A very common belief — yo-yo dieting is worse than staying heavy, so don’t bother trying to lose weight. Three sources now anchor the answer, across two outcome families: a narrative review (Montani 2015, tier moderate) foregrounding the controversy; a gold SR/meta-analysis of 23 cohort studies / 441,199 participants (Zou 2019) pooling the CVD-and-mortality magnitude Montani could not; and a second gold SR/meta-analysis of 14 cohorts / 253,766 participants (Zou 2020) adding the type-2-diabetes-incidence arm. The converging verdict is deflationary in the direction that matters for the decision: body-weight fluctuation is associated with higher all-cause and CVD mortality and with incident T2D across cohorts, but the association is observational, inconsistent, and — for the CVD/mortality outcome — not detected once fluctuation is restricted to the INTENTIONAL kind that an obese person choosing to diet would incur (3 reports; P values only, no pooled RR or CI), and not significant in the frankly obese (BMI>=30) for the T2D outcome (RR 1.47, 0.96-2.24; 3 reports, I2=88.8% — compatible with no effect or with a penalty as large as the overweight stratum’s) — the very stratum the decision is about. One signal runs the other way: Zou 2020’s intentional-cycling T2D subgroup is positive, but rests on 3 reports the authors say limit the conclusion; the source does not identify the three studies or their BMI mix (worked in the T2D section below) (corrected 2026-10-06: absent / null / 3 mixed-BMI reports -> not detected / not significant / BMI mix unreported; self-critique). No held source licenses deterring an obese patient from attempting weight loss out of fear of cycling — Zou 2019’s authors decline that conclusion: «Hence, we cannot conclude whether obese individuals should be deterred from efforts to control their body weight by concerns about the hazards of weight cycling.» (Zou et al., 2019) — which is not the same as showing the attempt is cycling-safe (corrected 2026-10-06: the authors say so in as many words -> the authors decline to conclude either way; Zou 2020 intentional-positive T2D signal added, Zou 2020 chunk 01). (Montani et al., 2015; inferred from Zou et al., 2019, 2020)
The two Zou meta-analyses are from the same Wuhan / Tongji-Huazhong research group (seven shared authors; Zou 2020 cites Zou 2019 as ref 46), so they are complementary by outcome, never independent confirmation — the agreement below is NOT a type-E robustness claim (see the same-group note in the T2D section and in Limits).
The evidence is contested, and the associations attenuate on adjustment
Montani foregrounds the controversy rather than asserting a harm: «the topic of health consequences of weight cycling has been the source of considerable controversy … with ongoing debates about whether it is prudent to even recommend that over- weight or obese adults should try to lose weight». (Montani et al., 2015)
- Early epidemiological studies suggested weight-fluctuation harm; later studies did not confirm it. In NHS II and a Framingham subset the weight-cycling -> type 2 diabetes association was no longer significant after adjusting for overall weight status; several large studies failed to confirm a weight-cycling -> CV-mortality link.
- There is no universally accepted definition of weight cycling (amplitude, number of cycles, intentional vs unintentional), so cross-study comparison is weak — a measurement problem, not a clean null. (Montani et al., 2015)
Who the signal is actually in — normal-weight, not obese
The decision-relevant nuance: «some of the potentially negative health consequences of repeated dieting and weight cycling are more readily seen in people of normal body weight rather than in those who are overweight or obese». The metabolic-syndrome / hyperinsulinaemia / dyslipidaemia associations reached significance only in BMI<25 subgroups and were lost above BMI 25; the author’s summary is that it is «weight cycling resulting from repetitive intentional weight loss in the young and primarily normal-weight population groups that seems to be more strongly associated with risks». (Montani et al., 2015)
This population split matters because the driver differs: normal-weight cyclers diet for body image (media/social pressure, weight-sensitive sport), not for a metabolic indication — a confounding structure quite unlike an obese patient losing weight on medical advice. It also predicts future weight gain in the lean (dose-dependent with cycle count). (Montani et al., 2015) The confounding contrast with medically-indicated loss is the wiki’s reading, not Montani’s.
