The measure you pick changes who counts as at-risk. INTERHEART — a standardised case-control study of first acute myocardial infarction, 27,098 people (12,461 cases / 14,637 matched controls) across 52 countries and all major ethnic groups — found that a marker of abdominal adiposity (waist-to-hip ratio; waist circumference) relates to MI far more strongly and consistently than BMI, the conventional measure. «Waist-to-hip ratio shows the strongest relation with the risk of myocardial infarction worldwide.» (Yusuf et al., 2005) This is a which-marker question (Layer-1 baseline-risk / route (a): measure the better prognostic marker) — NOT a claim that fat causes MI specifically through the WHR channel.
Contested by prospective design (added 2026-09-04). A later pooled individual-participant analysis of 58 prospective cohorts (ERFC 2011, 221 934 people) found the three measures have similar association strength with CVD, and that none adds predictive value once conventional risk factors are known — and it names INTERHEART as the claim it refutes. The “strongest relation / measure WHR not BMI” headline below therefore stands only for the case-control MI question in an all-ethnicities sample, and is attenuated for prospective CVD prediction in developed-country populations. See The prospective-cohort contest section and BMI vs Abdominal-Adiposity Markers - Which Predicts CVD.
The crux — BMI’s association is largely redundant with fat distribution type-F
Top vs bottom quintile OR for MI, following the adjustment cascade — this is the load-bearing number:
| BMI top vs bottom quintile | OR (95% CI) |
|---|---|
| Before adjustment | 1.44 (1.32-1.57) |
| After adjustment for waist-to-hip ratio | 1.12 (1.03-1.22) — substantially reduced |
| After adjustment for the other 8 risk factors | 0.98 (0.88-1.09) — non-significant, gone |
BMI’s apparent MI signal collapses once you know a person’s waist-to-hip ratio, and disappears entirely after full risk-factor adjustment — while WHR, waist, and hip stay highly significant after the same adjustment. So most of what BMI captures about MI risk is fat distribution it measures only crudely; the abdominal marker carries the information, and BMI adds little on top of it (adding BMI to WHR had «only a modest effect»; adding WHR to BMI was highly significant; ROC area WHR 0.601 > waist 0.571 > BMI 0.559). (Yusuf et al., 2005)
The graded WHR relation and the opposing waist/hip effects
WHR is monotone over the studied range with no threshold located. Each successive quintile carried a significantly higher OR than the last: Q2 1.15 (1.05-1.26); Q3 1.39 (1.28-1.52); Q4 1.90 (1.74-2.07); Q5 2.52 (2.31-2.74) (adjusted age/sex/region/smoking); top-vs-bottom 1.75 (1.57-1.95) after full risk-factor adjustment. «The risk of myocardial infarction rose progressively with increasing values for waist-to-hip ratio, with no evidence of a threshold». (Yusuf et al., 2005) Read as the dose-response prior says: no knee is shown over the studied quintile range — not that none exists -> The U-Shaped Association Artifact (the shape claim is bounded to the studied range, and rests on a quintile display, not a modelled curve).
