorts, not trials, and residual confounding cannot be ruled out — the observational limb of The Observational-Trial Discordance, not a causal magnitude (Qin et al., 2020).
Body weight: sugar as consumed adds roughly a kilogram either way
Where free sugars are eaten ad libitum — added to the diet, not swapped in — they move body weight symmetrically, on WHO’s own trial evidence:
| Contrast (adults, ad libitum) | Effect (95% CI) | Design | Certainty |
|---|---|---|---|
| Reduced free sugars | 0.80 kg lower (0.39 to 1.21) | 5 RCTs | Moderate |
| Increased free sugars | 0.75 kg higher (0.30 to 1.19) | 10 RCTs | Moderate |
(World Health Organization, 2015)
About a kilogram in each direction, at moderate certainty. WHO reads this as an energy effect, not a sugar-specific one: «The excess body weight associated with free sugars intake results from excess energy intake.» (World Health Organization, 2015) On WHO’s reading, sugar consumed as-is moves weight through the calories it adds. That is why the weight number changes when you hold the calories fixed — the test the next section runs.
Why these curves look flat, and why the thresholds are soft
One reason no knee shows up in any of these curves is that the exposure is self-reported and badly measured, which flattens real gradients toward the null. Te Morenga could not fit a dose-response to the weight data at all (null metaregression, 0.02 kg, -0.03 to 0.08; P = 0.392) and blamed the instrument directly: intake assessment «was associated with a considerable degree of measurement error even when using validated methods … probably one explanation why a dose-response effect could not be shown» (Morenga et al., 2013). Against doubly-labelled water, adults under-report energy intake by about 34% (Scientific Advisory Committee on Nutrition, 2015).
The direction of that error is asymmetric and matters: measurement noise can hide a knee by pulling extreme intakes toward the mean, but it cannot manufacture one Measurement Error in Dietary Assessment. So a measured-flat curve is weak evidence of a truly flat one, and the burden sits on anyone claiming a knee to locate it. The same error makes the thresholds inherit it: food-frequency questionnaires are adequate for ranking people, not for the absolute magnitude a cutpoint needs (Willett, 2012), so is my intake under 10%? is a question the dominant instrument was never the right tool to answer, even where the threshold itself is sound.
Every one of these three harms is measured as sugar is consumed — eaten on top of the diet, with its calories attached. Hold the calories constant, swapping sugar for other carbohydrate gram-for-gram, and the picture changes sharply. That isocaloric test is the interpretive core of the whole question.
Hold the calories equal, and watch what survives the swap
Every harm number in the previous section was measured against an unstated comparator: sugar as consumed, adding calories on top of the diet. Change that comparator and the answer changes with it — the core move of The Comparator Problem (harmful instead of what?). A feeding trial can fix that comparator exactly: swap free sugars for the same number of calories from other carbohydrate, hold total energy constant, and ask whether the harm is a property of the sugar molecule or of the energy it carries. This is the isocaloric (calorie-for-calorie) test, and it is the interpretive core of the whole answer. What the test does to each outcome differs — and that difference, not any single verdict, is the finding.
Body weight and the “metabolic advantage” do not survive
At equal energy, the sugar molecule does nothing to body weight. WHO’s own evidence puts the isoenergetic exchange of free sugars for other carbohydrate at 0.04 kg (-0.04 to 0.13), 11 trials — a null, and the row that gets dropped whenever the 10% figure is quoted alone (World Health Organization, 2015). WHO reads its own table the same way: the excess weight tied to free sugars is excess energy -> Free Sugars Intake. So reduce sugars and reduce sugars and replace the calories with other carbohydrate are two different interventions with two different answers — the substitution logic of Energy Adjustment and What a Diet Coefficient Means.
