Cohort studyindependent cross-checkfull text checked★ key study I-DRY-10 in-depth review
Yogurt linked to lower diabetes risk
Dairy consumption and risk of type 2 diabetes: 3 cohorts of US adults and an updated meta-analysis.
Overview
Chen M et al.. BMC Med 2014 · PMID 25420418 · DOI 10.1186/s12916-014-0215-1
Each daily serving of yogurt: about 18% lower risk of type 2 diabetes in the meta-analysis.
Does the study support the specific claim? No · Methodological quality: high
Plain-language summary
Study explained simply
- What did the researchers want to find out?
Dairy types and diabetes.
- How was the study done?
Three cohorts + meta-analysis.
- How many people, animals or samples were studied?
459,790 people in meta-analysis.
- What was actually found?
Yogurt inversely associated with diabetes; total, low-fat and high-fat dairy neutral; small positive associations for cheese and whole milk disappeared in a sensitivity analysis.
- How large was the observed effect?
RR 0.82 (0.70-0.96).
- What do the results mean in everyday life?
Yogurt is not harmful and may be beneficial.
- What are the study's limitations?
Observational; mostly white US health professionals; self-reported diet; yogurt eaters may live healthier.
- What does it say about Elawady's specific claim?
Contradicts banning yogurt.
- Do the results apply to Germany and Europe?
Moderate-high.
- What can explicitly not be concluded from this study?
Causation.
Scientific in-depth review
1. Full reference
- Authors
- Chen M et al.
- Journal
- BMC Med
- Year
- 2014
- Identifier
- PMID 25420418 DOI 10.1186/s12916-014-0215-1
- Verification
- verified – PubMed efetch 2026-10-25
- Retracted
- no
- Corrections
- none
- Extent checked
- full text
- Conflicts of interest and funding
Funded by NIH grants DK58845, P01 CA87969, P01 CA055075, R01 HL034594, HL60712 (funders had no role); Q. Sun NHLBI career award. COI: D. Mozaffarian member of Unilever North America Scientific Advisory Board; other authors none.
2. Research question
Total and specific dairy and type 2 diabetes.
3. Study design and population
- Study design
- Cohort study – Three prospective cohorts (HPFS, NHS, NHS II) with time-dependent Cox models on cumulative-average FFQ intake, pooled by fixed-effect meta-analysis; plus an updated random-effects meta-analysis of 14 prospective cohorts (11 prior studies for total dairy, 6 for yogurt, plus the 3 cohorts).
- Population
US men (HPFS) and women (NHS, NHS II)
- Country
USA (+ meta-analysis multinational)
- Period
1980-2010
- Sample size
Cohorts: 41,479 HPFS men (abstract states 41,436), 67,138 NHS and 85,884 NHS II women; 15,156 incident T2D cases over 3,984,203 person-years. Meta-analysis: 14 cohorts, 459,790 participants, 35,863 cases.
- Inclusion and exclusion criteria
Excluded at baseline: diabetes (incl. gestational), cardiovascular disease or cancer; >70 of 131 FFQ items blank; implausible energy intake (men <800 or >4,200 kcal/day; women <500 or >3,500 kcal/day); missing baseline dairy or diabetes-date data. Meta-analysis: prospective cohort, case-cohort or nested case-control studies of dairy and T2D (two prior meta-analyses plus MEDLINE/EMBASE search June-October 2013).
4. Methods and endpoints
- Exposure / intervention
Total dairy, low-fat and high-fat dairy, and individual items (skim/low-fat milk, whole milk, cheese, yogurt, ice cream, cream, sherbet) from validated FFQs repeated every 4 years (FFQ vs diet-record correlation 0.57 for hard cheese to 0.97 for yogurt).
- Comparator
Lowest quintile/category of intake; continuous per serving/day increment.
- Endpoints
Incident T2D confirmed by validated supplementary questionnaire (NDDG criteria; ADA criteria after 1998); 97-98% confirmed by medical record in validation samples.
- Statistical methods
Time-dependent Cox regression with 3 nested models (Model 3 adjusted for age, BMI, energy, race, smoking, physical activity, alcohol, menopausal/hormone/OC use, family history, baseline hypertension/hypercholesterolaemia, trans fat, glycaemic load, red/processed meat, nuts, SSB, coffee, and other dairy types); diet updating stopped at CVD/cancer diagnosis; sensitivity analysis stopping at hypertension/hypercholesterolaemia diagnosis; fixed-effect pooling of cohorts; random-effects meta-analysis with I2, Begg's test.
5. Actual results
- Results
Pooled cohorts, fully adjusted, per serving/day: total dairy HR 0.99 (95% CI 0.98-1.01) (age/BMI/energy model 0.96, 0.94-0.98); low-fat dairy 1.00 (0.98-1.02); high-fat dairy 1.01 (0.98-1.03); yogurt 0.83 (0.75-0.92), >=2/week vs <1/month 0.88 (0.83-0.93); ice cream 0.78 (0.71-0.86); skim/low-fat milk 1.02 (0.99-1.04); cheese 1.07 (1.03-1.11); whole milk 1.10 (1.04-1.16). After additionally stopping diet updating at diagnosis of hypertension/hypercholesterolaemia: skim milk 1.01 (0.99-1.03), cheese 1.03 (0.99-1.07), whole milk 1.03 (0.99-1.07), ice cream 0.89 (0.83-0.96), yogurt 0.86 (0.78-0.94). Plain yogurt 0.96 (0.88-1.06) and flavoured yogurt 0.88 (0.77-1.01), both non-significant. Meta-analysis (random effects): total dairy RR 0.98 (0.96-1.01), I2 58.8%; yogurt RR 0.82 (0.70-0.96), I2 63.2%; fixed effects 0.99 (0.98-1.00) and 0.84 (0.78-0.90). No publication bias by Begg's test (P = 0.19 and 0.92).