The pooled magnitude — a gold SR/MA that confirms the SHAPE and cashes the confounding
Zou 2019 pools 23 cohort studies (441,199 participants) and gives the numbers the narrative could not (Zou et al., 2019):
| Outcome | RR (95% CI) | Note |
|---|---|---|
| All-cause mortality | 1.41 (1.27-1.57) | I2=78% (high heterogeneity); corrected to 1.18 (1.05-1.32) after trim-and-fill for the one outcome with publication bias |
| CVD mortality | 1.36 (1.22-1.52) | I2=32% |
| CVD morbidity | 1.49 (1.26-1.76) | secondary outcome |
| Hypertension | 1.35 (1.14-1.61) | secondary outcome |
| Cancer mortality | 1.01 (0.90-1.13) | NULL — a specificity signal against a generic frailty confounder |
Three features carry the decision, and all three deflate the naive cycling-is-dangerous reading:
- Publication bias inflates the headline. All-cause mortality was the ONLY outcome flagging bias (Egger P=0.001, Begg P=0.014); trim-and-fill deflates it from RR 1.41 to «RR, 1.18, 95% CI 1.05-1.32» — the authors note this «suggested the existence of potentially negative studies.» (Zou et al., 2019) State the corrected figure, not 1.41, at point of use.
- The inconsistency is in the data, not just the framing. 21 of 28 reports were positive, but the 7 null reports held >60% of participants and events — so the pooled point sits on top of a genuinely split literature. (Zou et al., 2019)
- Intentional cycling: no association detected — but in a 3-report subgroup. Restricted to intentional fluctuation, no association was detected (3 reports; P values only, no pooled RR or CI reported): all-cause mortality P=0.49, CVD mortality P=0.12. The authors’ «only 3 reports each of intentional and unintentional» qualifier binds both subgroups — this is insufficient evidence, not absence (corrected 2026-10-06: NO association … vanish -> not detected, 3 reports; self-critique). «Limited studies demonstrated that intentional weight fluctuation was not associated with an increased risk of all- cause and CVD mortality or hypertension in our study.» The positive signal concentrates in unintentional fluctuation (all-cause P=0.016) — but from only 3 reports / 2,975 participants, which the authors say «limited the conclusion.» (Zou et al., 2019) Read on Zou alone, this is consistent with confounding by indication: illness-driven (unintentional) loss, not chosen dieting, may carry the mortality signal — the prior MAs Zou cites show intentional loss lowering mortality and unintentional loss raising it. This is NOT Montani’s position: Montani names intentional-vs-unintentional only as a measurement difficulty, and places the stronger signal in repetitive intentional weight loss in the young and normal-weight — a different stratum from Zou’s mixed populations (corrected 2026-10-06: exactly Montani’s confounding-by-indication concern cashed out -> Zou-only reading; self-critique). (inferred from Zou et al., 2019)
Montani (narrative) vs Zou (SR/MA) — matched parameters
| Parameter | Montani 2015 (narrative) | Zou 2019 (SR/MA) | Same quantity? |
|---|---|---|---|
| Overall harm verdict | controversial, attenuates on adjustment, inconsistent | pooled RR 1.36-1.49 significant BUT 7 null reports hold >60% of events; corrected all-cause RR 1.18 | PARTIAL — both foreground inconsistency; Zou adds a magnitude and confirms the inconsistency Montani asserted qualitatively (F-refinement, not a clash) |
| Intentionality | named as a measurement/definition difficulty; stronger signal placed in repetitive intentional loss in the young, normal-weight | intentional subgroup not detected (3 reports); unintentional carries the signal (3 reports) | NO — a distinction (Montani: intentional cycling in the lean; Zou: intentional not detected in mixed populations); not a test of Montani’s hypothesis (corrected 2026-10-06; self-critique) |
| Reverse causation | not discussed in held chunk | «the fluctuation in weight could be the consequence, and not the cause» | NO — Zou only (corrected 2026-10-06: foregrounded / both name it; self-critique) |
| Where the signal sits by BMI | normal-weight (<25); metsyn/insulin/lipid markers lost above BMI 25 | all-cause: at ALL BMI; CVD mortality: normal/overweight and <60y, NOT BMI>=30 | NO — different outcomes/designs (Montani = metabolic-syndrome markers, incl. a cited 7-year prospective French cohort of 3,553; Zou = cohort MORTALITY). A distinction, not a joined tension; they partially converge in that neither locates the strongest signal in the frankly obese |
| Decision for the obese patient | reports the «ongoing debates», no verdict | «we cannot conclude whether obese individuals should be deterred … by concerns about the hazards of weight cycling» | PARTIAL — neither issues a deterrent verdict; Montani reports a debate, Zou declines to conclude; not a demonstration that attempting is cycling-safe (corrected 2026-10-06: YES, same decision; self-critique) |