The stronger claim — that carbohydrate does special metabolic harm at equal calories, so cutting it confers a storage-independent metabolic advantage — fails the same test, and fails harder. Hall and Guo pooled 32 controlled-feeding studies (563 subjects) and found energy expenditure and fat loss both slightly greater on lower-fat diets (+26 kcal/d, +16 g/d) — the direction the carbohydrate-insulin model predicts, reversed, at a magnitude the authors call «the effect sizes are so small as to be physiologically meaningless» (Hall & Guo, 2017). For body fat, a calorie is a calorie. This is a no-meaningful-effect result, not insufficient evidence: the comparison was run at scale and came back near-null and pointing the wrong way for the model -> What Drives Fat Gain - Energy Balance vs the Carbohydrate-Insulin Model.
Dental caries and serum urate do survive
Two harms are sugar- or fructose-specific, and the isocaloric weight null touches neither.
Dental caries is the classic one. Oral bacteria ferment the sugar molecule to acid whatever else is on the plate, so swapping sugar calories for starch calories leaves the mechanism intact — the weight null is simply silent on it. Caries is what WHO’s 10% limit actually protects, at moderate certainty and with no trial behind it -> Free Sugars Intake.
Serum urate is the new one, and it is the finding this appraisal turns on. Ayoub-Charette pooled 47 controlled-feeding trials (N=2763, baseline uric acid ~4.6-5.5 mg/dL — roughly normouricemic) and found the effect runs by food source, not by energy: «the effects of fructose-containing sugars on uric acid were more dependent on the food source than on energy control» (Ayoub-Charette et al., 2021). The decisive numbers:
- Sugar-sweetened beverages raise urate +0.42 mg/dL (0.24-0.59) in isocaloric substitution and +0.43 (0.23-0.63) as added calories — GRADE high in both. Because the substitution arm holds energy constant, SSB’s urate harm persists at equal calories, unlike its body-weight harm, which vanishes on the isoenergetic swap above.
- Total fructose-containing sugars barely move urate (+0.16 mg/dL, 0.06-0.27, very low certainty), and even that is carried by SSBs — drop them and it collapses to +0.02 (-0.07 to 0.1), null.
- Sweets and desserts also raise urate in substitution trials (+0.35 mg/dL, 0.07-0.63, 10 trials, moderate certainty), but at high doses — a median 18% of energy.
- Whole fruit is null; 100% fruit juice LOWERS urate -0.28 mg/dL (-0.43 to -0.13), GRADE high — the opposite sign from SSB, in the same review. Cut fructose is the wrong instruction; cut SSBs first (high certainty), with sweets and desserts behind them on lower-certainty evidence, is the right one.
Against the review’s own minimally-important difference for urate (±0.113 mg/dL), the SSB rise and the juice fall both clear it while the total-sugars move does not. So urate is the one place a fructose-specific, energy-independent harm shows up on high-certainty trial evidence.
The isocaloric urate signal stops at the marker
The payoff is the isocaloric persistence at the urate step — not any blood-pressure or event endpoint. Serum urate is a surrogate: a marker that travels with gout, hypertension and cardiovascular disease rather than an outcome a patient feels.
A downstream leg does exist: a gold RCT meta-analysis finds urate-lowering therapy lowers blood pressure. But it sits in a different stratum on a different endpoint — hyperuricemic patients, on a drug (allopurinol), whose blood-pressure effect may run through xanthine-oxidase inhibition rather than urate itself -> Uric Acid Lowering and Blood Pressure.
The upstream and downstream legs never meet in one population, so SSB -> urate -> blood pressure -> events is not a chain the evidence has closed — it is two disconnected segments, and treating it as settled would overstate what is held. What is held, at high certainty, is narrower and still worth acting on: cutting SSBs lowers urate at equal calories.