- Absolute risks
Absolute rates not presented as such; category-specific cases/person-years are tabulated (e.g. pooled yogurt categories in NHS: 3,118/608,342 for <1/month vs 1,031/308,184 for >=2/week). No absolute risk differences reported.
- Authors' conclusion
Higher yogurt intake is associated with a reduced risk of T2D, whereas other dairy foods and total dairy are not appreciably associated with T2D; RCTs are warranted to examine causal effects.
6. Strengths
- Very large cohorts, >90% follow-up, up to 30 years
- Repeated validated FFQs with cumulative averaging; yogurt FFQ validity high (r = 0.97)
- Validated case ascertainment
- Sensitivity analysis explicitly addressing reverse causation from diet change after hypertension/hypercholesterolaemia
- Updated meta-analysis with heterogeneity and publication-bias assessment
7. Weaknesses
- Observational; residual confounding acknowledged, especially for yogurt (marker of healthy lifestyle)
- Health professionals of mainly European ancestry (91-99% white); limited generalisability
- Self-reported FFQ intake; yogurt type/brand not assessed; plain and flavoured yogurt separately non-significant
- Reverse causation demonstrated for skim milk and ice cream (associations changed after stopping diet updating)
- Many dairy subtypes tested without multiplicity adjustment
- Meta-analysis heterogeneity substantial (I2 ~59-63%); meta-analysis search short (adds to prior reviews) and not formally risk-of-bias assessed
- One author with industry advisory role (Unilever)
8. Risk of bias
- Orienting tool
- ROBINS-E (cohort component); meta-analysis component judged descriptively
- Type of assessment
- structured assessment based on the full text by a single reviewer (no formal double review)
- Full formal assessment
- no
- Overall risk of bias
- some concerns
- Rationale
Simplified single-reviewer judgement from the full text; a formal ROBINS-E needs two independent reviewers and a protocol. Meta-analysis part did not assess risk of bias of included cohorts.
| Domain | Judgement | Rationale |
|---|---|---|
| Confounding | some concerns | Extensive adjustment (lifestyle, diet, family history); residual healthy-user confounding for yogurt acknowledged by authors. |
| Exposure measurement | some concerns | Repeated validated FFQs with cumulative averaging; still self-reported, yogurt type not captured. |
| Selection of participants | low | Prevalent diabetes, CVD and cancer excluded at baseline; follow-up >90%. |
| Post-exposure interventions / changes in exposure | some concerns | Diet change after hypertension/hypercholesterolaemia diagnosis shown to distort skim-milk and ice-cream estimates; addressed by sensitivity analysis; yogurt association robust. |
| Missing data | low | High follow-up; exclusions for incomplete FFQs at baseline only. |
| Outcome measurement | low | Supplementary questionnaire validated against medical records (97-98%). |
| Selection of reported result | some concerns | Multiple dairy subtypes and models reported in full, but no prespecified protocol for the cohort analyses and no multiplicity adjustment. |
Full-text review (25 October 2026): Added subtype results from tables (cheese HR 1.07, whole milk 1.10, ice cream 0.78 per serving/day, attenuating to null/near-null in reverse-causation sensitivity analysis; yogurt robust at 0.86), meta-analysis heterogeneity (I2 58.8%/63.2%), inclusion/exclusion, funding and an industry advisory COI (Mozaffarian/Unilever) not in prior entry; HPFS n 41,479 in methods vs 41,436 in abstract noted. Bias overall unchanged (some_concerns).
9. Assessment of the statistical interpretation
Yogurt shows a consistent, precise inverse association (about 17-18% lower hazard per serving/day) robust to the reverse-causation sensitivity analysis; total, low-fat and high-fat dairy are null. Modest positive associations for cheese and whole milk in the primary model attenuated to null when diet updating stopped at hypertension/hypercholesterolaemia diagnosis. Relative measures only; absolute effects not reported.
10. Relevance to the specific claim
Contradicts C-DRY-11 (yogurt forbidden).
11. Independent contrary findings
None.
12. Transferability to Germany and Europe
Moderate-high.
13. Overall methodological judgement
Well-conducted large cohort analysis with updated meta-analysis: yogurt is inversely associated with T2D and no dairy type shows a robust adverse association; causality not established.
| Multidimensional assessment | |
|---|---|
| Authenticity of the source | verified |
| Methodological quality | high |
| Risk of bias | some concerns |
| Precision of results | high |
| Directness | direct |
| Reproducibility | replicated |
| Consistency with the overall evidence | consistent |
| Regional transferability | moderate |
| Adequacy of the interpretation (by the archive) | not applicable |
14. Open questions
RCT confirmation; which yogurt components (probiotics, fermentation) matter; generalisability to non-US, non-white populations and to Egyptian fermented products (laban rayeb, labneh).