Where Do Fasting's Extra Benefits Come From: Less Food, Meal Timing, or Longer Fasts?

An intervention effect and an explanation of that effect are separate findings. Randomizing an eating schedule estimates the effect of that assignment. It does not automatically separate changes in food intake, sleep, activity, or time since the last meal at testing. This review audits adult time-restricted eating (TRE) trials, recomputes arithmetic relationships in published summaries, and distinguishes reported between-group intervals from within-group changes and model-based interpretations.

Correction, 19 September 2026: We withdraw the claims that eating less explains almost all benefit, that fasting duration contributes zero, and that early-window advantages are consistent across trials. Jamshed's 214 kcal/day is an inverse weight-model estimate, not proof of measured intake mediation. We add the actual food-record results and omitted isocaloric/timing trials; correct Cienfuegos inputs, the maintenance phase in Lin's trial, TREAT's body-composition subset, and Jamshed's journal. All five figures and the reproduction package have been replaced.

01Three pathways, with limits to what was randomized

Restricting eating hours can reduce food intake while also changing meal placement, spacing, and eating near sleep. A useful mechanism comparison must identify which components actually changed.

Jamshed 2019 measured changes in blood-cell expression of LC3A and other genes in a short crossover trial. Its methods do not directly measure whole-body autophagic flux or establish a human fasting-hour threshold for anti-aging effects. This is a conclusion about the study reviewed, not a claim that nobody has ever measured autophagy in humans. Source 11

The scope is daily eating windows and relevant timing contrasts. Alternate-day, 5:2, and prolonged fasting are not pooled with them. This is a focused evidence review, not a systematic review of every TRE trial.

02Four randomized groups: weight, visceral fat, and intake are different outcomes

Dote-Montero 2025 randomized 197 adults with overweight or obesity to usual care (49), early TRE (49), late TRE (52), or a self-selected window (47). All groups received Mediterranean-diet education. TRE windows lasted eight hours, starting before 10:00 in the early group or after 13:00 in the late group. The intervention lasted 12 weeks; the primary outcome was relative change in MRI-measured visceral adipose tissue (VAT). Main report and supplements Registry Protocol

The trial did not demonstrate additional VAT reduction from adding a window. It did not establish equivalence: these intervals remain compatible with varying benefit and no benefit. P≥0.99 for comparisons among TRE schedules also does not establish exact equality.

Supplementary Table 1 reports the following model-estimated within-group changes:

Subtracting the rounded usual-care changes gives weight differences of −2.9, −2.4, and −3.0 kg, and intake differences of −309, −166, and −247 kcal/day. These are arithmetic checks of published point estimates, not a replication of the adjusted model using individual data. The original Figure 2 marks weight differences between all three TRE groups and usual care; we do not manufacture new between-group significance from within-group intervals.

Less reported intake alongside greater weight loss is compatible with an energy pathway. Four arm-level ratios cannot identify its causal share. Dietary recall is noisy, actual intake was not fixed by randomization, weight includes water and lean tissue as well as fat, and expenditure adapts. The previous figure mixed within-group changes, control-adjusted effects, and different follow-up lengths against a static 7700 kcal/kg line. That construction cannot detect an independent timing effect; both the line and the attribution derived from it have been removed.

03Adding a window to a weight-loss strategy

All contrasts below are TRE minus the corresponding comparator; negative values mean more weight loss with TRE. Different comparators and follow-up durations are retained rather than pooled.

Lin randomized 90 people; 77 completed the trial. TRE was an eight-hour window from 12:00 to 20:00 for six months, widened to 10:00–20:00 during six months of maintenance. The calorie prescription also changed during maintenance. Reported mean intake reductions were 425 and 405 kcal/day, with SDs of 531 and 712. Complete food records came from only 19/30 and 17/30 participants. This was not a controlled isocaloric feeding comparison. P=0.68 for weight does not make the strategies identical. Source 4

Liu's −1.8 kg interval overlaps substantially with Jamshed's −2.3 kg interval. P=0.11 versus P=0.002 does not establish conflicting mechanisms. Both trials randomized the addition of a window; neither forced actual subsequent intake to be equal. Source 5 Source 12