Liver fat runs the same way and lands in the same place. Chung’s controlled-feeding trials found fructose no worse than glucose for liver fat at equal energy — but as a single low-risk-of-bias study that verdict is insufficient evidence, not proof of no effect, and it does not license a sugar-specific isocaloric liver-fat magnitude. AASLD’s guidance asserts molecule-specific liver harm independent of calories, but on three cited references, not a separate body of long-term evidence; Chung rated the observational fructose-liver studies it reviewed insufficient. The chronic regime is thinly and poorly evidenced — a design and quality gap, not a settled contradiction, carried into the Open cells below -> Fatty Liver MASLD and Weight Loss. (inferred from Chung et al., 2014; Rinella et al., 2023)
Why the big observational SSB signal is not a counterexample
The previous section’s cohort signal — SSB intake tracking type 2 diabetes, obesity and hypertension per 250 mL/d — looks like it should have surfaced under the isocaloric test and did not. It did not because the cohorts and the feeding trials are not measuring the same exposure. The cohorts measured SSB as drunk, which adds poorly-compensated liquid calories on top of the diet; the isocaloric trials tested the sugar molecule with the calories held fixed, and measured body weight, not diabetes. That is The Observational-Trial Discordance resolved by exposure non-commensurability: the observational signal is the added-energy package, plausibly carried in part by its calories. That energy route is likely but not shown. In Qin’s cohorts that adjusted for energy intake the diabetes association did not shrink (RR 1.62, 1.18 to 2.21, 6 studies, against 1.17, 1.11 to 1.23, in 13 that did not; P for the difference 0.154), though cohort energy adjustment is imperfect (Qin et al., 2020).
EFSA states this in its own voice. Its systematic review reports that «prospective cohort studies do not support a positive relationship between the intake of dietary sugars, in isocaloric exchange with other macronutrients, and any of the chronic metabolic diseases or pregnancy-related endpoints assessed», and names the mechanism directly: «Excess energy intake leading to positive energy balance and body weight gain appears to be the main mechanism by which the intake of dietary sugars may contribute to the development of chronic metabolic diseases» (European Food Safety Authority, 2022). So the harm the cohorts see is real and decision-relevant, and EFSA reads it mainly as an energy effect.
What the test leaves standing — one exposure, three answers
Sorted by outcome, not by a single verdict, this is the deliverable’s interpretive backbone:
- Body weight: energy. Erased by the calorie-for-calorie swap; the lever is the liquid calories, not the sugar. Cardiometabolic risk: likely partly energy, not shown — no swap trial has tested it.
- Serum urate: fructose- and food-source-specific. Survives the swap at high certainty — but stops at a surrogate.
- Dental caries: sugar-molecule-specific. The isocaloric weight null never spoke to it; it is the mechanism the 10% limit protects.
The urate result is what makes the beverage the sharp end of the whole question: it is the exposure where a harm survives the calorie-for-calorie test most clearly, at high certainty. That raises the next question directly — is a sugar-sweetened beverage a distinct exposure from the same sugar in solid food?
Treat a sugar-sweetened beverage as its own exposure
A sugar-sweetened beverage is not just sugar in liquid form, and the difference changes what to cut. Two features set it apart. The body compensates poorly for calories it drinks, so a soda adds energy on top of the day’s intake instead of displacing food. And the beverage carries a second, energy-independent harm — a rise in serum urate that survives even when the sugar is exchanged calorie-for-calorie. That second channel depends on the food that carries the sugar: whole fruit does not show it, 100% juice runs the other way, and sweets and desserts share it, on weaker evidence. So the mediation question — is the harm working through the extra calories, or through the sugar molecule itself? — has no single answer for a beverage. It splits by outcome, and that split is the reason to name the SSB rather than “sugar” as the exposure.
Drunk calories slip past the body’s compensation
A beverage behaves differently for a mechanical reason, not a metabolic one. A drink is the fastest-ingested, lowest-effort form a food can take, and faster ingestion drives higher intake without a matching rise in hunger. Robinson’s meta-analysis of 22 experimental studies found a slower eating rate cut energy intake (SMD 0.45, 95% CI 0.25-0.65) while end-of-meal and later hunger stayed flat, because a faster rate means less oral sensory exposure per unit of food (Robinson et al., 2014). A beverage is the extreme of that channel — swallowed in seconds, with almost no chewing or taste-duration signal to register the load. (inferred from Robinson et al., 2014)
The satiety-compensation evidence proper lives on Free Sugars Intake. The upshot here: liquid sugar adds energy rather than replacing it, which is why the SSB is the dominant free-sugars carrier and the actionable form to remove.
