If you have seen the claim that polished rice raises the odds of metabolic syndrome nearly eightfold, it traces to one paper, from one city, published in 2009. The figure is real.

An odds ratio of 7.83 from the Chennai Urban Rural Epidemiology Study, with a companion paper from the same cohort putting the odds of newly detected type 2 diabetes at 4.25 for the highest quartile of dietary glycaemic load. Both are cross-sectional. When the same population was followed forward for ten years, the diet signal came back at less than half that size. What follows is what the two papers measured, how much weight each number carries, and what the grain-quality evidence supports.

• CURES-57 found an odds ratio of 7.83 (95% CI 4.72 to 12.99) for metabolic syndrome in the top quartile of refined grain intake, in a cohort where refined grains averaged 46.9% of calories.
• CURES-59 found an odds ratio of 4.25 (95% CI 2.33 to 7.77) for newly detected type 2 diabetes in the top quartile of glycaemic load, from 156 cases.
• Both analyses are cross-sectional, so diet and disease were measured at the same moment and neither can establish which came first.
• Following the same cohort for ten years produced a relative risk of 2.14 (1.26 to 3.63) for an unfavourable diet score. Prospective international estimates for white rice cluster between 1.2 and 1.6.
• Neither abstract reports the quartile cut-points, so the exposure contrast cannot be translated into everyday terms.

CURES is a population-based study of diabetes and metabolic risk in Chennai, recruiting adults aged 20 and over by systematic random sampling across city zones. Two of its analyses generate almost all the risk-and-risk material in circulation, and both used cross-sectional baseline data.

CURES-57

Published: Metabolism, 2009Sample: 2,042 adults aged 20 and overExposure: Refined grain intake by semi-quantitative FFQ, in quartilesOutcome: Metabolic syndrome, modified ATP III with Asian-Indian waist cut-offsHeadline estimate: OR 7.83 (4.72 to 12.99)

CURES-57 also found graded associations across quartiles with the individual components of metabolic syndrome:

• Triglycerides: 36.5% higher in the top quartile than the bottom
• HDL cholesterol: 10.1% lower
• HOMA-IR: 13.6% higher

CURES-59

Published: British Journal of Nutrition, 2009Sample: 1,843 adults aged over 20Exposure: Total carbohydrate, glycaemic load and glycaemic index, in quartilesOutcome: Newly detected type 2 diabetes, 75 g OGTT in all participants, WHO criteriaHeadline estimate: OR 4.25 (2.33 to 7.77) for glycaemic load

CURES-59 tested quantity and quality side by side, and both came back raised. Against the lowest quartile, the highest carried:

• OR 4.98 (2.69 to 9.19) for total carbohydrate
• OR 4.25 (2.33 to 7.77) for glycaemic load
• OR 2.51 (1.42 to 4.43) for glycaemic index
• OR 5.31 (2.98 to 9.45) for refined grains
• OR 0.31 (0.15 to 0.62) for dietary fibre (protective direction)

A corrigendum published in 2010 corrected the reported percentage of newly diagnosed diabetes in the fourth quartile.

An odds ratio approaching 8 for a dietary exposure is extreme by the standards of nutrition epidemiology. The reasons to discount it are structural rather than a matter of the authors having done anything wrong.

• The design cannot order the events: Diet and metabolic measurements were taken at the same visit. Someone already living with dysglycaemia or raised blood pressure may have changed what they eat, or been told to, and a single baseline questionnaire cannot see that.
• The referent is a quartile of the same city, not a healthy comparison group: The bottom quartile in a cohort where refined grains averaged nearly half of calories is still a high-refined-grain diet. A large odds ratio between two points inside a narrow, elevated range describes contrast within that range, not the effect of eating differently.
• Under-reporting works in one direction: Self-reported diet under-captures snacks, sweets and visible fat, more so among people with higher BMI or a known metabolic problem. If non-staple energy goes unreported in the people who also have the outcome, refined grains take up a larger apparent share of their calories.
• Adjustment was good, not complete: CURES-57 adjusted for age, sex, BMI, physical activity, total energy and other dietary factors. CURES-59 adjusted for age, sex, BMI, income, physical activity, family history, smoking, alcohol and fibre. At this effect size, even modest unmeasured confounding moves the estimate a long way.
• The intervals are wide because the case counts are small: CURES-59 rests on 156 cases, and its glycaemic load interval spans 2.33 to 7.77, more than a threefold gap between its bounds.

Both papers are strong signals about heavy reliance on refined grains in a sedentary urban population, and poor estimates of how much risk an individual would shed by changing staples.

The most useful comparison comes from the same cohort. A ten-year follow-up, CURES-142, found the top quartile of an unfavourable diet risk score carried a relative risk of 2.14 (1.26 to 3.63) for incident diabetes. Abdominal obesity and physical inactivity contributed more to population-attributable risk than diet did.

• Pooled analysis, 2012: Across 352,384 participants and 13,284 cases, each additional daily serving of white rice was associated with an 11% higher risk. Comparing the highest intake with the lowest, the relative risk was around 1.27.
• Asian meta-analysis, 2021: Across six cohorts and 12,395 cases, the highest white rice intake compared with the lowest had a relative risk of 1.25 (1.17 to 1.33). The estimate was 1.58 (1.26 to 1.99) in women and 1.30 (0.85 to 1.98) in men.
• PURE, multinational: In South Asia, the highest white rice intake category compared with the lowest had a hazard ratio of 1.61 (1.13 to 2.30). The overall estimate was 1.20 (1.02 to 1.40).
• Pooled dose-response, 21 prospective studies: High versus low glycaemic load or glycaemic index was associated with a relative risk of around 1.3.