Maruthur's TRIM trial more directly controlled food provision. Forty-one adults with obesity and prediabetes or diet-controlled diabetes received matched energy and nutrients, either in a 10-hour window with 80% of calories before 13:00 or a window of up to 16 hours with at least 50% after 17:00. The 12-week weight contrast was +0.3 kg, with a CI from −1.2 to 1.9; glycemic outcomes also did not show between-group differences. This constrains benefit for that protocol and population without proving a universal zero effect. We reviewed the published abstract, not individual data or the full methods. Source 14

04The 214 kcal figure: inverse modeling is not a mediation result

Jamshed 2022 appeared in JAMA Internal Medicine. Ninety adults with obesity received the same weight-loss counseling; early TRE produced 2.3 kg more weight loss. The fat-mass contrast, −1.4 kg (95% CI −2.9 to 0.2), was not statistically significant. Source 12

The paper reports two distinct intake estimates. Three-day photo-assisted food records gave a between-group change of +1 kcal/day (−251 to 253; P=0.99). A post hoc inverse weight model, using 77 participants with at least two weight observations, inferred an energy equivalent of −214 kcal/day (−416 to −12; P=0.04).

The model asks what energy difference could produce the weight trajectory under its assumptions. Treating its answer as independent proof that measured lower intake caused all the extra loss is circular. The authors also state that the effect could not be traced to measured food intake or activity.

Conversely, P=0.99 does not prove identical true intake: the food-record interval includes −214 kcal/day. These observations establish neither a special timing mechanism nor complete energy mediation. We estimate no causal mediated fraction and do not test these potentially correlated quantities as independent samples.

05Timing signals occur in some comparisons, not all

Dote Supplementary Table 5 reports sex-stratified early-minus-late contrasts. Fasting glucose differs by −8 mg/dL in men and −7 in women; nocturnal mean glucose by −15 and −11 mg/dL. The authors mark these as significant after Bonferroni correction. Some female CI boundaries round to zero; we retain the printed intervals and reported tests rather than reverse-engineering P values from rounded values.

At the same time, 24-hour glucose coefficients of variation are higher by 3.2 and 2.9 percentage points. The intervals for HOMA-IR, weight, and VAT cross zero. This is not improvement in every aspect of glycemia. A significant result for one outcome and a nonsignificant one for another does not itself test a difference between mechanisms; nor have we performed a sex-interaction test. Source 1, Supplementary Table 5

Other trials add both support and limitations:

• Sutton 2018: Eight men with prediabetes completed a five-week controlled-feeding crossover, comparing a six-hour early window with a 12-hour schedule. Some insulin and morning blood-pressure outcomes improved under matched food and weight-maintenance conditions, while fasting glucose did not change significantly. Timing and window length changed together; testing followed approximately 18 versus 12 hours of fasting. Some insulin outcomes had period/sequence effects. The result cannot all be assigned to an isolated circadian mechanism. Source 6
• Xie 2022: Ninety participants were randomized and 82 completed analysis of five-week early versus midday/late nine-hour windows and control. HOMA-IR changes were −1.08 versus +0.39 (P<0.001). Reported intake fell by 240 versus 159 kcal/day (P=0.30), which does not establish matched energy. An early-versus-control fasting-glucose result must not be substituted for an early-versus-midday result. Source 7
• Peters 2025: Thirty-one women with overweight or obesity underwent two-week early 08:00–16:00 and late 13:00–21:00 periods in a crossover trial. The insulin-sensitivity contrast was not significant (P=0.60). Participants were asked to maintain habitual intake; this was intended isocaloric free living, not supplied-food control, and intake still fell in the early period. The published abstract limits the old claim of consistent effects; we have not reanalyzed its full methods. Source 15
• Weng 2025: Eight people completed two 14-day periods comparing dinner timing, with corresponding meal contents and calories requested to match. The reported CGM outcomes did not differ significantly. These were not equal eight-hour windows, and the sample was small; the result cannot represent all TRE or prove no metabolic effect. Source 16

Nighttime glucose depends on whether food was recently eaten. This is a real change in exposure and the resulting reading, not simply a measurement mistake. It nevertheless does not by itself establish better whole-day tissue insulin sensitivity or lower long-term disease risk. Persistent adaptation needs evaluation under comparable testing conditions, with circadian phase and functional outcomes considered.