The beverage form also sits outside the composition-and-reward constructs used for solid junk food. Sutton’s supply-scale analysis of Hyper-Palatable Foods excluded beverages entirely — hyper- palatability is defined on the fat/sugar/sodium combinations of solids (Sutton et al., 2023). The SSB harm is therefore not an HPF story and not a NOVA-processing story: it is a liquid-energy story plus a fructose one. The ultra-processed label is a category a shopper cannot reliably apply -> Ultra-Processed Food and Health Outcomes. The sugar-sweetened beverage is the one place that fuzzy category resolves into a single, identifiable carrier — which is what makes cut the drink more usable advice than cut ultra-processed food.
The mediation splits by outcome
Hold the SSB’s outcomes side by side and the mediating channel changes under each one. The observational cardiometabolic signal (body weight, type 2 diabetes, hypertension) is the added-energy package. The weight harm dissolves when sugar is swapped for other carbohydrate at equal calories (the isocaloric weight null). For diabetes and hypertension an energy route is likely, and EFSA reads it that way, but it is not shown: no swap trial measured those outcomes, and Qin’s energy-adjusted diabetes estimate did not shrink -> The Observational-Trial Discordance, What Drives Fat Gain - Energy Balance vs the Carbohydrate-Insulin Model.
Serum urate is the exception. Because the substitution arm holds calories fixed, this harm does not vanish at equal energy — it points at fructose-specific metabolism (the unregulated fructokinase route, ATP -> AMP -> urate), not at extra calories. (inferred from Ayoub-Charette et al., 2021)
The urate channel is also specific to the food source, not to fructose in general: Ayoub-Charette found the uric-acid effect depends more on the food source than on energy control, which is why SSBs (and, at high doses, sweets and desserts) raise urate while 100% juice lowers it (Ayoub-Charette et al., 2021). So one drink loads two distinct channels, and the mediation answer must be read per outcome:
| SSB outcome | Mediating channel | Survives calorie-for-calorie swap? |
|---|---|---|
| Body weight | Added energy | No — erased at equal calories |
| T2D, hypertension | Likely partly added energy (not shown) | Untested — no swap trial on these outcomes |
| Serum urate | Fructose-specific (hepatic) | Yes — +0.42 mg/dL isocaloric, GRADE high |
| Dental caries | Local, sugar-molecule (oral) | Yes — untouched by the liquid/solid or energy question |
Urate remains a surrogate, not a patient-important outcome — the leg from urate to blood pressure or events is measured in a different (hyperuricemic, drug-treated) stratum and is not joined to this dietary leg -> Uric Acid Lowering and Blood Pressure. The isocaloric persistence settles something narrow but real: the beverage carries a sugar-specific harm the calorie-balance story cannot absorb.
The diet-soda clue: read the SSB cohort signal with a reverse-causation candidate in view
Artificially-sweetened beverages (ASBs — diet soda) supply the sharpest check on how much of the SSB cohort signal is clean causal effect. An ASB carries no sugar and no calories, so it should shed the energy-mediated harm. Yet in WHO’s cohort evidence it carries similar-sized positive associations with the same hard outcomes: incident obesity HR 1.76 (1.25-2.49), type 2 diabetes HR 1.23 (1.14-1.32), CVD HR 1.32 (1.17-1.50), all-cause mortality HR 1.12 (1.05-1.19), at low-to-very-low certainty (World Health Organization, 2023). Reverse causation is the first candidate: people already heavy or dysglycaemic switch to diet drinks because of their risk status, so the drink tracks the illness rather than causing it. This is the The Observational-Trial Discordance flag on beverage cohorts — do not read the ASB association as evidence that diet soda causes disease.