Waist and hip pull in opposite directions, and both are independent of BMI — this is why the ratio outperforms either alone (top vs bottom quintile, adjusted for BMI): waist 1.77 (1.59-1.97) harmful; hip 0.73 (0.66-0.80) protective. Per 1 SD (adjusted for BMI/height): waist 1.25 (1.21-1.30), hip 0.87 (0.84-0.89), WHR 1.37 (1.34-1.41) — the strongest single marker, BMI 1.10 (1.07-1.13) the weakest. (Yusuf et al., 2005)
Sizing the rock — the population burden BMI hides
Population-attributable risk of MI for the top two quintiles (~40% prevalence): 24.3% (22.5-26.2) for WHR vs only 7.7% (6.0-10.0) for BMI — a ~3-fold larger share of MI attributable to abdominal adiposity than the BMI cut-point captures. (Yusuf et al., 2005) For Layer-1 ranking this matters twice over: abdominal adiposity is a genuine big rock, and measuring it by BMI systematically under-ranks it. (In the parent INTERHEART risk-factor paper the dominant MI levers were smoking and the ApoB/ApoA1 lipid ratio; abdominal obesity sits in the next tier — sized here, not asserted to top the list.) (inferred from Yusuf et al., 2005) -> Layer 1 - Ranking Interventions for a Stratum
Consistency across strata
- Ethnicity — BMI was weakest in all 8 ethnic groups and NON-significant in South Asians (0.99), Arabs (1.00), and mixed-race Africans (1.07); WHR was significant in every group and the strongest marker in 6 of 8 (waist strongest in Chinese and Black Africans). A marker of abdominal obesity beat BMI in every group. (Yusuf et al., 2005)
- BMI fails exactly where risk concentrates — no BMI-MI association in those with raised ApoB/ApoA1 or with hypertension, whereas WHR held there. WHR also held across diabetes, lipids, smoking, and sex, and was steeper in the young (<55 y men / <65 y women 1.46 vs 1.32 older). WHR predicted MI even in the very lean (BMI <20). (Yusuf et al., 2005)
Decision relevance
- Measure waist / waist-to-hip ratio, not BMI alone, when ranking a person’s MI risk from adiposity. A normal-BMI person with a high WHR is not low-risk; a high-BMI person with a low WHR carries less of the abdominal signal. This is the sharpest decision-change: BMI alone under-ranks abdominal-adiposity risk, most starkly in South Asian, Arab, and mixed-race strata where BMI carries no MI signal at all.
- It refines, it does not replace, the BMI mortality curve. BMI still tracks all-cause mortality with a real above-nadir gradient -> BMI and All-Cause Mortality (different outcome, different — cohort — design); INTERHEART says BMI is the wrong instrument for the specific decision of who is at MI risk, cashing that page’s own BMI cannot separate fat distribution limitation.
- The lever question is downstream and only partly answered here. INTERHEART is a marker study; it does not show that lowering WHR lowers MI. It flags a two-pronged target — reduce abdominal fat AND preserve muscle (hip) — because weight loss that also strips skeletal muscle may forfeit some benefit -> Does Weight Loss Reduce Cardiovascular Events. (inferred from Yusuf et al., 2005)
The prospective-cohort contest — ERFC 2011 refutes the WHR-superiority claim type-D
INTERHEART’s reverse-causation exposure (flagged in Limits below) is not hypothetical: a pooled analysis of 58 prospective cohorts with concomitant BMI, WC and WHR in the same people, and serial measurements for regression-dilution correction, directly overturns the marker-superiority claim for prospective CVD. This is the higher-design test INTERHEART itself could not be — and the two are joined: ERFC names INTERHEART as the report it refutes. Filed as BMI vs Abdominal-Adiposity Markers - Which Predicts CVD.
- The measures are similar, not 3x apart. Per-1-SD HRs for CVD (age/sex/smoking-adjusted, BMI
=20) were BMI 1.23 (1.17-1.29), WC 1.27 (1.20-1.33), WHR 1.25 (1.19-1.31) — near-identical, against INTERHEART’s ~3-fold WHR-over-BMI gap. «BMI, waist circumference, and waist-to-hip ratio each have a similar strength of association with cardiovascular disease risk» (Collaboration, 2011). ERFC’s own verdict: «Our findings reliably refute previous recommendations to adopt baseline waist-to-hip ratio instead of BMI as the principal clinical measure of adiposity» (Collaboration, 2011).
- The design-bias diagnosis is explicit. ERFC attributes the discrepancy to «the greater susceptibility of retrospective studies of acute myocardial infarction to some biases (eg, selection biases, reverse causality)» than long-term prospective studies (Collaboration, 2011) — the exact hazard the Limits section anticipated. The INTERHEART BMI-MI odds ratio was 1.12 per 5 kg/m2 vs an ERFC CHD HR of 1.32 per 5 kg/m2, i.e. the case-control under-stated BMI, widening its apparent gap below WHR.