The BMI-split row is a distinction, not a tension: Montani’s normal-weight signal is on metabolic-syndrome markers (cross-sectional studies plus a cited 7-year prospective cohort), Zou’s is on cohort mortality — different outcomes on different designs, so they are not joined (they cannot be the same quantity). They do converge weakly in that neither places the strongest harm signal in the frankly obese, which is the population the decision is about. (Montani et al., 2015; inferred from Zou et al., 2019)
The T2D-incidence arm — a second gold SR/MA (same group, different outcome)
Zou 2020 pools 14 cohort studies / 253,766 participants / 8,904 new-onset diabetes events (21 reports, since Yokomichi 2017 supplied 8) and asks the diabetes question the CVD/mortality MA did not (Zou et al., 2020):
| Stratum | RR (95% CI) | Note |
|---|---|---|
| Overall (all reports) | 1.23 (1.07-1.41), P=0.003 | I2=73.9% (high heterogeneity) |
| Overall, omitting Yokomichi | 1.41 (1.25-1.59), P<0.001 | Yokomichi’s 8 null reports used a small (>4%) cycle amplitude that «might not have shown any adverse effect» |
| Overweight (25<=BMI<30) | 1.62 (1.31-2.00), P<0.001 | I2=0 — where the T2D signal concentrates |
| Obese (BMI>=30) | 1.47 (0.96-2.24), P=0.55 (full, as printed); P=0.08 omitting Yokomichi | not significant, 3 reports, I2=88.8% — the decision-relevant stratum; the printed P=0.55 is inconsistent with the CI (which implies about 0.08 — wiki arithmetic; the abstract reports P = 0.08) |
| Intentional weight loss | 1.23 (1.16-1.31), P<0.001 | I2=0 but only 3 reports — «the small number of studies limits the conclusion» |
Three features carry the decision:
- The obese stratum is not significant — the cell the decision is actually about. «the association between weight cycling and the risk of developing diabetes was not observed in obese participants (body mass index >=30 kg/m2; P = 0.08)», and the null «was still detected in obese individuals (BMI >=30 kg/m2), even after removing the Yokomichi et al. study (P = 0.08)». (Zou et al., 2020) The authors’ gloss: «suggesting that obesity might have an effect on the risk of diabetes similar to that of weight cycling.» (Zou et al., 2020) Reading that as obesity saturating the risk is the wiki’s mechanism, not the authors’. But the stratum is RR 1.47 (0.96-2.24), not significant, 3 reports, I2=88.8% — compatible with no effect or with a penalty as large as the overweight stratum’s (1.62). For an obese person choosing to diet this MA neither shows nor excludes a weight-cycling T2D penalty; state the obese-stratum estimate with its CI, not the pooled 1.23, at point of use (corrected 2026-10-06: shows no measurable weight-cycling penalty -> neither shows nor excludes; self-critique). (inferred from Zou et al., 2020)
- No publication bias — unlike the CVD arm. «Egger’s (P = 0.53) and Begg’s (P = 0.81) tests suggested that there was no significant publication bias.» (Zou et al., 2020) (Contrast Zou 2019, where all-cause mortality flagged bias and needed trim-and-fill.) The T2D point estimate therefore needs no headline correction, but heterogeneity stays high (I2=73.9%).
- Reverse causation is left open by the authors themselves. «weight cycling might be a marker of potential illnesses that have worse prognoses … weight cycling could be the early sign and consequence, and not the cause, of the health endpoints, including diabetes», with depression / bipolar disorder named as disorders that both drive cycling and raise diabetes risk. (Zou et al., 2020) The conclusion is explicitly non-causal: «Future physiological studies are required to show the causal links». (Zou et al., 2020)
Symmetric-standards note: Zou 2020’s own summary language is stronger than the CVD paper’s — its abstract calls weight cycling «a strong independent predictor of new-onset diabetes». (Zou et al., 2020) The pooled association is statistically significant and survives leave-one-out sensitivity analysis, but is null in the higher-quality (NOS>7: RR 0.94, 0.77-1.16) and >=10-year follow-up (RR 0.94, 0.74-1.21) subgroups, and meta-regression names study quality (P = 0.007) as a heterogeneity source (Zou et al., 2020) (corrected 2026-10-06: the pooled association is real -> significant but null in the higher-quality and longer-follow-up subgroups; self-critique). It is also observational, causally disclaimed by the same authors, and not significant in the obese — so it does not transport to the decision as a deterrent. Report the association and its obese-stratum (not significant, 3 reports) and reverse-causation caveats together.