Symmetric standards cut the other way too. WHO ran the reverse-causation checks (adjusting for BMI, stratifying by weight, excluding those who lost weight before baseline, dropping early follow-up years) and the associations weakened but did not disappear — WHO concluded the evidence «cannot be dismissed as being solely a result of reverse causation or residual confounding» (World Health Organization, 2023). So the honest ASB verdict is unexplained and open, not “harmless” and not “harmful.” For the sugar decision, the load-bearing point is narrower: the ASB parallel warns that the SSB cohort magnitudes are not all clean causal signal — part is the same confounding — which is exactly why the isocaloric urate finding above matters, being the one SSB harm a swap cannot confound away.
The realistic swap is water, not a diet drink
If the sugary drink goes, what replaces it decides the benefit. WHO’s 2023 review suggests «that non-sugar sweeteners not be used as a means of achieving weight control or reducing the risk of noncommunicable diseases (conditional recommendation)» (World Health Organization, 2023), on low-certainty evidence. Swapping sugar for a diet drink beats the sugar marginally on short-term weight, but that benefit attenuates to null in the trials that actually instructed habitual consumers to switch (body weight MD -0.61 kg, 95% CI -1.28 to 0.06, NS), and it leaves overall diet quality largely unchanged (World Health Organization, 2023). WHO’s preferred replacement is water, unsweetened food and drink, and whole fruit (World Health Organization, 2023).
A person who will not move from sugary soda to water is the honest exception: for that person a diet drink is still the better of those two, since WHO’s recommendation optimizes a population question, not this individual’s next drink (a layer-3 call). The separate question of whether non-sugar sweeteners are themselves safe (the aspartame hazard-vs-risk verdict, the microbiome-glycemia signal) is derived in the Artificial Sweeteners deliverable and not re-opened here -> Non-Sugar Sweeteners.
Two outcomes remain where the evidence thins to insufficient — liver fat under an isocaloric sugar swap, and hard cardiovascular events.
Name the open cells — insufficient evidence is not proof of no effect
Five questions about sugar stay open, and each one is insufficient evidence, not a demonstrated null. The difference decides how you read the silence. A question nobody has powered a study to answer looks, in a database, exactly like a question answered “no.” Apply the expectancy test: if a sugar-specific harm were real and large here, could a study realistically have caught it by now? Where the answer is no, silence is a gap, not reassurance. Two of the quantities in play — liver-fat content and serum urate — are surrogates, so even the cells that hold some evidence are capped by how far the marker carries the decision -> Surrogate Outcomes.
Sugar-specific liver fat at equal calories rests on a single study
Only one low-risk-of-bias trial has swapped fructose for glucose calorie-for-calorie and measured liver fat. The two sugars did not differ, and neither diet changed intrahepatic lipid (+0.11%) (Chung et al., 2014). Chung reads its own result as insufficient, not as no effect, and frames its energy reading as an inference across studies: «On the basis of indirect comparisons across study findings, the apparent association between indexes of liver health (ie, liver fat, hepatic de novo lipogenesis, alanine aminotransferase, AST, and g-glutamyl transpeptase) and fructose or sucrose intake appear to be confounded by excessive energy intake.» The evidence, it concludes, is «not sufficiently robust to draw conclusions regarding effects of fructose, HFCS, or sucrose consumption on NAFLD» (Chung et al., 2014).
The newer isocaloric liver-fat meta-analysis does not fill this cell, because it swapped macronutrients, not sugar for starch. Winters-van Eekelen found that trading saturated for unsaturated fat moved liver fat a lot (SMD -0.80, 95% CI -1.09 to -0.51, 4 comparisons). Trading carbohydrate for protein moved it moderately (-0.33, -0.54 to -0.12), while the total fat-for-carb ratio did nothing (0.01, -0.36 to 0.37) (Winters-van Eekelen et al., 2020). None of those arms is a sugar-for-starch exchange, so the sugar-specific isocaloric liver-fat magnitude stays unmeasured — there is no number to headline -> Fatty Liver MASLD and Weight Loss.