South Asians do appear more susceptible than the pooled average, and that is the part of the CURES story which survives replication. The eightfold figure does not. Nothing prospective, in India or elsewhere, has reproduced an effect of that size for a grain exposure.

This is where CURES earns its place, because it tested both in the same models. Glycaemic load and index remained associated with newly detected diabetes after accounting for overall carbohydrate intake, which means how the carbohydrate is packaged carries information the total does not.

Glycaemic index describes how fast a food raises blood glucose against a reference. Glycaemic load combines that rate with the amount of available carbohydrate in what is eaten.

Glycaemic index values for common Indian staples vary widely:

• Surti Kolam white rice: Glycaemic index 77, tested in healthy participants.
• Sona Masuri white rice: GI 72, tested in healthy participants.
• Ponni white rice: GI 70.2, tested in healthy participants.
• Millet-based roti: GI 53, tested in healthy participants.
• Indian branded basmati: GI 54.9, tested in healthy participants.
• Millet-based dosa: GI 37, tested in healthy participants.
• Whole-wheat chapati: GI 44.6, tested in healthy participants.
• Whole-wheat chapati: GI about 84, tested in participants with type 2 diabetes.

The chapati row illustrates the most important caveat on any such table: the value is a property of a food tested in a particular group of people, not a fixed attribute of the food.

Degree of processing matters at least as much as grain identity:

• A 2021 meta-analysis of millet studies found a mean GI of 52.7 (SD 10.3) for millets overall, against 71.7 for milled rice and 74.2 for refined wheat.
• A controlled trial on Indian parboiled variety BPT-5204 compared polishing directly: brown rice at 57.6, under-milled at 73, and white rice at 79.6. The lowest 24-hour glycaemic and insulin responses were on brown rice.
• Minimal polishing was enough to move a medium-glycaemic grain into the high range.

Two Indian trials sit behind the "switch your rice" advice, and they say different things.

The short one is a randomised three-period crossover in overweight Asian Indians comparing white rice, brown rice and brown rice with legumes for five days each:

• Five-day average glycaemic response was 19.8% lower on brown rice than white.
• It was 22.9% lower on brown rice with legumes.
• Fasting insulin fell substantially on both alternatives.

The sample was small, and the published participant count is inconsistent between the summary and full report, so it reads as proof of concept.

The longer one is harder on the story: A three-month crossover trial in 166 overweight adults aged 25 to 65 in urban South India substituted parboiled brown rice for white rice at two meals a day, six days a week. Across the full sample:

• No statistically significant differences in fasting glucose, fasting insulin, HbA1c, HOMA-IR or lipids.
• What appeared was confined to subgroups: among participants with metabolic syndrome, HbA1c fell by 0.18% on brown rice against a rise of 0.05% in those without.

Millet evidence is thinner still:

• A 2021 systematic review found millet-based diets lowered fasting and post-meal glucose against white rice, refined wheat or maize controls, with variable effect sizes across small trials.
• A six-month randomised trial in Delhi-NCR replacing 40% of daily refined cereals with millets and legumes reported a fall in HOMA-IR of 0.40 units, but it ran in adults aged 18 to 25 and is reported in a minor journal.

CURES describes urban Chennai: heavy reliance on polished rice, low intake of whole grains and millets, and a particular stage of economic transition. Rural populations, wheat-based North Indian diets and communities with higher traditional millet intake start from different baselines, and nothing in these papers supports assuming the same associations hold at the same size elsewhere.

Two further limits apply:

• Access to brown rice, good millets and dietitian-guided substitution is uneven, and in the Chennai trials study rice was prepared by trained staff with adherence monitored, which is not how food happens in a household.
• The cohort recruited from age 20 upward, so cross-sectional prevalence reflects cumulative lifetime exposure more than recent eating, and the baseline data predate the current environment of ultra-processed snacks and sugary drinks, which now form part of the same glycaemic load.

Two 2009 papers from one city, with an OGTT on every participant and careful adjustment, are good evidence that a diet built on polished rice and refined wheat travels with metabolic syndrome and undetected diabetes in urban South Indians.

They are poor evidence for how much any individual gains by changing grain. The honest range from prospective work is a reduction in the order of tens of percent rather than multiples.

What holds up is unglamorous: processing, and the company a grain keeps in the same meal, move glycaemic load more reliably than swapping one staple for a fashionable other.

Claims that a single millet substitution transforms metabolic risk run ahead of the trial data.

None of it decides what belongs on your plate, which depends on your own glucose readings, the rest of what you eat, and what a registered medical practitioner or registered dietitian finds when they look at both.

Glucose readings that are not moving despite diet changes? A Voy clinician can read the full metabolic picture. Book an assessment.

This article is for general information and education only and is not medical advice, diagnosis, or treatment. GLP-1 and other medications referenced are prescription-only and are appropriate only for certain people under the supervision of a qualified clinician. Do not start, stop, or change any medication based on this article. Please consult a registered medical practitioner about your individual circumstances. Information reflects what was available at the time of review and may change.