Intentional cycling: T2D DIFFERS from CVD/mortality — do not merge them
The intentional-weight-loss subgroup runs the opposite way across the two Zou outcomes, and the different outcome means this is a distinction, not a contradiction to reconcile. For CVD/mortality (Zou 2019) no association was detected for intentional fluctuation (3 reports; all-cause P=0.49, CVD P=0.12). For T2D (Zou 2020) the intentional subgroup is positive — «individuals with intentional weight cycling had a higher risk of diabetes (P < 0.001) … however, the small number of studies limits the conclusion». (Zou et al., 2020) Two guards keep this from over-reading: (i) it rests on only 3 reports (I2=0); the source does not identify the three studies or their BMI mix (corrected 2026-10-06: drawn from mixed-BMI, not-predominantly-obese populations -> unreported; self-critique); (ii) it is a different endpoint from the CVD-arm intentional result (not detected, 3 reports), so the two cannot be pooled or read as a single intentional cycling is safe/unsafe verdict. For the obese decider, both decision-relevant T2D cells — the intentional-positive and the obese-stratum RR 1.47 (0.96-2.24) — are 3-report subgroups; neither settles the question.
Zou 2020 vs Zou 2019 — matched parameters (same group, different outcome; NOT type-E)
| Parameter | Zou 2019 (CVD/mortality SR/MA) | Zou 2020 (T2D SR/MA) | Same quantity? |
|---|---|---|---|
| Primary outcome | all-cause + CVD mortality/morbidity, hypertension | new-onset type-2 diabetes | NO — different endpoints |
| Pooled overall effect | all-cause RR 1.41 -> 1.18 corrected; CVD mort 1.36 | T2D RR 1.23 (1.07-1.41) | NO — not comparable across endpoints |
| Publication bias | present on all-cause (Egger P=0.001), trim-and-fill applied | none (Egger P=0.53, Begg P=0.81) | NO (different outcome); both tested, opposite result |
| Obese subgroup (BMI>=30) | CVD mortality NOT in BMI>=30 | T2D RR 1.47 (0.96-2.24), not significant, 3 reports | PARTIAL — both not significant in the obese, on different outcomes; not one quantity |
| Intentional subgroup | not detected, 3 reports (all-cause P=0.49, CVD P=0.12) | POSITIVE 1.23 (1.16-1.31), 3 reports | NO — different outcome, opposite direction |
| Reverse causation | foregrounded | foregrounded («consequence, and not the cause») | YES, same axis — both name it |
| Research group | Wuhan / Tongji-Huazhong | SAME group; Zou 2020 cites Zou 2019 (ref 46) | SAME — non-independent (NOT type-E) |
Same-group NOT-E guard (load-bearing). Zou 2019 and Zou 2020 share the Wuhan / Tongji-Huazhong group (seven overlapping authors) and Zou 2020 cites Zou 2019 as its own antecedent (ref 46), so the two agreeing on non-significant obese strata and on reverse causation is shared-lineage self-consistency, not independent corroboration — no type-E robustness claim spans them. They are a different-outcome coverage extension (type-F/coverage: CVD/mortality + T2D), which is why the “same quantity?” column above is almost entirely NO. A T2D meta-analysis from a non-overlapping group would be needed before any confidence gain from convergence. (Zou et al., 2019, inferred from 2020)
Absolute magnitude. The MA reports relative risk only; crude pooled incidence is 8,904 / 253,766 = 3.5% over a median follow-up ranging 2.5-32 years, so RR 1.23 implies on the order of +0.8 percentage points at that (very heterogeneous) baseline — a rough figure, not a per-year absolute risk, and undefined for the obese stratum where the effect is not significant. (inferred from Zou et al., 2020) No dose-response curve is estimable: weight cycling is categorical with «no single definition», and the one low-amplitude (>4%) cohort (Yokomichi) drove the null reports — the authors offer low amplitude, low Japanese diabetes prevalence, and ethnicity as candidate explanations, so amplitude is one possibility among three, not a locatable knee (corrected 2026-10-06: a hint that amplitude matters; self-critique).
The proposed mechanism — repeated overshoot (a theory, not an outcome)
Montani’s mechanistic frame: weight cycling drives fluctuations of BP, heart rate, sympathetic activity, renal filtration, glucose and lipids, with overshoots above normal during the regain phase that may not be compensated by the loss-phase decreases (the oscillations are asymmetric), adding cardiac load and possible glomerular / vascular injury — invisible in a single cross-sectional snapshot. (Montani et al., 2015) This is a physiological hypothesis with small human corroboration, not an outcome finding.