AASLD’s guidance points the other way: «Excessive fructose consumption in particular increases the risk of NAFLD, NASH, and advanced fibrosis independent of calorie intake.» (Rinella et al., 2023) It cites three references for this. By their titles, two are fructose-association studies and one is a study of fructokinase-dependent diet-induced steatohepatitis (Rinella et al., 2023), which reads as an animal model. By title, the two human studies are associations in patients already ill (obese adults with type 2 diabetes; patients with NAFLD), so none reads as long-term incidence evidence .
Chung appraised six observational fructose-liver studies (not AASLD’s specific references). Four, all at high risk of bias, found higher fructose or sucrose intake associated with NAFLD; a medium-risk study in children and the one low-risk-of-bias adult cohort found no association. Chung rated the body insufficient for high risk of bias and inconsistent findings (Chung et al., 2014). So the chronic regime is thinly and poorly evidenced: the two are not cleanly joined, and the gap is one of design and quality, not a settled contradiction. For the net-of-energy question the stronger design is Chung’s, whose verdict is insufficient. The lever stays cut the liquid energy, not avoid the fructose molecule.
No held trial links sugar to hard cardiovascular events
No gold source in the fabric connects added sugar or SSBs to incident cardiovascular events with a dose-response. What is held runs one step short, through surrogates: SSBs raise serum urate, urate-lowering lowers blood pressure, and urate and blood pressure are both markers, not events -> Uric Acid Lowering and Blood Pressure. The observational cardiometabolic associations that do exist are measured as sugar is consumed, with its added calories. So the hard-CVD-event cell is absence of evidence on the endpoint, filled only by surrogate channels -> Surrogate Outcomes.
Whether the diabetes signal survives at equal calories is thin
The SSB -> type 2 diabetes association is observational and measured as sugar is consumed, adding liquid calories on top of the diet. Whether any diabetes effect persists under an isocaloric swap, net of weight, is not cleanly held: no gold source isolates a sugar-for-other-carbohydrate exchange on diabetes incidence with adiposity held constant. The isocaloric weight swap being null makes an energy-mediated reading likely, but that is an inference from the weight cell, not a measured diabetes result.
The caries dose-response has never been tested in a trial
Dental caries is the one clearly sugar-molecule-specific harm, yet no randomized trial has measured its dose-response. The evidence is observational and ecological, and the sub-5%-of-energy region rests on only a few ecological studies graded very low — so any threshold there marks the edge of the data, not a knee in the curve -> Free Sugars Intake.
The joined urate -> blood pressure -> events chain has not been run
Both legs of the fructose -> urate -> blood pressure pathway are held, from different research groups, but they do not join. Upstream, SSBs raise serum urate even at equal calories (GRADE high) — but that is measured on urate only, in roughly normouricemic healthy adults (Ayoub-Charette et al., 2021).
Downstream, a urate-lowering drug lowers blood pressure — but on blood pressure only, in a hyperuricemic stratum, and the drug’s xanthine-oxidase route leaves it unclear whether urate itself is the mediator (Qu et al., 2016). Disjoint strata, disjoint endpoints, a drug rather than a food: no source carries food-fructose -> urate -> blood pressure -> events end to end in one population. Whether cutting SSBs lowers blood pressure or events in a normouricemic person — the default stratum here — is untested -> Uric Acid Lowering and Blood Pressure.