- BMI is measured far more reproducibly than WHR — regression dilution ratio 0.95 (0.93-0.97) for BMI vs 0.63 (0.57-0.70) for WHR (WHR compounds two circumference errors; between-study heterogeneity I2=99%). A noisier marker attenuates more, so a fair comparison should favour WHR — yet it still shows no discrimination edge. This inverts a clinical intuition: the “better” abdominal marker is the harder one to measure well.
- Neither marker adds predictive value over the standard risk factors. Adding BMI / WC / WHR to a model with SBP, diabetes and lipids changed the C-index by -0.0001 / -0.0001 / +0.0008 and NRI by -0.19% / -0.05% / -0.05% — all null. «Simple adiposity measures provide little or no additional information on cardiovascular risk» once conventional factors are known (Collaboration, 2011). This is a different question from INTERHEART’s marker-vs-marker ROC (which had no risk factors in the model) — a distinction, not part of the tension — but it caps the whole debate: for anyone whose BP/lipids/ diabetes are known, the marker choice barely moves the prediction.
- What survives the contest. (i) Abdominal adiposity remains a real, modifiable CVD determinant — ERFC is explicit the null is about prediction on top of intermediates, not aetiology, and the effect runs through BP/lipids/diabetes (the sharp HR attenuation on adjusting for them is that mediation). (ii) The ethnic-heterogeneity signal is NOT refuted: ERFC was 90% European descent and states more data are needed in non-European populations, so INTERHEART’s finding that BMI carries no MI signal in South Asians / Arabs / mixed-race Africans stands untested here — a genuine transportability gap, not a contradiction. (iii) For the specific case-control MI decision in an all-ethnicities sample, INTERHEART’s within-dataset ranking is unchallenged on its own terms.
The net decision-change: for CVD risk-ranking in a developed-country adult whose conventional risk factors are known, BMI is an adequate and more-reproducible clinical adiposity measure; the case for switching to WHR does not survive prospective design. The WHR-over-BMI case is strongest where conventional risk factors are unmeasured and in the non-European strata ERFC could not test. (inferred from Collaboration, 2011)
Limits — read the design before the ORs
- Case-control, concurrent measurement — these are marker-prediction ORs, NOT cohort causal effects. Anthropometry was measured around the acute MI in cases, so the design carries reverse-causation and survival-bias exposure that a prospective cohort would not: fatal MIs are excluded (survivor selection), and acute illness or pre-MI weight/muscle change can distort the measures. The hip-protective finding is the most exposed — acute-illness muscle loss lowers hip circumference and thereby raises WHR in cases, which could inflate both the hip protection and the WHR gradient. The marker-ranking crux (WHR outperforms BMI within the same measured dataset) is more robust to this than any absolute magnitude. (inferred from Yusuf et al., 2005) — the paper itself notes only that a case-control design «cannot elucidate the relation between the different measures of obesity on other outcomes», not the reverse-causation exposure. (Yusuf et al., 2005)
- WHR is a surrogate for a surrogate. «Waist circumference and waist-to-hip ratio are simple and crude surrogate measures for visceral obesity, which is probably the key determinant of metabolic abnormalities» — and visceral fat is itself a marker for intra-organ fat -> Ectopic Fat and Depot-Specific Risk. So WHR steers toward the depot doing damage but is two steps removed from it; the paper says its WHR-MI relation «might be an underestimate of the true contribution of visceral fat». (Yusuf et al., 2005)
- Single source, single outcome. One (landmark) study, MI only; the finding is a marker-prediction claim, and its confidence field is omitted (single-source reference). Corroboration from a prospective cohort or Mendelian-randomization WHR instrument on hard CVD outcomes is the gap.
- Coherence, not validity (R1). Internally sound and source-faithful; no operation here grades the marker against a realized outcome.