A separate strand: older dieters regain fat but not fat-free mass, so repeated cycling risks relative sarcopenic obesity (Montani et al., 2015) — a function/muscle concern that may transport better than the CV one -> Weight-Loss Maintenance and Metabolic Adaptation (audit 2026-10-06: firmer dropped — narrative-review claim, not pooled).
Decision relevance
- Fear of weight cycling is not a good reason for an obese person to avoid intentional weight loss — harm is not shown, which is not the same as shown safe. Two gold SR/MAs (same group), neither of whose authors draws the deterrent conclusion. Zou 2019 (CVD/mortality): «we cannot conclude whether obese individuals should be deterred from efforts to control their body weight by concerns about the hazards of weight cycling.» (Zou et al., 2019) Zou 2020 (T2D incidence): the diabetes association is not significant in the obese stratum (BMI>=30, RR 1.47, 0.96-2.24, 3 reports) — the very population the decision is about — and its authors disclaim causation. Across both outcomes the harm signal is observational, and for the obese decider it is either not detected in a 3-report intentional subgroup (CVD/mortality) or not significant in a 3-report obese stratum (T2D) — insufficient evidence of harm, not evidence of no harm — while the benefits of sustained loss in the obese/metabolically-ill are established (glycaemia, remission, MASLD, function) -> Does Weight Loss Reduce Cardiovascular Events, Total Diet Replacement and Type 2 Diabetes Remission, Lifestyle vs Metformin for Diabetes Prevention. The honest framing: pursue sustainable loss; the cost of trying-and-partly-regaining is not shown to exceed the benefit of the loss achieved. (inferred from Zou et al., 2019, 2020)
- The population where caution is better warranted is the opposite of the intuitive one: lean, young, body-image-driven dieters (and weight-cutting athletes) — for whom repeated dieting predicts weight gain and carries the clearer metabolic signal, and for whom there is no metabolic indication to diet at all. (inferred from Montani et al., 2015)
- Prioritise maintenance design over loss magnitude. Since the plausible harm is in the cycling, the lever that matters is a structured maintenance phase that prevents regain, not a bigger initial deficit -> Weight-Loss Maintenance and Metabolic Adaptation.
Limits
- Confidence moderate — two gold SR/MAs now anchor the page, but the loop stays observational. Zou 2019
pools 441,199 participants (CVD/mortality) and Zou 2020 pools 253,766 (T2D incidence), together lifting the
page well off
low; it does NOT reachhighbecause both are observational (no causal inference possible), Zou 2019’s all-cause estimate is publication-bias-inflated (corrected RR 1.18) with high heterogeneity (I2=78%), Zou 2020’s overall estimate carries I2=73.9% and its own authors disclaim causation, the two are from the SAME group (no independence gain — see NOT-E guard below), and the decision-relevant subgroups are thin (CVD intentional = 3 reports; T2D obese = 3 reports, not significant). The magnitude that WOULD settle it — fluctuation vs maintenance-of-obesity vs stable loss, head to head — «no cohort studies have directly compared.» (Zou et al., 2019) - The repeated overshoot mechanism is a hypothesis; no hard-outcome trial isolates weight cycling from the confounders (baseline health, unintentional loss from illness, dieting motivation). Zou explicitly keeps reverse causation open alongside the metabolic-disturbance pathway.
- Lean-mass loss is unmeasured. Zou tracked only weight, not waist circumference or body composition, so the sarcopenic-obesity strand -> Weight-Loss Maintenance and Metabolic Adaptation is not addressed by its pooling. (Zou et al., 2019)
- Coherence, not validity (R1): the loop is open on whether cycling per se changes any patient-important outcome.
- Same-group NOT-E guard: Zou 2019 (CVD/mortality) and Zou 2020 (T2D) share the Wuhan / Tongji-Huazhong research group (seven overlapping authors), and Zou 2020 cites Zou 2019 as its own antecedent (ref 46). They are therefore complementary by outcome, not independent corroboration — their agreement on obese strata and on reverse causation is shared-lineage self-consistency, NOT convergent confirmation, and no type-E robustness claim spans them. A T2D meta-analysis from a non-overlapping group is the gap.
- T2D obese stratum is small-n, not a proof of no effect (R1 / four-states). BMI>=30 rests on 3 reports: RR 1.47 (0.96-2.24), I2=88.8% in the full analysis (Table 1, printed P=0.55), and P=0.08 after omitting Yokomichi (Table S5, not held); this is insufficient-evidence-of-harm in the obese stratum, not established no-effect. It licenses no measurable T2D deterrent for the obese decider, not a claim that cycling is proven safe. (Zou et al., 2020)