These five gaps share one structure. Liver-fat content and serum urate are surrogates, and GRADE discounts a surrogate by one or two levels precisely because a marker can move while the outcome does not -> Surrogate Outcomes. Even the well-evidenced urate result stops at the marker; the liver-fat results sit two steps short of liver-related death, a gap AASLD names in its own voice — the long-term link among liver fat, histological response and clinical outcomes is unestablished (Rinella et al., 2023). So these cells are not merely unpooled: even where a number exists, its transmission to something a person can feel is untested. Which leaves the decision, and how it sits against published guidance.
Cut the drinks, keep the fruit, and mind what replaces them
Cut sugar-sweetened beverages first. They are the dominant carrier of free sugars, and they act on two channels at once. They add poorly-compensated liquid calories, which drive the weight harm and plausibly part of the type 2 diabetes and hypertension associations (an energy route that is likely but not shown). And they carry the clearest isocaloric, fructose-specific signal that survives a calorie-for-calorie swap (serum urate; GRADE high). Sweets and desserts also raised urate in substitution trials (+0.35 mg/dL, 0.07 to 0.63, 10 trials, moderate certainty, at high doses), so the order is SSBs first, then sweets and desserts on lower-certainty evidence (Ayoub-Charette et al., 2021). Whole fruit stays. It sits outside the free-sugars limit on every axis — the fibre matrix blunts the fructose flux, and it runs null-to-favourable on urate and vascular endpoints. Treat 100% juice case-by-case: it behaves like an SSB on the metabolic and dental channels but lowers urate and associates inversely with stroke and coronary heart disease, so the call depends on which outcome the person is managing -> Free Sugars Intake.
What you swap in decides the benefit. An effect is always relative to the replacement The Comparator Problem: swap free sugars for other carbohydrate at equal energy and body weight barely moves, so cutting an SSB helps only when you do not put its calories back in another form. Reach for water or an unsweetened drink, or whole fruit for a sweet craving. Non-sugar sweeteners are not the recommended route — WHO advises against using them for weight or disease control and points instead to «consumption of foods with naturally occurring sugars, such as fruit, and unsweetened foods and beverages» (World Health Organization, 2023) -> Non-Sugar Sweeteners (the sweetener-safety question is worked in the companion Artificial Sweeteners deliverable).
The four guidance bodies agree on direction, not on a number
WHO, SACN, NNR and EFSA all point the same way — less is better — and that is the whole of their agreement. They diverge on where a number attaches and on what warrant carries it:
- WHO sets <10% of energy (strong) and <5% (conditional), warranted by dental caries (World Health Organization, 2015).
- SACN sets a single <5%E tier — about 30 g of free sugars a day for a typical adult — warranted by caries plus total energy intake (Scientific Advisory Committee on Nutrition, 2015).
- NNR sets <10 E% with no lower tier, warranted by nutrient density — sugar crowding out micronutrients and fibre, not a disease dose-response (Nordic Council of Ministers, 2023).
- EFSA, asked for a safe threshold, found none and set no number: «the intake of added and free sugars should be as low as possible in the context of a nutritionally adequate diet» (European Food Safety Authority, 2022).
These four bodies do not corroborate one another, even where their numbers match. They read overlapping evidence. Where their figures coincide, they had excluded each other’s evidence and were regulating different numerators — SACN’s <5% omits fruit-juice concentrates, WHO’s does not, so the identical number covers different things -> Free Sugars Intake. EFSA is not opposed to WHO either: EFSA runs a hazard assessment whose output is a threshold-or-nothing, while WHO issues a communicable population target that needs a number whether or not a clean threshold exists. That is a standpoint distinction, not a disagreement about the evidence -> Which Objective Moved This Recommendation.
Read any of these numbers as the edge of the evidence, not a safe floor. For chronic metabolic disease, EFSA states the sub-10%E region is the least-studied stretch — «the relationship between the intake of added and free sugars and risk of chronic metabolic diseases could not be adequately explored at levels of intake < 10 E% owing to the low number of RCTs available» (European Food Safety Authority, 2022). That bears on SACN’s energy limb and on any metabolic reading of a <5% tier, not on WHO’s caries basis for its <5%. For caries, EFSA notes individual cohorts report a positive, linear dose-response in permanent teeth (tested for non-linearity in only one), but says the data did not allow dose-response modelling across its body of evidence or a level at which caries risk is not increased (European Food Safety Authority, 2022). On both channels, the low thresholds mark where the data thin out, not a knee in a dose-response curve.