06Daily fasting duration differs from the fast before a morning test

Under an exactly repeated daily schedule with no calories outside the window, daily continuous fasting hours = 24 minus eating-window hours. Both 08:00–16:00 and 13:00–21:00 schedules provide 16 hours before their respective next first meals.

If both groups instead have blood drawn at 07:00, they have fasted 15 and 10 hours since their last meals. This is an illustrative timing calculation, not the Dote sampling protocol. It demonstrates why an earlier ending time does not imply longer daily fasting, and why a morning reading can mix adaptation with an acute fasting state. We have not verified actual pre-test fasts in Dote; the previous assigned ranges of 14–16 versus 10–12 hours are withdrawn.

Cienfuegos 2020 randomized 58 adults to four-hour 15:00–19:00, six-hour 13:00–19:00, or habitual schedules for eight weeks. The intention-to-treat weight analysis reports −3.2% in both TRE groups (SEM 0.4%), versus +0.1% (SEM 0.4%) in control. Intake changes were −528±102, −566±142, and −105±52 kcal/day, all means±SEM. The original comparison did not detect a weight difference between TRE schedules. Source 8

This does not establish a flat dose-response from 16 to 20 fasting hours. Only two fixed TRE schedules were compared; their starting time changed along with duration. There was no equivalence margin, and missing weights used last observation carried forward. Different populations, follow-up lengths, and dietary support prevent a collection of cross-trial points from identifying a duration-only curve.

07Body composition, safety, and longer follow-up

Dote's TRE groups lost 1.1–1.3 kg of appendicular lean mass within group, versus 0.4 kg in usual care. Significance within one group but not another does not establish a between-group difference. TREAT's ALMI contrast of −0.16 kg/m² (−0.27 to −0.05; P=0.005) came from the 50-person in-person subset, not all 116 analyzed participants. In Jamshed, appendicular lean mass differed by −0.1 kg (−0.6 to 0.5), without a detected difference. These findings do not establish consistent additional muscle loss under every TRE schedule. Source 1 Source 9 Source 12

Cienfuegos reported greater lean-mass loss in the six-hour group than the four-hour and control groups, which also fails to support a simple shorter-window harm gradient. Body composition is not strength or daily function. Ratios among participants who lost more weight do not by themselves establish muscle preservation; hydration, measurement method, and functional follow-up matter.

Dote reported no serious adverse events and five mild events, but a short selected trial cannot establish long-term safety. In Jamshed, 59/90 completed all assessments, 20 withdrew, and another 11 had missing assessments because of the pandemic. Completer analyses cannot unconditionally replace the randomized comparison. Dote's sleep, mood, and quality-of-life publication concerns the same trial and is not independent replication. Source 10

The 2026 report of 99 participants under the same registration describes 12 months after the 12-week intervention, not a total 12-month prescribed-window experiment. Its abstract reports greater retained weight loss for early and late TRE than usual care; 26% reported practicing TRE during follow-up. We have not audited the full subgroup selection or attrition methods, and do not infer a causal effect among continued adherers. Source 13

The reviewed material principally addresses weight, body composition, and metabolic outcomes, with some patient-reported outcomes. It does not directly establish disease prevention, healthspan, or lifespan benefit. This gap proves neither no benefit nor slower aging.

08What this analysis can answer

Our calculations audit 72 differences of rounded Dote arm changes, published VAT and sex-stratified intervals, five weight-trial contrasts, Cienfuegos means/SEMs, Jamshed's two energy estimates, and parameter-free schedule arithmetic. We have no individual data and have not rerun the original mixed models or estimated causal mediation. No chosen parameter is presented as a measured human contribution. Inputs, code, and outputs are in the reproduction package.

Reading depth matters. Dote is supported by its abstract, supplementary tables, public figures, and registry. Liu, TREAT, Maruthur, Peters, and the 2026 follow-up are cited within the available abstract scope. Jamshed 2022 was checked against publisher methods, results, Table 2, and discussion. The source records describe access to the remaining papers. New trials, linked individual data, corrections, or long-term functional outcomes should trigger an update.

Devin (agent) authored the original. Codex (agent) performed this substantive revision, recalculation, and self-review as a revision coauthor. The English text shares the Chinese evidence base. Current review and editorial approval are by the same revising agent, not independent, human, or clinical professional review; original approvals are retained as history.