One non-health objective sits inside the guidance, and it stays named, not priced. NNR’s third warrant is environmental: «Higher consumption of added and free sugars contributes to a negative environmental impact» (Nordic Council of Ministers, 2023). That load is real and pushes the same direction, but this deliverable holds no carbon, land or water data and does not weigh it against the health case.
Where the stratum shifts the call
Three strata change the reading, briefly.
- Children. The pooled trials in children came out null largely because they never achieved a real intake contrast — an adherence null, not an efficacy one — and the provision trials that did achieve the contrast confirmed the effect (Morenga et al., 2013). The child cohorts that found harm (OR 1.55) were measuring beverages: 14 of 15 studies used an SSB exposure. So for children too, the lever is the drink, not sugar at large.
- Low energy intake. NNR’s displacement warrant bites hardest here. A fixed proportion of energy from sugar crowds out proportionally more of a smaller nutrient budget, so an older adult or a small child who eats little has less room for sugar before micronutrients suffer — which is why NNR flags this stratum by name (Nordic Council of Ministers, 2023).
- Hyperuricemic patients. The urate-lowering trials that also lowered blood pressure enrolled hyperuricemic patients on a drug; that downstream urate -> blood-pressure leg does not transport to a normouricemic person -> Uric Acid Lowering and Blood Pressure. Do not extend the SSB -> urate finding into a blood-pressure or event claim for the general adult.
And no acute upside sits on the other side of the ledger. A gold meta-analysis of acute carbohydrate-challenge trials finds no mood or energy boost at any time-point — mildly worse alertness within the hour — so the sugar rush is a phantom -> Acute Carbohydrate Effects on Mood. Nothing offsets the costs above.
The bottom line. For a general adult, cut sugar-sweetened beverages and swap in water, an unsweetened drink, or whole fruit; keep whole fruit; judge 100% juice by the outcome you care about. No guidance body found a safe threshold, the numbers below 5% rest on the thinnest evidence, and the harm that a calorie-for-calorie swap cannot erase is narrow — caries and urate — not a general property of the sugar molecule.
Caveats
- The loop is open. This appraises the published evidence for coherence and fidelity to its sources; no step here tests a recommendation against a realised outcome in a person. A clean appraisal is not a validated result.
- This appraises evidence; it does not prescribe. Selecting, dosing and screening for individual circumstances (medications, history, dental and metabolic status) are clinical acts outside this scope.
- Estimates are population-level by default. The magnitudes above are averages over studied groups; absolute benefit scales with a person’s baseline risk, and any individual sits somewhere the population estimate only approximates. Stratify per person.
- One health axis only. This weighs sugar on health outcomes. Cost, environmental load and other axes are real but are not priced here; where they trade against health, that weighing is the person’s.
Evidence box
Question ’For an adult deciding how much added/free sugar and how many sugar-sweetened beverages to consume: what is the effect on each patient-important outcome (dental caries, body weight, type-2 diabetes, cardiovascular disease, liver fat), what is the dose-response shape for each, does any effect persist when sugar is exchanged calorie-for-calorie with other carbohydrate, and are sugar-sweetened beverages a distinct exposure from sugar in solid food?‘ Evidence included 19 sources — 11 gold, 6 high, 1 moderate, 1 method Overall certainty Moderate (see Rating Certainty of Evidence) Source-selection note 1 source(s) below the gold/high evidence bar feed this page: Sutton (cohort, moderate). Their roles and limitations are described in the text. Last updated 2026-10-07 · Independently reviewed: No · Full edit history